Guidelines for mechanical weeding: Developing weed control lines through point extraction at maize root zones

Abstract

Precision agriculture advancements are epitomised by precision mechanical weeding, which contributes significantly to sustainable farming practices. Traditional leaf-recognition technologies fail to meet the stringent requirements of precision weeding because they do not adequately guide weeding tools that operate close to seedling roots, such as finger weeders, to minimise crop damage. To address this issue, a novel method is developed to delineate paths for weeding tools, thereby preventing harm to seedlings. This method employs an advanced version of YOLOv8Pose to detect weeding areas around maize seedlings by pinpointing key points on the maize seedlings. To enhance the detection accuracy, a multi-scale dilation attention (MSDA) module and a lightweight reparameterisable EfficientRep module were used. The root connection line of the maize row was obtained by sequentially connecting the key point positions. The guide line for the weeding component was then determined by correcting this root connection line using the median absolute deviation (MAD) as the threshold. The approach demonstrated a remarkable precision in guiding weeding lines with an angular error of only 0–3° and a recognition rate of 100 FPS. In actual weeding operations, the effective weeding rate was 95.6%, which was far better than the 74.2% obtained by the leaf recognition-based method. This innovative method not only enhances weeding precision but also significantly reduces crop damage risk, thereby fostering more effective and sustainable agricultural practices.