Abstract:The rapid development of crop phenomics demands high-efficiency, intelligent, and cost-effective technologies and systems for large-scale data acquisition and analysis, as well as for germplasm phenotyping. To address these challenges, multidisciplinary innovations was integrated to overcome key technical bottlenecks in high-throughput data acquisition and intelligent traits extraction for crop phenomics. A suite of proprietary technologies was developed, including lightweight and agile multi-sensor arrays, universal imaging box, and both fixed and mobile high-throughput phenotyping platforms adaptable to diverse environments, together with corresponding algorithms and software systems. These developments culminate in the Crop Phenomics Big Data Factory (CPBDF). CPBDF is a comprehensive technology and equipment framework that conceptualizes farmlands, greenhouses, and growth chambers as “factories”, where phenotyping platforms function as “production lines”, and the output is high-quality phenomics big data. The system integrated field-based and facility-based autonomous phenotyping platforms, organ- and microscopy-level phenotyping systems, automated cultivation control devices, crop modeling systems, a digital-twin intelligent management platform, and a big data computing center. It enabled automated, multi-source, and multi-scale data acquisition with high throughput, precision, and integration, supporting three-dimensional reconstruction and quantitative phenotypic analysis across crop populations, individuals, organs, and microstructures. The proposed framework established a paradigm for the production, processing, and application of crop phenomics big data. It provided foundational infrastructure for digital breeding and smart cultivation, and served as a key enabler for AI for Science-driven research platforms and factory-style germplasm phenotyping.

Abstract:High-throughput and precise acquisition and analysis of crop phenotypic information are fundamental components of modern agricultural breeding and precision cultivation systems. However, traditional manual measurements in complex field environments are limited by low efficiency, high labor intensity, and strong subjectivity, making it difficult to meet the growing demand for large-scale, multi-trait, and time-series phenotyping. To address these challenges, an in-field phenotypic data fusion and analysis method was proposed based on a ring-shaped unmanned vehicle phenotyping platform and a multimodal homogeneous sensor array. The platform integrated multiple homogeneous sensors, including RGB and depth cameras, enabling multi-angle and three-dimensional in situ crop observations. A systematic multi-source heterogeneous data fusion workflow was designed, consisting of image preprocessing, depth information extraction, 3D reconstruction, temporal tracking, and feature analysis, to achieve accurate extraction and dynamic reconstruction of key phenotypic traits such as plant height, canopy structure, and spatial distribution. Field experiments were conducted on maize plants at multiple growth stages. The results demonstrated that the proposed platform can stably and continuously acquire high-quality multimodal phenotypic data. The reconstructed plant height measurements showed a high correlation with manual measurements, with an average error within 5cm, verifying the accuracy and robustness of the method. Compared with conventional single-view or mechanically rotating observation methods, the proposed platform exhibited superior adaptability to field environments, allowing rapid deployment and efficient operation, thereby providing an effective technical foundation for large-scale in-field phenotyping. Furthermore, the platform’s advantages were discussed in terms of portability, scalability, timeliness, and automation, and envisions future developments toward embodied intelligence and autonomous phenotyping. The proposed ring-type unmanned vehicle platform and multimodal data fusion method can provide a high-throughput, low-disturbance, and scalable technical solution for in-field crop phenomics, supporting modern crop breeding and precision agriculture.

Abstract:In greenhouse vegetable phenotyping, uneven terrain induces high-frequency vibrations in the 3~6Hz range, leading to small-angle vibration and hysteresis in the acquisition device, which degraded the resolution of collected images. A stabilized platform was developed based on a composite control system incorporating gravity-compensated dual angular acceleration feedforward and an improved cascade PID controller. A single-arm stabilized platform with dimensions of 300mm×280mm×250mm was constructed to accommodate a narrow line spacing of 30cm and support the integration of multi-source sensors. The stabilized platform had a self-weight of 5kg and can carry a payload of 15kg. It was equipped with X-Y-Z axis sliding rails for center-of-gravity adjustment, limiting the center-of-gravity deviation to within ±5mm and maintaining gravitational torque variation below 0.5N·m. A gravity compensation feedforward model was established by linear fitting, achieving a determination coefficient R2 of 0.9912. An improved cascade PID structure was implemented, combining an inner velocity loop with an outer position loop. Key enhancements included integral separation activated when the error exceeded 1°, integral limiting, and a resetting mechanism triggered by error zero-crossing. These measures effectively suppressed integral saturation during fine adjustments, achieved a steady-state error of 0.1°. In addition, angular acceleration feedforward from dual IMUs mounted on both the vehicle and the stabilized platform compensated for inertial disturbances caused by vehicle start-stop and turning accelerations of 2~3m/s2. The verification test results showed that the composite control strategy reduced the system step response time by 80% without overshoot. When the vehicle operated at 0.5m/s, the stabilized platform’s triaxial angular oscillation was constrained to ±0.5° in roll, ±0.3° in pitch, and ±0.2° in yaw. These outcomes confirmed that the system satisfied the requirements for high-accuracy phenotyping acquisition.

Abstract:High-throughput 3D crop phenotyping is one of the core methodologies in modern crop phenomics research, providing crucial data support for holistic morphological structure analysis, precise evaluation of plant architectural traits, and genotype-phenotype association analysis. Aiming to address the challenges of low efficiency and limited data accuracy inherent in traditional manual measurements, a high-throughput 3D crop phenotyping data acquisition platform was developed based on an unmanned ground vehicle (UGV). The performance of four mainstream sensors (FLIR visible light camera, Kinect DK, Velodyne VLP-16, and Livox Avia) and their corresponding 3D reconstruction algorithms for crop phenotyping were systematically investigated. Specifically, it was compared the 3D reconstruction from visible light images based on structure-from-motion (SfM) and multi-view stereo (MVS), 3D reconstruction from RGB-depth images based on iterative closest point (ICP), point cloud reconstruction from solid-state LiDAR leveraging LiDAR-inertial odometry (LIO) and point cloud stitching from mechanical rotating LiDAR by using uniform velocity frame superposition. Experiments were conducted on potted lettuce plants in a greenhouse, where point cloud data acquired by the four methods underwent standardized processing. An automated processing pipeline was developed, enabling precise extraction and analysis of key phenotypic parameters, such as plant height and maximum canopy width. This research thoroughly explored and analyzed the characteristics, advantages, and disadvantages of each method. Their applicability was comprehensively evaluated based on point cloud quality, reconstruction efficiency, phenotypic trait accuracy and system cost. The findings can not only provide experimental basis for sensor selection and algorithm development of 3D phenotyping UGVs but also can offer valuable references for breeders and agronomists in selecting efficient and accurate phenotyping data acquisition approaches.

Abstract:The rapid development of high-throughput crop phenotyping acquisition equipment has provided modern data collection means for breeding and cultivation research, while spawning massive multi-modal and unstructured phenotypic data. Traditional structured data storage models can no longer meet the efficient access requirements of such data.A hybrid access framework was proposed based on distributed technology, which used HBase and HDFS to build a structured and unstructured fusion storage engine, integrated client-side cache and Redis cache to design an efficient retrieval mechanism, and optimized core issues: aiming at the inherent defects of native HDFS in storing phenotypic data, a modal aggregation-based MCH storage framework was designed. By classifying and merging phenotypic data according to modalities and constructing local indexes by using double-layer hashing technology, it effectively reduced NameNode memory pressure while improving access efficiency and storage space utilization of single-modal data. For high-concurrency data reading scenarios, a double-layer cache mechanism based on data popularity was constructed. It optimized hot data reading efficiency through metadata hierarchical caching and innovatively proposed a data popularity evaluation model combining access frequency and time characteristics, which effectively improved cache hit rate. Experimental results showed that when the data scale was 1.0×105, the proposed distributed access method reduced the NameNode memory occupancy rate by 31.2% compared with the optimal native solution (SequenceFile), and the retrieval time by 25.4% compared with the optimal native solution (MapFile), providing technical support for the storage and retrieval of massive multi-modal phenotypic data.

