Abstract:As an important strategic material in China, the digital upgrading of cotton industry is of great significance to the realization of agricultural modernization and sustainable development of border economy. The research progress of digital cotton technology system was systematically reviewed, and the conceptual framework of “digital cotton” was proposed, covering the whole industrial chain links such as planting management, yield measurement and harvesting, acquisition and processing, fair inspection, warehousing and logistics, and whole-process quality traceability. Focusing on the analysis of the application status and key technological breakthroughs of digital equipment in precision seeding, intelligent water and fertilizer control, unmanned aerial vehicle plant protection, machine cotton picking monitoring and other links through the integration of Internet of Things, remote sensing, big data and artificial intelligence technology, the problems of insufficient equipment adaptability, data islands and cost-benefit imbalance in the current technology application were revealed. It can provide theoretical support for the quality and efficiency improvement and high-quality development of the cotton industry.

Abstract:In response to the issue of estimating the recovery rate of residual film in cotton fields by current residual film recovery machines, a lightweight residual film recognition method named DCA-YOLO 11 was proposed, which enabled rapid and accurate identification of residual film on cotton field surfaces in natural environments. Taking the residual film on cotton field surfaces after the operation of the 4JMLE-210 residual film recovery machine as the research object, totally 900 images of residual film were collected at different time periods. Through preprocessing steps such as perspective transformation, image cropping, data cleaning, and data augmentation, a dataset of 5215 residual film sample images was constructed, which was divided into training and test sets at a 4∶1 ratio. To enhance the model’s performance, a depthwise convolution (DWConv) module was added to the backbone network of YOLO 11 to replace a standard convolution (Conv) module, thereby reducing computational complexity and the number of parameters. Additionally, a CBAM attention mechanism module was incorporated at the end of the detection output to improve the model’s perception capability and reduce interference from edges and backgrounds. Furthermore, the ADown module was used to replace the Conv module in the backbone network, enabling downsampling between different layers of the residual film feature maps, reducing the spatial dimensions of the feature maps while retaining key information to improve the accuracy of residual film target detection. Experimental results demonstrated that the DCA-YOLO 11 model achieved a precision (P) of 81.9%, a recall (R) of 80.9%, and a mean average precision (mAP) of 86.7% (at an IoU threshold of 0.5) in complex natural environments. The model has about 2.20 million parameters, and an FPS of 80f/s. Comparative experiments with other models showed that DCA-YOLO 11 outperformed YOLO v10, YOLO v9 and YOLO v8 in precision by 2.9 percentage points, 2.3 percentage points, 3.8 percentage points. In terms of recall, it was improved by 2.0 percentage points, 1.0 percentage points, and 1.8 percentage points compared with that of YOLO v10, YOLO v9, and YOLO v8, respectively. While its processing speed was slightly lower than than that of YOLO v10, and it surpassed YOLO v9 and YOLO v8 by 12.7% and 14.2%. DCA-YOLO 11 achieved the smallest model size and the fewest parameters while maintaining high accuracy, demonstrating its lightweight design and superiority. Through generalization test, the model’s detection results on the validation dataset showed an R2 of 0.72, a mean absolute error (MAE) of 4.92 pcs and a root mean square error (RMSE) of 2.72 pcs, indicating good generalization. The research result can provide a theoretical foundation and data support for the precise and efficient collection of residual film by recovery machinery in complex environments, as well as for the visual estimation of the recovery rate of residual film recovery machines.

Abstract:Aiming to address the challenges in accurately assessing residual film coverage due to interference from multiple similar non-target scenarios, complex background textures in target scene images, and the small size, high fragmentation, and irregular contours of residual films during the operational process of residual film recovery machinery, a residual film recognition method was proposed based on vehicle-mounted imaging and deep convolutional neural networks. A multi-feature-enhanced SE-DenseNet-DC classification model was developed by integrating channel attention mechanisms before and after the nonlinear combination functions in each dense block of the DenseNet121 architecture, the model enhanced the weighting of effective feature channels. Additionally, the first-layer convolution of the original model was replaced with multi-scale cascaded dilated convolutions to expand the receptive field while preserving sensitivity to fine details, enabling effective extraction of target scene images. Furthermore, a CDC-TransUnet segmentation model was constructed with enhanced detail information and multi-scale feature fusion. In the encoder of the TransUnet framework, CBAM modules were introduced to capture finer and more precise global features. DAB modules were embedded in the skip connections to fuse multi-scale semantic information and bridge the semantic gap between encoder and decoder features. CCAF modules were then incorporated into the decoder to mitigate detail loss during upsampling, achieving precise segmentation of residual films against complex backgrounds in target scenes. Experimental results demonstrated that the SE-DenseNet-DC classification model achieved classification accuracy, precision, recall, and F1 score of 96.26%, 91.54%, 94.49%, and 92.83%, respectively, for target scene image classification. The CDC-TransUnet segmentation model achieved an average intersection over union (MIOU) of 77.17% for surface residual film segmentation. The coefficient of determination (R2) between the predicted and manually annotated film coverage was 0.92, with root mean square error (RMSE) of 0.23%, and average relative error of 2.95%. The average evaluation time was 0.54 s per image. This method demonstrated high accuracy and rapid processing capabilities for real-time monitoring and evaluation of residual film coverage post-recovery, providing robust technical support for quality assessment in residual film recovery operations.

Abstract:The construction of a standard system is a fundamental project for achieving the standardization of smart farms. In response to the key issues existing in the current construction of smart farms for cotton production, such as fragmented standard systems, low levels of standardization, and incomplete data sharing mechanisms, based on a systematic review of the current status and standardization needs of smart farms for cotton production, the core principles and paths for the construction of the standard system were established. Through integrating modified Checkland methodology and Hall’s three-dimensional structure model, a novel three-dimensional architecture encompassing hierarchical, procedural, and professional dimensions was proposed. The developed framework comprised five standardized clusters: fundamental and general standards, data standards, product standards, methodological standards, and management/service standards, achieving vertical integration across standard levels and horizontal coverage of operational processes. To validate system efficacy, a fuzzy analytic hierarchy process (FAHP) comprehensive evaluation model incorporating four primary and seven secondary indicators was established, with evaluation results demonstrating good applicability (grade Ⅱ) and significant implementation value. This research provided theoretical foundations for standard system development in China’s smart cotton production.

Abstract:Aiming to monitor the process of cotton precision seeding and laying film operation for real-time, and improve the intelligence level of the cotton precision film-laying hole seeder, employing white light source color code sensors and high-definition network cameras as primary monitoring components, a cotton precision film-laying hole seeder operation quality monitoring system was designed based on Vision Assistant. The system consisted of three modules: the seeding monitoring module, film-laying monitoring module and visualization module. Among them, the seeding monitoring module included a seeding status monitoring sensor, a proximity switch sensor, etc. the film-laying monitoring module was mainly composed of a high-definition network camera; the visualization module includes data acquisition module, industrial control computer, etc. Using Labview software graphical programming, equipped with a multi-functional industrial computer, this system achieved real-time monitoring of seeding quality (seeding amount, missed seeding amount), film-laying quality (lighting surface width, lighting surface soil covering width, lighting surface damage area), and operation conditions by running specific function programs through function selection controls. The results of the quality monitoring system bench and field test results showed that the system worked stably and reliably, the monitoring accuracy rate of seeding amount reached over 92%, the monitoring accuracy rate of lighting surface width reached over 94%, the monitoring accuracy rate of lighting surface damage area reached over 81%, and the monitoring accuracy of lighting surface soil covering width reached over 90%. It met the actual requirements of the cotton precision film-laying hole seeder operation quality monitoring system and provided technical support for quality evaluation film-laying sowing operation. It was of great significance to improve the quality and efficiency of cotton cultivation.

Abstract:Aiming at the problems of sticky soil in the cotton area after wheat in the Yangtze River basin, large amount of straw stubble left on the ground surface, the ground wheel driving method is prone to winding slippage and congestion phenomenon, which leads to leakage of sowing, broken strips, and the lack of integrated device for cotton seed guiding and performance detection, thus a radar speed measurement based seeding measurement and control system of cotton precision direct seeding machine was designed. The cloud-end architecture model of 3-channel independent DC motor speed control and 3-channel real-time detection of cotton seed flow was constructed; the seeding measurement and control cloud platform was designed to realize real-time collection, display, storage and multi-device control of the seeding operation parameters; and the fuzzy PID controller was designed to realize real-time regulation and control of the specified grain spacing at the same time with the speed. Matlab was utilized to carry out a simulation comparison test between the fuzzy PID controller and the PID controller. The results showed that the fuzzy PID algorithm model reduced the overshooting amount by 28.95%, the rise time by 28.57%, and the steady state time by 22.22% than the PID algorithm model. The motor control accuracy test was carried out, and the results showed that the maximum speed error between the actual motor speed and the theoretical speed was 1.8%, and the average error was 1.1%. The results of the bench test showed that compared with the JPS-16 computer vision test bench, the difference between the qualified index, omission index and reseeding index detected by the measurement and control system was lower than 1.3, 0.7 and 0.8 percentage points, respectively. The results of the road test showed that the maximum error between the actual and theoretical number of seed rows was 2.4% at the operating speed of 3.6~9.2m/h; the accuracy of single-road detection was not less than 97.44%. The results of field test showed that at the operating speed of 3.6~9.2m/h, the seeding qualification index controlled by this system was higher than 90.81%, the leakage index was lower than 4.64%, and the coefficient of variation of grain spacing was lower than 14.38%. Demonstration application showed that this cotton precision direct seeding machine row seeding measurement and control system can effectively improve the quality of cotton direct seeding machine sowing after wheat in the cotton area of Yangtze River Basin.

Abstract:Aiming at the problem that the signal generated in capacitive cotton seeding monitoring contained noise and thus the seeding information was not easy to be extracted, the CEEMDAN-wavelet threshold joint noise reduction method was proposed. Firstly, according to the detection principle of cotton seedling quality, the noisy simulation signal was constructed, and the empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) and complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) denoising effects of three traditional methods were compared on normal seeding, missed seeding, and repeat seeding simulation signals. Secondly, the wavelet threshold denoising method was integrated into the CEEMDAN denoising method, and the threshold formula of the correlation coefficient was designed to differentiate a large number of intrinsic mode function (IMF) componented with a large number of noisy and IMF components with effective signals, and the noise in the noisy IMF components was removed and more of the shape characteristics of the original signal were retained, and the signal-to-noise ratio(SNR) of the omitted rebroadcasting was increased by 4.9509dB and 6.8493dB, respectively. The similarity of the curve (NCC) was increased by 0.0280 and 0.0549, and smoothness(SR) was decreased by 0.0024 and 0.0045, respectively, which improved the problem of the poor noise reduction effect of the CEEMDAN denoising method alone on the omitted replay signal. Finally, a rowing signal acquisition test platform was built to validate the proposed method, and the results showed that the method had good noise reduction and signal feature reduction capability, and after noise reduction, it implemented a noise reduction effect on the number of distinguished seeds. The results showed that the method had good noise reduction and signal feature reduction ability, and the noise reduction could realize monitoring of number of sown seeds.

Abstract:A process-based cotton growth model could precisely and dynamically simulate the biomass accumulation and yield formation of cotton, so as to provide technical support for smart agricultural decision-making. A dynamic prediction method for cotton growth and yield was developed by integrating meteorological data with the APSIM-Cotton model. Firstly, model parameters were calibrated based on field trial data (2023—2024). Secondly, short-term weather forecasts (ECMWF Open Data) were incorporated for 9d growth simulations. Thirdly, climate analogue years were used to construct seasonal meteorological datasets to enable the dynamic yield prediction throughout the growing season of cotton. The results showed that the APSIM-Cotton model could accurately simulate the phenology dates (NRMSE was 5.18%), biomass (NRMSE was 19.60%), and yields (NRMSE was 6.08%) of cotton under various planting densities (9~27 plants/m2) in Changji, Xinjiang. Short-term biomass predictions achieved the highest accuracy within 1~3d (NRMSE was 1.3%), then the errors were increased to about 3.24% at a 9d forecast. Integrated meteorological data (the dynamic integration of historical meteorological data, short-term weather forecasts, and historical climate analog year data) enabled seasonal yield prediction. Using 18 optimal analogue years minimized prediction errors, stabilizing yield forecast errors below 4%. However, prediction accuracy fluctuated significantly between 90d and 115d after sowing (maximum relative error was 10%), which necessitated cautious application of the prediction results during this period.

