This paper takes a harvester as an example to analyze the components of a precision operation control system for agricultural machines. A target detection model (SSD) is introduced for pedestrian detection in the farmland environment. The SSD-Mobilenet network model is used to improve the detection real-time. Further fusion of Feature Pyramid Network (FPN) realizes multi-scale feature extraction for detecting targets and enhances target detection accuracy. Design a depth sensor-based combination of far and near field stubble multi-information detection scheme to improve the accuracy and stability of agricultural land environment detection. Verify the effectiveness of the method in this paper through model training and simulation experiments, etc. The results show that the threshold value is set to 0.55, and the sensor detection effect is the best. In the model training, the value of the intersection and merger ratio is close to 1, and the loss value is close to 0. The enhanced target detection algorithm has a higher recognition effect than the 2 comparative algorithms in all 6 types of pedestrian states. In the special scene simulation experiments, the average reward value of this paper’s algorithm is finally stabilized in the interval of 0.1 to 0.3, and the path length is stabilized at about 100 steps with less fluctuation, and the precise operation assistance effect is better than the comparison algorithm.