Abstract:Efficient and accurate monitoring of maize ear height (EH) is critical for anti-lodging breeding. The traditional manual measurement approach is labor-intensive and time-consuming, while existing automated approaches often lack robustness under varying field conditions or involve high costs. To address these limitations, an iOS application (APP) was developed based on the you only look once (YOLO) model and augmented reality (AR) technology for real-time, accurate, efficient, and low-cost maize EH measurement. It comprised two modules: a maize ear detection model and a height measurement module. The ear detection model was trained and validated on a dataset comprising 1000 field images collected from maize fields during the filling stage, under various lighting and occlusion conditions. Among different object detection models, the YOLO v5s model demonstrated the most robust performance with a precision of 0.844, a recall of 0.724, and an AP0.5 of 0.814. The trained detection model had been integrated into a maize EH measurement system, which utilized the AR technology for real-time measurement. It demonstrated excellent compatibility and performance on iOS devices, with response time below 0.3 s. Field evaluation results indicated a high correlation between the EH measured by the app and manual measurements (R2=0.750~0.864, RMSE=0.10~0.13m). The app was optimized for solo operation. To finish measuring a plot with over 10 maize plants only took less than 2 minutes, which was over 6 times faster than that of the traditional measurement with the leveling rod. This app significantly improved the efficiency of maize EH measurements while maintaining accuracy, providing real-time and precise data support for field management and breeding programs.

Abstract:Image semantic segmentation technology is one of the key methods for obtaining phenotypic information of maize plants. Traditional fully supervised semantic segmentation methods typically rely on a large number of pixel-level labels. However, maize exhibits significant morphological variability across different growth stages, leading to high costs associated with image annotation and limiting the practical application of such models in real-world production scenarios. To eliminate the need for manual annotation during model training, a self-supervised few-shot semantic segmentation network for maize plant images (MSDANet) was proposed based on self-supervised learning, aiming to improve the semantic segmentation accuracy and model generalization capability of maize plant images across different growth stages. MSDANet utilized a superpixel-based self-supervised learning method to generate pseudo labels, enabling the construction of preliminary supervision signals for the support set images without manual annotation. It designed a mixed masking mechanism (MM) that applied pseudo label-based semantic masking to construct diverse masked samples in the feature space, promoting the model to learn more robust feature representations and thereby improving segmentation accuracy in complex backgrounds. To address the complex morphological issues of corn plants in images, such as bending, overlapping, and occlusion, a multi-scale deformable large kernel attention mechanism (MS-DLKA) for the model was designed. By integrating multi-scale receptive fields and deformable convolutions, it can flexibly perceive important structural information of corn plants at different scales, effectively improving semantic segmentation accuracy. When validated on a small sample dataset, MSDANet achieved mIoU and FB-IoU of 75.63% and 87.12%, respectively, in the 1-shot setting;in the 5-shot setting, mIoU and FB-IoU reached 76.04% and 87.21%, respectively, both outperforming other models of the same type proposed in this study. Additionally, compared with current mainstream fully supervised few-shot semantic segmentation models, mIoU was improved by 2.9 and 2.93 percentage points under 1-shot and 5-shot settings, respectively. The results demonstrated that the MSDANet model can achieve high-precision semantic segmentation of corn plant images without human labels and with few samples, providing technical support for corn image analysis and plant phenotyping at different growth stages.

Abstract:Maize is one of the most important staple crops. Early-stage management during the seedling period is crucial for its yield formation. Accurate and rapid monitoring of maize seedling growth is essential for early-stage interventions such as replanting and water-fertilizer management. Traditional seedling monitoring methods rely heavily on manual field surveys, which are often inefficient and subject to strong observer bias. Leveraging RGB imagery and computer vision techniques to perform large-scale, rapid, and accurate crop monitoring has become a key trend in smart agriculture. An automated method for evaluating field crop uniformity was proposed based on seedling counting and leaf age estimation results from RGB images. The method firstly performed image-based row detection and missing seedling detection to extract six indicators: seedling missing rate, plant spacing, row spacing, leaf age, missing plant rate, plant bounding box area, and plant coverage. The seedling missing rate and the coefficient of variation (CV) of the other five variables form the six key indicators for seedling uniformity assessment. The entropy weight method was employed to determine the weight of each indicator, and the TOPSIS multi-criteria decision-making model was used to calculate the overall uniformity score. Based on expert knowledge, the uniformity was then classified into three discrete levels. Validation results showed that the classification results of the proposed evaluation system were highly consistent with expert grading, achieving an overall classification accuracy (OA) of 0.92. The method also achieved high accuracy (OA was 0.94 and 0.96) in two independent datasets, demonstrating its adaptability and generalizability. The research result can provide a technical foundation for the standardized and automated evaluation of crop uniformity in the field.

Abstract:The high-throughput and non-destructive acquisition of panicle phenotypic parameters is a key step in rice breeding and phenomics research. To address the limitations of traditional manual methods—which are inefficient and destructive—and the poor flexibility of existing image-based methods that rely on manual priors, a graph structure-guided approach for skeleton parsing and non-destructive measurement of key phenotypic parameters was proposed. Firstly, building on the YOLO v9 framework, a more robust key point detection model was trained by incorporating mixed-background data augmentation and a Wise-IoU (WIoU) loss function to improve the detection of panicle nodes and neck nodes. Next, threshold segmentation and thinning were applied to panicle images to extract skeletal structures and construct an undirected graph topology. The detected key points were then deeply integrated with the skeleton topology to classify key point types and assist graph-based algorithms in automatically identifying the rachis, primary branches, and secondary branches. Physical scale conversion was achieved by using a calibration object. Experimental results demonstrated that the improved detection model increased mAP for neck nodes and panicle nodes by 4.5 and 2.4 percentage points, respectively, compared with the baseline, while recall was improved by 7.8 and 4.0 percentage points, and the correct detection ratio of key points was increased by 4.6 and 5.0 percentage points. In structural counting, panicle node counting achieved zero error, and the mean relative errors for primary and secondary branch counts were within 0.39% and 2.38%, respectively. For dimensional measurements, the mean relative errors of primary branch length, secondary branch length, rachis length, and internode length were controlled within 3.2%, 7.5%, 3.1%, and 5.2%, respectively, with mean absolute errors not exceeding 2.9mm, 2.3mm, 2.3mm, and 1.6mm. The research result achieved automatic and non-destructive extraction of key rice panicle phenotypic parameters, providing a viable technical solution for high-throughput panicle phenotyping.

Abstract:Plant phenotyping plays a vital role in precision agriculture, crop breeding, and production management, among which maize phenotyping research is of particular significance for yield improvement, quality enhancement, and agricultural modernization. With the advantages of high precision and rich structural information, 3D point cloud technology has emerged as an important tool in plant phenotyping. Compared with traditional 2D image-based methods, point clouds provide a more accurate description of plant organ morphology, thereby enabling precise monitoring of maize growth and extraction of phenotypic traits. Nevertheless, existing point cloud segmentation methods still face challenges in maize stem-leaf analysis, especially in recognizing newly emerging leaves, segmenting overlapping or closely spaced leaves, and delineating stem-leaf boundaries, which restricted the accuracy of phenotypic parameter measurement. To address these issues, a distance field-based stem-leaf segmentation method for maize point clouds was proposed. Specifically, Quickshift++ and Minkowski distance fields were integrated with a constrained median-normalized region growing algorithm for precise stem extraction. Furthermore, the segmentation framework based on skeleton and optimal transport distance has been refined, enhancing the accuracy of boundary recognition between stems and leaves. Experiments were conducted on both self-collected and public maize point cloud datasets. The results demonstrated that the proposed method significantly improved segmentation accuracy and enhanced the precision of phenotypic trait extraction, including stem height, stem diameter, leaf length, and leaf width. The research result can provide methodological support for maize phenotyping and offer valuable references for intelligent agriculture and precision crop management.