Abstract:Accurate and nondestructive detection of cotton aphids is crucial for effective pest control and enhancing cotton yield and quality. Aiming to propose a multi-feature fusion method for cotton aphid damage level (CADL) monitoring, spectral feature wavelengths, vegetation indices, and cotton canopy texture characteristics were integrated to enhance the accuracy of cotton aphid damage level determination. A UAV-mounted hyperspectral imaging system was employed to collect hyperspectral image data of cotton canopy. Pre-processing of the extracted spectral data involved Savitzky-Golay smoothing (SG smoothing) and multiple scattering correction (MSC). Support vector machine (SVM) modeling was applied to the pre-processed spectral data, results revealed that MSC performed better than SG smoothing in pre-processing. Thus the spectral data pre-processed by MSC was used for characteristic wavelengths extraction. Characteristic wavelengths extraction was conducted by using the competitive adaptive reweighting algorithm (CARS) and the shuffled frog leaping algorithm (SFLA), totally 31 and 37 characteristic wavelengths were extracted by CARS and SFLA, respectively. Subsequently, the successive projections algorithm (SPA) was utilized for secondary characteristic wavelengths extraction. Ultimately，six sensitive wavelengths at wavelengths of 650nm, 786nm, 931nm, 938nm, 945nm and 961nm were extracted. Based on six secondarily extracted characteristic wavelengths, nine vegetation indices and eight texture features were calculated, followed by correlation analysis between these vegetation indices/texture features and CADL. Four machine learning models (LightGBM, XGBoost, SVM, RF) were developed to evaluate the classification performance by using characteristic wavelengths alone, vegetation indices alone, texture features alone, combined characteristic wavelengths and vegetation indices, and integrated characteristic wavelengths, vegetation indices, and texture features. Results indicated that vegetation indices (RDVI, SAVI, MSAVI, OSAVI) and texture features (MEA, VAR, DIS, HOM) exhibited strong correlations with CADL. The XGBoost model incorporating the tri-feature combination (characteristic wavelengths, vegetation indices, texture features) achieved optimal CADL classification performance, yielding an overall accuracy (OA) of 86.99% and a Kappa coefficient of 0.8371 on the test set. Compared with models by using characteristic wavelengths alone, vegetation indices alone, texture features alone, or the dual-feature combination (characteristic wavelengths, vegetation indices), this integrated approach improved OA by 4.88, 27.64, 21.95, and 2.44 percentage points, respectively.

Abstract:Accurate above ground biomass (AGB) estimation is a key technology for crop growth monitoring and precision agriculture decision making. Aiming to address the two limitations of traditional unmanned aerial vehicle (UAV) remote sensing methods in cotton AGB estimation—models based on vegetation indices (VIs) were susceptible to the interference of canopy spectral saturation effects, and it was difficult to quantify the spatio-temporal heterogeneity of the dynamics of three-dimensional canopy structure and AGB accumulation—the spatial analysis of three-dimensional UAV point clouds and the temporal characteristics of canopy cover were integrated to construct a multi-dimensional estimation model based on plant height×canopy cover (PH×CC). By designing a comparative experimental framework, the performance differences between the PH×CC model and four types of traditional models were investigated: VIs combined with random forest (RF), gradient boosting (GB), support vector machine (SVM) and backpropagation neural network (BPNN) were systematically evaluated. The results showed that the PH×CC model had significant advantages on the test set. Its coefficient of determination of estimation accuracy (R2) was 0.93, and the root mean square error (RMSE) was 15.30g/m2, which was an improvement of 22.3% compared with that of the optimal traditional model (RF: R2=0.76, RMSE was 23.35g/m2) (P<0.01). The mechanism analysis showed that the PH×CC parameters can analyze 83% of the variation in canopy structure (only 57% for the traditional VIs model) by synergistically representing the dynamic coupling relationship between the vertical expansion of PH and the horizontal expansion of canopy width, significantly improving the model’s ability to explain the interaction mechanism between AGB and structure. The research result can provide a method to overcome the technical bottleneck of “spectral-structural” information fusion in UAV agricultural situation monitoring, and at the same time it can provide a quantifiable modelling tool to analyze the biological mechanism of cotton canopy growth dynamics.

Abstract:Aiming to address the challenges of accurate obstacle detection in complex cotton field environments due to occlusions and the computational limitations of edge devices, a field obstacle detection method based on improved YOLO 11n model was proposed. Firstly, the lightweight StarNet network was adopted as the primary feature extraction network, and the dynamic position bias attention block module (DBA) was introduced to reconstruct convolutional block with parallel spatial attention (C2PSA) to enhance multi-scale feature interaction. Secondly, Kolmogorov-Arnold generalized network convolution (KAGNConv) was used to replace the bottleneck structure in the cross stage partial with kernel size 2 module (C3k2) of the baseline model, enabling fine-grained feature extraction while improving model flexibility and interpretability. Finally, the separated and enhancement attention module (SEAM) was integrated into the detection head to enhance the model’s detection capability in occlusion scenarios. The experimental results showed that, compared with the baseline model, the improved YOLO 11n-SKS achieved increases of 2.3, 2.1, 1.3, and 1.4 percentage points in precision, recall, mAP50, and mAP50-95, reaching 91.7%, 88.3%, 91.9%, and 62.3%, respectively. The model’s floating-point operations were reduced to only 4.4×109 FLOPs, and the number of model parameters was decreased by 17.1%. This study achieved a favorable balance between performance and computational complexity, meeting the real-time detection requirements of cotton harvesting operations while lowering the computational demands for deployment on edge devices, thereby providing technical support for the autonomous and safe operation of cotton pickers.

Abstract:In order to solve the problems of severe shaking and poor stability of the topping device in the process of cotton topping operation, according to the planting mode of machine-picked cotton field in Xinjiang region, the mechanically-picked cotton field was selected as the research object, a crawler walking system of cotton topping machine based on Beidou navigation was designed, and the design structure and parameters of the chassis and key components were determined. STM32F103 was used as the main controller, equipped with a real-time dynamic difference Beidou navigation system to realize automatic navigation control, and the kinematic model of the tracked chassis was established to determine the preview point, and the curvature deviation between the positioning point and the target path was obtained, at the same time, the speed of the chassis geometric center was introduced as the input of the controller, and a self-adjusting double-input fuzzy PID control algorithm was proposed to realize the automatic tracking of the operating path of the chassis. The results showed that the maximum absolute deviation was not more than 39mm, and the standard deviation was not more than 18.5mm, the average absolute deviation was 15.3mm. The average heading angle deflection angle, the average roll angle deflection angle and the average pitch angle of the topping device were 0.38°, 0.33°and 0.26°, respectively. The self-propelled crawler chassis had strong driving stability, accurate alignment and small tracking error, and can be applied to cotton topping walking operation.

Abstract:Cotton boll count is a critical phenotypic trait for estimating cotton yield and plays a vital role in precision agricultural management. However, accurately detecting cotton bolls in densely planted fields remained challenging due to complex backgrounds, occlusion, and varying illumination conditions. High-resolution UAV imagery was employed to capture cotton field scenes in a densely planted area of Xinjiang. A comprehensive dataset was developed through image segmentation and augmentation techniques, ensuring diverse representations of field conditions. To address the trade-off between detection accuracy and computational efficiency, an improved lightweight detection model IML-YOLO was proposed. The model integrated a novel GRGCE module that combined efficient ghost convolution with a RepGhostCSPELAN structure for feature extraction, a CAHSFPN feature fusion mechanism to enhance multi-scale representation, and a Focaler-MPDIoU loss function to refine localization accuracy. Extensive experiments demonstrated that IML-YOLO reduced computational complexity by 32.1%, decreased model size by 47.5%, and lowered parameter count by 50% compared with that of the baseline YOLO v8n, while boosting mean average precision by 10.1 percentage points. Furthermore, when applied to cotton yield prediction, the model achieved an average relative error of only 7.22%. These findings indicated that the proposed IML-YOLO model and yield prediction methodology can offer an effective solution for real-time cotton boll detection and significantly contribute to the advancement of intelligent cotton management.

Abstract:With the widespread adoption of mechanical cotton harvesting in Xinjiang, challenges such as data silos, delayed information sharing, and inadequate whole-process monitoring have hindered precise and transparent quality traceability from seed cotton to lint cotton. To address this, a blockchain-based quality traceability model for machine-harvested cotton was proposed by leveraging the decentralized and tamper-resistant features of blockchain technology and the real-time data acquisition capabilities of IoT devices. A logistic regression-based off-chain/on-chain collaborative data query optimization was introduced to achieve intelligent pre-caching of high-frequency data. Additionally, an access control model integrating reinforcement learning and elliptic curve cryptography was designed to enhance data security and privacy protection. The quality traceability system was developed on the ChainMaker open-source blockchain platform. Performance tests demonstrated that the system reduced query latency from 72.37ms to 60.14ms in regular scenarios and further decreased it to 32.75ms in high-frequency scenarios, with optimization efficiency improving as data volume increased, meeting real-time user query demands. In addition, through plaintext sensitivity and key sensitivity tests, confirming average ciphertext change rates of 87.78% and 82.68%, respectively. These results ensured the privacy and security of data during cross-institutional collaboration. The model established a closed-loop architecture of “IoT data collection-blockchain notarization-smart contract verification-multi-level access control” fulfilling enterprises’ requirements for privacy data permission management and secure sharing while enhancing information retrieval efficiency.

Abstract:The moisture regain rate significantly affects the test results of cotton quality indicators. Accurately measuring the moisture regain rate is of great significance for cotton grading. Aiming at the compensation and correction of the moisture regain rate during the detection of the breaking tenacity of cotton bundle fibers, a method for detecting the moisture regain rate based on the resistance method was proposed. By building a resistance-image synchronous acquisition platform and using image features to represent the fiber thickness, the influence laws of the electrode distance, temperature, and fiber thickness on the resistance measurement during the moisture regain rate measurement were explored, and a multiple prediction model with resistance and temperature as input variables was established. Experiments showed that the gray-scale features of the image were highly correlated with the resistance value and showed a non-linear relationship, and the influence law of the fiber thickness on the resistance measurement was ascertained. The resistance value had a significant positive correlation with the electrode distance within the range of 2~12mm. The intrinsic mechanism of the increase in electrode spacing leading to the expansion of resistance measurement error was explained. Based on this, an electrode distance of 2mm was determined as the optimal detection parameter, and it was verified that there was no significant difference in the resistance of different quality fibers under this parameter (P>0.05). Through experiments on 32 groups of cotton samples with a moisture regain rate of 4.44%~12.2%, the results showed that the random forest (RF) model had the best prediction accuracy, with R2 of 0.99 and RMSE of 0.24%. This study broke through the limitations of traditional moisture regain detection methods for loose cotton fibers and enabled rapid measurement of bundled fibers. It can provide reliable technical support for the precise compensation and correction of physical property indicators such as the breaking tenacity of cotton, and promote the development of intelligent cotton quality detection.

