Abstract:Diseases and insect pest are the most restricting factors affecting the crop health, the improvement of the crop yield and quality. It is of great significant to strengthen the development of crop disease and insect pest monitoring. Therefore, to undertake the precise prevent and control on the crop disease and insect pest is key for ensuring the food safety, and improve the yield and quality of crop. The traditional disease and insect pest monitoring mainly relies on the manual field investigation, with low efficiency and quality, which can no longer meet the needs of efficient, intelligent and professional modern agriculture. With the development of information technology, the monitoring of crop diseases and insect pest has gradually developed from the traditional manual monitoring to remote sensing monitoring. Crop monitoring platform, monitoring sensor technology, data analysis and processing technology are key technologies for the development of remote sensing monitoring of crop diseases and insect pest. The development of the above technologies determined the development of remote sensing monitoring technology of crop disease and insect pest. The research progress of monitoring platform, monitoring sensor technology, data analysis and processing technology for crop disease and insect pests were summarized. In terms of monitoring platform, the research status of ground machinery platform, aircraft platform, and satellite platform was summarized. In the monitoring sensor technology, the research progress of radar sensor, image sensor, thermal imaging sensor and spectral sensor for crop diseases and pests monitoring was summarized. In data analysis and processing technology, the research achievements of classical statistical algorithms, computer image processing algorithms, machine learning algorithms and deep learning algorithms in crop diseases and insect pests monitoring were expounded. Furthermore, recommendations were proposed for further promoting the development of crop diseases and insect pests monitoring, including building multi-scale integrated application monitoring platform, promoting the development of multi-scale data fusion sensor, and continuous optimizing multidisciplinary theory and algorithm structure research.

Abstract:Aiming at the problems of easy loss of GNSS positioning signals and poor robustness of traditional SLAM algorithms in forest and orchard environments, the problems of easy loss of GNSS positioning signals and poor robustness of traditional SLAM algorithms in forest and orchard environments was addressed. The proposed method was based on the factor graph for multi-source constrained IMU odometry construction, real-time output of high-frequency position information. The IMU odometry factors and pre-integration factors were used to optimize LiDAR odometry and provide a priori constraints on IMU bias. The LiDAR odometry was optimized by the odometry factor and pre-integration factor, which provided a priori constraints on the IMU bias of the position. The local point cloud map was introduced to participate in feature point cloud coarse matching and non-feature point cloud progressive matching to further densify the source point cloud and improve the performance of the LiDAR odometer. The map construction by fusing GPS signals with LiDAR/IMU tightly coupled framework can obtain accurate and high-frequency continuous position information and improve the reuse rate of point cloud maps. The experimental results showed that the positioning accuracy was maintained at around 0.05m and the root mean square error was 0.0162m compared with algorithms such as LIO-SAM. The algorithm presented enabled the robot to achieve higher accuracy, real-time performance and robustness, effectively reducing the cumulative error of the system and ensuring the global consistency of the constructed maps.

Abstract:To address the problems of inadequate environmental representation in single sensor map construction and inability to provide a complete environmental map for autonomous navigation of mobile robots, a more complete and accurate raster map was constructed by complementary fusion of environmental information obtained from LiDAR and depth cameras. Firstly, the traditional ORB-SLAM2 algorithm was enhanced to have the functions of dense point cloud map construction, octree map construction and raster map construction. Secondly, in order to verify the performance of the enhanced ORB-SLAM2 algorithm, it was tested in the fr1_desk1 dataset and real scenes, and the data showed that the absolute position error of the enhanced ORB-SLAM2 algorithm was reduced by 52.2%, and the camera tracking trajectory grew by 14.7%, which made the localization more accurate. Then the D435i type depth camera adopted the enhanced ORB-SLAM2 algorithm and the Gmapping-Slam algorithm adopted by LiDAR, and constructed the global raster map by complementary fusion according to the rules of Bayesian estimation. Finally, an experimental platform was built for validation and compared with the map building effect of the two sensors, depth camera and LiDAR, respectively. The experimental results showed that the fusion algorithm had a stronger ability to recognize the surrounding obstacles, which can obtain more complete environmental information, and the map construction was more clear and precise, which met the needs of mobile robot navigation and path planning.

Abstract:The foundation and premise of implementing precision weeding and intelligent agriculture is the real-time detection and precise identification of weeds in the corn seedling stage. A method for weed detection in straw-covered farmland suitable for the deployment of Jetson TX2 mobile terminal was proposed. This method addressed the issue that the surface environment of conservation tillage mode was complex, weeds were primarily covered by straw residues on the surface, and the detection speed of existing algorithms was not ideal. Building a corn seedling weed identification model by extracting and analyzing the highlevel semantic information from corn seedling weed photos by using deep learning technology. Based on the YOLO v5s model, the network model’s width was decreased to make minor adjustments that balance the model’s detection speed and accuracy. The network model was analyzed by using the TensorRT reasoning acceleration framework, and the integration of the dimensional tensor into the reasoning network allows for the reconstruction and optimization of the network structure while also lowering the computational demand for the model to operate. Each model was trained and tested before migrating and deploying it to the Jetson TX2 mobile platform. The test findings demonstrated that the lightweight enhanced YOLO v5ss, YOLO v5sm, and YOLO v5sl models, which had accuracy rates of 85.7%, 94%, and 95.3%, respectively. The detection speed were sequentially 80f/s, 79.36f/s, 81.97f/s. The YOLO v5sl model’s detection accuracy was 93.6% after Jetson TX2 embedded reasoning acceleration, and its average running time for a single frame image was 35.3ms, which was 77.8% faster than it was before acceleration. It can achieve the detection of corn seedlings while guaranteeing the accuracy of the detection. The real-time detection of weed targets provided technical support for precise weeding operations in fields with limited hardware resources.

Abstract:To obtain the parameters of different soil in the maize field tillage layer, the typical soil of the maize field tillage layer was divided into ordinary soil (PT) that was not in contact with maize stubble and soil (GT) that combined with maize stubble to form a root-soil complex. The discrete element parameters were calibrated by combining physical experiments and discrete element simulation. Based on Hertz-Mindlin (no slip) contact model, a four factor and five level simulation test was conducted with the central composite design of experiments method and the soil accumulation angle as the target value. Based on the Hertz-Mindlin with bonding model, the Plackett-Burman test, steepest climbing test and Box-Behnken test were designed by using Design-Expert software. The significance parameters were optimized with soil hardness as the target value, and the optimal solution combination for PT was obtained as follows: normal stiffness per unit area was 4.37×107N/m3, shear stiffness per unit area was 1.46×107N/m3, critical shear stress was 3.24×105Pa. The optimal solution combinations for GT as follows: normal stiffness per unit area was 5.19×107N/m3, shear stiffness per unit area was 4.25×10.7N/m3, and critical normal stress was 4.52×105Pa. A soil direct shear validation test was conducted based on the parameters calibrated for two types of soil. The results showed that the relative error of maximum shear stress between the simulation and measurement of the two types of soil was less than 10%, indicating that the simulation parameters were reliable. The soil particle modeling method and calibration method proposed as well as the calibrated parameter values were accurate and reliable, and can provide a theoretical basis for the construction of soil models in maize fields.

Abstract:Aiming at the lack of intrinsic parameters problem of Panax notoginseng stem, contact parameters between Panax notoginseng stem and operating equipment, when using the discrete element method for simulation analysis of key working processes, such as Panax notoginseng combined harvesting and stem killing. Panax notoginseng stem was taken as the object, the discrete element Hertz-Mindlin/Hertz-Mindlin with bonding of Panax notoginseng stem was established by software EDEM. The parameters of discrete element were calibrated by stacking angle experiment and virtual simulation test, and the seedling killing device model of Panax notoginseng stem was established. The intrinsic parameters of stem of were determined by mechanical properties tests. The pile angle of Panax notoginseng stem was tested by cylinder lifting method, the cylinder lifting method was used to test the stacking angle of Panax notoginseng stem, the stacking angle of Panax notoginseng stem in the physical experiment was 44.53° by performing contour fitting on the stacking angle image with Origin software. The Placktt-Burman experiment, steepest climb test and the Central-Composite experiment were used to determine the contact parameters of Panax notoginseng stem and the operating equipment, and the reliability of the model was verified by the stacking angle test, and the shear test. The results showed that the optimal values of collision recovery coefficient, static friction coefficient and rolling friction coefficient between Panax notoginseng stem and operation equipment were 0.319, 0.25 and 0.029, respectively. The optimal values of collision recovery coefficient, static friction coefficient, and rolling friction coefficient were 0.4, 0.29 and 0.032, respectively. The normal stiffness Kn of the Hertz-Mindlin with bonding model was 3.26×108N/m3, the tangential stiffness Ks was 2.17×108N/m3, the normal critical stress σ was 2.27MPa, the tangential critical stress γ was 9.65MPa, and the bonding radius Rd was 0.1mm. The relative error in the accumulation angle verification test was 0.29%;the relative error in the shear verification test was 1.52%, and the error was small. The discrete element model of Panax notoginseng stem was basically consistent with the actual situation, and the model of Panax notoginseng stem and the calibration of discrete element simulation parameters were reliable, which can provide a reference for the research of discrete element simulation of Panax notoginseng stem.