Abstract:Peanut (Arachis hypogaea L.), a critical oilseed crop, plays a crucial role in ensuring food and oil production security. Accurate, nondestructive, and real-time phenotypic monitoring is essential for optimizing peanut production management. Multispectral data acquired by an unmanned aerial vehicle (UAV) platform during key growth stages were leveraged to extract canopy multispectral (MS), structural (CHM), and textural (TEX) parameters. Four machine learning algorithms, partial least squares regression (PLSR), support vector machine (SVM), artificial neural network (ANN), and random forest regression (RFR), were employed to construct estimation models for plant height, SPAD values, and aboveground biomass. Results demonstrated strong correlations between peanut aboveground biomass/plant height and the near-infrared band (Pearson correlation coefficients were 0.77 and 0.69, respectively). The random forest model, integrating textural, structural, and spectral features, achieved optimal biomass estimation accuracy (R2=0.96). For plant height inversion, the PLSR model combining textural and spectral features performed best (R2=0.94). SPAD estimation using PLSR with fused textural and structural features yielded moderate accuracy (R2=0.39, RMSE=3.06, nRMSE=0.062, RPD=1.30). The research identified feature-specific requirements for machine learning-based estimation of distinct peanut phenotypic traits and established a UAV multi-source data fusion framework capable of accurate, nondestructive, and efficient assessment of plant height and biomass. These findings can provide a robust technical approach for growth monitoring and precision management in peanut cultivation systems.

Abstract:The detection of inflorescences from the world’s three major cereal crops (rice panicles, wheat spikes, and maize tassels) is a fundamental task in precision farming and cereal crop phenotyping. However, accurate detection remains challenging due to dense distributions, significant scale variations, and small-target in complex environments, which substantially compromise the precision of detection. To tackle these issues, gated attention-DETR (GA-DETR), an architecture based on RT-DETR was proposed, which introduced three novel components: aiming at the delicate tip features of cereal inflorescences, a gated mechanism C2F (GMC2F) module was proposed to enhance backbone feature discrimination through dynamic channel weighting and cross-stage local feature integration. To address the scale mismatch caused by differences in the shapes of cereal inflorescences, an attention upsample scale sequence feature fusion (AUSSFF) module was proposed, which enhanced the robustness of multi-scale dependency modeling through 3D convolutions. For difficult small target in UAV images, a FPIoU loss function was proposed, which combined target-size adaptive weighting and difficulty-aware stratification to optimize the performance on hard samples. GA-DETR performed better than the baseline RT-DETR and five mainstream detection models on the RiceR dataset, GWHD dataset, and MTC-UAV dataset, including rice panicles, wheat spikes, and maize tassels, achieving mAP@0.5 of 92.8%, 91.7%, and 91.3%, respectively, while for RiceR dataset reducing model parameters by 32.5% and floating-point computational load by 14.4%. The proposed framework surpassed five state-of-the-art frameworks in inflorescence counting on GWHD dataset, achieving an MAE of 5.650 and an RMSE of 7.383. It effectively balanced accuracy and efficiency, providing a cross-species feature modeling paradigm for the universal detection framework of cereal inflorescence morphology. Compatible with diverse cereal crops (e.g., wheat, rice, maize) and data from different acquisition platforms (ground cameras, UAVs), it supported automated high-throughput field phenotyping monitoring of cereals, further advancing precision agriculture.

Abstract:Tomato crops are prone to be attacked by various diseases during their growth process. The computational load of disease detection strategies based on deep learning models was usually substantial. To address this issue, a lightweight deep learning model called ResDepSepNet was proposed. This model was constructed based on residual modules to alleviate the gradient vanishing problem that may occur during model training, thereby improving the overall performance of the model. To reduce the model’s computational load, depthwise separable convolutions were introduced, and downsampling was achieved by increasing the stride of the convolutional operations. Additionally, an squeeze-and-excitation (SE) attention module was introduced to enable the model to focus more on feature information crucial for disease identification, thereby enhancing its disease recognition capability. The ResDepSepNet model was tested by using the PlantVillage tomato disease dataset, and the test results were compared with the MobileNetV2 model and the TrioConvTomatoNet model. The test results showed that the overall accuracy of the ResDepSepNet model was 4.8 and 1.1 percentage points higher than that of the MobileNetV2 and TrioConvTomatoNet models, respectively. Moreover, its floating-point operations count was merely 3.5×107, approximately 1/18 and 1/7 of those of the MobilenetV2 and TrioConvTomatoNet models, respectively. The research result can provide a technical reference for disease detection in tomato crops.

Abstract:Plant phenotyping is one of the key bottlenecks restricting the modernization of agriculture and forestry. Traditional phenotyping methods suffer from limitations such as low efficiency and complex operation, making it difficult to achieve large-scale, dynamic monitoring of plant physiological responses under environmental stress. With the rapid development of high-throughput phenotyping technology, multi-source sensor data fusion has become an important means of studying plant health and stress adaptation. However, existing systems are unable to cope with the phenomenon of varying plant height and large phenotypic variations at different growth stages, resulting in poor adaptability of data acquisition equipment and limited operational efficiency, which restricted the accurate capture of dynamic physiological responses. To address this, a gradient nutrient stress experiment (normal, mild, and severe) on slash pine was conducted, designing and constructing a self-propelled high-throughput phenotyping monitoring system. This system integrated multi-source imaging sensors such as visible light and multispectral sensors, and can automatically adjust the spatial position of the sensors according to dynamic changes in plant height, achieving efficient collection of plant phenotypic information from 360 samples. At the algorithmic level, the system introduced a genetic algorithm-recursive feature elimination with cross-validation (GA-RFECV) method to screen sensitive features highly correlated with nutrient stress, and combined this with a machine learning model to construct a classification framework for the nutrient stress response of slash pine. Experimental results showed that the GA-RFECV method improved the model’s monitoring accuracy, with the random forest (RF) model achieving accuracy, precision, recall, and F1 score of 0.694, 0.695, 0.694, and 0.685 on the validation set, respectively. After further hyperparameter optimization, the extreme gradient boosting (XGBoost) model optimized by differential evolution (DE) achieved the best overall performance on the validation set, outperforming other models. Accuracy, precision, recall, and F1 score improved to 0.759, 0.770, 0.759, and 0.756, respectively, validating the effectiveness of the hybrid feature selection and hyperparameter optimization strategy in plant nutrient stress classification. The self-propelled high-throughput phenotypic monitoring system proposed and constructed demonstrated significant advantages in the accurate and efficient tracking of plant nutrient stress, providing reliable technical support and research methods for precision fertilization, stress-resistant variety breeding, and large-scale forest nutrient monitoring.

Abstract:A lightweight tobacco aphid detection algorithm called GEB-YOLO v8n was proposed to address the problems in field image acquisition, such as dynamic changes in ambient light and image blurring. Firstly, GSConv and the efficient channel attention (ECA) mechanism were innovatively introduced into the backbone network, and the rich image feature information and target-oriented ability of tobacco aphids were jointly output. Secondly, the bidirectional feature pyramid network (BiFPN) was introduced into the neck network, and the semantic expression ability and spatial information quality of the model for detecting tobacco aphid feature maps were enhanced. Finally, WIoU was introduced as the bounding-box regression loss function, and the model was enabled to better generalize to new and challenging tobacco aphid detection scenarios by dynamically focusing on complex samples. After the model structure re-parameterization and hyperparameter optimization, a network architecture for field tobacco aphid detection was formed. The results showed that the mean average precision (mAP) and F1 value of the improved model reached 91.8% and 90.4%, respectively, the number of parameters was reduced by 42.8%, the model memory footprint and floating point operations (FLOPs) were reduced to 3.5MB and 4.1×109, respectively, and the average inference time reached 3.6ms. A system for tobacco aphid recognition and counting in small-scale fields was developed based on the GEB-YOLO v8n model. The system had the dual functions of online image detection and video detection, and can intuitively display the detection results of the number of tobacco aphids on the interface, meeting the requirements of real-time detection of tobacco aphids in small-scale fields and mobile-end deployment. The improved lightweight GEB-YOLO v8n model can provide a method reference for the identification and phenotypic analysis of tobacco plant diseases and pests in the field environment.