Abstract:Aiming to address the challenges in accurately assessing residual film coverage due to interference from multiple similar non-target scenarios, complex background textures in target scene images, and the small size, high fragmentation, and irregular contours of residual films during the operational process of residual film recovery machinery, a residual film recognition method was proposed based on vehicle-mounted imaging and deep convolutional neural networks. A multi-feature-enhanced SE-DenseNet-DC classification model was developed by integrating channel attention mechanisms before and after the nonlinear combination functions in each dense block of the DenseNet121 architecture, the model enhanced the weighting of effective feature channels. Additionally, the first-layer convolution of the original model was replaced with multi-scale cascaded dilated convolutions to expand the receptive field while preserving sensitivity to fine details, enabling effective extraction of target scene images. Furthermore, a CDC-TransUnet segmentation model was constructed with enhanced detail information and multi-scale feature fusion. In the encoder of the TransUnet framework, CBAM modules were introduced to capture finer and more precise global features. DAB modules were embedded in the skip connections to fuse multi-scale semantic information and bridge the semantic gap between encoder and decoder features. CCAF modules were then incorporated into the decoder to mitigate detail loss during upsampling, achieving precise segmentation of residual films against complex backgrounds in target scenes. Experimental results demonstrated that the SE-DenseNet-DC classification model achieved classification accuracy, precision, recall, and F1 score of 96.26%, 91.54%, 94.49%, and 92.83%, respectively, for target scene image classification. The CDC-TransUnet segmentation model achieved an average intersection over union (MIOU) of 77.17% for surface residual film segmentation. The coefficient of determination (R2) between the predicted and manually annotated film coverage was 0.92, with root mean square error (RMSE) of 0.23%, and average relative error of 2.95%. The average evaluation time was 0.54 s per image. This method demonstrated high accuracy and rapid processing capabilities for real-time monitoring and evaluation of residual film coverage post-recovery, providing robust technical support for quality assessment in residual film recovery operations.

Abstract:Machine-picked cotton has been the main cotton planting mode in China, especially in the main cotton producing areas. Cotton pickers are key equipment for harvesting machine-picked cotton and are also typical representatives of high-end agricultural machinery. Starting from the operation process and principle of cotton pickers, the literature on unginned cotton picking, unginned cotton collection and transportation, unginned cotton compression and packaging, chassis driving and walking, and intelligent control of the whole machine were introduced, and the topics, difficulties, problems, and shortcomings of current research were analyzed. After the development of government guidance and free market competition, the research, development, and manufacturing systems of cotton pickers have been formed in China, based on the introduction, digestion and absorption of foreign advanced technologies. However, there were still some basic scientific problems in cotton pickers, such as the blocking mechanism of the picking head, the dynamic evolution of the pneumatic conveying flow field, and the chassis load spectrum, which have not been solved. The structural design schemes such as the baling mechanism were not innovative and original enough and were still subject to foreign patents. The operation status monitoring and intelligent operation levels were high, but there was a lack of high-precision special sensors such as unginned cotton flow, cotton bale density/humidity. In view of the above problems and shortcomings, the future research direction was analyzed and prospected from the aspects of unginned cotton picking mechanism, high-fidelity simulation analysis of pneumatic conveying, dynamic analysis and optimization of walking chassis, innovative design of baling mechanism, and intelligent control, which can provide reference for the optimization and design, production and manufacturing, and the control of cotton pickers.

Abstract:Tropical fruits are significant export products for many tropical and subtropical countries, playing a crucial role in regional economic development. With the continuous advancement of agricultural intelligent equipment and the integration of emerging technologies such as artificial intelligence, machine learning, and computer vision, automated and intelligent tropical fruit harvesting robots have gradually become a research hotspot in the field of agricultural harvesting. The application and research progress of harvesting robots in the tropical fruit industry was comprehensively reviewed, and the current status of tropical fruit harvesting both domestically and internationally was outlined. It highlighted the growing demand for efficient and sustainable harvesting solutions due to labor shortages and increasing production costs. The technical characteristics of harvesting machinery for different tropical fruits were analyzed, such as bananas, pineapples, and mangoes, based on the complexity of their growing environments and the uniqueness of crop growth patterns. The key technologies of various subsystems in tropical fruit harvesting robots were also discussed, including fruit identification and localization, robotic arm manipulation, and intelligent control systems. The current technical challenges were examined, such as improving recognition accuracy in complex environments and enhancing adaptability to different fruit varieties. Finally, it was looked forward to the opportunities presented by high-tech advancements such as artificial intelligence, big data analytics, and IoT in empowering agricultural equipment. It proposed that future tropical fruit harvesting robots would increasingly move towards intelligent and unmanned harvesting, with potential applications in precision agriculture and smart farming systems.

Abstract:In response to the low permeability of sodic saline soils and the limited adaptability of conventional reclamation equipment, a specialized machine was designed for saline-alkali soil improvement based on deep straw incorporation. Soil column experiments revealed that filling the 20~40cm soil layer with straw increased permeability by 59.52%. A stepped drag-reduction optimization of the trenching device was conducted by using the discrete element method (DEM) and central composite design (CCD). The results showed that at subsoiler shovel working depth of 38.4cm, penetration angle of 24.7°, and blade angle of 60.1°, the trenching shovel resistance reached 9752.5N, while the total system resistance was 12401.9N, representing decreases of 27.7% and 0.4% compared with that of the control, respectively. Field validation of the optimized parameter combination revealed that the total system resistance reached 14500.5N, a 2.6% increase compared with that of the control, whereas the trenching shovel resistance dropped by 16.5% to 11801.8N. The average trenching depth was 39.1cm (coefficient of variation 5.06%), and the average trenching width was 9.4cm (coefficient of variation 5.16%), demonstrating a high degree of operational consistency. The straw burial qualification rate was 80%, indicating that the equipment was capable of effectively accomplishing the deep burial task. Through structural innovation and parameter optimization, the research can effectively address the high-resistance, low-efficiency issues inherent to trenching in sodic saline soils, providing a reliable technical and equipment solution for saline-alkali land improvement.

Abstract:There is a lack of soil loosening components suitable for grassland ploughing. To better break the compacted structure of grassland soil, a special soil loosening component for grassland was designed: a folded-wing subsoiler. Optimization of the combination of structural parameters was carried out by using the shank slip angle, wing width and tip entry angle of a folded-wing subsoiler as test factors, and tillage resistance, furrow contour area and ridge contour area as target parameters. The working performance (ridge contour area, pit contour area, soil disturbance coefficient, soil bulkiness, turning rate, and surface flatness) of the folded-wing subsoiler and three traditional subsoilers (chisel point, diamond point, and arrow point) in grassland soil was compared. The test results showed that the optimal combination of the structural parameters of the folded-wing subsoiler was as follows: the wing width was 40mm, the shank slip angle was 20°, and the tip entry angle was 20°. At this point, the optimal solution for the objective values was 6140N of tillage resistance, 160cm2 of pit contour area, 68cm2 of ridge contour area. Compared with the three traditional plough points, the surface of the grassland after the operation of the folded-wing subsoiler was more flat, and there were fewer turning pieces, and all of them were small pieces of soil; at the same operating speed, the pit contour area and soil disturbance coefficient caused by the folded-wing subsoiler were the largest, while the ridge contour area, soil bulkiness, turning rate and surface flatness were the smallest. Therefore, the grassland loosening performance of the folded-wing subsoiler was better than that of the traditional chisel point, diamond point, and arrow point. The research results can provide technical reference for the creation of key tillage components suitable for mechanized improvement of compacted grassland.

Abstract:In order to solve the problems of low seed feeding accuracy and unstable seed feeding caused by the fast seed conveying speed of the air-fed rice-wheat dual-purpose high-speed seeder, a seed reducer based on the principle of cyclone deceleration was designed. The CFD-DEM coupling simulation method was used to carry out a single factor test to determine the main structural factors and select the appropriate size range. In order to determine the structural parameters of the seed reducer, the Box-Bhnken orthogonal combination simulation test was carried out based on the single factor test results, and the results showed that the optimal structural dimensions were 82.352mm of cylinder diameter D, cylinder length Ht of 101.364mm, exhaust port diameter DP of 25.0002mm, cone length Hz of 67.9025mm, and the indica seed outlet flow velocity V1 and seed vertical velocity V2 were 5.212m/s and 0.462m/s, respectively. The seed flow velocity V1 and seed vertical velocity V2 of japonica rice were 5.339m/s and 0.473m/s, respectively. The flow velocity V1 and seed vertical velocity V2 of wheat seeds were 5.341m/s and 0.408m/s, respectively. The results of bench verification test showed that the vertical velocity of indica rice seeds was 0.411m/s, japonica rice seeds was 0.452m/s, and wheat seeds are 0.457m/s at the seed outlet, which was consistent with the simulation test results. It can be seen from the bench test of strip sowing performance that the effect of strip arrangement with seed reducer was significantly better than that without seed reducer, and the coefficient of variation of displacement uniformity of indica, japonica rice and wheat was decreased from 41.61%, 25% and 37.84% without seed reducer to 9.10%, 8.42% and 8.49%, respectively, which met the performance requirements of seed reducer. The results can provide guidance for improving the seeding performance of air-fed rice-wheat seeder in the future.

Abstract:A stubble cutter was designed with manually variable disc diameter adjustment for the problem that the passive disc stubble cutter on the no-till planter in the no-tillage seeder in the Huang-Huai-Hai wheat and corn area has a low straw cutting rate in the case of large amount of straw cover or insufficient soil strength. The variable-diameter stubble cutting blade was equipped with a disc diameter adjustment mechanism, which can adjust the disc diameter according to the soil characteristics of different fields and the amount of straw stubble cover on the ground surface in order to realize efficient straw cutting. Combined with theoretical analysis, it was determined that the angle of two edges of star teeth of blade was 110°, the minimum radius of variable diameter disc stubble cutter was 230mm, the maximum radius was 280mm, and when the stubble cutter was adjusted to a certain diameter, the variable diameter mechanism had a self-locking function to ensure that the diameter of blade was fixed. The results of soil bin test showed that the diameter of 460mm variable diameter stubble cutting disc, flat disc, notched disc, and corrugated disc, under the same working conditions, the disc into the soil depth was deeper than the shallow cutting performance, into the soil depth of 10cm, variable diameter stubble cutting disc straw cutting performed better than other types of discs, at this time, the straw cutting rate was 75.16%, the traction force was 355.27N. Soil bin performance tests showed that the larger the diameter of the variable diameter stubble cutting disc was, the better the straw cutting performance was. In the disc diameter of 560mm, the straw-cutting rate was 93.25%. Field validation tests showed that the variable-diameter stubble-cutting disc device in the disc diameter of 560mm, the straw cutting rate was 98.33%, which was good to meet the no-tillage sowing operation of the Huang-Huai-Hai regions of wheat agronomic and technical requirements. The research result can provide a reference for the design of the stubble-cutting device for no-tillage implements.

