Automatic Crop and Weed Detection using Deep Learning: A Comparative Analysis of YOLO Models

Managing weeds efficiently is essential to securing optimal crop production, particularly in the face of rising herbicide resistance and the push toward environmentally sustainable farming. This study evaluates the performance of advanced deep learning-based object detection models for identifying crops and weeds in realistic field conditions. Over 1,400 RGB field images were gathered via drone platforms, covering multiple crop growth phases and challenging visual scenarios, including variable lighting and partial plant coverage. Four models—YOLO-NAS, YOLOv11, YOLOv12, and RF-DETR—were trained and tested under the same experimental setup. The models were assessed using standard object detection metrics such as mean Average Precision at 0.5 (mAP@0.5), precision, and recall. Among all tested models, YOLOv12 achieved the highest performance, recording $89.0\%$ mAP@0.5, $89.4\%$ precision, and $81.9\%$ recall, indicating strong suitability for real-time detection tasks. YOLO-NAS followed closely, offering a well-balanced combination of accuracy and detection consistency, enhanced by its Neural Architecture Search driven structure. RF-DETR achieved the top precision score but lagged in recall. YOLOv11 delivered comparable accuracy alongside rapid inference times. Overall, YOLOv12 emerged as the most promising choice for scalable, real-time deployment using drone imagery. The findings emphasize the benefits of combining drone acquired imagery with automated AI detection tools to enhance precision farming and support sustainable land use practices.