Weed detection in soybean fields using improved YOLOv7 and evaluating herbicide reduction efficacy

• DOI: 10.3389/fpls.2023.1284338
• Publication Date: 2024-01-11T04:35:41Z
• Authors: Jinyang Li; Wei Zhang; Hong Zhou; Chuntao Yu; Qingda Li
• Keywords: Adaptability; Metribuzin

With the increasing environmental awareness and demand for sustainable agriculture, herbicide reduction has become an important goal. Accurate efficient weed detection in soybean fields is key to test effectiveness of application, but current technologies methods still have some problems terms accuracy efficiency, such as relying on manual poor adaptability complex environments. Therefore, this study, weeding experiments with reduced including four levels, were carried out, unmanned aerial vehicle (UAV) was utilized obtain field images. We proposed a model-YOLOv7-FWeed-based improved YOLOv7, adopted F-ReLU activation function convolution module, added MaxPool multihead self-attention (M-MHSA) module enhance recognition weeds. continuously monitored changes leaf area dry matter weight after reflection growth at optimal application levels. The results showed that level electrostatic spraying + 10% could be used fields, YOLOv7-FWeed higher than YOLOv7 YOLOv7-enhanced all evaluation indexes. precision model 0.9496, recall 0.9125,