White radish is an important economic vegetable in China. It has the advantages of short growth cycle, strong adaptability and high yield. Based on the growth characteristics and planting patterns, a pulling white radish combine harvester is designed. The harvester is mainly composed of cherry blossom pulling device, soil ripping device, clamping device, rhizome separation device and collection devices, which can perform the combined harvesting operation including lifting, scarifying, conveying, cutting and collecting in series. In order to ensure the smooth entry of white radish stems and leaves into the gripper and avoid the stems and leaves from being pulled off, a lifter with conical opening is designed. By adjusting the conical opening angle of the lifter through active and underactuated combination, the adaptability of the prototype is improved under the condition of low planting line straightness and uneven ridge surface. In order to realize stable flexible pulling and gripping conveying, and avoid stem and leaf fracture in the process of pulling and conveying, a flexible gripper and gripper force adjusting device is designed. The gripper belt is composed of rubber in which the antiskid pattern is printed on the working surface. According to the growth environment, soil mechanical properties and operation stress characteristics of white radish, the structural parameters of the ripper are designed. Through finite element analysis and strength check, the performance and reliability of ripper are improved. To avoid the cutting damage caused by the shaking of white radish during the cutting process, and achieve the effect of smooth cutting and neat incision, a double disc cutter is designed. The field performance tests indicate that the working performance is stable, with the damage rate of 2.7%, the missed rate of 1.8% and the efficiency of 0.08~0.26hm2/h. The popularization and application of pulling white radish combine harvester will effectively reduce labor intensity and improve harvest efficiency, which is helpful for improving the mechanization level of white radish in China.

In order to solve the problem of poor uniformity of sowing operation, a wheat mechanized sowing control system based on STM32 single chip microcomputer is designed. The wheeled driving robot could travel with variable speed. According to the real-time driving speed of the wheeled driving robot, the speed of the seeding motor is controlled to realize the variable speed and uniform seeding. The system used multi-stage control of DC motor speed. The real-time speed signal of the wheeled driving robot is used for The first-stage control parameter, which is controlled by PID. The real-time current and speed of the seeder motor are used for secondary control parameters, which is controlled by fuzzy PID. Simulation results show that the control algorithm has short response time, small overshoot and good control effect. The results of sowing experiment have shown that the control accuracy of seeding amount under constant speed condition is 96.8% and that under variable speed condition is 95.1%.

There are some problems in sorting clementine in our country, such as great difference in fruit size, low sorting efficiency, high manpower cost and low automation level. In order to solve the automatic sorting mature clementine, this paper introduces a machine which uses machine vision technology to identify, detect and sort mature clementine. The machine is composed of MCU module, vision recognition module, Stepper Motor Drive Module, actuator module, humancomputer interaction module, LCD module. Firstly, the machine vision module OpenMV is used to get the color image of the fruit, and the image is binarized. Secondly, the threshold editor in the OpenMV IDE software is used to set the LAB value representing the orange color of the mature sugar orange fruit. The SCM can judge whether the color of the sugar orange fruit is orange through calculation and comparison. Finally, the diameter of clementine is calculated by the relationship between the pixel and the diameter. The results show that the device can automatically identify the color, size and sorting of the target fruit. When the sorting rate is set at 120 fruits per minute, the error of the fruit diameter is less than 1.6 mm, the accuracy is 96.19%, the accuracy of classification and sorting is 95.4%, and the detection rate of unqualified fruits is 97.2%.

Chinas tea industry is still a laborintensive industry. Among them, fresh leaf picking consumes a lot of labor, which is the most laborintensive and labortime link in tea production. The mechanized tea harvesting is the only way for my countrys tea development. Starting from the current situation of tea picking in my country, this paper summarizes the tea picking standards and agronomic requirements of machine picking in China. The research and application status, advantages and disadvantages of single tea picking machine, double tea picking machine, riding tea picking machine and tea picking robots are analyzed from the perspective of onetime picking principle and selective picking principle respectively. The problems that restrict the realization of comprehensive mechanized tea picking in my country are analyzed,and the development opinions such as the deep integration of agronomy and agricultural machinery, the combination of basic research and advanced technology, and the improvement of the versatility of tea machinery are proposed in the future related to tea production in China, so as to provide reference for the further research of tea picking equipment in China.

