Chassis assembly is one of the core components in mobile agricultural power machinery. A universal chassis can carry the functional tools to realize the various activities in agricultural production. Among them, the oil-operated track chassis are widely used in recent years, due to their low cost, mature technology, sufficient power, high range, and easy maintenance. But their low handling accuracy and severe vibration have posed great challenges to the application of automatic navigation. In this study, the concept of clutch-braked track chassis was proposed as a low-cost agricultural oil-operated track chassis platform. Stronger traction was achieved during steering, compared with differential track chassis vehicles. The RTK-GNSS-based automatic navigation system was built with the STM32F303 as the main controller. The working path of agricultural machines generally consisted of a straight line in the actual operation. The process of tracking was divided into the desired path of the track chassis platform online and the straight line tracking, according to the online point (the online point referred to the sampling point with a lateral deviation less than 3.0 cm, heading deviation less than 2.0°, and the closest distance to the starting position). The kinematics model was established using the transmission and motion of the platform vehicle. A virtual search light pathfinding tracking (VSPT) was proposed to represent the pseudo-code of the VSPT. The fuzzy pre-scanning control was enhanced to search for the target points within the searchlight's field of view in real time. The steering of the track vehicle was controlled to expose the target points within the field of view at all times. In addition, the field of view angle was dynamically adjusted for the tracking oscillation caused by the lateral deviation and speed change. The lateral deviation index and a virtual target point judgment were utilized with a positive proportional relationship of speed. The logical process was designed for the data flow diagram. Direct memory access (DMA) data transfer was used with the idle interrupt and cyclic mode to complete the reception of variable-length byte data. The operation efficiency of the single-chip computer was improved to design the logic flow for the data flow diagram. The improved model was verified to calibrate the relevant parameters by simulation tests. The better navigation was achieved at the chassis forward speed of 0.4 m/s, where the parameters lateral deviation index λ, field-of-view gain k₁, and target gain k₂ were taken as 1/4, 0.005 rad∙m and 6.0 s⁻¹, respectively. Subsequently, the calibrated parameters after simulation were used for the field tests. The results showed that: the average online distance was 1.64 m, while the average lateral and heading deviations were 0.44 cm and 1.57°, respectively, for 6 different initial positions under concrete road conditions. The navigation tests showed that the increasing speed led to a decrease in the navigation accuracy. The appropriate correction of parameters was conducted to maintain better navigation. The average lateral and heading deviations were 0.75 cm and 1.05°, respectively, under 3 speeds, where the average number of corrected deviations was 4.7 times. The number of corrected and heading deviations were reduced under the field dirt road conditions, due to the increasing adhesion coefficient of the dirt road and relatively smooth steering. But the navigation effect was similar under the same parameters in both road conditions. Therefore, the VSPT performed better tracking and adaptability for the control of the clutch-braked track platform. This finding can provide an efficient and stable control scheme for the clutch-braked crawler platforms with navigation algorithms

Manual planting and picking cannot fully meet the large-scale industry of green Sichuan pepper in recent years. It is crucial to accurately segment the field scene of green Sichuan pepper, and then extract the navigation path of agricultural machinery. Field management can be advanced in the intelligence of agricultural machinery in green Sichuan pepper fields. In this study, field images were collected from three planting demonstration bases of green Sichuan pepper in Jiangjin District, Chongqing Municipality, in the various planting stages. A total of 400 images were gathered to divide the dataset and test set, according to a 3:1 ratio. The open-source annotation tool Labelme was utilized to annotate the images. A navigation dataset was constructed between the rows of green Sichuan peppers, followed by data enhancement. Given the complex scenes of a green Sichuan pepper field, a lightweight network, Mobile-Unet, was proposed for the semantic segmentation of five scene types: road, trunk, tree, sky, and background. U-Net network was taken as the base semantic segmentation framework and then MobileNetV2 as the feature extraction network. The last three layers of the original MobileNetV2 network were omitted to adapt MobileNetV2 for semantic segmentation. 