Abstract:SPAD is a key indicator for evaluating the growth potential and nitrogen nutrition status of millet. To achieve high-precision and wide-coverage dynamic monitoring of the relative chlorophyll content of millet during key growth stages (jointing stage, heading stage, filling stage, and maturity stage), taking the field scale as the research area and integrated multi-source remote sensing data from unmanned aerial vehicles (UAV) and satellites, an SPAD inversion model for optimizing long short-term memory neural networks (LSTM) based on the improved grey wolf optimization algorithm (IGWO) was proposed. During the research process, UAV remote sensing data, satellite remote sensing data and ground point-like SPAD measured values of each key growth period were obtained simultaneously. Through the longitude and latitude coordinates of the ground measured sample points, the point-like SPAD values were spatially matched and associated with the UAV image pixels and satellite image pixels at the corresponding positions to construct an image-measured corresponding data set. The point spread function (PSF) method was adopted for scale upward inference of UAV images. The mean-variance method was combined to preliminarily correct the satellite data. Then the support vector regression (SVR) was used to establish a collaborative correction model for multi-source remote sensing data of UAV and satellite. Based on the corrected high-precision satellite data, the spectral characteristic parameters sensitive to SPAD were screened through Pearson correlation analysis and XG-Boost feature importance ranking. The nonlinear convergence factor was introduced to enhance the hyperparameter optimization ability of the grey wolf optimization algorithm, and finally the IGWO-LSTM SPAD inversion model was constructed. The results showed that compared with other resampling methods, the point spread function method had the least information loss during the scale-up process. The average value and standard deviation of the processed image pixels were 0.103 and 0.056, respectively. The satellite remote sensing data corrected by the histogram matching method effectively retained the original spectral shape, and the spectral angle mapping value (SAM) was as low as 0.062°. The SVR algorithm had the highest model accuracy in the B/G/R/NIR bands during the critical growth period, and the determination coefficients of the four bands were 0.920, 0.961, 0.963 and 0.900, respectively. The coefficient of determination (R2) of the IGWO-LSTM model in the inversion of SPAD during the critical growth period reached 0.985, and the root mean square error (RMSE) was 0.111, which was significantly superior to that of traditional models such as partial least squares regression (PLSR), BP neural network (BPNN), and random forest regression (RF). The research achieved the precise dynamic inversion of SPAD during the key growth period of millet, which was of great significance for the intelligent monitoring of crop growth and the precise application of nitrogen fertilizer.

Abstract:Aiming at the problems of serious clay and high resistance to furrow opening when the seeding coulter operates under the environment of clay soil, a core-share seeding coulter suitable for clay soil was optimally designed. Through theoretical analysis and empirical design, it was determined that the furrow coulter had an entry angle of 45°, a gap angle of 5°, a bevel angle of 30°, a start slip angle of 23°, a termination slip angle of 45°, and a with of 45mm. In order to improve the performance of the furrow opener in viscosity reduction and desludging, basing on the principle of viscosity and drag reduction of non-smooth surfaces of biomimicry, it was determined that the surface of the plough body of the furrow opener was locally textured reshaped by increasing the convex ribs, and the movement of the furrow coulter in the soil groove was simulated by EDEM software. Using EDEM software to simulate the movement of the furrower in the soil groove, taking the surface characteristic geometric parameters as the test factors, the soil adhesion amount and furrowing resistance as the evaluation indexes, and adopting the response surface test to analyze the influence law of each geometric parameter on the evaluation indexes, it was determined that the optimal combination of the surface characteristic geometric parameters of the furrower body was as follows: width of the convex rib was 7.623mm, thickness of the convex rib was 1.344mm, and the spacing between the adjacent convex ribs was 11.782mm, at this time, the soil adhesion amount of the furrower was 159.88g, the furrowing resistance was 60.065N. The soil trench test showed that under the same working conditions, the soil viscosity reduction rate of the textured reshaping furrow opener was 16.33%, compared with the conventional furrow coulter, the working resistance was reduced by 2.91%~4.45%, achieving the expected viscosity reduction and resistance effect.

Abstract:In order to solve the problems of single function, large volume and poor effect of swing plate closing device for rice seedling raising tray in the field, a small integrated device of swing plate closing for rice seedling raising tray with both swing plate and closing function was designed. The device was composed of swing plate closing integrated mechanism, connecting rod plate splitting mechanism, conveyor belt and chain plate collecting mechanism, which could realize the switching of swing plate closing mode and complete swing plate and closing efficiently. The basic structure and working principle of the wobble plate closing mechanism and the connecting rod type plate mechanism were expounded, the working parameters of the key mechanism were determined, and the feasibility of the mechanism was verified by ADAMS. A prototype was designed and built for testing. The closing test showed that the closing success rate of the device was not less than 90% when the closing speed of the device was 4~6s/tray. The orthogonal experiment of three factors and three levels was carried out with the height of the tray, the operation period and the number of initial trays as the experimental factors, and the success rate of the tray as the experimental index. The quadratic regression model was established, and the primary and secondary order of the influence significance was obtained: the height of the tray, the number of initial trays, and the operation period. The optimal parameter combination was the height of the tray was 5cm, the operation period was 5s/tray, and the number of initial trays was 8 trays/group. The success rate of the tray was 93% under these parameters. The whole machine ran stably and met the design requirements, which can realize fast and effective tray swing and closing operations.

Abstract:Horizontal transplanting of sweet potatoes can enhance yield and quality, but both are significantly influenced by the transplanting depth. And mechanized transplanting has extremely high requirements for the shape of sweet potato seedlings. To meet the agronomic requirements of this method, a scraper soil covering device was designed for the horizontal transplanter of sweet potatoes with field seedlings to ensure the transplanting depth. Initially, the structure of the scraper soil covering device was designed, and the theoretical analysis of the movement process of soil particles during its operation was conducted. Factors such as the installation inclination angle α, scraper inclination angle β, scraper linear speed vb, scraper length L, width W, and spacing D were identified as key factors affecting soil loading and particle movement speed and direction. The installation position of the scraper soil covering device, values of α, W, and D, and the acceptable range of vb, β, and L were determined through theoretical analysis. Eventually, a coupling model of the ridge-scraper soil covering device was established by using RecurDyn-EDEM simulation. The Box-Behnken experimental design method was adopted, with vb, β, and L as experimental factors, and the average covering thickness as the evaluation index. The influence of each experimental factor and their interaction on the average covering thickness was analyzed. The prediction model of the regression equation was obtained by using Design-Expert software, and the response surface analysis was carried out. Experimental results determined the optimal parameters for the scraper soil covering device: when the scraper inclination was 100.35°, the scraper linear speed was 1.74m/s, and the scraper length was 150.32mm, the performance was the best, and the average soil covering thickness was 50mm. The field experiment showed that under the optimal parameter combination, the average soil covering thickness was 48mm, the qualified rate of transplanting depth was 96%, and the standard deviation of soil covering thickness was 4.6mm.