Aiming at the problem that the transplanting effect of the transplanting machine body is poor due to the incline of the transplanting machine body in hilly and mountainous areas, an automatic leveling control system of the transplanting machine body is designed, which is composed of a state measurement unit, a singlechip embedded system and an electric drive system. First, considering the advantages and disadvantages of position error leveling and angle error leveling, a multisensor coupling leveling strategy is proposed. Secondly, the fuzzy PID controller is used to control the electric push rod in the actuator, and the Kalman filter is used to filter out the engine vibration interference from the transplanter chassis. Then, the fuzzy PID controller and Kalman filter designed are simulated and analyzed by MATLAB/Simulink. The simulation results show that under the same initial parameters, fuzzy PID control has better control performance than traditional PID control. The system adjustment time is reduced by 62.86%, the rise time is reduced by 47.64%, the peak time is reduced by 45.54%, and the maximum overshoot is reduced by 12.23%. Kalman filter can effectively suppress interference and jitter signals. Finally, the static and dynamic tests are carried out for the automatic leveling control system of the car body. The test results show that the adjustment time of the automatic leveling system is less than 3.5 s and the maximum leveling error is less than 0.5° when the vehicle body tilt angle is within -8°~8°. When the transplanter is running at a speed of 3.6km/h, the vehicle body tilt angle is controlled within ±3° under the road with large undulation, and the vehicle body tilt angle under the road with small undulation is basically maintained at 0°. The leveling speed is fast and the effect is good, which is of great significance to improve the transplanters transplanting effect.

In view of the difficulties of farmers and grassroots plant protection personnel in identifying crop diseases, the identification model is established by using VGG16 and Resnet50 for 18 crop diseases of apple, corn, grape and tomato as the research object. Through data pretreatment, data enhancement, model parameter optimization and model cross validation, the single crop multidisease identification and multi crop multidisease identification models are constructed. The performance comparison results show that VGG16 has better recognition performance than Resnet50, and the recognition accuracy of VGG16 model is more than 96%. After analyzing the VGG16 recognition model, it is found that the recognition performance of the single crop multidisease identification model has the best recognition performance. Therefore, based on the method of establishing single crop multidisease identification model, combined with smart phone, Web technology and network programming technology, this paper is proposed to develop an intelligent identification system of crop diseases. The system can provide users with accurate identification results, disease knowledge and prevention methods. The Socket network service of the system can be used as an independent module to provide a unified interface for crop disease identification for agricultural robots, intelligent agricultural machinery, unmanned aerial vehicle, agricultural expert systems, and so on. This study can provide technical support for the informatization and intellectualization of agricultural plant protection.

This article reviews the development of major economic crop industries and mechanization in China from 2019 to 2023, including cotton, oil, sugar, vegetables, fruits, tea, and medicine, and proposes new trends and urgent problems in crop mechanization. The results showed that the cotton planting area has been shrinking year by year, the yield per unit has been increasing year by year, and the machine yield has exceeded 77%. Domestic cotton picking machines have achieved new development, and mechanized plant protection and control are the shortcomings. Rapeseed continues to expand its area and increase yield, with a machine planting rate of 45%. Mechanized transplantation has formed supporting solutions in different regions, but machine harvesting losses are still significant. The planting area and yield of peanuts have reached a new high, with a comprehensive mechanization rate of over 68%. Peanut harvesters are in short supply, harvesting and sowing with machines in southern China is a difficult task. The planting area and yield of sugarcane are basically stable, and stepbystep machine harvesting and combined machine harvesting are advancing synchronously. The problem of “one cut harvest with manual” urgently needs to be solved. The planting area and yield of vegetables have steadily increased, with a comprehensive mechanization rate of 42.61%. The concept of integrating harvesting, seed selection, and machine technology has been recognized, and there is an urgent need to research and promote the application of key planting and harvesting equipment for ripening. The area and yield of garden fruits have steadily increased, with a comprehensive mechanization rate of only 27.3%. Modern orchards are constantly experiencing typical mechanization throughout the entire process, and the level of fruit harvest loss reduction and disaster prevention and mitigation equipment urgently needs to be improved. The area and yield of tea gardens continue to increase, with a comprehensive mechanization rate of 34.09%. The level of machine harvesting and transportation, as well as the utilization rate of summer and autumn tea, have significantly improved. Processing is developing towards intelligent complete sets of equipment, and there is a lack of specialized equipment for intercropping and fertilization. The planting area of commonly used Chinese medicinal materials in clinical practice has shown a restorative growth, with a comprehensive mechanization rate of 42.90%. Suitable varieties in key provinces have basically achieved full mechanization, while the mechanization of Chinese medicinal materials in mountainous understory and flower leaf fruit categories is basically blank.