8-layer 5-times downsampling structure was then aligned with the U-Net architecture. Additionally, the LeakyReLU activation function was employed in the convolutional units to avoid neuron death during training. After segmentation, a navigation line extraction was then introduced to incorporate dual characteristics of roads and tree trunks. Experimental results demonstrate that the dataset and Dice Loss as the loss function effectively enhanced the prediction accuracy of the model. Compared with the two lightweight networks, Fast-Unet and BiseNet, Mobile-Unet has achieved the higher segmentation accuracy on the test set, with a pixel accuracy of 91.15%, mean pixel accuracy of 83.34%, and mean intersection over union of 70.51%. Compared with U-Net, the recognition accuracy was slightly reduced, but the complexity of the model was significantly reduced, with a 92.17% decrease in the memory occupation, and the inference speed of nearly 10 times faster. Additionally, the tests were conducted on 100 test set images for navigation line extraction. A total success rate of 91% was achieved for the extraction. The average deviation of yaw angle was 2.6° and 6.7°, respectively, to extract the navigation line using road contour and tree trunk features. The accuracy requirements were fully met in the field navigation. The finding can offer a valuable reference to explore the visual navigation in green Sichuan pepper fields.

As the cornerstone of agricultural production, soil provides various nutrients and moisture essential for crop growth, directly influencing crop yield and quality. Field soil information management constitutes a pivotal component in the development of modern agriculture. Currently, manual soil sampling coupled with offline parameter testing remains the primary approach in field management, which is plagued by issues such as high labor costs, intensive workload, and inefficiencies in collection and measurement. To mitigate these challenges and enhance the cost-effectiveness and efficiency of field management, a soil automatic sampling and real-timesoil parameter detection device was developed in this study. Furthermore, a prediction method based on the BPNN (Back Propagation Neural Network) is proposed for estimating soil firmness and mass moisture content. Firstly, based on the requirements for automatic soil sample collection and parameter measurement, a direct-pressure, two-stage, and step-by-step soil sampling and inspection mechanism, an unloading mechanism, and an indexing soil sample collection mechanism were designed. These mechanisms were analyzed and verified, resulting in the determination of a 400 mm movement stroke and a maximum soil penetration thrust of 800 N. A dual-layer control system architecture was also established, utilizing the Jetson TX2 embedded computer and STM32F3 series microcontroller, integrated with GNSS positioning. This system enabled functions such as automatic soil sampling, autonomous navigation, information recording and transmission, soil sampling self-protection, and dynamic prediction of soil firmness and mass moisture content. Secondly, a three-layer BPNN neural network prediction model was constructed to establish a regression relationship between easily measurable parameters including volumetric water content, soil sampling current, sampling speed, and sampling depth, with soil firmness and mass water content. The model was trained and tested using 275 experimental samples, resulting in an optimal number of hidden layer nodes determined as 10. The average percentage errors for the predictions of soil firmness and mass moisture content were 7.74% and 1.53%, respectively. Finally, to validate the machine's overall performance, field tests were conducted at 10 sampling points along an inspection path in an orange grove. The evaluation criteria included machine sampling time, soil penetration depth of the temperature and humidity sensor probe, absolute error of soil weight measurement, and relative errors of predicted soil firmness and mass moisture content. The results indicated that the average soil sampling time per operation was 60.5 s, the average soil penetration depth of the sensor probe was 64.7 mm, the average absolute error of soil sample weight measurement is 1.53 g. and the average relative errors of predicted soil firmness and mass moisture content at the 10 sampling points were 6.37% and 5.00%, respectively. These results satisfied the requirements for soil sampling and parameter detection. Additionally, by combining geographical location information, a field distribution map of soil firmness and mass water content was provided. This study provides a reference for intelligent soil collection, real-timeparameter detection, and visual management of field soil information distribution.