Abstract:Aiming to address the issues of high energy consumption and low efficiency in the hydraulic steering systems of traditional self-propelled sprayers, taking the 3WPZ-1800G self-propelled sprayer as the research object, the energy losses in its hydraulic steering system were analyzed, including throttling and overflow losses. Based on this analysis, an energy-saving hydraulic steering system was designed and implemented, which employed a load-sensitive variable pump as the core component combined with an electro-hydraulic proportional valve. This system dynamically matched the output pressure and flow of the steering hydraulic pump with the load demand of the steering cylinder. To verify the effectiveness of the proposed system, simulation models of the original and energy-saving systems were constructed by using AMESim and Matlab/Simulink software, respectively. A fuzzy PID controller was designed for steering synchronization control, and the energy consumption of both systems under different working conditions was compared through simulation. The results demonstrated that the energy-saving system reduced energy output by approximately 90.1%, 71.3%, and 66.7% compared with the original system. Additionally, real vehicle tests were conducted, the experimental results showed that the energy-saving system reduced energy output by about 90.7%, 60.7%, and 68.7% compared with the original system, which aligned well with the simulation results. These findings confirmed the excellent energy-saving performance of the proposed load-sensitive hydraulic steering system.

Abstract:Currently premium teas are primarily harvested manually, as mechanical picking faces challenges such as the reddening of tea stem cut surfaces, which affects quality, and the large size of end effectors, impacting precision in picking. An end effector that simulated the action of human fingers in gripping and lifting tea stems was designed. The design of the picking mechanism and collection mechanism components was constrained by the geometric parameters and biomechanical properties of tea leaves, which were measured. Kinematic simulations using Matlab and Solidworks software were conducted to verify the dimensional parameters of the mechanism components. The key factors influencing the picking success rate—gripper thickness, picking height, and gripper opening angle—were identified, and their parameter ranges were determined. The Box-Behnken response surface analysis method was used to establish a quadratic regression model, with the picking success rate as the response value, to explore the interactive effects of these factors on picking success. The significance of each factor’s impact on the picking success rate was ranked as follows: gripper opening angle, gripper thickness, and picking height. By optimizing these factors with the picking success rate as the objective, the optimization results were obtained as follows: gripper thickness was 6mm, gripper opening angle was 59°, and picking height was 3mm. Field experimental tests using the optimized parameters indicated a picking success rate of 91.67%, with the error between the experimental and predicted values being less than 5%, thereby confirming the reliability of the optimization results, indicating that the designed picking end-effector can meet the requirements for efficient tea picking.

Abstract:Panax notoginseng is a valuable traditional Chinese medicine in Yunnan. The mechanized harvesting of Panax notoginseng has severe digging resistance. Reducing digging resistance has a significant impact on improving the efficiency of mechanized harvesting of Panax notoginseng. To address this issue, a composite bionic shovel was proposed. The factors affecting the digging resistance of composite bionic shovel were determined through theoretical analysis as follows: the length, width and height of the bionic knife, the width, length and bevel angle of the shovel blade. By combining the above factors and the physical parameters obtained from the calibration test, an orthogonal test was designed by using the Hertz-Mindlin (no-slip) model. The combination of composite bionic shovel structural parameters with the smallest digging resistance was obtained. Simulation tests of tracking the particle displacement flow direction were designed to evaluate the resistance reduction capability and excavation effect of the proposed composite bionic shovel. Based on verifying the simulation test through field tests, the best combination of working parameters of the composite bionic shovel was determined by using orthogonal test methods in the field test under the shed environment of the team as follows: the entry angle was 15°, the forward speed was 0.3m/s, and the center distance of the shovel blade was 80mm. The composite bionic shovel exhibited an average digging resistance of 1094.51N. This value was reduced by 25.30%, 19.55%, and 10.76% compared with the following shovels that were effective in Panax notoginseng excavation: the flat shovel, second-order shovel, and combination shovel, respectively. The above results showed that the proposed composite bionic shovel can meet the requirements of reducing the digging resistance and improving the mechanized harvesting efficiency of Panax notoginseng.

Abstract:Aiming at the problem that corn stubble is large and easily entangled with residual film, which makes it difficult for film rolls to be baled and easy to be scattered, a tilting angle kneading residual film baling device suitable for corn stubble crops was designed. Through the design and optimization of the baling support frame and baling space, the friction force on the film rolls in the baling process was improved to make the film rolls easier to be baled. The force analysis model of the baling device was established. By analyzing and optimizing the experimental parameters through field tests and Design-Expert 13, it was determined that the primary and secondary orders affecting the bale formation rate were baling belt friction coefficient, baling space angle and forward speed, and the primary and secondary orders affecting the film roll density were baling space angle, forward speed and baling belt friction coefficient. The optimized theoretical best parameter combinations were forward speed of 6.758km/h, baling belt friction coefficient of 1.378, baling space angle of 55°, under which can be concluded that the bale formation rate reached 100%, the film roll density was 131.835kg/m3. Under the condition of advancing speed of 6.8km/h, the pattern of baling tape was a word pattern (μ=1.38), and the baling space angle was 55°, the actual bale formation rate was 100%, the density of film rolls was 127.31kg/m3, and the film roll density error was 3.55%, which confirmed that the model was in line with the actual situation, meeting the operational requirements, and providing practical basis for the technological innovation of residual film recycling.

Abstract:The drag force model of Euler-Euler two-phase flow numerical method is an important factor affecting the calculation results of the solid concentration distribution in suspended load sediment solid-liquid two-phase flow. The existing interphase drag force model does not consider the influence of the turbulence intensity change of the surrounding fluid caused by the particle dynamic scale on the movement and diffusion of particle, resulting in the calculated error of the solid concentration compared with the experimental value. To this end, the improved drag force modified model PDS-MTE-Wen-Yu model was obtained by modifying the MTE-Wen-Yu model with the correlation expression of the fluid turbulence intensity change ratio and particle dynamic scale for sediment-laden flow. The improved drag force modified model was verified by the solid-liquid two-phase flow numerical simulation in circular pipes. The results showed that for the numerical simulations of solid-liquid two-phase flow with different particle diameters, different flow velocities and different solid concentrations, the solid concentration distribution calculated by the PDS-MTE-Wen-Yu model was more consistent with the experimental values and the calculation accuracy was higher compared with that of the Wen-Yu model and MTE-Wen-Yu model. The calculation accuracy of PDS-MTE-Wen-Yu model and MTE-Wen-Yu model was basically the same in the turbulent core region, while the PDS-MTE-Wen-Yu model owned higher calculation accuracy in the near-wall region. However, the calculation error of the solid concentration distribution obtained by PDS-MTE-Wen-Yu model was gradually increased with the increase of particle diameter because of the increased inertia of large size particle and the decrease influence of turbulence intensity change on large size particle. For the pressure field of solid-liquid two-phase flow, the pressure drop calculated by the MTE-Wen-Yu model and the PDS-MTE-Wen-Yu model was basically the same. Both of them were closer to the experimental values compared with the Wen-Yu model, but there were still some errors with the experimental values. Therefore, the PDS-MTE-Wen-Yu model owned higher calculation accuracy. It was more suitable for the solid concentration calculation in the suspended load sediment solid-liquid two-phase flow with small diameter particle and the pressure field calculation of the solid-liquid two-phase flow with the small flow velocity and the low solid concentration.

Abstract:Aiming to achieve rapid prediction of the unsteady gas-liquid two-phase flow field in liquid ring pumps, an unsteady periodic flow field prediction method was proposed based on deep learning. This method can realize high-precision and fast prediction of the flow field within a certain period in the future after the sample set. A flow field dataset was established using flow field snapshots at each time step obtained from unsteady CFD results of liquid ring pumps. The features of these flow field snapshots were extracted by convolutional neural network (CNN), and a time series neural network prediction model was constructed by combining long short-term memory neural network (LSTM). The prediction results were compared with CFD numerical simulation results. The analysis showed that the CNN-LSTM model can realize high-accuracy prediction of unsteady flow fields in the future. The average relative errors of the prediction results for the phase field, pressure field, and temperature field were 1.37%, 1.28%, and 1.78%, respectively. When LSTM was used to predict the pressure pulsation of the shell and inlet, the average relative errors on the impeller rotation time of one week after the sample set were 1.61%, 0.09%, and 0.20%, respectively. The prediction performance of CNN-LSTM was better than that of the proper orthogonal decomposition (POD) method on the prediction set outside the sample space. Although the prediction accuracy of the extrapolated time series gradually decreased with the increase of time, it maintained good prediction accuracy throughout the entire time history and had a significant advantage in predicting internal flow field results.

