Precision Weed Management for Straw-Mulched Maize Field: Advanced Weed Detection and Targeted Spraying Based on Enhanced YOLO v5s

Straw mulching in conservation tillage farmland can effectively promote land utilization and conservation. However, in this farming mode, surface straw suppresses weed growth, affecting weed size and position distribution and obscuring the weeds, which hampers effective weed management in the field. Accurate weed identification and localization, along with efficient herbicide application, are crucial for achieving precise, efficient, and intelligent precision agriculture. To address these challenges, this study proposes a weed detection model for a targeted spraying system. Firstly, we collected the dataset of weeds in a straw-covered environment. Secondly, we proposed an improved YOLO v5s network, incorporating a Convolutional Block Attention Module (CBAM), FasterNet feature extraction network, and a loss function to optimize the network structure and training strategy. Thirdly, we designed a targeted spraying system by combining the proposed model with the targeted spraying device. Through model test and spraying experiments, the results demonstrated that while the model exhibited a 0.9% decrease in average detection accuracy for weeds, it achieved an 8.46% increase in detection speed, with model memory and computational load reduced by 50.36% and 53.16%, respectively. In the spraying experiments, the proposed method achieved a weed identification accuracy of 90%, a target localization error within 4%, an effective spraying rate of 96.3%, a missed spraying rate of 13.3%, and an erroneous spraying rate of 3.7%. These results confirm the robustness of the model and the feasibility of the targeted spraying method. This approach also promotes the application of deep learning algorithms in precision weed management within directional spraying systems.