Abstract:The Allium chinense, a perennial plant of the Allium genus in the lily family, is predominantly found in the Yangtze River basin and southern regions of China, with a cultivation area exceeding 6.6×104hm2. It boasts a unique flavor and high medicinal and edible value, being extensively exported to countries such as Japan and South Korea. It has become a characteristic industry and foreign exchange product for implementing the “Rural Revitalization” strategy in some areas, with broad market prospects. Currently, the cultivation of Allium chinense largely relies on manual labor, which is physically demanding and costly, thus hindering the industry’s large-scale development. There is an urgent need to develop and apply mechanized planting equipment for Allium chinense. Aiming to address the agronomic specifications regarding the orientation of bud scales and the engineering requirements for low-position discharging during the mechanized planting of Allium chinense, a flexible belt clamp directional discharging device for Allium chinense, based on the spoon clip type seed metering device was engineered. The mechanism comprised a feed deflector, conveyor systems, electric motors, synchronous pulleys, and flexible belts, among other components. The operational principle of the flexible belt clamp seed discharging device was subjected to theoretical analysis, with a focus on the feeding, seed posture correction, and seed discharging processes. This analysis facilitated the determination of the structural design and parametric specifications for critical components. A coupled simulation model was constructed, integrating multi-flexible body dynamics (MFBD) and the discrete element method (DEM). Using response such as the seed horizontal discharging rate, horizontal seeding rate, and qualified rate of hole distance, a significance screening of the coupled simulation experiments was performed, considering five key factors: the clamp belt angle, theoretical conveying speed, belt speed differential ratio, clamp belt spacing, and feeding radius. Subsequently, a regression orthogonal field test was executed, focusing on the clamp belt angle, theoretical conveying speed, belt speed differential ratio as experimental variables. Employing the Plackett-Burman design and the Box-Behnken central composite design, regression models were formulated for the seed horizontal discharging rate and the qualified rate of hole distance. These models were then utilized for parameter optimization, yielding an optimal parameter set: a clamp belt angle of 65°, a theoretical conveying speed of 0.38m/s, and a belt speed difference ratio of 1.64. Field test were conducted under the optimized parameters, and the findings indicated that at a forward speed of 0.16m/s, the average seed horizontal discharging rate and the average qualified rate of hole distance achieved by the device were 61.11% and 78.89%, respectively. The experimental results demonstrated deficits of 4.17 and 1.15 percentage points relative to the model-predicted optima. The outcomes of this research can offer valuable insights for the development and design of mechanized orientation seed planting equipment for Allium chinense.

Abstract:A pneumatic seeding downforce regulating device was designed to reduce pressure fluctuations caused by surface undulations during mechanized sowing operations, which reduced the stability of seeding depth. The motion process of downforce regulating device was analyzed, and the torsional deformation process of the air spring which was the main working component was clarified. The main structural parameters of the air spring that affected the seeding downforce stability were determined through analyzing the influencing factors of seeding downforce and the deformation process of the air spring, including cord angle, piston radius, and piston angle. In order to determine the optimal parameter combination, a finite element simulation model for gas-solid coupling of air springs was established. Taking improving the downforce stability as the optimization index, a quadratic rotation orthogonal combination simulation experiment was conducted, and a regression model of test indicators and influencing factors was established. Following the principle of reducing the vertical stiffness and ensuring that the vertical output downforce of the air spring met the requirements, the optimal parameter combination for the downforce air spring were determined by simulation experiment: cord angle was 38°, piston radius was 42mm, piston angle was 23°. To verify the effectiveness of theoretical analysis and simulation experiments, the field experiments were conducted on the pneumatic seeding downforce regulating device under the optimal parameter combination, the experimental results showed that the pneumatic seeding downforce regulating device can effectively improve the stability of ditch depth compared with the spiral spring seeding downforce regulating device. When the operating speedwas 4km/h, 8km/h, and 12km/h, the qualified rates of ditch depth were increased by 8, 3 and 11 percentage points respectively, reducing the coefficient variation of ditch depth by an average of 2.58 percentage points, which significantly improved the consistency of seeding depth during mechanized seeding.

Abstract:Aiming to address the issue of unstable seed orientation and inconsistent seed positions within the single-seed chamber of a belt-type seed delivery device during high-speed seeding (13~16km/h), a seed orientation correction element for the belt-type seed delivery device was designed. This element consisted of five parallel transverse ridges, each with a height of 1mm, and was made from nitrile rubber. When soybeans came into contact with the first ridge, they were adjusted so that their long axis was oriented perpendicularly to the direction of the seed belt movement, thereby ensuring stable seed delivery. By analyzing the state of the seeds during the collision deformation phase and the collision recovery phase, the theoretical two-dimensional seed delivery position within the correction zone was clearly defined. Using the EDEM discrete element simulation software, a simulation experiment was conducted to determine the optimal height of the correction ridges. The qualification rate of seed inclination and the qualification rate of seed displacement were used as evaluation indicators. Through single-factor experiments, the displacement trajectory of the seeds from the seed-limiting to the correction phase was analyzed, clarifying the effect of the correction ridge height on the lateral movement of the seeds. The results showed that with a ridge height of 1.00mm, the average qualification rate of seed inclination was 95.7%, and the average qualification rate of seed displacement was 98.2%. High-speed camera technology was used to conduct single-factor comparative experiments, with the orientation variability coefficient and plant spacing variation coefficient used as indicators to compare the correction effects. The comparative experiments demonstrated that the belt-type seed delivery device equipped with the correction element had lower orientation variability coefficients and plant spacing variation coefficients than the device without the correction element. With the correction element featuring a ridge height of 1mm, the average orientation variability coefficient was 16.45%, and the average plant spacing variation coefficient was 12.78%, meeting the requirements for high-speed precision seeding operations.

Abstract:Aiming at the problems of insufficient seed filling and difficult seed cleaning caused by irregular shape and uneven size of Pinellia ternata seeds, based on the characteristics of chain flipping motion, a method of using rotational inertia force for seed cleaning was proposed. A chain spoon upturned tail self-cleaning Pinellia ternata precision seeder was designed, which improved the seed filling rate and achieved rapid seed cleaning through the spoon upturned tail structure and water droplet shaped holes. By analyzing the force and motion state of seeds during the working process of the seeder, the working principle of the chain spoon self cleaning seeder with upturned tail was explained. Through theoretical calculations and kinematic analysis, simulation experiments were conducted based on DEM-MBD coupling to analyze the effects of different spoon filling angles, chain tension forces, and spoon shaped hole structural parameters on the performance of the seeder. The structural parameters of the seeder were determined. A quadratic regression orthogonal rotation combination simulation experiment was designed to determine the optimal structural parameter combination for spoon shaped holes: spoon shaped hole length was 18.6mm, spoon shaped hole width was 14.1mm, and spoon shaped hole retention depth was 8.6mm. To determine the optimal operating parameters of the seeder, a quadratic regression orthogonal rotation combination bench test was conducted with the driving sprocket speed and seed layer height as experimental factors. The experimental results showed that when the driving sprocket speed of the seeder was 39.2r/min and the seed layer height was 206mm, the operational performance of the seeder was optimal, with a qualification index of 93.37%, a replay index of 2.17%, and a leakage index of 4.46%, the research results can provide a reference for the design of seeders for bulbous Chinese medicinal materials.

Abstract:Cistanche deserticola is a herbaceous plant that parasitizes deep in the roots of Haloxylon ammodendron and Tamarix ramosissima. Due to its small seed volume, high price, and the requirement that the seeds need to be attached near the roots of host plants during sowing, the mechanized sowing of Cistanche deserticola demands precise seed metering and deep sowing. According to the agronomic planting requirements of Cistanche deserticola and combined with manual sowing methods, a Cistanche deserticola seeder capable of ditching, sowing, and covering soil operations was designed and manufactured. By analyzing the stress state of the ditching cutterhead, the main structural parameters and the blade layout pattern of the cutterhead were determined. Based on the principle of constant volume of soil throwing and covering, the structural parameters of the soil guide cover were designed. Combined with finite element analysis software, modal analysis was mainly carried out on the ditching cutterhead and the main drive shaft, and the working parameters of the machine were determined. The sowing flow rate of Cistanche deserticola was matched according to the liquid spraying volume and the working speed, and the form and size range of the seed outlet were determined. Taking the blowing air pressure at the nozzle and the opening degree of the seed spraying port as experimental factors, and the coefficient of variation of sowing uniformity as the experimental index, field experiments were carried out. The results showed that when the blowing air pressure at the nozzle was 0.2MPa and the opening degree of the seed outlet was 1.5mm, the coefficient of variation of sowing uniformity was 13.50%, and the sowing depth stability coefficient was not less than 91.17%, which met the current agronomic planting requirements of Cistanche deserticola.

Abstract:According to agronomic requirements, most leafy vegetables adopt a dense transplanting mode with a plant spacing and row spacing of less than 200mm. However, the automatic seedling picking and throwing devices developed domestically are mostly suitable for transplanting with large plant spacing and row spacing. The gantry arm type seedling picking and throwing device can replace manual operations, but due to the long movement path of the seedling picking claws, the speed of seedling picking and throwing is slow. Therefore, this structure cannot meet the requirements of dense transplanting of vegetables. For the A5-1200 semi-automatic high-density transplanter, an oblique automatic seedling picking and throwing device was designed to address the issue of low transplanting efficiency. The device consisted of a seedling tray displacing mechanism, a picking and throwing arm, and eight claws. The seedling tray and the arm were both arranged at a 45° angle to the horizontal direction, and the claws moved back and forth in a straight line between the seedling picking position and the feeding position, shortening the seedling picking stroke. The installation positions of the claws on the picking and throwing arm were fixed, corresponding to the distance between the seedlings and the distance between the seedling feeding cup, omitting the seedling separation step. The orderly and smoothly operation process was ensured by the PLC control system. The ball screw module was driven by the closed-loop stepper motor in the transmission parts for both the picking and throwing arm and the seedling tray displacing mechanism. To realize the whole row interval seedling picking, the tray displacing mechanism completed a single action within 1s, moved the seedling tray in the horizontal and longitudinal directions, and located the position of the seedlings precisely. The influences of the return speed of picking and throwing arm, the insertion depth and the penetration angle of claw on the picking and throwing effect were analyzed. The Box-Behnken response surface experiments were designed to optimize the working parameters. The results showed that when the return speed of the picking and throwing arm was 300mm/s, the insertion depth of claw was 31mm, and the penetration angle of claw was 10°, the actual success rate of picking and throwing was 97.0%. Equipped with the oblique automatic taking and throwing device, the A5-1200 transplanter had a theoretical transplanting capacity of 7200 plants/h, which met the technical requirements of dense transplanting.

Abstract:Aiming to address the challenges of poor soil clearance, high operational resistance, and elevated damage rates in ginger harvesting machines, the design and development of a shovel-screen combined ginger excavation and soil clearance device was introduced. The device primarily consisted of a digging shovel and a soil-shaking screen. Theoretical analysis and simulated experiments were conducted on the ginger excavation and soil clearance device, leading to the preliminary determination of a shovel face angle of 18°, a shovel face length of 160mm, and a swinging frequency of 4Hz. Field experiments on the ginger harvester were carried out based on the Box-Behnken experimental design principle, with forward speed, shaking screen swing amplitude, and shaking screen swing frequency as experimental factors, and ginger soil content and damage rate as experimental indicators. Variance analysis was performed on the experimental results, and regression models between ginger soil content, damage rate, and significant factors were established. The optimization of the regression model’s objective function yielded the optimal parameter combination as follows: forward speed of 0.39m/s, swing amplitude of 30°, and swing frequency of 3.901Hz, resulting in a predicted ginger soil content of 9.85% and a damage rate of 1.79%. Field validation experiments demonstrated that the average ginger soil content harvested by the ginger excavation and soil clearance device was 10.31%, with an average damage rate of 1.86%, both showing relative errors of less than 5% compared with the model’s predictions. Compared with the original ginger excavation and soil clearance device, the average ginger soil content harvested by the device was decreased by 2.39 percentage points, and the average damage rate was decreased by 1.38 percentage points. The operational resistance of the ginger excavation and soil clearance device was approximately 1240N, representing a reduction of about 11.43% compared with that of the original device.

