With China’s urbanization, land circulation has emerged as an avenue for efficient and intensive land use. Embracing technology as a primary driver, the unmanned farm technology model represents a bold attempt at sustainable agriculture development. This review focuses on the frontier and development trajectory of key core technologies, and addresses the significant industrial challenges of insufficient accumulation of data on unmanned farms, unknown interaction mechanisms among the environment, plants, and equipment, and lack of multi-parameter integration and regulation strategies for intelligent equipment. Technologies such as autonomous positioning and navigation, online professional sensor, work obstacle information perception, path planning, decision-making, multi-machine collaboration, autonomous operation, and variable operation technology have been implemented in this study. The research shows that, in future, realizing unmanned farms requires special sensors for agricultural intelligent equipment, precise operational decision control systems, and practically intelligent equipment. At meanwhile, addressing such challenges necessitates a focus on core technologies such as insufficient basic big data accumulation of unmanned farms, modeling of environment-plant-equipment interaction mechanism, and intelligent decision control algorithm. This comprehensive improvement in automation, intelligence, and environmental sustainability of agricultural production marks the future development trajectory of unmanned farms.

Aiming to solve the problems of insufficient capacity and high consumption of electric tractors, an optimal control strategy of composite power supply system based on battery plus super capacitor was proposed. By analyzing the topological structure of the electric tractor complex power system and the resistance characteristics of the electric tractor under various operating conditions, a complex power system structure scheme suitable for electric tractor was selected and a test condition suitable for electric tractor was designed. At the same time, the power demand of the electric tractor drive motor under the test condition was determined. According to the charging and discharging characteristics of the main power components of the electric tractor composite power system, the relevant mathematical model was established, and the simulation model of the electric tractor complex power system was built in Simulink. In addition, according to the load demand of the electric tractor under the test condition and the discharge characteristics of the complex power supply system, a fuzzy control strategy of the complex power supply system was formulated. Taking the minimum current at the maximum power discharge of the battery as the optimization goal, the improved genetic algorithm was used to optimize the fuzzy control parameters. The models of the composite power supply system under different conditions such as single power supply and logic threshold control strategy was built for comparative simulation. The results showed that: compared with the single power supply, logic threshold control strategy and fuzzy control strategy, the peak current battery with the improved genetic algorithm can be reduced by 63%, 62.9% and 43.9% respectively, the average current was respectively reduced 24.8%, 3.2% and 8.5%. It is also shown that this strategy is beneficial to reduce the peak current battery and prolong the service life of the battery.

Agricultural machinery autonomous navigation is an effective approach to achieve precision agriculture, and high-precision path tracking control is a critical assurance for the reliability of intelligent agricultural machinery autonomous operations. To improve the adaptability of path tracking algorithms under different path curvatures and initial error conditions, a geometric path tracking combined algorithm based on adaptive switching of two geometric path tracking methods was proposed. The steady-state errors and convergence characteristics of the pure pursuit (PP) and Stanley model in U-shaped path tracking were analyzed, and their optimal geometric path tracking parameters were respectively set at a constant velocity. A path tracking method switching logic was designed based on tracking errors and path curvature as input, and a boustrophedon path was used to test the geometric path tracking combined algorithm. A mobile cart platform was built to verify the effectiveness of the proposed algorithm. The tracking results showed that when the initial error is 2.5 meters, the Stanley model had a faster convergence rate than the PP algorithm. When operating at speeds of 1.0 m/s for straight paths and 0.7 m/s for curved paths, the PP and Stanley model exhibited similar tracking errors for straight segments, but when the curvature changed abruptly, the maximum tracking errors for the PP and Stanley model were 12.0 cm and 11.0 cm respectively. The path tracking combined algorithm effectively reduced the tracking error during curvature changes, with a maximum tracking error of 9.0 cm under the same speed configuration for the reciprocating shuttle path, representing a reduction of 25.0% and 18.2% compared to the PP and Stanley model respectively. The integration of geometric-based path tracking algorithm reduced the computational load and complexity of online optimization for forward distance and gain coefficient, and thus effectively improved the adaptability of path tracking algorithms for agricultural machinery under complex soil condition.

