Remote Aerial Vehicle Solutions for Weed Detection in Precision Agriculture

This study presents a novel Remote Aerial Vehicle-based approach for detecting pigweeds in soybean (Glycine max) fields using a combination of deep learning and advanced image processing techniques. A custom, high-resolution dataset comprising RGB (red, green, and blue) and multispectral images was collected from USDA operated fields and manually annotated for target pigweed detection. Beyond incorporating YOLOv8 variants for real-time weed classification, this study integrates a comprehensive image processing pipeline, incorporating global thresholding, k-means clustering, 3D surface mapping, and spectral signature analysis to enhance interpretability and detection accuracy. A comparative evaluation of YOLOv8 nano, small, medium, and large models, along with Faster R-CNN, confirmed YOLOv8’s balance between detection accuracy and real-time performance for deployment in precision agriculture. The YOLOv8 nano model emerged as the most balanced in terms of precision (75.6%), recall (81.7%), and mAP@0.5 (81.5%), demonstrating effective weed detection performance under real field conditions. In addition, quadrant-level weed coverage and spatial heatmaps were generated to support targeted interventions. This study advances the current state of RAV based weed detection by providing a field-ready, explainable, and resource-efficient solution, contributing to sustainable farming and data-driven weed management practices.