Aiming at the problems of high tillage resistance, high power consumption and large amount of soil movement during the operation of the trencher of the current orchard ditching machine, a rotary trencher was designed and its structural parameters were optimized by orthogonal test. Combining theoretical analysis and empirical design, it was preliminarily determined that the range of cutting edge length of the trencher was 100-110mm, the cutting edge width ranges was 30-50mm and the value range of the angle of penetration was 19°-23°. The order of primary and secondary factors affecting power consumption obtained from orthogonal test were as follows: angle of penetration, width and length of cutting edge, the optimum combination were 21° angle of penetration, 40mm edge width and 105mm edge length. The order of the main and secondary factors affecting the cross section area of soil disturbance were as follows: angle of penetration, length of cutting edge, width of cutting edge, the optimal combination were 21° angle of penetration, 40mm edge width and 110mm edge length. The optimal combination of cutting edge length 105mm, cutting edge width 40mm and sinking angle 21° was obtained by comprehensive weighted scoring method. When these parameters combination were obtained in the field validation test, the power consumption of the trencher was 3.09kW and the cross section area of soil disturbance was 9 531.21mm2, which was better than the orthogonal test results of each group.

The precise identification and positioning of pruning points is the basis for the intelligent pruning of grapevines in winter, the segmentation of the critical structure of the grapevine is an important prerequisite for reasoning about the precise pruning point. Aiming at the problem that the existing cutting method is greatly affected by the background, resulting in the loss of critical structures of the grapevine, and inaccurate identification and positioning of pruning points, a segmentation method of grapevine critical structure based on Mask R-CNN was proposed, the grapevine pruning model and the critical structure data sets were established. Through the comparative experiment of backbone feature extraction network and segmentation performance, the optimal Mask R-CNN model structure was obtained and its fitting and generalization ability and segmentation performance in different natural backgrounds were verified, The results showed that the Mask R-CNN model with ResNet 101+FPN as the backbone feature extraction network proposed had better fitting and generalization ability, compared with the control group model, the accuracy rate was increased by 7.33% and 8.89%, the recall rate was increased by 9.32% and 9.26%, and the average precision was increased by 12.69% and 12.63% respectively, it could overcome various natural planting background factors, the edge of the segmentation target was complete, and the connection relationship between the critical structures of the grapevine was correct.

The reliable operation of agricultural machinery involves the construction of national food security industrial belt,  and it is very important to analyze its research status and development trend to follow up the engineering development process of agricultural machinery and equipment in time. Citespace is used to visually analyze the literature related to the reliability of agricultural machinery in China from 1980 to 2021, and the specific contents of the relevant literature are reviewed in combination with the literature research method. The results show that the reliability research of agricultural machinery in China mainly has four respects, such as reliability test, reliability analysis, reliability design and reliability evaluation. From the perspective of future development trend, the research on reliability in the field of agricultural machinery is mainly integrated with automation and informatization, and faces the diversified direction of research and development, manufacturing and service.

In order to solve the problems of low efficiency, high labor intensity and high breeding cost caused by manual operation in the identification of caged dead chickens in largescale breeding farms at present, a dead chicken identification algorithm based on temperature judgment is designed based on robot technology, infrared thermal imaging technology and image processing technology, taking laminated caged broilers as the research object. In terms of recognition algorithm, firstly, Otsu algorithm is used to segment chickens from the background, then open operation is used to remove some small noise areas, and finally the maximum temperature of the remaining area is extracted to determine whether there are dead chickens in the cage. In the aspect of hardware design, through the function analysis of the robot, the main hardware selection and system development are completed. The dead chicken detection test was carried out on the dead chicken identification robot system. The results showed that the identification rate of dead chicken in the upper coop was 83.0%, the identification rate of dead chicken in the lower coop was 77.0%, and the overall recognition rate was 80.0%.