Abstract:Aiming at the problems such as high digging resistance, high power consumption and easy soil blockage in the soil layer of residual film recovery equipment, the residual film-soil-straw excavating and conveying device with vibrating chain tooth bar is designed, in which the rotary tillage excavating mechanism reduces the excavation resistance and solves the soil blockage problem, and the vibrating chain tooth conveying mechanism improves the residual film-soil-straw conveying efficiency. The force model of the material particles on the surface of the conveyor chain is established, the relationship between the forward speed and the speed of the conveyor chain is analyzed, and the speed of the shaking wheel and the conveyor chain is calculated. The content and distribution of residual film and straw in soil profile are measured and a virtual simulated soil tank is established to simulate the content and distribution characteristics of residual film and straw in cotton soil. The simulation model of excavating and conveying device is built and set the depth of excavation shovel to 150mm in EDEM. Under different forward speed (0.75m/s, 1m/s, 1.25m/s), rotary blade speed (210r/min, 230r/min, 250r/min), conveyor chain speed（65r/min, 85r/min, 105r/min） combination conditions, simulation of soil backwater effect and particle velocity variation in the process of excavating and conveying residual film-soil-straw. According to the simulation results, it can be seen that the blockage problem is easy to occur under the condition that the forward speed of the excavating and conveying device is high, and the motion speed of the residual film of soil layer, soil and straw particles is less than 5m/s. The results of the field test are basically the same as those of the simulation test. Under the conditions of different combination of factors, the height of the soil blockage measured in the field test ranges from 71mm to 246mm. The field test shows that when the height of soil blockage is less than 90mm, the problem of excavation resistance and stoppage will not occur. The excavating and conveying device and the operation effect analysis method designed in this paper can provide reference for designing a new type device of excavating and conveying residual film-soil-straw.

Abstract:Combine harvester is an important tool of agricultural production. The feed rate of harvester has always been a hot topic in the area of automatic control. By analyzing the operation mode of harvester, the change model of feed rate of harvester was established. A monitoring system for state parameters of harvester was designed and developed. The system includes grain flow sensor, header height sensor, grain moisture content sensor, feed rate sensor and global navigation satellite system. Taking wheat as the experimental object, a field experiment was carried out in North China to verify the monitoring accuracy of feed rate in different yield moisture content plots. The parameters such as feed rate, yield, moisture content and operation speed were also collected simultaneously. The average relative error of feed rate monitoring was 8.55%. Taking the feed rate of harvester as the control target state and the operating speed of harvester as the control quantity, the feed rate of harvester was simulated by using the method of model prediction. Grey wolf optimization algorithm was used to optimize the weight matrix of quadratic programming. The simulation results showed that the MAE of feed rate control was less than 0.1kg/s after the weight matrix optimization, which was decreased by 38.1% on average. The error of feed rate control was inversely proportional to the yield of the harvest area and proportional to the moisture content. The control effect was better in harvest areas with even wheat growth.

Abstract:Rapeseed stalks are nutrient rich vegetable, but the harvesting of rapeseed stalks mainly relies on manual labor, because it is lack of mechanized harvesting equipment for rapeseed stalks，which seriously restricts the development of the rapeseed stalks production. In order to realizing mechanized harvesting for rapeseed stalks，a row-following rapeseed stalks orderly harvester was designed. The machine was mainly composed of narrow track chassis, divider, single disc cutter, flexible conveying device and motors to realize the functions of cutting, clamping, conveying, collecting and laying stalks. The overall structure and working process of harvesting were expounded. The parameters of the rotational speed and installation position of disc cutter were designed based on the clamping cutting theory. According to the kinematics and mechanical analysis of rape stalk in the process of conveying, the speed of conveyor and clearance between the two conveyors were designed. Based on the force balance conditions of the rapeseed stalk during the laying process, the range of the inclination angle of the guide plate was determined. The rapeseed stalks orderly harvester was developed, and field experiments with the cutting damage rate and variation coefficient of laying angle as the evaluation indexes were carried out. The quadratic polynomial regression models between the evaluation indexes and factors were established, and the influence of each factor on the evaluation indexes were analyzed. The optimal model for evaluation indexes was obtained. The results showed that the comprehensive cutting effect and orderly laying effect was the best when the machine forward speed was 0.5m/s, the cutting blade speed was 910r/min, the conveying speed was 0.75m/s, and the inclination angle of the guide plate was 49°. The verification test showed that under the optimized parameter combination conditions, the cutting damage rate and variation coefficient of laying angle results were 4.95% and 9.55%, respectively, which met the requirements of rapeseed stalks orderly harvesting. The research result can provide reference for the design and parameters optimization of rapeseed stalks orderly harvesting equipment.

Abstract:To address the problem of high digging resistance during the excavation of Ning-guo Radix peucedani, a bionic digging shovel was designed with the shark dorsal fin as the research object;according to the soil stress analysis in Moore-Coulomb theory, the soil was more likely to reach the rupture state when the shark dorsal fin structure was selected as the raised structure of the bionic shovel. The shark specimen was scanned by 3D scanner, the shark dorsal fin 3D model was got, the projection structure of the bionic shovel was determined according to the dorsal fin 3D model, and the bionic digging shovel 3D model was created by NX12.0;the shape profile characteristics of the rootstock of Ning-guo Radix peucedani were obtained by 3D scanner, and the discrete element model of the rootstock of Ning-guo Radix peucedani was created, and the discrete element composite model of Ning-guo Radix peucedani rootstock-soil was established by Hertz-Mindlin with JKR;through the discrete element simulation comparison, the average values of particle displacement and excavation resistance in X, Y and Z directions were obtained from the simulation test, the drag reduction mechanism of excavating shovels was analyzed, and the resistance of the bionic shovel was reduced by 14.37% compared with the plane shovel in the excavation process;by conducting soil trench tests and comparing the rhizome excavation effect, it was basically consistent with the better displacement performance of Ning-guo Radix peucedani rhizome in X, Y and Z directions after excavation with the bionic digging shovel derived from simulation tests, and the average values of excavation resistance of the bionic shovel and the plane shovel were 1342.28N and 1622.73N, respectively. The resistance of the bionic shovel was 17.28% lower than that of the flat shovel in the excavation process, which was very close to that of the simulation test and met the requirements of resistance reduction in the excavation process of Ning-guo Radix peucedani.

Abstract:Vibratory harvesting is widely used for olive harvesting. In vibratory harvesting, reasonable vibration parameters will greatly improve the harvest rate of the fruit. To improve the vibration harvesting efficiency of oil olive, the current situation of small operating space in oil olive orchards was addressed. A single eccentric block suspension type olive harvester was designed, where the centrifugal force generated by the rotation of the eccentric block caused the fruit tree to vibrate under pressure to achieve harvesting. The fruit shedding acceleration can be reduced to the ratio of the fruit stalk binding force to the fruit mass, and the 90% quantile of the shedding acceleration was measured to be 1113.35m/s2. The relationship between the fruit tree shape parameters (trunk diameter X1, trunk height X2, main branch angles X3 and X4) and the excitation parameters (excitation frequency f, excitation force F) was investigated through response surface simulation, the vibration frequency range was obtained by modal analysis, from which, the optimal vibration frequency was figured out by sweeping frequency analysis in rigid-flexible coupling simulation, and analysis of simulation and experimental correlation. The analysis showed that the average correlation coefficient between simulation and experiment was 0.73 and the average relative error was 26.5%, which showed that the simulation could better express the experimental results, and the trunk diameter and trunk height had a significant effect on the excitation parameters. Field experiments were carried out on olive (Leccino) by using the harvesting prototype. The measurements of amplitudes at the clamping point showed that amplitudes at the clamping point varied a lot under different frequencies, namely, rising with the increase of frequency. According to the measurements of accelerations at four monitoring points on the trunk and the lateral branches showed that, the resonant frequencies of the three points located on the lateral branches were similar to the results of the response surface. The optimal vibration parameters obtained by the response surface model were used to carry out the harvest test, which showed that the average harvesting percentage reached about 91.22%, and there was no fruit damage. The research result can provide a reference for the design of vibration harvesting equipment.

