关键词: 玉米苗间杂草检测, YOLOv5, 注意力模块, BoTNet模块, 超绿特征, OTSU阈值分割

Abstract: To address the problems of diverse weed species, high detection complexity, and slow detection speed in the detection of weeds among maize seedlings， this paper proposes a weed detection method based on optimized YOLOv5 algorithm. The SE parameter-free attention module is introduced into the convolutional layers of the YOLOv5 backbone network, with the SE-integrated C3 module replacing the original C3 module to better focus on detection targets. The traditional residual neural network is replaced with the BoTNet module, and global multi-head self-attention is used to replace 3×3 spatial convolution in the last three bottleneck blocks of ResNet, thereby improving the accuracy of detecting small targets. The improved target detection algorithm is used to detect weeds, with non-maize seedling areas in the field labeled as weeds. The super-green feature, combined with the OTSU threshold segmentation algorithm, is used to segment the soil background and identify the foreground areas of weeds, effectively solving the problem of weed detection in maize seedling fields. The results show that the improved YOLOv5 algorithm achieves a target detection precision of 97.5% for maize seedlings, which is 7.4% higher than the original YOLOv5 algorithm. The detection speed reaches 40 ms, thereby improving the detection accuracy and model robustness to meet the needs of real-time detection.

Key words: weed detection among maize seedlings, YOLOv5, attention module, BoTNet module, super-green feature, OTSU threshold segmentation