With the continuous development of information and intelligent manufacturing technology, agriculture has entered the era of intelligent and automated production. Unmanned smart farms, as an important direction of modern agricultural development, are facing opportunities and challenges. As one of the key supporting technologies for unmanned farms, unmanned aerial vehicle (UAV) swarm flight control technology provides crucial technical support for agricultural production in fields such as field inspection, livestock management, and irrigation control. This study aimed to analyze the data interaction methods of various components in UAV swarm simulation flight control, establish and optimize a multi-rotor UAV swarm flight simulation environment to meet the application requirements of UAV swarm flight control in unmanned smart farms. In this paper, by analyzing the interaction logic among the ROS system, PX4 flight controller, MAVROS communication module, and Gazebo simulation environment in UAV swarm simulation flight control, we built a multi-rotor UAV swarm flight simulation environment based on the open-source XTDone simulation platform. We also realized model construction and flight control of UAV swarm based on the ROS system, PX4 flight controller, and Gazebo. Furthermore, using a laser radar to collect environmental information, we optimized software algorithms in the ROS distributed framework, achieving simultaneous localization and mapping (SLAM) based on scan matching algorithm and navigation based on optimal path planning algorithm. Theoretical simulation and experimental results demonstrated that the platform had advantages such as open-source, low cost, scalability, and modularity. The constructed simulation environment can achieve UAV swarm flight control, construction of 2D maps in enclosed environments, and autonomous navigation flight. Analysis revealed that the unit flight accuracy of the UAV swarm under this simulation platform was approximately 76%, and the regression model constructed from cumulative flight distance and flight error had an R² of 0.830 9. The research results demonstrated the feasibility of using UAV formation collaborative flight operations to meet the common agricultural operation needs in unmanned smart farms, demonstrated the application effect and advantages of UAV formation in unmanned smart farms, and provided technical ideas for deep optimization of simulation flight environments and expansion of modern agricultural production application scenarios. It also showed certain reference value and reference value in research and practice in related fields.

In order to solve the problems of high difficulty and low efficiency of manual counting in scientific research of cotton aphid population prediction and control, a cotton aphid image recognition algorithm based on YOLO neural network was proposed and developed into software. First, the images of artificially inoculated cotton aphids were taken by mobile phone for 15 consecutive days, 50 clear images were selected and cut into 6 sub-images, and then the training set and test set were obtained by using LabelImg software for manual labeling. Then, 10 models from the series of YOLOv5 and YOLOv8, whose training parameters were set as the same (batch size was 32, iteration was 100 rounds, initial learning rate was 0.01, and periodic learning rate was 0.01), were selected and trained by using the server of the AutoDL platform. Finally, the trained models were tested, and the YOLOv8l model showed the best overall performance, with mAP50 reaching 0.926. In order to provide users with convenient and easy-to-use man-machine software, the front end of the software was developed by using PYQT5, to realize the functions of reading and counting of cotton aphid pictures, visualizing results and exporting results to Excel. The back end of the software adopted an image processing method of "splitting-detection-merging", which ensuring the efficient detection of YOLO model on small targets. After testing, the software had an average precision of 0.945 for counting dead cotton aphids and live cotton aphids, which was comparable to manual counting and had good practical value. This research may provide an intelligent detection tool for researchers related to cotton aphid control and also provided key operation information for precise operation in scenes such as unmanned farms.

Seed distribution information detection is an essential prerequisite for ensuring the performance of seeders, providing primary data for precise and efficient management of crop growth processes such as seed quality detection, depth determination and fertilization, field weeding, and harvesting. Computer vision technology has been widely applied in multiple fields by extracting meaningful information from image and video data and analyzing and explaining physical phenomena. Based on a systematic analysis of existing literature, this article explored the research progress of computer vision technology in seed distribution information detection from the perspectives of technology application scenarios, hardware devices, software processing algorithms, etc. The study found that computer vision technology still faces challenges in accuracy and stability in seed distribution information detection, and seed distribution information detection requires real-time and large-scale data processing capabilities. Traditional algorithms and devices are difficult to meet these needs, and data fusion and analysis capabilities still need to be improved. Accurate interpretation of seed distribution information requires comprehensive utilization of multiple sensor data and effective data processing and analysis to obtain more comprehensive and accurate results. In response to the above issues, we proposed the following suggestions: Firstly, the widespread application of deep learning and machine learning algorithms is crucial. By utilizing large-scale annotated data for model training, the accuracy and robustness of the algorithm can be improved; Secondly, it is necessary to improve the adaptability of the algorithm and optimize it for different soil conditions, seed types, and light changes to ensure that the algorithm can work effectively in various environments; At the same time, developing multi-sensor data fusion technology can provide more comprehensive and accurate seed distribution information by fusing image data with other sensor data, such as laser scanning and thermal infrared images. This study aimed to provide a reference for the future development of computer vision in seed distribution information detection.

With the increase in labor costs, using intelligent modern equipment to replace humans to complete a series of repetitive, boring, and even high-intensity labor is an important direction in current robot research. Sorting robots, as an important branch of agricultural robots, have good development potential. In order to replace manual, this article used the STM32 microcontroller, OpenMV machine vision module, and ESP32 microcontroller to design an intelligent fruit sorting robot so as to identify and sort fruits with different maturities and sizes. The OpenMV machine vision module collected the color and size information of the fruit, and judged whether the fruit was ripe. The STM32 microcontroller controlled the robot's movement and the robotic arm executed the corresponding sorting program to sort the fruit into the designated container. The ESP32 microcontroller uploaded the sorting data to the cloud. The robot used 3D software for structural design, established 3D models, and conducted simulation analysis during the design process. The simulation results indicated that sorting automation, informatization, and efficiency can be achieved to a certain extent. In the research, multiple and two sorting points were debugged on the physical object, and a total of 24 fruits were sorted. The success rate of fruit sorting reached 87.5%. The research results showed that the sorting robot can achieve precise sorting and achieve a certain degree of reliability and practicality, which can replace human beings to complete the work of precise fruit sorting.