Abstract:In order to achieve uniformity and stable seed discharge, a uniform seeding control system for ratoon cane reseeding machines was developed and designed to address the problem of pre-cut ratoon cane reseeding, which consisted of a roller rake, a cane pressure plate, a seed box and an electronic control system. By using EDEM to conduct simulation analysis and virtual tests on the seeding process of the seeding mechanism, the impact of the main working parameters of the system on the seeding performance was discussed, and the mechanism was optimized. Through the one-factor test and Box-Behnken response surface orthogonal test, the relationship between the influence of roller harrow rotational speed, cane pressurization pressure, the number of cane seeds in the seed box, and the activity angle of the cane pressurization plate on the uniformity of seed discharging was investigated, the response surface equation of the seed discharging performance of the seed discharging mechanism was established, and the control model and control algorithm for the adjustment of the parameters of the seed discharging system were constructed. Taking the remaining amount of sugarcane seeds in the seed box as the input of the controller, and the controller adjusts the roller rake speed and the pressure of the cane press to realize the uniform seeding of the seeding discharge system. The test results showed that when the parameter self-tuning seeding control system was used for seeding operations, the seeding pass rate was 94.44%, which was an increase of 8.88 percentage points compared with that of the seeding system without parameter self-tuning, and the seeding uniformity calculated based on variance was 0.46pcs2. The research result can provide theoretical technical support to improve the uniformity and stability of ratoon cane replanting.

Abstract:Aiming to find a more accurate method to assess the effect of agricultural machinery compaction on soybean yield, data of soil physical properties and soybean yield in different compaction environments were obtained by carrying out different numbers of compaction walks with different types of tractors. Soybean yield forecast models were developed from mechanical factors, soil factors, and composite factors which affected soybean growth, respectively. To find out the differences of models built by different types of machine learning algorithms, multiple linear regression (MLR), random forest (RF), adaptive boosting (AdaBoost), and artificial neural network (ANN) were used in modeling. In addition, the importance of model features was comprehensively analyzed. The results showed that the relationship between mechanical operation and crop yield was complex, and the models built by integrated learning algorithms (AdaBoost and RF) had a better fit and higher coefficient of determination. Among the machine learning algorithms used, the best performance of the models built was AdaBoost, followed by random forest, artificial neural network and multiple linear regression. The model built using composite factors for soybean yield had the best fit, followed by mechanical and soil factors. The AdaBoost-based composite factor for soybean yield forecast model had the optimal fit with R2 of 0.92, MAE of 1.33% and RMSE of 1.86%. Mechanical factors and soil factors all had an effect on the variation of soybean yield. The number of mechanical compaction, soil penetration resistance in the surface and subsurface layers were the important factors affecting soybean yield. Therefore, the effects from mechanical compaction can be relieved by reducing the number of mechanical operation and loosening soil penetration resistance of the surface and subsurface soils.

Abstract:In response to problems of poor seed cutting quality and low degree of automation in seed potato cutting machine,based on the design idea of integrating seed cutting agronomy and agricultural machinery, an automatic cutting device suitable for seed potatoes was developed. The seed potato sorting and positioning were completed by sorting-positioning conveying mechanism, and the seed potato cutting process was completed by the combination of the clamping-picking and cutting mechanism. The machine was controlled by PLC to realize the cutting action process and realize the automation of the cutting process. Combined with typical seed potato geometric size parameters, the key structure design of the seed potato cutting device was completed, the process of seed potato cutting was theoretically analyzed, and the main factors affecting the seed potato cutting effect and the value range of each factor were clarified. The qualified rate of seed cutting and blind rate of seed cutting were taken as evaluation indexes, and the center distance of round table roll set, chain conveying speed and V-shaped cutter angle were taken as experimental factors. The response surface test of three factors and three levels was conducted. The analysis of variance and interaction of the test results were conducted by Design-Expert 12.0.3 software. The optimal parameter combination conditions of the test were determined by optimizing the module design of the software, and the seed potato slices were verified under the optimal parameter combination conditions. When the center distance of the round table rolls was 101.60mm, the chain conveying speed was 0.019m/s, and the angle of the V-type cutting tool was 49.50°, the qualified rate of seed cutting was 97.56%, the blind rate of seed cutting was 1.27%, and the relative error with the optimized value was less than 5%, which indicated the optimized parameter combination had high reliability and can meet the requirements of potato seed cutting.

Abstract:Cotton is an important economic crop and strategic reserve material in China, timely and accurate acquisition of cotton spatial distribution information is of great significance for cotton yield prediction and agricultural policy development and adjustment. In order to address the problems of the difficult availability of high-resolution remote sensing data and insufficient usability of feature information by traditional machine learning, a CBAM-U-HRNet classification model was established to extract cotton planted area, where U-HRNet and CBAM attention mechanism were combined, and Tumxuk City in the southern Xinjiang was taken as an study area. Firstly, the Sentinel-2 remote sensing data were pre-processed and annotated. Secondly, the attention mechanism CBAM was introduced into U-HRNet to enhance the important features for cotton classification, suppress the relatively unimportant features, and reduce the interference caused by complex background information. Finally, U-Net, HRNet and U-HRNet were selected to compare with CBAM-U-HRNet model to test their performance in the classification of cotton planted area. During this process, two different spatial resolution datasets such as Sentinel-2 (10m) and GF-2 (1m) were used, and the advantages of CBAM-U-HRNet model were evaluated by using the best feature subset. The results showed the CBAM-U-HRNet model that using Sentinel-2 remote sensing data had the best classification accuracy for cotton planted area, with mIoU and mPA reaching 92.78% and 95.32%, respectively. Comparing with the Sentinel-2 dataset, the GF-2 data had higher spatial resolution and achieved higher accuracy by using HRNet, U-Net and U-HRNet networks. For the two datasets with different spatial resolutions, the classification accuracies of cotton planted area using the CBAM-U-HRNet model was comparable to each other. The CBAM-U-HRNet model can reduce the misclassification induced by the difference in spatial resolution of the two datasets. Comparing with the random forest algorithm, the CBAM-U-HRNet model had higher accuracy in the classification of cotton. The research results can provide technical support for the classification of cotton, and the fast and objective extraction of vegetation planted area in arid regions.

Abstract:In order to monitor and evaluate the growth information and environment of winter wheat, a multilayer non-homogeneous hybrid electromagnetic scattering model was constructed for winter wheat overlaying farmland surface by introducing a three-phase mixed-media model to characterize the vegetation layer and a Gaussian random roughness surface to characterize the rough surface of the farmland. Firstly, the validity of the model was analyzed and verified by comparing the backward scattering coefficient prediction results of the proposed multilayer non-homogeneous hybrid model with those of the water cloud model and the Oh model in the nodulation and tasseling stages of winter wheat. Subsequently, the model equivalent dielectric constant was analyzed and the electromagnetic scattering and radiative transfer equations were solved to obtain the effects of the vegetation growth information, the water content of vegetation and the soil roughness on the surface of the covered farmland. The effect of factors such as vegetation growth information, vegetation water content and soil roughness on the surface equivalent dielectric constant and radar backscattering coefficient of the covered farmland was obtained. The results showed that the multilayer non-homogeneous hybrid model proposed was in good agreement with the prediction results of the water cloud model and the Oh model, and in good agreement with the equivalent dielectric constant R2 of the wheat layer obtained by the double dispersion model with the values of 0.9817, 0.9922, 0.9863, 0.9711, respectively. Moreover, the R2 of the model proposed for the prediction results of the water content of the wheat at the stage of pulling out and the stage of spiking were the same as the actual measurement values, and the R2 of the predicted results were the same as the predicted results of the model. The RSME of the prediction results and the actual measured values were 0.88% and 4.65%, respectively, and the model can better simulate the electromagnetic scattering characteristics of the surface of the overlying farmland, which provided a solid theoretical basis for the subsequent UAV microwave inversion of winter wheat growth and soil moisture information.

Abstract:In order to achieve automation of pitaya harvesting, an improved U-Net based method for pitaya image segmentation and pose estimation was proposed. Firstly, a concurrent spatial and channel squeeze and channel exception (SCSE) module was introduced into the skip connection (connection operation between the encoder and decoder feature maps) of the U-Net model. At the same time, the SCSE module was integrated into the residual module double residual block (DRB) to enhance the network’s ability to extract effective features while improving its convergence speed, obtaining a pitaya image segmentation network based on attention residual U-Net. By using this network to segment mask images of fruits and their accompanying branches, image processing techniques and camera imaging models were used to fit the contours, centroids, minimum bounding rectangle boxes, and three-dimensional bounding boxes of fruits and their accompanying branches. Then based on the positional relationship of fruits and their accompanying branches, three-dimensional pose estimation of pitaya was performed. A test set was obtained in pitaya plantations to evaluate the performance of this algorithm. Finally, field picking experiments were conducted in a natural orchard environment. The experimental results showed that the average intersection and union ratio (mIoU) and the mean pixel accuracy (mPA) of image segmentation for pitaya fruit reached 86.69% and 93.89%, respectively. The average error of threedimensional pose estimation was 8.8°. The success rate of pitaya fruit picking robot in orchard environment was 86.7%, and the average picking time was 22.3s. The research results indicated that this method can provide technical support for developing an intelligent pitaya picking robot to achieve automated and precise picking.

