Introduction: The accurate extraction of navigation paths is crucial for the automated navigation of agricultural robots. Navigation line extraction in complex environments such as Panax notoginseng shade house can be challenging due to factors including similar colors between the fork rows and soil, and the shadows cast by shade nets.Methods: In this paper, we propose a new method for navigation line extraction based on deep learning and least squares (DL-LS) algorithms. We improve the YOLOv5s algorithm by introducing MobileNetv3 and ECANet. The trained model detects the seven-fork roots in the effective area between rows and uses the root point substitution method to determine the coordinates of the localization base points of the seven-fork root points. The seven-fork column lines on both sides of the plant monopoly are fitted using the least squares method.Results: The experimental results indicate that Im-YOLOv5s achieves higher detection performance than other detection models. Through these improvements, Im-YOLOv5s achieves a mAP (mean Average Precision) of 94.9%. Compared to YOLOv5s, Im-YOLOv5s improves the average accuracy and frame rate by 1.9% and 27.7%, respectively, and the weight size is reduced by 47.9%. The results also reveal the ability of DL-LS to accurately extract seven-fork row lines, with a maximum deviation of the navigation baseline row direction of 1.64°, meeting the requirements of robot navigation line extraction.Discussion: The results shows that compared to existing models, this model is more effective in detecting the seven-fork roots in images, and the computational complexity of the model is smaller. Our proposed method provides a basis for the intelligent mechanization of Panax notoginseng planting.
In order to explore the optimal vibration parameters for the selective harvesting of coffee fruits, a high-velocity dynamic photography monitoring system was developed to analyze the vibration-assisted harvesting process. This system identified the optimal vibration position on coffee branches and facilitated theoretical energy transfer analysis, obtaining a mathematical formula for calculating the total kinetic energy of coffee branches. A single-factor experiment was conducted with the vibration position as the experimental factor and the total kinetic energy of coffee branches as the response variable. The results showed that the total kinetic energy of the branches was the highest at Vibration Position 2 (the position between the third and the fourth Y-shaped bud tips on the branch). Therefore, Vibration Position 2 was determined as the optimal vibration position. Further analysis established a mathematical model linking coffee cherry motion parameters to theoretical detachment force. A factorial experiment was conducted with vibration frequency and amplitude as test factors, using detachment rates of green, semi-ripe, and ripe cherries as indicators. The results showed that at 55 Hz and 10.10 mm amplitude, the detachment rate of ripe cherries was highest (83.33%), while green and semi-ripe cherries detached at 16.67% and 33.33%, respectively. A field validation experiment, with Vibration Position 2, 55 Hz frequency, 10.10 mm amplitude, and 1 s vibration duration, yielded actual detachment rates of 15.86%, 35.17%, and 89.50% for green, semi-ripe, and ripe cherries, respectively. The error margins compared with the theoretical values were all below 10%. These results confirm the feasibility of optimizing vibration harvesting parameters through high-velocity photography dynamic analysis.
This study addresses the challenges of coordinating tracking accuracy and steering control stability in agricultural differential-drive tracked vehicles, particularly their adaptability to varying working conditions. An adaptive preview Model Predictive Control (MPC) path-tracking algorithm is developed using an electronically controlled modified differential-drive tracked vehicle as the test platform. A preview MPC controller is constructed based on kinematic analysis and a vehicle preview tracking error model, with a fuzzy controller optimizing the objective function’s weight matrix. A speed- and path-dependent preview-distance selection criterion is formulated, and an improved Sparrow Search Algorithm (SSA) provides an optional numerical implementation of preview-distance selection. The corresponding fixed-state objective admits a closed-form analytical solution. A tracked vehicle model was built based on a Robot Operating System (ROS), and a path tracking controller was built based on Simulink. Simulations and field tests at 0.5, 1.0, and 1.5 m/s evaluate lateral tracking error and steering angular-velocity statistics against separate fixed-preview baselines. Relative to the 3.0 m baseline, mean absolute lateral error decreases by 30.13–56.00% in simulation and 46.05–72.57% in field tests. Relative to the 1.0 m baseline, lateral-error changes depend on speed, while angular-velocity standard deviation decreases by 22.23–40.67% in simulation and 24.10–46.39% in field tests. These results characterize the tested controller configurations and show a condition-dependent trade-off between lateral error and steering-response variability.