Research on improved YOLOx weed detection based on lightweight attention module

• DOI: 10.1016/j.cropro.2023.106563
• Publication Date: 2023-12-15T19:59:07Z
• Authors: Huibin Zhu; Yuanyuan Zhang; Danlei Mu; Lizhen Bai; Xian Wu; Hao Zhuang; Hui Li
• Keywords: False positive rate; Feature (linguistics)

Accurate weed detection is essential for precise control in farmland, and machine vision serves as an effective method identifying targeting these unwanted plants. An advanced YOLOx model, incorporating a deep network connection lightweight attention mechanism, has been proposed to effectively identify distinct types of weeds maize seedling fields. The module connected the YOLOx-Darknet, which weakens channel noise effect residual computation thus makes model more efficient. A deconvolution layer introduced improve small size feature extraction capability. Generalized Intersection over Union (GIoU) replaced (IoU) minimize positional discrepancy between predicted frame actual frame, thereby enhancing learning capacity model. Compared with original algorithm, improved one achieved average accuracy 94.86% performance evaluation, 0.07% better F1 value, 1.16% AP value. weeding robot algorithm had 92.45% rate seedlings 88.94% recognition at 0.2 m s−1. results this study can provide technical references real-time robotic precision weeding.