Abstract:Aiming at the problems of labor intensity, strong subjectivity and low detection efficiency in the current manual sensory inspection of cotton preparation, a machine vision-based cotton preparation inspection system was designed. The system consisted of cotton pressing mechanism, image acquisition mechanism, detection processor, detection control board and touch screen. Firstly, a low-angle direct lighting system and an image acquisition mechanism were designed, where the LED light source was illuminated at an angle of 45° to the normal of the inspection window, and the industrial camera collected cotton images through the optical glass. Then the system adopted image texture features to express the appearance morphology of cotton, and established a relationship model between image texture features and appearance morphology by measuring the angular second moment of cotton preparation sample standards. In the adaptive filtering and Canny algorithm, it integrated the noise point evaluation and the high and low threshold adaptive methods for image filtering and segmentation identification, and the ginning quality level determination was made according to the Euclidean distance. Finally, cotton samples were selected for system performance test verification. The results showed that the angluar second moment of the ginning quality physical standards P1, P2 and P3 were [0.8932, 1], [0.6891, 0.7761], [0.2136, 0.5873], respectively, and the difference in the texture eigenvalues of the angular second moment between the grades was obvious, which verified the feasibility of the image texture to express the appearance and morphology of cotton. The relative deviation of the inspection of the number of defects index of the system was 0.15, and the separation effect of defects and background was obvious. Compared with the national standard inspection method, the detection accuracy of the preparation visual system reached 94.20%, and the detection deviation was not more than 1 preparation grade, which was in high consistency with the national standard inspection results. The detection time of single cotton sample system was 1.2s, and the detection efficiency was improved by 77.36%. The system can meet the requirements of field use, and provide a technical reference for the instrumental detection of cotton preparation indexes.

Abstract:Intelligent recognition of crop diseases is a hot topic in the intersection of artificial intelligence and agriculture. At present, the crop disease identification system has a single function and lacks a system upgrade mechanism, and the cost of manual upgrade system is large. To solve the above problems, tomato disease was taken as an example, automatic tomato leaf image labeling algorithm was proposed based on OpenCV and an improved YOLO v5 tomato disease recognition model was constructed. Combining the ideas of automatic data set division, automatic model training and evaluation, and automatic creation and update of mobile phone APP were combined, and a tomato disease recognition system that can be automatically upgraded was designed. The expert review and correction mechanism was introduced to improve the reliability of the system identification results. The experimental results showed that the system realized the identification of the healthy leaves of tomato and the nine kinds of disease leaves, it can automatically expand the tomato disease image data set while identifying tomato diseases through the mobile phone APP in practical application, and automatically start the upgrade and optimization process of the system according to the number of data expansion, so as to continuously improve the tomato disease recognition performance of the system. The design of the system can provide a convenient and reliable tool for tomato disease identification in tomato production.

Abstract:Rice blast is one of the three major rice diseases in the world, which poses a serious threat to food security of China. In order to reduce yield loss, it is urgent to establish a rapid and accurate method for monitoring and identifying rice leaf blast. Rice in northeast China is taken as the research object. Based on a plot experiment, hyperspectral images of rice leaves with different degrees of disease after infection by rice blast fungus were obtained through hyperspectral image analyzer, and spectral data was extracted. Firstly, the SG smoothing method was used to preprocess the spectral data, and then principal component analysis (PCA), Pearson correlation coefficient analysis (PCCs), and PLS-VIP method were used to reduce the dimensionality of the spectral data. An SVM classification detection model based on Logistic chaotic mapping PSO optimization (LMPSO-SVM) was proposed. To verify the effectiveness of the proposed method, classification models based on artificial neural network (ANN), support vector machine (SVM) and particle swarm optimization-support vector machine (PSO-SVM) were established by using feature variables extracted by different dimensionality reduction methods, and were compared and analyzed. The simulation results showed that each model had the best detection performance for level 4 samples. For these five levels of diseases, the prediction accuracy of SVM and ANN classification models fluctuated relatively large, and the effect of disease prediction was not ideal. The LMPSO-SVM classification model established under different feature selection had high accuracy for disease prediction at all levels, and the accuracy fluctuated less. The average accuracy of the model based on PCA extraction of feature variables and the whole band as input was very similar, with 96.49% and 96.12%, respectively. However, the number of input variables extracted by PCA was only 5, which greatly simplified the model complexity, reduced the difficulty and time of training. Comprehensive analysis showed that the PCA-LMPSO-SVM model had the best training effect and could be considered as the best disease classification model. The accuracy rates for the five levels of diseases were 94.29%, 96.43%, 93.44%, 9.30% and 100%, respectively. Therefore, the proposed method could further improve the accuracy and reliability of rice blast classification detection, and the results could provide a certain theoretical basis and technical support for the occurrence of rice blast diseases.

Abstract:For the problems of low accuracy of rice disease identification methods and slow convergence of models, a highperformance lightweight rice disease identification model was proposed, referred to as coordinate attention (CA)-MobileNetV3. The training of the model was optimized by fine-tuning the migration learning strategy, and the convergence speed of the model was improved. Firstly, a ten species dataset was created, containing nine rice diseases and healthy rice leaves. Secondly, the CA module was also used to embed spatial coordinate information in the channel attention to improve the feature extraction and generalization ability of the model. In addition, the improved MobileNetV3 network was used as the feature extraction network and the SVM multi-classifier was added to improve the model accuracy. The experimental results showed that on the rice disease dataset constructed, the initial MobileNetV3 recognition accuracy was only 95.78% and the F1 score was 95.36%, and then the recognition accuracy and F1 score were improved to 96.73% and 96.56%, respectively, after adding the CA module, and then the SVM multiclassifier was added, and the recognition accuracy and F1 scores reached 97.12% and 97.04%, respectively, the number of parameters and the time taken were only 2.99×106 and 0.91s, which were significantly better than that of other models. The experimental results showed that the CA-MobileNetV3 rice disease recognition model proposed can effectively recognize rice leaf diseases and achieve a lightweight, high-performance and easy-to-deploy rice disease classification and recognition algorithm.

Abstract:Accurately identifying the severity of strawberry leaf disease is essential for precise disease control. However, methods based on image classification had a rough division of disease severity and fuzzy classification boundary, while methods based on semantic segmentation had high computational costs and long inference time. To address these problems, a real-time strawberry disease diagnosis method was proposed based on interactive bilateral feature fusion network (IBFFNet). The IBFFNet was a lightweight model containing a context path and a spatial path to extract semantic and detail features from the input image, respectively. Furthermore, an attention spatial pyramid pooling module was constructed to extract multiscale semantic features from the context path, and an edge enhancement module was designed to enrich edge detail information in the spatial path. Finally, the multiscale semantic feature and detail information were aggregated for precise leaf and lesion area segmentation. The percentage of lesions in the leaf area was the estimated severity. The method achieved a promising trade-off between accuracy and speed on the strawberry leaf disease diagnosis dataset. On the strawberry leaf disease diagnosis dataset, the mIoU of IBFFNet2_Seg was 77.8% with 40.6f/s on a single NVIDIA GTX1050. In the test set, an R2 value (coefficient of determination) of 0.98 was achieved, which denoted that the IBFFNet2_Seg could accurately predict the severity of the three diseases. This study paved the way for the precise control of strawberry disease.