Abstract:In order to improve the hydraulic efficiency of the magnetic levitation centrifugal pump, a certain type of magnetic levitation centrifugal pump was selected as research object, and the maximum value of the pump efficiency was taken as the optimization target under the working condition of flow rate of 15L/min and rotational speed of 6000r/min, and the most significant geometrical parameter impact on the efficiency was screened out by using the Plackett-Burman experimental design based on the basic equations of the pump and the inlet side of the blade was selected to be the intersection point of inlet side. Finally, the intersection point of the inlet side of the blade was selected as the optimization variable, which included the diameter of the pitch circle, the angle between the tangent line of the pitch circle and the tangent line of the working surface, the radius of the working surface of the blade, the radius of the backside of the blade, and the angle between the projection line of the axial surface of the front cover plate and the vertical direction. The optimal Latin hypercube design method was used to design 50 groups of test schemes, and the corresponding head and efficiency values were calculated by combining with numerical simulation, RBF neural network was introduced for training to get the approximation model between the optimization variables and optimization objectives, and finally the improved GWO was used for optimization searching. The results showed that after optimization, the head of the magnetic levitation centrifugal pump was increased by 0.06m, the hydraulic efficiency was increased by 0.56 percentage points, and at the same time, the flow-head curve became smoother, which made the operation of the pump more stable;the impeller channel became wider, and the pressure gradient inside the channel became smaller, the vortex shrank in the radial direction, and the vortex in the working surface of the blades almost disappeared;the distribution of the turbulent kinetic energy inside the impeller channel was more reasonable, and at the same time, the low turbulence kinetic energy distribution was more reasonable. The distribution of turbulent kinetic energy in the impeller channel was more reasonable, and at the same time, the area of low turbulent kinetic energy was increased, the flow loss was reduced, thus the work capacity of the blades was improved, as a result, the hydraulic efficiency was improved.

Abstract:Model evaluation using benchmark datasets is an important method to measure the capability of large language models (LLMs) in specific domains, and it is mainly used to assess the knowledge and reasoning abilities of LLMs. Therefore, in order to better assess the capability of LLMs in the agricultural domain, Agri-Eval was proposed as a benchmark for assessing the knowledge and reasoning ability of LLMs in agriculture. The assessment dataset used in Agri-Eval covered seven major disciplines in the agricultural domain: crop science, horticulture, plant protection, animal husbandry, forest science, aquaculture science, and grass science, and contained a total of 2283 questions. Among domestic general-purpose LLMs, DeepSeek-R1 performed best with an accuracy rate of 75.49%. In the realm of international general-purpose LLMs, Gemini-2.0-pro-exp-02-05 standed out as the top performer, achieving an accuracy rate of 74.28%. As an LLMs in agriculture vertical, Shennong V2.0 outperformed all the LLMs in China, and the answer accuracy rate of agricultural knowledge exceeded that of all the existing general-purpose LLMs. The launch of Agri-Eval helped the LLM developers to comprehensively evaluate the model’s capability in the field of agriculture through a variety of tasks and tests to promote the development of the LLMs in the field of agriculture.

Abstract:Agricultural machinery spare parts are the foundation for the repair of agricultural machinery and are essential for timely maintenance of machinery failures and the normal operation of agricultural production. Therefore, accurate forecasting of the demand for agricultural machinery spare parts is crucial. However, the demand for agricultural machinery spare parts is characterized by non-stationarity, non-linearity, multiple zero values, and large fluctuations, making the prediction task challenging. A time-series efficient convolution network (TECNet) was proposed based on convolutional neural networks for predicting the demand of agricultural machinery spare parts. The model firstly extracted the periodicity of the original one-dimensional sequence by using fast Fourier transform, then a two-dimensional time series convolution module for feature extraction was constructed based on the periodicity, and finally the two-dimensional features back was reshaped to one-dimensional features and the predicted values were obtained by linear transformation. The sales data of four different spare part types from an agricultural machinery spare part supplier were used to evaluate and validate the model, and the root mean square scaling error was introduced as a measure to unify the prediction effect among different sequences. The findings from the experiment indicated that the predictive performance of the new model surpassed that of other comparative models. The root mean square errors for the projections of the four distinct spare parts-demand were 0.775, 1.349, 0.822, and 0.205, respectively, demonstrating a high degree of accuracy in predicting. The model was capable of analyzing the time-dependent relationships within time series data, effectively identifying nonlinear patterns. It performed well in predicting the demand for various agricultural machinery spare parts, offering valuable insights for the demand of predicting agricultural machinery spare parts.

Abstract:Aiming to address the heavy reliance on large amounts of annotated data in fruit object detection tasks such as pomelo, a pomelo tree image generative data augmentation method was proposed based on vision-language models. The approach required only 3~5 unlabeled real images to generate a large-scale labeled dataset, which can be used to train object detection models and enhance their performance in zero-shot and few-shot scenarios.The method consisted of the following three main stages. Firstly, real pomelo tree components (including fruits, leaves) were extracted from unlabeled images by using the grounded segment anything model (Grounded SAM). Secondly, stable diffusion was used to create diverse background images based on textual descriptions, increasing the complexity and variability of the training data. Thirdly, a modified fractal tree algorithm was employed to construct structurally diverse pomelo trees, integrating real components with synthetic backgrounds to produce a variety of tree images and corresponding automatic annotations. Experimental results on pomelo object detection by using the YOLO v10 model (Nano version) showed that the proposed method improved mAP50-95 performance by 662.3%, 24.9%, 13.7%, 8.8%, and 1.8% when the number of real training images was 0, 8, 16, 32, and 64, respectively. With 221 real and 512 generated images, the model achieved optimal performance: precision was 76.9%, recall was 62.7%, mAP50 was 70.3%, and mAP50-95 was 38.4%. When transferred to orange detection tasks under the same data conditions, performance gains were 212.9%, 16.5%, 14.0%, 5.2%, and 4.1%. With 1302 real and 512 generated images, the model achieved the best overall performance: precision was 90.3%, recall was 87.8%, mAP50 was 94.0%, and mAP50-95 was 54.0%, demonstrating strong generalization ability. Compared with tree images generated with blank backgrounds, the proposed method consistently outperformed across all training set sizes, whereas the blank-background approach only excelled in the zero-shot setting. Against traditional data augmentation techniques such as mosaic, this method performed better under low-shot conditions in pomelo detection, and although not the best in orange detection for every individual case, it achieved the best overall results under the default configuration of Ultralytics YOLO. In summary, the proposed method effectively mitigated the limitations caused by insufficient labeled data in fruit object detection model training and offered promising practical value and scalability.