Abstract:Aiming to address lack of accurate modeling in discrete element simulation analysis for cleaning devices in combined oil sunflower harvest, the combined-harvested oil sunflower extracts were taken as object. A discrete element method was used to categorize and calibrate contact parameters for various oil sunflower extract models. The randomly selected oil sunflower extracts were classified, its main components were identified, and the corresponding mass fractions were determined using digital calipers, a universal testing machine, and a custom test platform measure intrinsic and contact parameters of each oil sunflower extract. Plackett-Burman, the steepest ascent, and Box-Behnken tests were proceeded based on each extract’s physical repose angle. Parameters with significant effects on extract repose angle were identified and their valid ranges were defined. An optimization module in Design-Expert software was employed and physical repose angle of each extract was treated as the objective value. The optimal parameter sets were determined as follows: oil sunflower seed shear modulus was 7.35×107Pa, oil sunflower seed-steel restitution coefficient was 0.295, oil sunflower seed-oil sunflower seed static friction coefficient was 0.669, crushed oil sunflower head shear modulus was 1.94×107Pa, crushed oil sunflower head-steel restitution coefficient was 0.467, crushed oil sunflower head-steel static friction coefficient was 0.436, oil sunflower stalk shear modulus was 7.39×107Pa, oil sunflower stalk-steel static friction coefficient was 0.553, and oil sunflower stalk-oil sunflower stalk static friction coefficient was 0.775. Simulation stacking tests were conducted on oil sunflower seeds, crushed oil sunflower heads, stalks, and mixed extracts based on each optimal parameter set. Results showed that errors between simulated and physical repose angles were 0.66%, 0.96%, 0.64% and 1.15%, respectively. These findings can serve as a reference for discrete element simulation research of combined oil sunflower harvest.

Abstract:Aiming at the current domestic mechanized peeling of seed corn using ordinary corn combine harvester operation with large kernel loss, high cost of peeling production line construction, shortage of drying field peeling equipment and other problems, a mobile seed corn husker was designed, using flexible conveyor belt feeding device, full rubber segmented combination peeling rollers, gas-solid coupling type rapid screening device and separable mobile frame design, in order to improve the efficiency of peeling, reduce kernel damage, and facilitate transport and storage. Through theoretical analyses and calculations on the peeling process of seed corn, the main factors affecting the peeling effect and the structural parameters of key components were determined. The Box-Behnken experimental design principle was adopted to conduct a three-factor, three-level experiment by using peeling roller rotational speed, peeling roller inclination angle and peeling roller bias angle as experimental factors, and bract stripping rate, kernel shedding rate and kernel breakage rate as performance indexes, and the verification test was carried out according to the actual working conditions at the end. The results of multi-objective optimization showed that the optimal combination of working parameters was the peeling roller rotational speed of 265.57r/min, the peeling roller inclination angle of 10.06°, and the peeling roller offset angle of 19.76°，at which time the bract stripping rate was 95.33%, kernel shedding rate was 1.471%, and the kernel breakage rate was 0.661%. The results of the validation test showed that at the optimal parameter combination of the bract stripping rate was 94.22%, the kernel shedding rate was 1.511%, and the kernel breakage rate was 0.675%, which was basically consistent with the results of the parameter optimization, meeting the operational requirements for mechanized peeling of seed corn. The research result can be used for the design and improvement of seed corn husker to provide a reference.

Abstract:A drum type fallen peanut picking up machine was designed to address the issues of excessive fallen peanut and difficulty in recycling after peanut harvest. The excavation and picking device, upward conveying device, screening and collecting device were designed and related parameters were calculated. The current research status and existing problems of fallen peanut picking up machine were introduced, providing ideas and key solutions for future design. Firstly, the entire machine was constructed to determine the main technical parameters of drum type fallen peanut picking up machine and introduce its working principle, then the key devices and parameters were designed and optimized. In the excavation and picking device, the excavation angle and shovel surface width of the excavation shovel were determined, and the structure and material design of the feeding roller and guide grid were carried out.The force analysis and kinematic analysis were conducted on the fallen peanut-soil mixture on the upward conveying device, and the theoretical value range of the angle between the upward conveying monomer and the upward conveying chain, as well as the operating speed of the upward conveying chain, were obtained. Through kinematic analysis of the fallen peanut-soil mixture in the screening and collecting device, the diameter D of the roller screen, the pore width K of the screen mesh, the rotational speed n of the roller screen, and the installation angle of the roller screen were obtained. Through EDEM simulation analysis, the angle between the upward conveying monomer and the upward conveying chain, speed of the upward conveying chain, the leakage and lifting situation of two factors were observed. Using pick up rate and trash content as evaluation indicators, through single factor and double factor simulation analysis of upward conveying device operation parameters, the angle between the upward conveying monomer and the upward conveying chain, the running speed of upward conveying chain, and the impact of their interaction on peanut lifting efficiency and trash content were clarified. The optimal parameter combination was obtained as follows: the angle between the upward conveying monomer was 45°, upward conveying chain running speed was 1.5m/s. According to the preliminary design, peanut varieties were pre-picked, and field tests were conducted after measuring the soil moisture content and the distribution of fallen peanut. The field test results of the prototype showed that the pick up rate of the whole machine was 95%, and the trash content was 8.5%, which met the design requirements of the whole machine.

Abstract:Aiming to address the issue of low hydraulic performance in low-specific-speed stamping centrifugal pumps, focusing on the CDL1 multi-stage stamping centrifugal pump, using its first-stage impeller as the research object, by combining numerical simulation and experimental testing methods, a comprehensive analysis of the hydraulic performance of the first-stage impeller was conducted. Given that the hydraulic performance of low-specific-speed impellers was influenced by multiple factors, Latin hypercube sampling (LHS) was employed to sample various design variables of the first-stage impeller, forming a sample space and obtaining the corresponding performance parameters. A Kriging surrogate model was then established to analyze the sensitivity of each parameter to the hydraulic performance of the impeller. The critical influence parameters of the impeller were selected as the input for the particle swarm optimization algorithm (PSO), and multi-parameter optimization design was carried out. On this basis, the hydraulic performance and internal flow mechanism of the impeller were investigated in depth. The results showed that the hydraulic performance of the optimized impeller was significantly improved compared with the original design, with the efficiency at the rated point increased by 2.8 percentage points and the single-stage head increased by 0.4m. Additionally, the optimization process revealed that the impeller’s blade angle, inlet and outlet diameters, and blade thickness were the most sensitive parameters affecting hydraulic performance. The improved design not only significantly enhanced the overall efficiency and head but also optimized the flow distribution, reduced turbulence, minimized energy losses, improved fluid dynamics, and increased operational stability, leading to better performance, reliability, and long-term durability in practical applications.

Abstract:Accurately predicting the erosion regions is always a challenging point of numerical cavitation erosion simulation, which is beneficial for designing and extending the lifespan of the hydraulic machinery. The density-corrected SST k-ω turbulence model and the Sauer cavitation model were used to simulate the unsteady cavitation around NACA0009 3D twisted hydrofoil. The accuracy of the current numerical method was verified by comparing the cavitation shedding frequency and transient cavity behaviors in the experiment. Considering the energy transfer efficiency, a propagation relationship of cavitation energy from the flow field space radiation to the wall was constructed, thereby predicting the wall erosion load. By comparing the erosion energy and erosion load on the hydrofoil surface at different instants, it was found that compared with cavitation energy, the erosion load, which comprehensively considered the influence of erosion energy from the whole flow field on hydrofoil surface, predicted a wider coverage erosion area. Moreover, the average wall surface erosion intensity distribution was obtained. By time-averaging the wall surface erosion intensity solved from each instantaneous time step within 12 cycles, and by comparing the average wall surface erosion intensity obtained from the erosion energy and erosion load with the experimental erosion results, it was demonstrated that the erosion area predicted by time-averaged erosion load was more agreeable to the experiment result, indicating that it was necessary to consider the energy transfer efficiency when predicting cavitation erosion.

Abstract:The improvement of crop remote sensing identification accuracy is a key driving force for the leapfrog development of precision agriculture and smart agriculture. The accuracy of crop remote sensing identification depends on three elements: samples, image features and classification methods. Aiming to reduce the classification error caused by the bottleneck of sample data, the accuracy of crop remote sensing identification by jointly enhancing the sample quantity and quality control was improved. Taking the Wulanbuhe Irrigation District in the Hetao Irrigation Area as the study area, the time-series image of NDVI during the crop growth period in 2023 was constructed. Combined with the NDVI time-series characteristics of the crops, sampling was conducted on the image to expand the number of crop samples, and then the unqualified samples were screened and removed to achieve sample quality control. A total of 801 pixels of field samples (pre-expansion samples), 17917 pixels of image samples (expanded samples), and 18718 pixels of total samples (post-expansion samples) were selected. Four machine learning classifiers were used to compare the crop classification effects before and after sample expansion. The results showed that the classification accuracy of crops was significantly improved after sample expansion, with the overall classification accuracy increased by approximately 5 percentage points and the Kappa coefficient rose by about 0.05. Among them, the classification accuracy of RF and NNC was relatively high, while that of CART and SVM was slightly lower. The crop remote sensing recognition was carried out after sample expansion by using CNN and LSTM deep learning models. The results showed that the classification accuracy of CNN and LSTM was higher than that of RF and NNC, which had relatively high classification accuracy.

Abstract:Soil background has an impact on the accurate estimation of maize leaf area index(LAI), and the traditional soil background removal method eliminates the area information of soil pixels thus resulting in a lower correlation between the target area spectrum and maize LAI. Therefore, a soil background removal method was proposed, which removed the spectral reflectance of soil pixels while retaining the area information of soil pixels. Based on this method, the multispectral image was preprocessed and 26 vegetation indices such as normalized difference vegetation index (NDVI) were extracted along with eight texture features such as Mean. Combined with crop growth covariates such as plant height/chlorophyll content, the above three different types of features were arranged and combined to form multiple input feature sets, and eight modeling algorithms were used to build multiple LAI estimation models, which were compared with those based on traditional soil background removal methods. The results showed that the soil background removal method proposed effectively eliminated the effect of soil spectral reflectance on vegetation spectral reflectance under the premise of retaining the area information of soil pixels and vegetation pixels, and the LAI estimation models based on this method were better than the traditional methods; the fusion of multiple types of features can improve the model estimation accuracy of LAI from multispectral images, and the estimation effect of texture features on LAI was better than that of the vegetation index; the machine learning model was better than the traditional statistical regression algorithm for LAI estimation, and the optimal model was the one-dimensional convolutional neural network (1D-CNN) model with vegetation index + texture features + plant height/chlorophyll content as inputs, which was pre-processed with the soil background processing method proposed. 1D-CNN model with testing set adjust coefficient of determination R2Adj, root mean square error （RMSE）, and mean absolute error (MAE） of 0.9515, 0.2421, and 0.1795, respectively. The research result may provide a method for rapid and accurate estimation of maize LAI.

Abstract:A deep learning model Font_MBAFF was proposed for the task of text similarity calculation, which was mainly applied to the matching of question pairs in Chinese agricultural short texts. In order to solve the problems of sparse semantic features and inadequate understanding of specialized vocabulary in agricultural short texts, it was firstly optimized in the feature representation stage. By introducing the unique font features of Chinese characters to expand the features, including side radicals and four corner numbers, thus enriching the semantic representation of features. In the feature extraction layer, the multi-scale convolution attention channel weighted network MSCN and the bidirectional long short-term memory network Multi_SAB based on multi-head self-attention mechanism were combined respectively, so that the model can further optimize the feature extraction from the spatial and temporal relationship sequences of semantic features. Finally, TEXTAFF, an improved attention fusion mechanism for text, was used in the intelligent fusion stage of features. The experimental results indicated that the Font_MBAFF model can effectively compensate for the lack of feature words in short texts, optimizing text feature extraction and feature fusion. The accuracy of semantic matching reached 96.42%. Compared with five other semantic matching models, including MaLSTM, BiLSTM, BiLSTM_Self-attention, TEXTCNN_Attention, and Sentence-BERT, the Font_MBAFF model demonstrated significant advantages, achieving a correctness rate that was at least 2.07 percentage points higher. Furthermore, the model proved resilient in experiments with datasets of different sizes, showing rapid response times during testing. Font_MBAFF deep learning model exceled at determining the similarity of Chinese agricultural short texts.

