Soybean Seedling-Stage Weed Detection and Distribution Mapping Based on Low-Altitude UAV Remote Sensing and an Improved YOLOv11n Model

Abstract

Seedling-stage weeds are one of the key factors affecting the crop growth and yield formation of soybean. Accurate detection and density mapping of these weeds are crucial for achieving precise weed management in agricultural fields. To overcome the limitations of traditional large-scale uniform herbicide application, this study proposes an improved YOLOv11n-based method for weed detection and spatial distribution mapping by integrating low-altitude UAV imagery with field elevation data. The second convolution in the C3K2 module was replaced with Wavelet Convolution (WTConv) to reduce complexity. A SENetv2-based C2PSA module was introduced to enhance feature representation and context fusion with minimal parameter increase. Soft-NMS-SIoU replaced traditional NMS, improving detection accuracy and robustness for dense overlaps. The improved YOLOv11n algorithm achieved a 3.4% increase in mAP@50% on the test set, outperforming the original YOLOv11n in FPS, while FLOPs and parameter count increased by only 1.2% and 0.2%, respectively. More importantly, the model reliably detected small grass weeds with morphology highly similar to soybean seedlings, which were undetectable by the original model, thus meeting agricultural production monitoring requirements. In addition, the pixel-level weed detection results from the model were converted into coordinates and interpolated using Kriging in ArcGIS (10.8.1) Pro to generate continuous weed density maps, resulting in high-resolution spatial distribution maps directly applicable to variable-rate spraying equipment. The proposed approach greatly improves both the precision and operational efficiency of weed detection and management across large agricultural fields, providing scientific support for intelligent variable-rate spraying using plant protection UAVs and ground-based sprayers, thereby promoting sustainable agriculture.