Abstract:In response to the existing deficiencies in the dynamic growth model of alfalfa root system and the root system model of slope protection plants in geotechnical engineering ecological protection, the traditional L system was improved. A dynamic growth root system modeling method based on the L system applicable in the field of geotechnical engineering ecological protection was proposed. By controlling the dry density, initial moisture content, temperature, and humidity of Ili loess, alfalfa planting was carried out. The main root length, main root diameter, number of lateral roots, lateral root diameter, lateral root length, and lateral root branching points of alfalfa at different growth times were recorded. Based on the growth parameters of alfalfa and L system and modeling technology, a dynamic growth root system model of alfalfa was constructed. The results indicated that the root growth of alfalfa followed the Logistic equation. The angle between lateral roots and main roots ranged from 15° to 60°. The number of lateral roots, branch spacing, and length of starting branches was increased with growth time. The diameter of the lower segment of the main root remained stable between 0.01mm and 0.04mm, while the diameter of the main root near the soil surface varied significantly with time. The lateral root diameter was small, and the diameter variation was also minor. Using the Logistic equation as the root growth model for alfalfa, a dynamic root growth model for alfalfa suitable for numerical simulation was constructed based on the L system combined with modeling techniques. The model was validated, and the results showed that the overall error of the root model was small. This L system-based dynamic growth root model of alfalfa successfully visualized the dynamic growth of alfalfa roots and can be applied to numerical simulation fields in agricultural production, botany, and geotechnical engineering. It filled the gap in the dynamic growth model of alfalfa roots and provided a reference for the establishment of plant dynamic growth root models in geotechnical engineering ecological protection.

Abstract:In order to solve the problems of low picking success rate and long planning path when the picking robot picks multiple target tomatoes continuously, a multi-objective tomato picking sequence optimization method based on visual guidance was proposed. A spatially heterogeneous binocular stereo vision positioning system was established to obtain the three-dimensional coordinates of multi-objective tomatoes, to judge the maturity and occlusion of tomatoes, establish the space and collection of tomato picking tasks based on visual guidance in non-enclosed space, and transform the continuous picking problem into a three-dimensional traveling salesman problem. A continuous picking sequence optimization method based on improved sparrow algorithm (VG-ISSA) was constructed, the population was initialized by cubic chaos mapping, and the sparrow population with high randomness and ergodic nature was obtained, the position of the explorer was adaptively adjusted by combining the particle swarm optimization strategy, the Levy flight strategy was added to enhance the traversal of the followers, and a visual information introduction strategy was proposed, so that the algorithm could carry out reasonable sequence optimization according to the actual occlusion. The results showed that compared with the genetic algorithm, particle swarm optimization and standard sparrow algorithm, the improved algorithm reduced the response time by 19.8%, 32.9% and 42.4%, and the path length was reduced by 25.8%, 24.0% and 16.24%, respectively. Experiments showed that the proposed method had certain advancement in the process of continuous tomato picking by picking robot.

Abstract:Aiming to address the inefficiency, high cost, and poor safety of traditional agricultural vehicle-based transport, a dynamic modified hippopotamus optimization (DMHO) was proposed for agricultural UAV path planning. The algorithm synthesized the advantages of Lévy flight, growth ratio mechanism, lens opposition-based learning (LOBL) algorithm with adaptive learning rate and stochastic diffusion to comprehensively improve the algorithm’s global search and exploration capabilities. Based on the test results of the algorithm on 23 classical benchmark functions, it was shown that dynamic modified hippopotamus optimization exhibited optimal performance on 21 of these functions and had the best optimization searching effect compared with eight algorithms such as the original hippopotamus optimization algorithm. The three-dimensional terrain of the unmanned aerial vehicle flight environment in the hilly planting area was constructed, the trajectory planning model of the agricultural unmanned aerial vehicle in this environment was built, and the trajectory planning cost function was designed to satisfy the multi-conditional constraints. In the three different complexity tasks, dynamic modified hippopotamus optimization had the lowest average fitness result among all the compared algorithms, and the standard deviation in the test results was decreased by 33.39%, 72.81% and 7.08%, respectively, in comparison with hippopotamus optimization algorithm. The dynamic modified hippopotamus optimization algorithm demonstrated remarkable superiority and stability in experimental evaluations.

Abstract:In complex unstructured farmland environments, accurate extraction of navigation lines is crucial for agricultural machinery and agricultural robots to achieve autonomous operation. Challenging factors such as variable lighting, undulating terrain, and weed interference that are common in agricultural environments, making traditional image processing methods perform poorly in terms of adaptability, accuracy, and real-time performance, it’s difficult to meet the visual navigation needs of smart agriculture. To address these issues, a ResAC-UNet deep learning network model was proposed based on an improved UNet. This model used the ResNet-50 network to replace the encoder structure of the traditional UNet to enhance feature extraction capabilities. The segmentation speed and real-time response capabilities were improved through optimized jump connections. The ASPP module was introduced in the bottleneck part of the network to achieve multi-scale receptive field modeling, while maintaining high-resolution features and capturing richer contextual information. In addition, the model integrate CBAM to enhance the accurate perception of crop row boundaries, effectively prevented the loss of key feature information, and further improved the segmentation quality. Based on the segmentation results, the row anchor method and RANSAC algorithm were used to achieve high-precision navigation line extraction and smoothing. The acquired front view image was converted into a bird’s-eye view to eliminate the perspective effect, and a top view of the crop row with ROI was generated and retained. The experimental results showed that the ResAC-UNet model achieved 99.23%, 95.44%, 85.23% and 94.71% in precision, MPA, MIoU and recall, respectively, which was better than the current mainstream segmentation networks such as Segformer, DDRNet, HRNet and DeepLabV3+. The average inference time of ResAC-UNet was 15.26ms, which met the real-time recognition requirements of intelligent agricultural machinery visual navigation. Three navigation lines can be extracted in the ROI area of the camera. The maximum angle error of the middle navigation line was only 0.96°, and the maximum pixel deviation was 4.3, which realized the reliable extraction of high-quality navigation lines. Compared with other navigation path extraction methods, the proposed method had higher accuracy and stability. The research result can provide an efficient and robust visual perception solution for the autonomous navigation of intelligent agricultural machinery in the field, which had certain practical value.

Abstract:Aiming to solve the steering instability and hysteresis of agricultural robots in the process of movement, a fusion PID control method of particle swarm optimization (PSO) and genetic algorithm (GA) was proposed. The fusion algorithm took advantage of the fast optimization ability of PSO to optimize the population screening link of GA. The Simulink simulation results showed that the convergence of the fitness function of the fusion algorithm was accelerated, the system response adjustment time was reduced, and the overshoot was almost zero. Then the algorithm was applied to the steering test of agricultural robot in various scenes. After modeling the steering system of agricultural robot, the steering test results in the unloaded suspended state showed that the PID control based on fusion algorithm reduced the rise time, response adjustment time and overshoot of the system, and improved the response speed and stability of the system, compared with the artificial trial and error PID control and the PID control based on GA. The actual road steering test results showed that the PID control response rise time based on the fusion algorithm was the shortest, about 4.43s. When the target pulse number was set to 100, the actual mean value in the steady-state regulation stage was about 102.9, which was the closest to the target value among the three control methods, and the overshoot was reduced at the same time. The steering test results under various scene states showed that the PID control based on the proposed fusion algorithm had good anti-interference ability, it can adapt to the changes of environment and load and improve the performance of the control system. It was effective in the steering control of agricultural robot. This method can provide a reference for the precise steering control of other robots.