Abstract:In unstructured environments, the cluster growth characteristics of Hangzhou white chrysanthemum lead to severe mutual occlusion, reducing detection accuracy for chrysanthemum detection algorithms. To address this issue, an improved YOLO v8n detection model for Hangzhou white chrysanthemum, called Hangzhou white chrysanthemum-YOLO v8n (Hwc-YOLO v8n), was proposed. Firstly, the model’s ability was enhanced to finely detect critical, similar features of the chrysanthemum by increasing the label categories from two to three. Secondly, a dynamic feature extraction module (C2f-Dynamic) was designed in the backbone network to strengthen the model’s adaptive response to missing features in occluded targets. Additionally, a 160 pixel×160 pixel detection head was added to the detection head section, allowing the model to detect small targets more effectively. Finally, the angle penalty metric loss (SIoU) was adopted to optimize the bounding box loss function, improving both detection accuracy and generalization capability. Experimental results from module placement and heatmap analysis demonstrated that the C2f-Dynamic module can dynamically adapt to feature changes in occluded targets. The improved Hwc-YOLO v8n model achieved a 1.7 percentage points increase in mean average precision and a 0.88 percentage points increase in mean recall rate for the occluded Hangzhou white chrysanthemum. Ablation and comparison experiments showed that the improved Hwc-YOLO v8n outperformed DETR, SSD, YOLO v5, YOLO v6, and YOLO v7 in detection of the chrysanthemum. Specifically, compared with DETR, SSD, YOLO v5, YOLO v6, and YOLO v7, the mAP was improved by 5.7, 12.6, 0.7, 0.75, and 11.25 percentage points, respectively. The mR was increased by 2.15 percentage points and 1.4 percentage points compared with that of YOLO v5 and YOLO v7, respectively. The research result can provide a technical foundation for future intelligent harvesting of Hangzhou white chrysanthemum.

Abstract:The binocular camera-based target localization system has the advantages of short starting distance, high accuracy, and low cost, which is suitable for strawberry target identification and localization applications in space-constrained greenhouse strawberry production environments. Accurate target matching is the guarantee of the effectiveness of binocular camera measurements, but the surface brightness and shadow areas of strawberries vary greatly in natural environments, and it is difficult to obtain stable and accurate matching results with the binocular matching method based on local features. A binocular strawberry target matching method was investigated based on image semantic features, which can maintain the stability of the target description under the conditions of large illumination changes, rich image texture, fruit occlusion, image blurring, etc., and therefore can improve the accuracy of binocular strawberry target matching. The semantic feature extraction method of the strawberry target region in the image was firstly designed, and secondly the strawberry target similarity calculation method was designed based on the semantic features and the geometric constraints of the binocular structure, and finally the binocular strawberry target matching in the greenhouse environment was realized. The experimental results showed that the correct rate of strawberry target matching applied to the greenhouse environment by the method was 96.3%, which can provide good target matching results for the strawberry target binocular localization system under the actual picking environment.

Abstract:Aiming to address the issue of overlapping occlusion caused by the varying sizes and large quantities of dragon fruit, a multi-scale weighted feature fusion network (MBS-YOLO v8) was proposed based on the YOLO v8 model. Firstly, the squeeze-and-excitation attention (SEAttention) mechanism was incorporated into the feature extraction module to enhance the network’s ability to capture critical details, thereby addressing the challenge of small object detection. Secondly, a multi-scale weighted fusion network (MWConv) was introduced to generate feature maps with varying receptive fields, improving the capture of global features within images. Finally, experimental results demonstrated that MBS-YOLO v8 achieved an accuracy of 92.5%, a recall rate of 90.1%, and a mean average precision (mAP50) of 94.7%. Compared with the YOLO v8n algorithm, MBS-YOLO v8 showed improvements of 2.1 percentage points, 5.9 percentage points, and 2 percentage points in accuracy, recall, and mAP50, respectively. The proposed MBS-YOLO v8 model exhibited high robustness, effectively integrating global feature information with low-dimensional local features to enhance the model’s understanding of image content. This approach effectively addressed challenges related to overlapping occlusion and small object detection, providing an improved methodology for detecting dragon fruit and other similar targets.

Abstract:The detection and segmentation of cucumber fruits are crucial for phenotypic analysis and the management of cucumber growth. However, in complex greenhouse environments, fruits are often occluded by stems and leaves, and their color may be similar to the background, making it difficult for traditional methods to accurately identify fruit boundaries and achieve efficient segmentation. To address this issue, an improved YOLO v8-based method for cucumber fruit segmentation was proposed. This method incorporated deformable convolution network v4 (DCNv4) to enhance the model’s spatial adaptability and utilized the RepNCSPELAN4 module in combination with an additional C2F module to refine feature extraction and fusion, thereby improving the model’s segmentation performance for cucumber fruit images in complex greenhouse environments. Experimental results showed outstanding performance across multiple categories in two experimental settings: a glass greenhouse and a plastic greenhouse. Specifically, in the glass greenhouse scenario, the model achieved a precision of 96.3%, recall of 93.1%, mean average precision (mAP50) of 96.2%, and mAP50-95 of 85.3%. In the plastic greenhouse scenario, the precision was 86.8%, recall was 81.9%, mAP50 was 90.0%, and mAP50-95 was 77.0%. The proposed method demonstrated stronger robustness and generalization in handling boundary issues, multiple occlusions, and multi-scale segmentation, enabling the model to adapt to diverse and complex cultivation environments and accurately segment cucumber fruits. Accurate fruit image segmentation facilitated the acquisition of phenotypic parameters and provides reliable technical support for further phenotypic analysis of cucumber fruits, thereby promoting the application of agricultural phenotyping robots and the intelligent development of agricultural production.

Abstract:Aiming to address the current limitations in detecting chestnut of varying scales under natural conditions, an innovative multi-scale chestnut detection method was introduced, YOLO 11-MCS, based on an improved YOLO 11 model. Firstly, a novel multi-scale key feature aggregation (MKFA) module was proposed, which was integrated into the C3k2 module to form the C3k2-MKFA feature extraction module, effectively capturing features at different scales, enhancing multi-scale feature extraction capabilities. Subsequently, the CGAFPN network was introduced, which incorporated a small object detection layer through a content-guided attention module and increased the contribution proportion of chestnut small object to multi-scale object, overcoming the deficiencies of the original algorithm in multi-scale and small object detection. Finally, a shared convolution separated batch normalization detection head (SCSB) was presented, utilizing shared convolution and separated batch normalization structures to efficiently extract cross-scale features and enhance feature consistency across different scales, effectively improved the performance of multi-scale object detection. Experimental results demonstrated that the improved model achieved a chestnut detection precision of 88.2%, a recall rate of 79.2%, and an average precision of 87.2%, which had improvements of 0.8, 5.9, and 5.5 percentage points, respectively, compared with the original YOLO 11 network. The model with channel-wise feature distillation achieved an average precision of 84.7%, with a model size of 6.0MB. When deployed on the Jetson Nano using the Infer inference library, the detection speed was 23ms per image, meeting the requirements for chestnut detection.

Abstract:Aiming at the problem that it is difficult to extract body measurement points efficiently and accurately in the automatic measurement of body size of group-raised pigs, an automatic measurement method of body size of group-raised pigs based on improved YOLO v5-pose was proposed. Firstly, the convolutional block attention module (CBAM) was integrated into the YOLO v5-pose backbone network to better capture the relevant features of the measurement points. Then the C3 traditional module of the Neck layer was replaced with the C3Ghost lightweight module to reduce the number of model parameters and memory usage. Finally, the dynamic head (DyHead) target detection head was introduced in the Head layer to enhance the model’s ability to represent the position of the measurement points. The results showed that the average accuracy of the improved model was 92.6%, the number of parameters was 6.890×106, and the memory usage was 14.1MB. Compared with the original YOLO v5-pose model, the average accuracy was increased by 2.1 percentage points, and the number of parameters and memory usage were decreased by 2.380×105 and 0.4MB, respectively. Compared with the current classic models YOLO v7-pose, YOLO v8-pose, real-time multi-person pose estimation based on mmpose (RTMPose) and CenterNet, this model had better recall rate and average precision and was more lightweight. Experiments were conducted on a dataset of 2400 group-raised pigs images. The results showed that the average absolute errors of the body length, body width, hip width, body height and hip height measured by this method were 4.61cm, 5.87cm, 6.03cm, 0.49cm and 0.46cm, respectively, and the average relative errors were 2.69%, 11.53%, 12.29%, 0.90% and 0.76%, respectively. In summary, the method improved the detection accuracy of body size measurement points, reduced the complexity of the model, and achieved more accurate body size measurement results, providing an effective technical means for the automatic measurement of body size of pigs in group-raising environments.

Abstract:The problem of lameness in pigs presents significant challenges to the production and management of pig farms, making accurate detection of pig lameness crucial. Currently, pig farms primarily rely on manual observation and recording, which is inefficient, time-consuming, and prone to subjective judgment errors. In light of this, a method for detecting pig lameness based on key points and walking characteristics was proposed. Firstly, key point information for pigs was defined and determined, including critical parts such as the legs, knees, and back. Based on these key points, an improved YOLO v8n-pose model was employed for detection. This model built upon the original YOLO v8n-pose by introducing a bidirectional feature pyramid network (BiFPN) at the neck for multi-scale feature fusion and incorporating a RepGhost network into the backbone to reduce the parameter count and computational complexity of the feature extraction network. Then using the coordinates of the detected key points, walking characteristics such as stride length, knee bending degree, and back curvature were calculated. These features were inputed into a K-nearest neighbors (KNN) algorithm to classify pigs as lame or non-lame. Experimental results showed that the improved YOLO v8n-pose model achieved a mean average precision (mAP) of 92.4%, which was 4.2 percentage points higher than the detection accuracy of the original YOLO v8n-pose model. Compared with other key point detection models (HRNet-w32, Lite-HRNet, ResNet50, ViPNAS, and Hourglass), the mAP was improved by 10.2, 11.6, 14.2, 11.8 and 12.5 percentage points, respectively. The KNN algorithm achieved a detection accuracy of 81.7% on the pig lameness test set, which was 1.5, 11.3 and 6.5 percentage points higher than that of the BP algorithm, Decision Tree algorithm, and SVM algorithm, respectively. These results demonstrated that the proposed method for detecting pig lameness was feasible and can provide technical support for pig farm detection.

Abstract:In recent years, the application of information technology in sheep farming has become increasingly sophisticated, necessitating more accurate individual identification and behavior monitoring. This, in turn, has placed higher demands on the accuracy of multiple object tracking (MOT) algorithms, which formed the foundation of these applications. However, existing MOT algorithms often underperformed in scenarios involving object occlusion and dynamic environments. Two novel tracking cues, depth modulated IoU (DIoU) and tracklet direction modeling (TDM), was proposed, aiming at enhancing the precision and robustness of multiple object tracking by supplementing the intersection over union (IoU) cue. DIoU improved the traditional IoU calculation by incorporating depth information of the objects. TDM focused on the movement trends of targets, predicting their future directions based on their historical movement patterns. The DIoU and TDM strategies were integrated into the BoT-SORT algorithm, resulting in an improved multiple object tracking algorithm. Evaluations on two datasets showed that the enhanced algorithm increased the multiple object tracking accuracy (MOTA) by 1.6 percentage points and 1.7 percentage points and the identification F1 score (IDF1) by 1.9 percentage points and 1.0 percentage points, respectively, compared with baseline methods. These results indicated that the improved algorithm significantly enhanced tracking continuity and accuracy in complex scenarios. This research provided insights and methods for multiple object tracking technology, holding significant implications for practical applications.