Abstract:Sweet peppers are prone to malformed fruits during the growth and development process. Machine replace manual identification and removal of deformed sweet peppers, on the one hand, it can improve the quality and yield of sweet peppers;on the other hand, it can solve the current problems of high labor costs and low efficiency. In order to realize the identification of sweet pepper fruits by robots, an improved YOLO v7-tiny target detection model was proposed to distinguish between normal and abnormal growth of sweet pepper fruits. The parameterfree attention module (SimAM) was integrated into the backbone feature extraction network to enhance the feature extraction and feature integration capabilities of the model;the original loss function CIOU was replaced with Focal-EIOU loss, Focal-EIOU can speed up model convergence and reduce loss value;the SiLU activation function was used to replace the Leaky ReLU in the original network to enhance the nonlinear feature extraction ability of the model. The test results showed that the overall recognition precision, recall rate, mAP0.5 and mAP0.5-0.95 of the improved model were 99.1%, 97.8%, 98.9% and 94.5%, compared with that before improvement, it was increased by 5.4 percentage points, 4.7 percentage points, 2.4 percentage points, and 10.7 percentage points, respectively, the model weight size was 10.6MB, and the single image detection time was 4.2ms. Compared with YOLO v7, scaled-YOLO v4, YOLOR-CSP target detection models, the model had the same F1 score as YOLO v7. Compared with scaled-YOLO v4, YOLOR-CSP was increased by 0.7 and 0.2 percentage points, respectively, mAP0.5-0.95 was increased by 0.6 percentage points, 1.2 percentage points and 0.2 percentage points, respectively, and the weight size was only 14.2%, 10.0%, 10.0% of the above model. The model proposed achieved small size and high precision, and it was easy to deploy on the mobile terminal, providing technical support for subsequent mechanized picking and quality grading.

Abstract:The precise identification and height positioning of the bifurcation points of sugarcane tips is one of the key technologies for achieving realtime control of sugarcane harvester cutters, and is also an important way to improve the mechanization level of sugarcane harvesting and reduce sugarcane impurity content. In response to the complex environment of sugarcane fields, significant changes in lighting, and mutual obstruction of sugarcane bifurcation points, the field investigations, on-site testing and analysis of the characteristics of sugarcane growth points, sugarcane bifurcation points, and their interrelationships were firstly conducted, statistical analysis of sugarcane bifurcation points in images was collected, and combined with on-site measurement and statistical analysis of the height of sugarcane bifurcation points, it was found that they all had obvious normal statistical characteristics. Secondly, a sugarcane tip bifurcation point recognition method was proposed based on improved YOLO v5s. In this method, monocular and binocular cameras were used to collect sugarcane image data in Fusui Agricultural Science Base of Guangxi University, and data preprocessing and labeling were carried out to build a data set of sugarcane tip bifurcation points. Then BiFPN feature fusion structure and CA attention mechanism were introduced into the backbone network of YOLO v5s to enhance the interaction and expression ability of different levels of features, and using GSConv convolution, Slim-Neck normal form design, and the Ghost module was introduced into the original model backbone network to replace the original ordinary convolution in Neck, in order to reduce the computational and parameter complexity of the model and improve its operational efficiency. Finally, the effectiveness and superiority of this method were verified through training and testing on on-site collected datasets. The experimental results showed that this method achieved an average accuracy of 92.3%, a recall rate of 89.3%, and a detection time of 19.3ms on the sugarcane tip bifurcation point dataset. Compared with the original YOLO v5s network, the average accuracy was improved by 5 percentage points, the recall rate was improved by 4 percentage points, the parameter quantity was reduced by 43%, the model size was reduced by 5.5MB, and the detection time was reduced by 0.7ms. Finally, based on the obvious normal statistical characteristics of sugarcane bifurcation points, this feature can be combined with binocular vision positioning algorithms to lay a theoretical and technical foundation for conducting research on feature recognition of sugarcane harvester cuttings, height positioning of cuttings, and real-time control.

Abstract:Fish activity intensity is one of the characteristic indicators of fish health and welfare farming. The fine-grained classification of fish activity intensity is beneficial to describe fish health status and assess fish welfare levels. The fine-grained classification of Atlantic salmon activity intensity where a smallscaled underwater video dataset was collected in the industrial recirculating aquaculture system was carried out. Firstly, the features of video frames were extracted through a small convolutional neural network with residual connections. Then the inter-frame features were obtained by performing differential and square operations between adjacent frames. Finally, the inter-frame features were inputted into the classification network IFDNet based on the external attention mechanism to obtain the video category. The experimental results showed that the classification accuracy of the CNN-IFDNet model proposed reached 97.72%, and the F1 score reached 97.42%. With low computational complexity, the three classification of the fish activity intensity video was realized. Compared with the laboratory environment, the algorithm research based on the real farming environment for fish activity intensity was more practical. The research result can provide a reference for elaborately describing the activity intensity of fish school and realizing intelligent monitoring of fish health status, which can help aquaculture workers discover abnormal conditions and investigate factors causing abnormal fish activity intensity, such as water quality environment and diseases.

Abstract:The laying duck behavior pattern is an important indicator for assessing the health and welfare status of ducks in cage farming. An object detection algorithm based on improved YOLO v4 (you only look once) was proposed to identify multiple behavior patterns in laying ducks by machine vision, and the different behavior patterns provided a basis for duck breeding management scheme. By replacing the backbone feature extraction network MobileNetV2 and using the depthwise separable convolution, this algorithm can improve the detection accuracy while reducing the number of model parameters and effectively improving the detection speed. The parameter-free attention mechanism SimAM module was introduced in the prediction output part to further improve the model detection accuracy. By using this algorithm to detect the cage-reared laying duck behavior validation set, the mAP value of the optimized model reached 96.97% and the image processing frame rate was 49.28f/s, which improved the mAP and processing speed by 5.03% and 88.24%, respectively, compared with the original network model. Comparing the effect with commonly used object detection networks, the improved YOLO v4 network improved the mAP values by 12.07%, 30.6% and 2.43% compared with Faster R-CNN, YOLO v5 and YOLOX, respectively. The improved YOLO v4 network proposed was experimentally studied. The results showed that this algorithm can accurately record the behaviors of cage-reared ducks at different time periods, helping identify abnormal conditions of ducks according to the different behavior patterns exhibited by ducks, such as some behaviors occurring for abnormal periods of time or during abnormal periods. The research result can provide valuable guidance for duck breeding management and enable technical support for implementing automated and intelligent management of duck houses.

Abstract:Aiming at the problem of low pig target detection accuracy in the complex environment in the current intelligent breeding of group-raised pigs, a lightweight target detection model for group-raised pigs based on improved YOLOX, Ghost-YOLOX-BiFPN was proposed. The Ghost convolution was used to replace the traditional convolution, which greatly reduced the number of model parameters. BiFPN was used as the model feature fusion network to effectively fuse the feature maps of pigs of different sizes, and Focal Loss function was added in the post-processing stage, increasing the learning of the model to the positive sample target, and reducing the rate of missed detection. The results showed that the improved model had a detection accuracy of 95.80% for pigs, and the number of model parameters were 2.001×107. Compared with the original YOLOX algorithm, the detection accuracy and recall were increased by 2.84 percentage points and 3.22 percentage points, respectively, and the number of model parameters were reduced by 63%. Finally, the proposed algorithm model was deployed to the Nvidia Jetson Nano mobile terminal development board. The actual operation on the development board showed that the model proposed can guarantee the recognition rate of pigs and realize the accurate recognition of pigs of different sizes and breeds. The research result can provide support for the subsequent establishment of intelligent pig breeding system.

Abstract:Aiming at the problems of overlapping triples and diverse entity expressions in maize breeding text data, a joint bidirectional encoder representations from transformers-conditional random field (BERT-CRF) maize breeding entity relation extraction method with embedded lexical information was proposed. Firstly, the expression characteristics of maize breeding corpus were analyzed, and a synchronous labeling strategy for entity boundary, relation type, and entity position information was adopted. Secondly, a BERT-CRF model with embedded lexical information was constructed for training and prediction, a selfbuilt dictionary of maize breeding knowledge was designed to enhance the semantic ability of the model by embedding lexical information in BERT, integrating character features and lexical features, and using CRF model to output the globally optimal label sequence, and an entity and relation triple matching algorithm (ERTM) was designed to obtain triples by mapping and matching labels. Finally, in order to verify the effectiveness of the proposed method, experiments were carried out on maize breeding data set. The results showed that the precision, recall and F1 value were 91.84%, 95.84% and 93.80%, respectively, which improved the performance compared with the existing models. This method can extract maize breeding knowledge effectively and provide data basis for constructing maize breeding knowledge graph and other downstream tasks.