Stem cutting is a key link in crop harvesting. Sugarcane, cassava and other crops stem thick, hard texture, difficult to cut, cutting device performance is the key factor affecting the quality and efficiency of harvester operation. In this paper, the research status of hard stem crop cutting equipment at home and abroad was introduced. The existing problems and research status of hard stem cutter were summarized from key parameters, cutting resistance and power consumption, cutter wear and other aspects. It is pointed out that there are some problems in the research of hard stem cutting device, such as too simple modeling of stem, insufficient research on cutting tool materials, separation of stem and cutting tool, and lack of structural innovation. At the same time, it is expected to strengthen the basic research, indepth research on cutting theory, carry out bionic cutter research, strengthen the combination of agricultural machinery and agronomy and other related development trends, in order to provide reference for the research of hard stem crop harvesting mechanization in China.

Computer vision is a field that involves enabling machines to “see”. This technology uses cameras and computers instead of human eyes to identify, track, and measure targets for further image processing. With the development of computer vision, this technology has found widespread applications in modern agriculture and has played a crucial role in its advancement. Firstly, the concept, components, and working principles of computer vision are detailed. Secondly, the research progress and applications of computer vision technology in areas such as aquaculture, livestock farming, crop growth monitoring, crop pest surveillance, and fruit and vegetable recognition, positioning, and harvesting are introduced both domestically and internationally. Through analysis, it is found that existing technology can promote the development of modern agricultural automation, realizing advantages of low cost, high efficiency, and high precision. However, future technologies will continue to expand into new application areas in modern agriculture, bringing about more technical challenges that need to be overcome. Finally, the paper systematically summarizes and analyzes the applications and challenges of computer vision technology in modern agriculture, discussing future opportunities and prospects, providing the latest references for researchers.

In order to solve the problems of long picking paths, low efficiency and low picking quality caused by old leaves, stems and other interferences in the picking process of famous tea leaves, an improved deep reinforcement learning method based on target recognition is proposed. After the image target is preprocessed, the HIS color model is used to obtain target objects of different depths, the picking position of the shoot is obtained through the setting of parameter channels, the shape characteristics of the picking object are analyzed, and the speed, angular velocity, and distance error are used as the guiding factors of the reward function to realize the improvement of deep reinforcement learning. The path planning design of the picking process is accomplished by establishing objective functions, objective networks, and empirical recovery to achieve intensive training of the planned paths. Gazebo simulation platform is used to carry out  reinforcement learning training of picking path, simulate obstacles to achieve the optimization of picking path, complete the verification of the planning algorithm, and get with the increase of training times, the improved deep reinforcement learning method is effective for picking path optimization, the localization cutting accuracy is controlled within 0.005m, and the efficiency of path optimization is improved by 3.6%.

In order to solve the problem of timeconsuming, inefficiency and low accuracy of manual sorting of Pleurotus eryngii, this paper proposes an intelligent sorting method and robot intelligent sorting system for Pleurotus eryngii, which is dualmodel parallel method of grading detection and grasp detection based on deep learning. The depth camera is used to collect the images of Pleurotus eryngii, and the robot is used as the sorting actuator. The intelligent sorting control software is designed based on ROS (Robot Operating System), python and C++ language, and the monitoring and management system is designed based on PyQt. The test results show that the sorting system can automatically realize the grading detection of Pleurotus eryngii and the robot sorting and grasping. The sorting detection of single Pleurotus eryngii takes 18ms, and the average accuracy of grading detection is 88.35%, the success rate of robot grasping is 98.33%, and the success rate of intelligent sorting is 88.35%, which confirms the feasibility and effectiveness of the system as a whole. It provides a new solution for the whole realization of intelligent sorting system of other agricultural products.