Abstract:In the research on smart pig breeding, the pig instance segmentation method is one of the key technologies to realize automatic detection of pigs. However, in actual segmentation scenarios, there is occlusion and adhesion phenomenon, which makes pig segmentation difficult. Aiming at the difficulty of piglet segmentation in the farrowing room, an instance segmentation model OD_SeGAN was proposed based on YOLO v5s and generative adversarial network(GAN). This method extracted the piglet target through the target detection algorithm YOLO v5s, and inputed it into the semantic segmentation algorithm GAN to achieve segmentation, and used dilated convolution to replace the ordinary convolution in GAN to expand the network receptive field; secondly, a squeeze-incentive attention mechanism was used module to enhance the model’s ability to learn the global characteristics of piglets and improve the model’s segmentation accuracy. Experimental results showed that OD_SeGAN’s IoU on the test set was 88.6%, which was 3.4, 3.3, 4.1, 9.7, and 8.1 percentage points higher than YOLO v5s_Seg, Cascade_Mask_RCNN, Mask_RCNN, SOLO, and Yolact, respectively. OD_SeGAN was applied to the piglet litter average weight estimation task, and the Pearson correlation coefficient between the piglet litter average weight and the number of piglet pixels was measured to be 0.956. The OD_SeGAN proposed had good piglet segmentation performance in actual production scenarios, and can provide a technical basis for subsequent research such as piglet litter weight estimation.

Abstract:The dynamic relationship between stem water content and sap flow is of great significance for understanding the mechanisms of plant water transport and transpiration regulation. However, traditional plant water monitoring methods are limited by spatial and temporal resolution, making it difficult to capture subtle changes in plant water dynamics. An intelligent monitoring system based on the i.MX6ULL chip was developed. By integrating advanced sensor technology, data acquisition, and analysis methods, real-time monitoring of key parameters such as stem water content, sap flow rate, transpiration, soil moisture, and air temperature and humidity was achieved. Long-term field monitoring of Ginkgo biloba trees verified the system’s stability and reliability. Statistical analysis results showed that the sap flow and stem water content of Ginkgo biloba exhibited significant trends at different growth stages, with sap flow rates ranging from 0.82cm/h to 20.52cm/h. During the growing season, the stem water content derivative showed a significant negative correlation with sap flow data (Pearson correlation coefficient was greater than -0.7). As sap flow was increased during the growing season, stem water content was decreased, and the rate of stem water change could reflect the trend of sap flow to a certain extent. Additionally, for every 1℃ increase in air temperature, sap flow rate was increased by an average of 8.6%, while for every 10 percentage points increase in air relative humidity, the sap flow rate was decreased by 27.3%. The research result can provide experimental evidence for the relationship between water transport in standing trees and offer scientific support for plant physiology research and ecological management.

Abstract:During freeze-thaw cycles, significant migration of soil moisture, heat, and salts occurs, which exacerbates soil salinization, thereby having a profound impact on agricultural production stability and the sustainability of soil fertility. Based on field experiments, biochar and straw were applied to the 0~15cm soil layer (BQ and CQ) and the 15~30cm layer (BS and CS), with a blank control group (CK) as a comparison. The moisture content, temperature, and salt concentration in the 0~15cm, 15~30cm, and 30~45cm layers were monitored during the freeze-thaw period to investigate the effects of biochar and straw applied at different depths on soil moisture, heat, and salt dynamics. A structural equation model was used to analyze the relationships between moisture, temperature, and salts across different soil layers. The results showed that during the experimental period, the application of biochar and straw significantly improved the water, heat, and salt characteristics of the soil. Specifically, the average moisture content in the 0~45cm soil layer for the BQ, BS, CQ, and CS treatments was increased by 2.85, 3.13, 1.56, and 2.15 percentage points, respectively, compared with that of the control group. All treatments effectively increased soil temperatures and reduced temperature fluctuations during the freeze-thaw period. The average salt concentration in the 0~45cm soil layer for the BQ and BS treatments was increased by 0.34g/kg and 0.40g/kg, respectively, compared with that of the control group. Furthermore, the application of biochar effectively suppressed salt migration by adsorbing salts. The structural equation model results indicated that moisture migration affected both heat transfer and solute movement, and the application of biochar and straw changed the correlations between water, heat, and salts across the different soil layers. These findings can provide theoretical and technical support for regulating soil ecological environments in regions with seasonal frozen soil.

Abstract:The accumulation of phytolith-occluded organic carbon (PhytOC) in soil is a potential pathway for long-term organic carbon sequestration. Silicon fertilizer, an exogenous silicon amendment, can enhance carbon sequestration by phytoliths in crops, while biochar plays an active role in reducing soil greenhouse gas emissions. To investigate the pathways through which combined application of silicon fertilizer and biochar affected the carbon sequestration capacity of phytoliths and its impact on soil greenhouse gas emissions, four treatments were established, including control (CK), silicon fertilizer (SF), biochar (BC), and a mixed application of silicon fertilizer and biochar (BS). The distribution characteristics of soil silicon fractions were analyzed through field experiments combined with laboratory phytolith stability grading tests to confirm differences in PhytOC sequestration and its stability in different maize organs. Additionally, the effects of mixed exogenous silicon on crop agronomic characteristics and the reduction of soil greenhouse gas emissions were elucidated. The results indicated that under the BS treatment, the contents of readily soluble silicon (CaCl2-Si), unstable silicon on the surface of inorganic soil particles (Acetic-Si), and unstable silicon on the surface of soil organic matter (H2O2-Si), exhibited an initial increase followed by a decrease, while the content of weakly crystalline silicates and amorphous silicon (Na2CO3-Si) showed a continuous decline. BS treatment significantly increased the phytolith content of the maize stem, sheath, and leaf by 54.75%, 5.68%, and 56.87%, respectively, compared with CK. The crop PhytOC production flux reached 57.79kg/(hm2·a). The stable phytolith content was increased by 16.32 percentage points, and the stable PhytOC production flux reached 34.12kg/(hm2·a). Furthermore, the global warming potential (GWP) under combined application was 3716.88kg/hm2. The results demonstrated that combined application of silicon fertilizer and biochar significantly enhanced the carbon sequestration capacity of crop phytoliths and reduction of soil greenhouse gas emissions, providing strategies and methodologies for achieving long-term carbon sequestration.

Abstract:Aiming to investigate the effects of combined straw and biochar application on soil nitrogen pools, crop yield, and nitrogen utilization in the black soil region of Northeast China, a split-plot experimental design was employed, incorporating three field return methods—control (CK, no return), full straw return (SF), and combined biochar and straw return (BS)—along with three nitrogen application rates: 75kg/hm2 (N1), 60kg/hm2 (N2), and 45kg/hm2 (N3). The results showed that the mixed application of straw and biochar alleviated the unfavorable impression of reduced nitrogen application on soil nitrogen content and soybean plant growth, and the effects were more pronounced at N2 nitrogen application. Compared with CK and SF treatments, BS treatment promoted the increase of soil ammonium nitrogen and nitrate nitrogen contents by 7.58%~78.08% and 19.02%~95.56%, respectively, which significantly enhanced the net photosynthetic rate and nitrogen utilization efficiency of soybean, and led to the increase of soybean yield by 38.62%~60.97%. Comprehensive evaluation using the entropy-weighted TOPSIS model identified the BSN2 treatment as the most effective, achieving an average two-year yield of 3058.48kg/hm2 with a nitrogen application rate of 60kg/hm2. This treatment also recorded high nitrogen use efficiency (0.99), agronomic efficiency (9.34kg/kg), nitrogen recovery rate (0.98), and nitrogen response index (2.18). These findings can provide a scientific basis for optimizing nitrogen fertilizer management and enhancing straw utilization in Northeast China’s black soil region.

Abstract:In order to investigate the simulation effect of machine learning model on actual evapotranspiration (ETa) of winter wheat during the reproductive period and the effect of solar-induced chlorophyll fluorescence (SIF) on the simulation accuracy of machine learning model in the absence of meteorological data, SIF was combined with meteorological indicators, crop physiological indicators, soil thermal conditions and other factors, and three classical machine learning models, namely the gradient boosting (GB), random forest (RF), and support vector machine (SVM) were constructed, combined with linear regression (LR) model to simulate winter wheat ETa and compared with the evapotranspiration ET_pm calculated by Penman-Monteith (P-M) model. The results showed that although SIF was significantly correlated with ETa, the fitting accuracy of the machine learning model constructed only by using SIF as a feature parameter was low; according to the importance ranking of the feature parameters based on the machine learning model as well as the simulation accuracy of the model under each scenario, it was known that SIF had an enhancement effect on the accuracy of the machine learning model in simulating ETa. The machine learning model fit better than the P-M model when there were enough feature parameters, and adding feature parameters to the average temperature, SIF, sunshine hours, leaf area index (LAI) and soil moisture content did not improve the simulation accuracy, so it was recommended to use the feature set composed of the five feature parameters mentioned above to construct a machine learning model to predict ETa. The R2 of the models were 0.92, 0.91 and 0.91, respectively, among which the GB model had the best fitting effect on the ETa of winter wheat during the whole reproductive period. The research result can provide a reference for the accurate simulation of local evapotranspiration and the development of rational irrigation system in the absence of meteorological data.

Abstract:Efficient and sustainable water use in agriculture, as viewed through the lens of water footprint analysis, plays a crucial role in enhancing regional food security and environmental sustainability. Focusing on the rice-wheat rotation system, a crop water footprint calculation model was developed based on water footprint theory. The model was applied to assess the efficiency of water use in the rice-wheat rotation system in Lianshui Irrigation District spanning from 1960 to 2019. The analysis revealed the temporal evolution and influencing factors of water use efficiency in this system. Results indicated that the generalized water system number was ranged from 0.50 to 0.76 over the study period with a multi-year average of 0.65, showing no significant overall trend. In contrast, the crop production water footprint exhibited a yearly average of 58.4m3/GJ, displaying a consistent decline. Specifically, the green water footprint accounted for 40.6%~80.4% of the overall water footprint, while the blue water footprint averaged 22.6m3/GJ. Meteorological factors, predominantly precipitation, significantly influenced both the broad water system number and crop water footprint. The study highlighted a negative correlation between crop production water footprint and agricultural inputs, as well as regional irrigation intensity. Factors such as agricultural mechanization and water-saving irrigation practices played a crucial role in shaping water use efficiency. Enhancing rainfall utilization and adopting advanced agricultural technologies were identified as effective strategies to optimize water resource management in agriculture. Findings from this research can offer valuable insights for developing regional agricultural water conservation standards.

Abstract:The surface of saline-alkali soil is often accompanied by a complex network of shrinkage cracks. Investigating the characteristics of shrinkage cracks in saline-alkali soil and the distribution of soil salinity during the dynamic evolution process is of significant importance for scientifically formulating leaching regimes to mitigate soil salinity. Indoor soil grids were employed to investigate the dynamic evolution characteristics of shrinkage cracks in saline-alkali soil and conduct salt leaching experiments. Three initial soil salinity levels were set at 2g/kg (S1), 5g/kg (S2), and 8g/kg(S3). Digital image processing technology and morphological algorithms were employed to obtain geometric parameters and connectivity indices of the soil cracks. The evolution process of shrinkage cracks during the drying-wetting cycles in soils with different initial salinity levels was analyzed, and simultaneous investigations were conducted into the dynamic variations of soil salinity during crack evolution. The results indicated that during the process of soil shrinkage and cracking (soil dehumidification), an increase in initial soil salinity corresponded to increases in the crack area ratio, mean width, length density, and connectivity index. Moreover, within a single wet-dry cycle, the crack area ratio and mean width form an “∞” ring shape. Concurrently, soil salinity gradually migrated toward the vicinity of the cracks, ultimately leading to a non-uniform distribution pattern with higher salinity at the edges of the crack network and lower salinity within the grid. During soil shrinkage and cracking, the coefficient of variation of soil salinity content within treatments S1, S2, and S3 was increased as soil moisture was decreased, reaching 0.235, 0.247 and 0.251, respectively, after crack development stabilized (at soil water content of approximately 5%). In the process of soil salinity leaching (soil hygroscopic), the crack area ratio in treatment S3 was increased by 8.565 percentage points and 4.208 percentage points compared with that of treatments S2 and S1, respectively, with corresponding increase in soil desalination rates of 20.4% and 67.3%. Overall, higher initial soil salinity resulted in a greater soil leaching desalination rate. The final soil desalination rates for treatments S3, S2, and S1 were 54.2%, 45.0%, and 32.4%, respectively (P<0.05). Overall, the research result elucidated the dynamic evolution characteristics of shrinkage cracks in saline-alkali soil and unveiled the intricate relationship between crack evolution and soil salinity distribution. The findings can offer valuable insights into the formulation of effective soil improvement strategies to mitigate soil salinity issues.