Abstract:In the anomaly recognition of animal body temperature, methods such as infrared temperature measurement are prone to system bias, making the results unreliable. Deep learning based anomaly detection algorithms has poor robustness and generalization performance on different temperature measurement devices, and is difficult to apply to non-standardized temperature measurement scenarios with low data volume, strong randomness, and inconsistent standards. Therefore, a method of animal body temperature anomaly recognition for non-normalized data sources was proposed. The abnormal animal body temperature detection could be completed by measuring the similarity between body temperature time series data. An improved dynamic time warping (iDTW) algorithm was proposed to solve the problem that the commonly used similarity measurement algorithms were not effective in sequence matching and sequence distance measurement. The Euclidean distance and the first derivative were integrated in the measurement between data points, which effectively solved the problem of sequence over-alignment. The sequence intersection ratio was used to represent the overall characteristics of the sequence, which improved the effect of sequence distance measurement. Aiming at the problem of anomaly detection of unequal length sequence based on similarity measure, an anomaly detection method based on sliding window and sequence equal division was proposed. The shorter sequence was used as the sliding window to traverse the longer sequence to obtain a set of sequence distance. According to the different stages of training and detection, the maximum or the minimum value was selected as the similarity measurement result to solve the problem of unequal length sequence matching. To solve the problem of excessive distance between normal samples and undetected anomaly caused by the long sequence, the long data sequence was equally divided into multiple sub-sequences, and the sum of the sub-sequence distance would be taken as the final similarity measurement result. Experimental results on the public dataset UCR showed that the iDTW algorithm outperformed Euclidean distance, dynamic time warping, derivative dynamic time warping and weighted dynamic time warping by an average of 6.0, 3.0, 5.2 and 2.5 percentage points on 10 time series datasets, respectively. Compared with the classical anomaly detection algorithms, the F1 score of the anomaly detection method based on sliding window and sequence equal division on three animal body temperature datasets were increased obviously.

Abstract:Moistube continuously fertigates the crop-root zone through the nano pores on the Moistube, as required by the crop water demands. To investigate the Moistube discharge and water movement, a series of Moistube irrigation experiments was conducted under different regulated working pressure heads (WPH) in Moistube. The scenarios of WPH increase (0→1m, 0→2m and 1→2m) and WPH decrease (1→0m, 2→0m and 2→1m) adjustments were designed. The Moistube discharge, wetting front advance and soil water dynamics were studied. The HYDRUS 2D model was used to simulate the Moistube discharge and soil moisture transport under regulated WPH, by assuming the Moistube as a porous medium clay. After the model performance was validated, the soil moisture dynamics of moitube irrigation under scenarios of multiple WPH adjustments were analyzed accordingly. The results showed that regulation of WPH significantly changed the cumulative infiltration volume and infiltration rate with respect to time. The curve of cumulative infiltration with time was manifested as a fold line comprising of two interaction lines. The slope of lines was increased or decreased regularly with WPH adjustment. WPH adjustments gave rise to sudden increases or decreases in infiltration rate, and the stabilized infiltration rate was linearly correlated with the adjusted WPH. With increase of WPH, the moisture content within the wetting front was sharply risen, displaying significantly positive feedback. When the WPH was decreased, the water content around the Moistube was slightly decreased, and then rose gradually as the moisture redistributed. The Moistube was treated as a clay porous medium, and the Moistube discharge and water flow transport was well simulated based on the HYDRUS 2D model. The model performances were rated as ‘good’, with coefficient of determination (R2) greater than 0.90, the Nash-Sutcliffe efficiency (NSE) was not less than 0.70 and RSR approached 0. Multiple WPH regulation scenarios (0→1→2m, 0→2→1m, 1→0→2m, 1→2→0m, 2→0→1m and 2→1→0m) were formulated. The soil-water dynamics around the Moistube under all scenarios were analyzed. It was found that the infiltration rate showed exponentially decreasing and subsequent stabilizing trend with respect to time after the WPH was increased. However, following a decrease in WPH, the infiltration rate showed an exponential increase followed by stabilization. The final cumulative infiltration volume was maximum under the consecutive incremental WPH scenario (for instance, 0→1→2m);and the consecutive decremental WPH scenario (2→1→0m) resulted in a minimum cumulative infiltration, which was reduced by 3.7% compared with the 0→1→2m treatment. It was feasible to regulate the wetting front advance and the moisture condition within the wetting front by adjusting WPH. The moisture condition around the Moistube was more sensitive in response to the regulation of WPH. The results provided scientific and theoretical basis for dynamically regulating the working pressure head of Moistube. The Moistube irrigation technique could also be integrated with intelligent irrigation by automatically adjusting the WPH, in order to maintain an appropriate water environment within the root zone and to conduct precise irrigation.

Abstract:Lycium barbarum (goji) fruit is widely used as a medicinal food in China. Aiming to investigate how different types of salt ions in brackish water affected the yield, appearance quality, and nutritional quality of Lycium barbarum. A field crossover experiment was conducted in the Hetao Irrigation Area, using five representative groundwater salinity types (NaCl, CaCl2, CaSO4, NaHCO3, Na2SO4) at four concentration levels (0.1g/L, 0.5g/L, 2.0g/L, 4.0g/L), along with a total of 21 control treatments. The irrigation amount was 100mm of brackish water in Wulat Front Banner, the main production area of Lycium barbarum. Throughout the growth period, Lycium barbarum was irrigated three times. The results revealed that NaCl had the strongest inhibitory effect on the osmotic regulation of Lycium barbarum among the tested salinity types. NaHCO3 had the most significant impact on the secondary stress. CaCl2 at concentrations below 2.0g/L helped alleviate osmotic stress, while both CaCl2 and CaSO4 reduced secondary stress. The yield and dry mass of 100 grains of Lycium barbarum was decreased with the increase of concentrations of NaCl, Na2SO4, and CaSO4, reaching their peak at 0.1g/L.The yield and dry mass of 100 grains was initially increased and then decreased with the increase of concentrations of NaHCO3 and CaCl2, reaching their peak at 0.5g/L. Total sugars, flavonoids, and total amino acids were increased and then decreased with the increase of concentrations of CaCl2 and CaSO4, reaching their peak at 0.5g/L, 2.0g/L, and 2.0g/L, respectively. Betaine and total amino acids were decreased with the increase of concentrations of NaCl, Na2SO4, and NaHCO3, and significantly decreased after exceeding 0.1g/L. Carotenoid content was increased and then decreased with the increase of concentrations of CaCl2 and NaHCO3, reaching its peak at 2.0g/L and 0.5g/L, respectively, which was significantly higher than that of the control treatment. The comprehensive scores indicated that under the same anion (Cl-, SO2-4) environment, Na+ inhibited the planting benefits of Lycium barbarum, while Ca2+ promoted them. Under the same cation (Na+) environment, the inhibitory strength of different anions on the planting benefits of Lycium barbarum was observed in the order of Cl-,SO2-4,HCO-3. Based on Gaussian regression, it was found that when Na+, Ca2+, Cl-, SO2-4 and HCO-3 were ranged from 18.6mmol/L to 19.2mmol/L, 12.2mmol/L to 13.0mmol/L, 63.0mmol/L to 68.4mmol/L,6.3mmol/L to 14.4mmol/L and 5.5mmol/L to 14.0mmol/L, respectively, the planting benefit compatibility of Lycium barbarum was higher (Ci>0.7). With the improvement of Ci standard, Na+, Cl-, SO2-4, HCO-3 approached the lower limit (18.6mmol/L, 63.0mmol/L, 6.3mmol/L, 5.5mmol/L), and Ca2+ approached the upper limit (13.0mmol/L). The planting benefits remained satisfactory (with an average fit degree greater than 0.46) within the simulated range. The upper concentration thresholds were determined as 34.8mmol/L for Na+, 81.6mmol/L for Cl-, 22.6mmol/L for SO2-4, and 21.4mmol/L for HCO-3. The lower concentration threshold for Ca2+ was determined as 9.8mmol/L. These research findings provided a scientific basis for guiding the cultivation of Lycium barbarum in different water-quality areas of the Hetao Irrigation Area. The research evaluated the effects of different salinity levels of brackish water salt ions on the irrigation of Lycium barbarum and proposed suitable concentration ranges of various sensitive ions for irrigating Lycium barbarum, providing a theoretical basis for the promotion of brackish water irrigation for Lycium barbarum. The main innovation lied in the establishment of the relationship between brackish water salt ion concentration and the comprehensive benefits of Lycium barbarum through cross experiments and numerical simulations. The research findings demonstrated strong applicability in moderately saline-alkaline soils, mainly including loess and loam soils.

