Optimization and Experimental Evaluation of a Deep Learning-Based Target Spraying Device for Weed Control in Soybean Fields

Weed management is a critical component of soybean production. Efficient weed con-trol can improve both yield and crop quality. However, conventional spraying tech-niques exhibit low pesticide utilization and contribute to environmental pollution. To address these issues, this study proposes a deep learning–based precision target spray-ing method. A lightweight YOLOv5-MobileNetv3-SE model was developed by modi-fying the backbone feature extraction network and incorporating an attention mecha-nism. Field images of weeds were collected to construct a dataset, and the detection performance of the model was subsequently evaluated. A grid-based matching spray-ing algorithm was developed to synchronize target detection with spray actuation. The system time delay, including image processing delay, communication and control de-lay, and spray deposition delay, was analyzed and measured, and a time-delay com-pensation strategy was implemented to ensure accurate spraying. Experimental results demonstrated that the improved model achieved an mAP@0.5 of 86.9%, a model size of 7.5 MB, and a frame rate of 38.17 frames per second. Experimental results showed that weed detection accuracy exceeded 92.94%, and spraying accuracy exceeded 85.88% at forward speeds of 1–4 km/h. Compared with conventional continuous spraying, pesticide reduction rates reduced 79.0%, 72.5%, 55.8%, and 48.6% at weed coverage rates of 5%, 10%, 15%, and 20%, respectively. The proposed method provides a practi-cal approach for precise herbicide application, effectively reducing chemical usage and minimizing environmental impact.