In order to address the problems of the over-reliance on large amounts of training data and the poor generalization of unseen disease identification in traditional potato disease leaf detection methods, a few-shot potato disease leaf detection model based on hierarchical feature alignment network is proposed. Firstly, a weakly labeled dataset containing various types of potato diseases were collected and annotated. Secondly, the multi-modal bi-modal feature semantic representations of textual and visual semantics in the support branch were established, and multiple candidate boxes were generated using a pre-trained region proposal network. Thirdly, a convolutional neural network was adopted to map the candidate box regions into deep feature space, and feature alignment was performed between textual and visual semantics using an unparameterized metric method. Finally, the similarity was computed between the unseen class disease images in the query branch and the multi-modal visual and textual semantic association set, and the disease category of the unseen new class was quickly provided according to the similarity value. Through testing on self-built potato disease leaf datasets and open source datasets, the proposed models can achieve recognition accuracy of 93.55% and 96.35% on the test sets, respectively, and 95.15% and 94.06% on the cross-domain datasets, which is superior to the current classical object detection models. The proposed method has certain practical application value.

Aiming at the problems of high computational complexity and slow computational speed of YOLOv5 model in rice disease leaf detection, an improved method for YOLOv5 model identification and detection of rice disease leaf based on SPP-x was proposed. Firstly, three MaxPool layers of different sizes (5×5, 9×9, 13×13) in the SPP module of the original BackBone network were replaced with three 5×5 MaxPool layers of the same size, and then the output feature dimension was adjusted by a 1×1 convolutional layer. Then the optimizer in the YOLOv5 network was replaced with Adam. Thus, a new YOLOv5 network structure was constructed. By comparing the convergence rate of SGD and Adam optimizer on the training set, the results showed that the operation time of the improved SPP-x module was only 50% of the original SPP, and the calculation accuracy reached 97%. The two indexes of mAP_0.5 and mAP_0.5:0.95 converged to 0.983 and 0.822, respectively. The experiment found that the detection speed of single image of YOLOv5 model improved SPP-x was 0.34 s, and the effect was good, which could effectively assist rice disease recognition.

A seriesparallel hybrid picking robot for passion fruit with drapetype planting process was developed in this paper. The overall structure was mainly composed of seriesparallel hybrid mechanical arm, end shear actuator, mobile car mechanism, deep camera, control electrical box, intelligent control terminals and mobile power supply. Aiming at the problem of locating the picking point of passion fruit in complex background, according to the connection relationship between passion fruit and fruit stalks, a recognition and location algorithm for passion fruit picking points based on YOLO v5s target detection and region search with depth information was proposed. There was a nonlinear motion coupling phenomenon in the lifting and stretching degrees of freedom of the seriesparallel hybrid picking manipulator. The inverse kinematics control model was obtained by using geometric dimension reduction and numerical iteration. The experimental results showed that the success rate of picking point location was 89.4%, the average time of inverse kinematics solution of dimension reduction iteration was 0.084 s, the success rate of picking was 93.3%, and the average time of single fruit picking was 30 s.

Plastic film mulching cultivation technology is widely used in China, but it is not easy to be collected. The accumulated residual plastic film in farmland soil in China has exceeded 1 million tons. In order to improve the collection efficiency, mechanical collection is imperative. Aiming at the problem that it is difficult to collect the old residual plastic film in the soil tilth, 1MSD-1.1 residual plastic film collector was designed and developed. The machine was designed with a twolevel residual plastic film collection mechanism to collect the surface film and deep residual film respectively. In this paper, the key components and parameters were designed and calculated. According to the design expectation, the forward speed of the collector, depth of the excavation, and the rotation speed of the drum type residual film pickup mechanism were taken as the test factors, and the surface cleaning rate, deep cleaning rate, and film wrapping rate were taken as the target values. The BoxBehnken test with three factors and three levels was conducted. After that, the operating parameters were optimized by DesignExpert. The optimal results was 84.97% surface cleaning rate, 79.06% deep cleaning rate and 10.32% film wrapping rate. It was found that too much excavation depth was not conducive to the collection of residual film. Before operation, the distribution of residual film depth in the operation area should be determined first before the operation begins.