Abstract:In order to explore the best water-fertilizer cooperative control system for apple tree planting in semi-arid northern China under the condition of integrated drip irrigation and fertilization, two experimental factors (irrigation and fertilization) were set, among which irrigation was 75%θf~90%θf (W1), 65%θf~80%θf (W2), 55%θf~70%θf(W3) and 45%θf~60%θf(W4) of field water capacity, respectively, and fertilization mass ratio of N, P2O5, K2O to air-dried soil was 0.9g/kg, 0.3g/kg,0.3g/kg (F1), 0.6g/kg, 0.3g/kg, 0.3g/kg (F2), and 0.3g/kg, 0.3g/kg, 0.3g/kg (F3), respectively. The effects of cooperative regulation of drip irrigation and fertilization on the growth physiological indexes, dry matter quality and yield, fruit quality and water and fertilizer use efficiency of apple trees were analyzed. Finally, an apple comprehensive index evaluation model aiming at high efficiency, yield and fruit quality was established based on TOPSIS method. The results showed that under the cooperative control of drip irrigation and fertilization, the maximum values of apple tree growth, basal stem growth and leaf area appeared in F1W2 treatment. The net photosynthetic rate, transpiration rate and chlorophyll content of apple tree were basically increased with the increase of irrigation amount and fertilizer application rate, and the maximum values appeared in F1W1 treatment. There was no significant difference between F1W1 and F1W2 in dry matter quality and yield during the whole growth period, and the maximum value of water use efficiency and water productivity appeared in F1W2 treatment, which was increased by 10.6% and 11.1% compared with that of F1W1 treatment, respectively. Fertilizer partial productivity was basically increased with irrigation increasing and fertilizer decreasing. Increasing irrigation amount was beneficial to increase apple color index, fruit shape index and sugar-acid ratio, and increasing fertilizer amount was beneficial to increase apple fruit quality, vitamin C content and soluble sugar content. The TOPSIS multi-objective comprehensive evaluation model was established by synthesizing 17 indexes of four categories, and the final weight value of each index was taken into the calculation. It was concluded that the closest degree of F1W2 treatment was 0.7653, and the comprehensive index evaluation of apple was the best under this treatment, while that of F3W4 treatment was only 0.2583. In conclusion, F1W2 treatment under the cooperative control of drip irrigation and fertilization was the best water and fertilizer management system for apple trees.

Abstract:In the context of ecological protection and high-quality development strategy in the Yellow River Basin, green water management in the upper reaches of the basin is of great significance for conserving water sources, protecting fragile ecology and alleviating water shortage in the basin. There are many measures for green water management, and their effect needs to be further studied. An SWAT hydrological model was constructed in Huangshui Basin in the upper reaches of the Yellow River Basin. Five green water management measures, including contour tillage, residue mulching, stone line, terraced field and farmland conversion to forest on slops above 15° were simulated, and the changes of water quantity, sediment and water quality were analyzed. Combined with the coordination level of water, sand and quality management, the effects and applicability of different measures were explored. The results showed that farmland conversion to forest on slops above 15° had the best effect on increasing water yield and groundwater recharge, which were 1.77×107m3 and 1.72×107m3, respectively. Contour tillage and stone line can effectively regulate annual runoff distribution and reduce runoff in flood season. All the five measures can reduce the sediment yield load, among which terrace and stone line had the most significant effect, and the reduction rate was 13.5% and 13.0% respectively. All the five measures can reduce the total nitrogen (TN) and total phosphorus (TP) load, and the reduction rate in dry years was higher than that in wet years. The reduction effect of terraced and contour tillage was better, which was 24.6% and 14.7% for TN and 45.3% and 21.9% for TP, respectively. Through the water-sand-quality management collaborative analysis, excellent coupling can be achieved in the other four scenarios except for the residue mulching which was high coupling. The order of coupling coordination degree was terraced field, farmland conversion to forest on slopes above 15°, stone line, contour planting, residue mulching. The results can provide reference for green water management measures and optimal allocation of water resources in the Yellow River Basin.

Abstract:Canopy-air temperature difference can indirectly monitor the variation of crop moisture, and the time lag effect between canopy temperature and atmospheric temperature will affect the monitoring effect. In order to explore the characteristics and influencing factors of the time lag effect between canopy temperature and atmospheric temperature, winter wheat from jointing stage to ripening stage was used as the research object. The infrared temperature sensor was used to continuously monitor the canopy temperature of four different irrigation treatments with irrigation upper limits of 95% (T1), 80% (T2), 65% (T3) and 50% (T4) of field water capacity, and simultaneously obtained meteorological data such as short-wave net radiation (RS), atmospheric temperature (TA) and relative humidity (RH). The time lag between canopy temperature and atmospheric temperature was calculated by dislocation correlation method, and its variation characteristics under different growth stages and different irrigation conditions were analyzed. The correlation analysis method was used to explore the correlation between the change rate and daily mean value of meteorological factors (RS, TA, RH) and time lag. Finally, the common influence of meteorological factors (RS, TA, RH), soil moisture content (SMC) and leaf area index (LAI) on time lag was discussed by path analysis. The results showed that the change of winter wheat canopy temperature was ahead of the atmospheric temperature under different growth stages and different irrigation conditions;under different irrigation treatments, the lag time of T1, T2 and T3 treatments was higher than that of T4 treatment, and the lag time was decreased firstly and then increased at different growth stages. The correlation between the change rate of shortwave net radiation (RSCR), the change rate of atmospheric temperature (TACR) and the change rate of relative humidity (RHCR) and the time lag was higher than that between the corresponding daily mean and the time lag. At the same time, the correlation between RSCR and lag time was the highest (R=0.718~0.806), followed by TACR (R=0.582~0.661) and RHCR (R=-0.534~-0.570). Path analysis showed that the lag time was mainly affected by RSCR, SMC and LAI, but the main factors affecting the lag time were different under different irrigation conditions. T1, T2 and T3 treatments were mainly affected by RSCR and LAI, while T4 was mainly affected by RSCR and SMC. The research result can provide a theoretical basis for monitoring crop water changes by using canopy temperature difference information.

Abstract:Decision support system for agrotechnology transfer (DSSAT) is increasingly used in agriculture, and the primary task in the localization of DSSAT is to estimate crop cultivar coefficients. Generalized likelihood uncertainty estimation (GLUE) coefficient estimator is a self-contained coefficient estimation tool for DSSAT, but the crop cultivar coefficients estimated by GLUE coefficient estimator are not always effective, and the simulation accuracy of the DSSAT with the estimated coefficents is often not high. Through using the field measured yield data of four cultivars of rice and the comparative analysis method, with the results of running the GLUE coefficient estimator as a reference, treating each particle of particle swarm optimization (PSO) was considered as a group of rice cultivar coefficients, calling DSSAT to simulate rice yield during the operation of the PSO, and modifying the particles according to the yield simulation error and the operation mechanism of PSO, thus verifying the feasibility of PSO to optimize the coefficients of DSSAT rice cultivar coefficients. The results showed that both algorithms can identify the DSSAT rice cultivar coefficients well, but the GLUE coefficient estimator had a higher frequency of estimating invalid coefficients. Compared with the GLUE coefficient estimator, the coefficients identified by the PSO were all efficient, and the accuracy of its optimized parameters for DSSAT simulated rice yield was higher, and the normalized root mean square error (NRMSE) was in the range of 5.98%~8.78%, which was significantly lower than that of the GLUE coefficient estimator, which was ranged from 6.89% to 18.06%, and the simulated rice yield was close to the measured yield.

Abstract:In large-scale pig farms, environmental quality is critical for the health and growth of pigs. To achieve optimal and real-time regulation of the pig building environment, an IoT-based pig building environment control system was developed by using an STM32 microcontroller as the core controller. The system included a PC terminal and an APP remote monitoring platform as well as a touch screen on-site monitoring platform, that can realize real-time control of pig building environment. Meanwhile, an optimal control strategy for pig building environment based on double deep Q-Network (Double DQN) was proposed. It was shown that the average temperature and relative humidity could be controlled at (20.53±1.72)℃ and (74.16±7.84)%. Compared with the control strategy on a single parameter of temperature, the temperature, relative humidity, NH3 concentration, and CO2 concentration in the pig building under the control of Double DQN were closer to the expected value (temperature was 19℃, relative humidity was 75%, NH3 concentration was 10μL/L, and CO2 concentration was 800μL/L). The maximum relative error of indoor temperature and relative humidity under the Double DQN control strategy were 3.7% and 2.5% lower than that under the temperature threshold control strategy, respectively. Furthermore, the average delay of sensor data upload and control instruction delivery were 226ms and 140.4ms, respectively, which achieved the control ability of small monitoring and control delay and high stability. Under the Double DQN control strategy, the total operation time of three fans in one day was 28.01h, and the total power consumption was 11.4kW·h, which could save about 7.39% of the power consumption compared with that of the traditional temperature threshold method. Therefore, the proposed IoT-based control system integrated with deep reinforcement learning strategy was helpful to improve the environmental quality of pig building and improve the level of automation and intelligent control of breeding environment.