Abstract:Under the background of dual-carbon goals, energy transition in rural areas has become a critical element in achieving green development. Traditional agricultural machinery in rural China, such as tractors and harvesters, heavily relies on fossil fuels, necessitating a shift toward low-carbon and sustainable development models represented by electric agricultural machinery. This transition promoted the utilization of renewable energy and provided an environmentally friendly solution for sustainable agricultural development. Considering the abundance of renewable resources like solar energy and biomass in rural areas, as well as the operational characteristics of agricultural machinery and flexible agricultural loads, a planning method for a rural energy system integrating biomass and solar energy with electrified agricultural machinery was proposed. Firstly, the framework of rural energy system considering electric farm machines was constructed, and the load of electric farm machines and the flexible load of agriculture were modeled and analyzed at the same time. Secondly, an optimization model was established with the objectives of minimizing economic costs and carbon emissions, and solved collaboratively by using the non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ) and the commercial solver Gurobi. Finally, a planning simulation was carried out for a farm in Northeast China. Results showed that introducing electric agricultural machinery and agricultural flexible loads reduced total system costs by 24% and carbon emissions by 46% under the condition of considering the purchase of electric agricultural machinery.

Abstract:The Box-Behnken experimental design was used in the experiment, and a total of 17 groups of pelleting experiments were carried out. The results showed that with the increase of replacement ratio of wheat, the bulk density, tap density, water solubility index, protein dispersibility index, pasting time of the mash feed were increased significantly (P<0.05), the angle of repose, water absorption index, peak-viscosity, through-viscosity and final viscosity were decreased significantly (P<0.05), and the pelleting rate was increased, power consumption per ton was decreased and the pellet durability index was increased first and then decreased. With the increase of conditioning temperature, the pelleting rate was firstly increased and then decreased. When the replacement ratio were 0 and 50%, the power consumption per ton was decreased, and when the replacement ratio was 100%, the power consumption per ton was firstly decreased and then increased and the pellet durability index was increased. With the extension of conditioning time, the power consumption per ton was increased, and the changes of pelleting rate and pellet durability index were relatively small. The results of variance analysis showed that the primary and secondary order of factors affecting the pelleting characteristics of pellet feed was replacement ratio, conditioning temperature, and conditioning time. It can be seen that the use of wheat replacing of corn can significantly improve the quality of pellet feed and reduce the power consumption per ton.

Abstract:Oyster agricultural products may carry transmissible viruses, bacteria, and other biological risk factors during their market circulation, which has caused frequent biosecurity incidents in many countries and seriously threatened people’s health. The privacy characteristics of biological risk detection data are significant, and the improper disclosure of biological risk detection information can pose a threat to public health security. A privacy-preserving data encryption and sharing method for oyster supply chain biological risk factors was proposed based on blockchain technology. By encrypting and storing biological risk factor detection data from various links in the supply chain on the blockchain, data security and sharing were ensured, and the source of the risk can be traced back. The method used attribute-based searchable encryption algorithms to perform access control and privacy protection on biological risk factor detection data, and an encrypted inverted index for query efficiency optimization was constructed. By combining searchable encryption algorithms and inverted indices, rapid location of relevant risk batch goods’ detailed data was achieved. The experimental test results showed that the average encryption time of biological risk detection data keyword was 31ms, the average trapdoor generation time was 32ms, and the average matching time of searchable ciphertext and trapdoor was 16ms. The average time for regulators to query risk oyster agricultural product-related data through encrypted inverted index was 385ms, and a prototype system was built on the Ethereum blockchain platform to realize the functions of biological risk factor encrypted privacy data storage and traceability query. The results showed that the method can meet the demand of privacy data sharing scenarios in oyster supply chain and provide technical support for oyster biological risk regulation.

Abstract:Aiming to address the challenges of prolonged inability to maintain constant temperature control and excessive reliance on manual assistance in the curing phase of white pepper primary processing production lines, a proportion integration differentiation (PID)-based control system was developed to control the curing temperature of the white pepper during processing. It is a high demand to maintain the constant curing temperature. Specifically, too high curing temperature can lead to the internal physicochemical properties of the destruction, whereas, too low curing temperature can lead to curing not complete, which makes the peeling rate decreased. The control system with an ST Microelectronics 32-bit Microcontroller (STM32) and a touchscreen was utilized to control the start/stop of the steam generator and the opening of the electric regulating valve. A temperature sensor was installed at the outlet of the curing machine, and a PT100 temperature sensor was employed to collect the curing temperature in real-time. Subsequently, the collected data was fed back to the STM32 microcontroller. The PID closed-loop control algorithm was applied to calculate the actuator, adjusting parameters appropriately to ensure stable control of the curing temperature by modulating the steam flow. A systematic analysis of the convective heat exchange process between white pepper and steam at temperature was conducted. A theoretical model of heat transfer was established by using the step response curve method, and the data curve was processed (R2=0.969) to derive the control model for the temperature inside the curing machine over time. Simulation analysis was performed by using the Simulink platform to determine the optimal parameters for PID control. Response curves from four PID parameter tuning methods, including the Ziegler-Nichols method, the decay curve method, the critical proportional method, and the sparrow search algorithm-based particle swarm optimization method (SSAPSO), were compared. Ultimately, it was found that the SSAPSO-based method yielded the best control effect in terms of dynamic performance indicators with PID parameters (proportional coefficient Kp=0.8759, integral coefficient Ki=0.02, and differential coefficient Kd=4.3255). The response time of the PID controller obtained by the SSAPSO-based method was approximately 40s with an overshoot of about 2.5%. Systematic experimental studies demonstrated that throughout the entire 8 minutes curing process, the current curing temperature was sampled every minute. Due to direct convective heat exchange between the curing machine and the air, the temperature remained stable within the range of (99±1.5)℃. The average relative error of the curing temperature was less than 1.2%, and the coefficient of variation was less than 1.3%, thereby achieving automated, precise, and efficient temperature control during the curing process.

Abstract:In order to accurately and effectively predict the traction performance of tracked undercarriages under the complex operating environment in hilly and mountainous areas, a prediction model for the traction performance of tracked undercarriages was established and experimentally verified based on the law of grounding pressure distribution. Firstly, a mathematical model of ground pressure distribution with multi-peak non-linear distribution was proposed by considering the slope angle and chassis parameters, and the ground pressure test was carried out at different slopes and postures, and the results showed that the average error of its prediction was about 4.7%, the model can better predict the distribution of grounding pressure of crawler chassis in slope environment. Secondly, based on the grounding pressure model and the track-ground interaction law, by considering the soil characteristics, slope and the change of the position of the center of gravity for the attitude adjustment of the crawler chassis, the traction-slip rate prediction model of the crawler chassis composed of the driving force characteristics of the evenly distributed vertical load and the non-uniform load control part was further constructed. Finally, the traction performance tests of contour driving and longitudinal climbing conditions were carried out based on the three types of crawler chassis to verify the prediction model, and the results showed that the average prediction error of the model was 3.6%, 5.4% and 6.3%, respectively, and the overall prediction error was small, which could provide theoretical basis and data support for the design and development of the applicability of the tracked chassis in hilly and mountainous areas and the optimization of maneuverability.

Abstract:The alternating flow distribution pump can directly output alternating fluid flow through the continuous rotation of its built-in rotating valve plate, driving the hydraulic cylinder to generate excitation motion. Due to nonlinear factors such as internal leakage and friction within the alternating flow distribution pump and pump-controlled hydraulic cylinder, the bidirectional fluid flow generated by the pump may not be entirely symmetrical. As a result, the displacement center of the pump-controlled hydraulic cylinder may shift during the excitation process, thereby affecting its practicality. To address above problem caused by asymmetric fluid flow, a pump-valve-cylinder cascaded deviate regulation system was designed by placing a servo valve in series between the alternating flow distribution pump and the hydraulic cylinder. Additionally, a half-cycle piecewise deviate regulation control strategy based on the phase of the rotating valve plate was proposed. Based on the dynamic characteristics of the pump and valve, an AMESim-Simulink co-simulation model was established. System parameters were identified by experiment. The control performance was analyzed through simulation. A test platform was constructed to verify the effectiveness of the deviate regulation method. The research result showed that this method can effectively control the displacement center of excitation hydraulic cylinder controlled by the alternating flow distribution pump. Moreover, the servo valve remained mostly at a large opening, minimizing throttling losses. This approach ensured the high efficiency of the pump-controlled excitation system while improving its practicality.

Abstract:In the automatic assembly line of agricultural machinery, the on-chip resources of its embedded control platform are extremely limited, and the parameter amount of the convolutional neural network-based deep learning detection system is too large, which is difficult to be directly transplanted to the embedded platform. Therefore, a deep learning real-time detection method based on improved ResNet18-SSD (single shot multi-box detector) and field programmable gate array (FPGA) acceleration engine was proposed. In order to improve the accuracy of the detection model while reducing the number of parameters, a deep learning fast detection model based on ResNet18-SSD was proposed, which utilized the optimized and improved ResNet18 network to replace the VGG16 predecessor network of the SSD model, and introduced a multi-branch isomorphic structure and an asymmetric parallel residual structure, so as to adapt to the complex scenes such as occlusion, dim light; and in the case of meeting the detection accuracy requirements, a dynamic fixed-variance network was used to meet the detection accuracy requirements. Under the condition of meeting the requirements of detection accuracy, the dynamic fixed-point quantization was adopted to reduce the model data volume to improve the execution efficiency of the detection model. Aiming at improving the convolutional layer in the ResNet18-SSD model, which consumed serious resources, an FPGA acceleration engine based on the Winograd algorithm was proposed to improve the real-time performance of the model detection, and through the software-hardware co-design, joint optimization was carried out from the perspectives of the hardware gas pedal and the lightweighting of the software network, so as to achieve a balance between the lightweighting, acceleration performance, and accuracy in the complex scene. Experimental results on the Xilinx FPGA embedded platform showed that the detection accuracy of the proposed method reached 93.5%, and the detection time of a single image under the operating frequency of 100MHz was 80.232ms, which met the real-time demand.

Abstract:Aiming at the problem that the traditional ORB algorithm fails to meet the high-precision localization requirements due to the high mis-matching rate in the binocular feature matching stage, a feature matching algorithm based on the improved binocular ORB-SLAM3 is proposed. The nearest neighbor matching algorithm (FLANN) is introduced in the feature point matching stage, and more accurate matching pairs are filtered out by setting the ratio threshold, and the adaptive weighted SAD-Census algorithm is introduced in the binocular ORB-SLAM3 three-dimensional matching, and the geometric distances between the cases are taken into account to recalculate the SAD values and merge them with the Census algorithm to improve the stability and accuracy of feature matching, while the adaptive weighted SAD-Census algorithm is introduced. At the same time, the adaptive SAD window sliding range is added to further expand the search distance, so as to filter out the correct matches to improve the accuracy of the system. Experiments are carried out in the EuRoC dataset and real indoor scenes, and the results show that compared with the pre-improved ORB-SLAM3 algorithm, the localization accuracy of the improved algorithm is improved by 23.32% in the dataset, and nearly 50% in the real environment, thus verifying the feasibility and effectiveness of the improved algorithm.