Abstract:The waterlogging limited the production potential of cultivated soil (waterlogged albic soil ) in the eastern Sanjiang Plain of Heilongjiang Province, which restricted the stable and high yield of this region. Focusing on the hard barrier layer, poor ventilation and permeability, and poor drainage of white clay in the flooded areas of the Sanjiang Plain, which can easily lead to waterlogging disasters issues. Self-developed mouse hole plows (RHPT) and straw deep buried plows (SDBT) were used for tillage and soil improvement operations, with conventional tillage (stubble removal and ridge formation) as the control (CK). Based on artificial simulation of waterlogging, maize during the silk emergence stage was subjected to 7 days of waterlogging stress. Two deep tillage and soil improvement methods were studied to break through the barrier layer, improve soil tillage structure, and resist the nature of waterlogging on maize root growth, dry matter accumulation, yield composition, etc. The experimental results showed that compared with conventional tillage (CK), the two tillage measures reduced soil hardness, improved soil permeability, and had a more reasonable three-phase ratio. Compared with CK, RHPT and SDBT treatments resulted in a maximum decrease of 25.9% and 19.3% in subsoil （20~40cm） hardness, a decrease of 7.9% and 9.2% in soil solid fraction, and an increase of 11.5% and 10.6% in liquid fraction, respectively. The permeability coefficient of the plow layer was increased by 451.1% and 407.1%, and the differences in various indicators were significant (P<0.05). These two cultivation measures can alleviate the growth stress of waterlogging on corn and increase corn yield. Compared with CK, RHPT and SDBT treatments increased maize root vitality by 25.0% and 28.0%, respectively, and increased dry matter mass by 11.1% and 11.8%, respectively;SPAD values were increased by 17.1% and 14.4%, respectively. The pollen vitality was increased by 26.1% and 22.7%, and the grain yield was increased by 11.3% and 12.0%, respectively. The research adopted two tillage and soil improvement measures to increase soil storage capacity, alleviate waterlogging stress on maize roots, improve crop stress resistance, which provided technical support for disaster reduction and yield preservation regulation of spring maize in Northeast China.

Abstract:Aiming to enhance the promotion of straw ball milling pretreatment technology and assess the feasibility of scaling up the ball milling pretreatment process, the energy consumption during straw ball milling pretreatment was predicted at a larger scale by using discrete element simulation, followed by verification of the pretreatment efficacy post-scaling. After increasing the cylinder size of the CJM-SY-B vibrating ball mill in laboratory conditions to 24 times its original volume, a three-dimensional geometric model was constructed. The EDEM software was employed to simulate impact energy dissipation within the ball mill by inputting physical property parameters for corn stalks both at initial stages and after milling, thereby estimating energy consumption for scaled-0up operations. Concurrently, particle size distribution and sugar production concentration from corn straw subjected to ball milling were compared across two different scales, validating the effectiveness of straw ball milling treatment. Results indicated that predicted energy consumption was 1.48kW·h/kg of straw while actual consumption measured at 1.65kW·h/kg with a relative prediction error of 10.3%. The difference in particle size span between laboratory-scale and scaled-up corn stalks was merely 0.8%, with total monosaccharide concentrations resulting from enzymatic hydrolysis recorded as 85.5g/L and 88.5g/L respectively;these findings suggested that enzymatic hydrolysis efficiency for corn straw remained largely unchanged despite scale enlargement in milling processes. This research substantiated the viability of utilizing discrete element methods for scaling up corn stalk processing and offered theoretical insights along with technical support for broader applications in stalk milling.

Abstract:In order to overcome the technical bottlenecks of strong subjectivity, over-reliance on manual experience and sensory evaluation in the process of traditional cigarette formula design and maintenance, an indirect correlation model of “near infrared spectroscopy-chemical composition-sensory indicators” was constructed, and an end-to-end tobacco sensory quality indicators prediction method was proposed based on near infrared spectroscopy and Transformer architecture. Firstly, three spectral preprocessing techniques, Savitzky-Golay convolution smoothing method (SG), first derivative method (D1), and multivariate scattering correction (MSC), were used to effectively eliminate baseline drift and scattering interference;then a Transformer prediction model oriented to spectral data features was designed to achieve accurate prediction of the three-dimensional evaluation system of tobacco sensory quality (style characteristics: freshness, sweet, and burnt;smoke characteristics: concentration and strength;quality characteristics: quality of aroma, volume of aroma, offensive taste, irritating, and pleasant aftertaste). The model was analyzed by using the SHAP method to enhance its interpretability. Results showed that the model’s mean absolute error for each sensory indicators test set was no more than 0.56, demonstrating good usability. For different sensory indicators, the model demonstrated strong capture of distinct spectral feature bands, effectively exploring the synergistic mechanism of spectral features and demonstrating good interpretability. Furthermore, a method for assisting tobacco leaf substitution was designed by combining multidimensional similarity analysis, providing quantitative decision support for tobacco leaf substitution and blend optimization.

Abstract:Detecting the level of adenosine triphosphate (ATP) in plants is of significant importance for evaluating plant growth and metabolism, monitoring plants’responses to environmental stresses, conducting research on plant pathology and physiological processes, as well as guiding agricultural production practices. A near-infrared fluorescent sensor for the sensitive detection of ATP was developed based on CuInS2@ZnS quantum dots (QDs). The CuInS2@ZnS QDs were prepared by using the hot decomposition method. Negatively charged MPA-modified CuInS2@ZnS QDs were able to interact with positively charged carboxymethyl chitosan, resulting in the formation of carboxymethyl chitosan-coated CHIT/CuInS2@ZnS nanocomposites. The aptamer of ATP, possessing a strong negative charge, induced aggregation of the positively charged nanocomposites through electrostatic and hydrogen bonding interactions, leading to fluorescence quenching. The sensor exhibited a linear relationship between the fluorescence intensity (I/I0) of the CHIT/CuInS2@ZnS nanocomposites and the logarithm of ATP concentration within the range of 5pmol/L to 10nmol/L. The detection limit of the sensor was determined to be 1.67 pmol/L. The research successfully established a CuInS2@ZnS quantum dot-based near-infrared fluorescent sensor for sensitively detecting ATP. This sensor has important theoretical significance and practical value for the growth and development of plants and the judgment of their stress responses to the external environment.

Abstract:In order to solve the problem of high-precision positioning of traditional mechanical systems in a wide range of motion, a precision macro-micro drive system was designed, which compensated for the positioning error of the macro-drive system through the micro-drive system to realize sub-micron precision positioning. Based on the principle of flexible hinge lever and the principle of balanced additional force, a micro-amplification mechanism was designed, which can precisely amplify the input displacement according to the design amplification ratio of 1.5 without additional displacement. In the macro-micro drive system, the servo motor and ball screw was combined as the macro-drive system, and the piezoelectric ceramic actuator was used to drive the micro-amplification mechanism as the micro-drive system, which was used to compensate for the positioning error of the macro-drive system to realize the large-stroke and high-precision motion. On the basis of completing the working principle design of macro-micro drive system, the positioning error of the system was analyzed and the error compensation scheme was put forward, and the macro-micro drive system positioning error compensation experiment was completed. The experimental results showed that the average positioning error of the macro-micro drive system was reduced from 14.49μm to 0.34μm after the positioning error compensation within the range of 2mm stroke, and the average positioning error was reduced by 97.65%, which verified the validity and accuracy of the design of the macro-micro drive system and the error compensation scheme.

Abstract:A class of three-degree-of-freedom parallel mechanisms was firstly proposed with alternately used moving platforms, the innovation of which lay in the adoption of alternately used moving platforms structure. This type of mechanism can alternately use two different moving platforms during different stages of the working process to generate two distinct modes of output motion—two translations and one rotation (2T1R) and three translations (3T)—thereby achieving different process operations. It can be regarded as a novel dual-mode output motion mechanism. Furthermore, the topological, kinematic, and dynamic performance of this three-degree-of-freedom parallel mechanism under the two different modes was analyzed. This included topological analysis based on position and orientation characteristics (POC), degree of freedom (DOF), and coupling degree (k);the derivation of symbolic forward and inverse position solutions based on its topological characteristics;workspace analysis based on forward position solutions;singularity analysis based on inverse position solutions;dynamic modeling of the mechanism using the virtual work principle and the ordered single-open-chain method, along with the solution of its driving force curves;and the optimization design of the mechanism’s dimensional parameters with the reachable workspace as the optimization objective. Finally, the application of this mechanism as an actuator in laser cutting processes was discussed, and the conceptual design for two process application scenarios—material handling in the 2T1R mode and cutting in the 3T mode—was elaborated. The research result can provide a theoretical basis for the design, analysis, and potential applications of parallel mechanisms with alternately used moving platforms, while also expanding the concept, design methods, and application scope of multi-mode mechanisms.