Abstract:To improve the data credibility and readability of the hazard analysis and critical control points (HACCP) national standard for aquatic products in terms of quality traceability, the enhanced HACCP credibility and visualization traceability model (EHCVTM) was proposed by using the example of the pasteurized crabmeat HACCP plan in the national standard. By combining semantic modeling and blockchain technology, the model comprehended the pasteurized crabmeat HACCP plan according to the national standard, a HACCP quality and safety data system (HQSDS) was established for the plan, and the knowledge representation was designed. A reasonable data storage structure and smart contracts for the generated enhanced HACCP plan was further devised, achieving a data classification storage mode of “risky data on-chain, risk-free data autonomous” and “high-risk data directly on-chain, low-risk data encrypted on-chain”. Then different visualization displays were implemented by using a graph database to meet various needs. The application prototype based on EHCVTM for HACCP traceability was tested and proven successfully. The results demonstrated that this model ensured the credibility of risk data in traceability, improved data readability during feedback, strengthened the platform’s warning capability, and enabled internal monitoring for enterprises, externally multi-party supervision, public risk disclosure for quality safety, and precise positioning of safety responsibility. Moreover, the blockchain-based system achieved a throughput of 300 transactions per second, which effectively met the traceability system’s business requirements. The research result can provide a perspective for credible and visual traceability of aquatic products quality and safety risk monitoring based on the HACCP national standard.

Abstract:A multi-layer tray-type hot air-assisted microwave drying machine was designed to address the issues of non-uniform airflow and electromagnetic field distribution, low drying efficiency, and poor drying quality during static drying of existing straw-based nutrient seedling-growing bowl tray (referred to as seedling trays). The microwave resonant cavity and airflow distribution chamber of the drying machine were optimized. ANSYS Electronics software was used to simulate different arrangements of the microwave resonant cavity feed ports, based on the uniformity of electromagnetic field intensity and the influence of S-parameters, to determine the arrangement and height of the feed ports. ANSYS Fluent software was used to optimize the height of deflector chamber and height of airflow distribution chamber, as well as diameter of the bottom edge of deflector. The performance and heating uniformity of the machine were experimentally verified by using the seedling tray as the test material. The results showed that the uniformity index of gas velocity at the outlet of the optimized airflow distribution chamber was increased by 21.26% compared with that before optimization. The V-L-L arrangement of the three feed ports in the microwave resonant cavity had the best electromagnetic field intensity uniformity and the smallest S-parameters.The reflectivity was the lowest when the feed port height was 160mm, decreasing by 78.13% compared with height of 70mm. Compared with hot air drying and microwave drying, the drying rate of the seedling tray under hot air-assisted microwave drying was increased by 291.31% and 86.48%, respectively, and the drying uniformity was better. The research result can provide a reference for the structural optimization of hot air-assisted microwave drying machines.

Abstract:Butter serum (BS) is a byproduct of anhydrous butter production and a natural milk byproduct. There are two main sources: one was the water phase obtained by melting and centrifugation, and the other was using 75% dilute cream to convert the phase to obtain butter serums. There were few studies on the components and their functional characteristics of butter serums at home and abroad, which limited the reuse value of butter serums. The component composition and application characteristics of butter serums were analyzed in this experiment. By centrifugation and other methods the cow butter serums was successfully isolated, and the basic and fat bulb membrane source components of butter serums were analyzed by using raw milk and buttermilk as controls. The results showed that the protein and fat contents of butter serums were increased significantly, and the total calcium content was decreased significantly, without lactose, moisture and pH value (P>0.05). Using liquid phase mass spectrometer, butter serums contained 56 kinds of fat globular membrane phospholipids, of which sphingomyelin content reached 0.352mg/mL;the fat globular membrane protein of butter serums accounted for 23.32% of the total protein. The effects of butter serum and buttermilk on emulsion emulsification were compared. The results showed that the emulsification of butter serum was significantly better than that of buttermilk (P<0.05). The emulsifying activity and stability of butter serum reached the maximum at 3% concentration, which were 2.682m2/g and 91.1%, respectively. In conclusion, butter serum contained abundant fat globule membrane source components, and the emulsification property of butter serum was better than that of buttermilk at the same concentration, which had higher application potential.

Abstract:A model for the dynamic process of wheeled tractor driving, which combined a dynamic model containing road slope disturbance and a tracking error model, to address the issue of reduced accuracy of path tracking control algorithms caused by lateral sloping farmland terrain in some regions was proposed. Based on this model, a linear model predictive control method was used to obtain the control law. As the predictive model included the influence of slope, which enabled feedback correction to be compensated in advance, effectively improving the path tracking performance of agricultural machinery on sloping land. Considering the different requirements for error and stability of agricultural machinery in different driving stages, an adaptive model prediction method was proposed, which allowed the Q and R values to vary dynamically to meet different needs. Here, the variation referred to the relative size of the two, rather than the absolute value. Experiments were conducted on the selection of prediction intervals and control intervals. Then, a comparative experiment was conducted on the model predictive control based on a simple kinematic model with or without adaptive for Q and R. Finally, comparative experiments were conducted with the proposed method and the model predictive control method based on simple kinematics on a fixed slope angle transverse slope road surface and a continuously changing slope angle transverse slope road surface, respectively. Experiments showed that adaptive control can significantly improve control effectiveness, the path tracking performance of the method proposed was significantly better than that based on simple kinematic models on lateral sloping roads, and the stability level was also significantly improved in steady-state.

Abstract:In both industrial and agricultural sectors, robots frequently encounter complex scenarios that consist of numerous discontinuous and discrete local paths, forming challenging trajectories. Rational motion planning serves as the primary foundation for robots to achieve their expected operational goals. A multi-objective comprehensive optimization method was proposed based on the non-dominated sorting genetic algorithmⅡ (NSGA-Ⅱ). The algorithm operated on the principle of hierarchical sorting based on the dominance relationship between individuals, and introduced a “crowding distance” index to characterize the diversity between individuals, thereby providing robust support for maintaining population diversity during the genetic process. Simultaneously, a kinematic model of the robot was established, and a path sequence optimization function was constructed to reduce the robot’s unloaded travel distance, motion time, and joint impact. Higher-order spline fitting and interpolation planning were implemented in Cartesian and joint spaces, significantly enhancing the smoothness and geometric characteristics of the spatial trajectory. The main contribution lied in generating a spatial Pareto optimal frontier solution set based on NSGA-Ⅱ, which effectively solved the multi-objective optimization problem under constraints such as short robot motion time, small joint impact, and optimal task path. After optimization, the robot’s travel path length was reduced by 74%, operational efficiency was improved by 33.44%, and joint stability was enhanced by an average of 50.97%. Through simulation and experimentation, the algorithm’s significant effectiveness in improving robot motion efficiency, continuity, and non-mutability was verified.

Abstract:As planar eight-bar linkages can achieve a more complex kinematic trajectory curve and satisfy the needs of practical applications better in the field of agricultural machinery, an automatic generation method for determining the crank existence region of an eight-bar linkage with a six-bar group was hereby proposed. Compared with the direct solution method of position equations, the numerical iteration solution of the type transformation method (NISTTM) was proved to be more simple in solution programme and more effective in solution time: firstly, all possible initial position solutions of the linkage were solved based on NISTTM;then, based on each initial position solution, NISTTM was also adopted for kinematic analysis, effectively avoiding the problem of higher-order nonlinear equations failing to determine the optimal solution for multiple solutions;finally, the loop without branch defects was screened by using the Jacobian matrix method. The crank can exist in the loop when the position solutions of the linkage can be obtained for the whole rotation cycle of the driving link. Based on the above theoretical basis, the crank existence region of an eight-bar linkage with a six-bar group could be automatically generated in the given region by a VC++ program combined with OpenGL. The numerical example justified the simplicity, effectiveness, and practicability of this method, which can help the designer master the distribution of the crank existence region and be guided to complete the analysis task with no difficulty quickly.

Abstract:To address the problem that traditional methods are difficult to reveal the sequence generation mechanism of dynamic error in machine tool multi-axis interpolation and the error time series features in each time dimension are interrelated, a cascaded dynamic error prediction model integrating chaotic representation (CR) and feature attention mechanism (FA) was proposed. Firstly, on the basis of proving that the time-varying evolution of multivariate dynamic error had chaotic characteristics, the phase space was reconstructed to represent the hidden information behind the time series of dynamic error parameters in the phase space. Then the fused feature attention mechanism further reshaped the dynamical state vector space of the original system by dynamically assigning phase point feature weights in the time dimension while reducing the redundancy of information in the high-dimensional evolution phase space. Finally, considering the long-range correlation of chaotic time-varying evolution, the bi-directional long short-term memory (Bi-LSTM) network model was used to approximate the dynamics in the chaotic phase space to achieve the effective prediction of dynamic error chaotic time series information. Compared with the Bi-LSTM model and the single cascade models CR-Bi-LSTM and FA-Bi-LSTM, the root mean square error of this algorithm was reduced by about 35%, 16% and 43%, respectively, as shown by the example of XK-L540 CNC milling machine with real data. The algorithm realized the phase space expression of dynamic error sequence generation mechanism in time dimension, and constantly played the main role of key phase point feature, with high prediction accuracy.