An Efficient Weed Detection Model using RCNN and YOLO Algorithms

Weed management is a crucial aspect of precision agriculture, directly influencing crop yield and resource efficiency. This paper presents a dual-task deep learning framework that addresses both Weed vs. Crop Classification and Object Detection using advanced computer vision techniques. A curated dataset of 1300 annotated images of sesame crops and various weed species was employed, with labels in the YOLO format. For the classification task, bounding boxes were extracted and cropped from the annotated images to create a separate dataset, enabling the training of a pre-trained ResNet50 model to distinguish between weeds and crops through transfer learning. Simultaneously, YOLOv5 and RCNN models were utilized for object detection to identify and localize weeds within full-field images. The dataset underwent comprehensive preprocessing steps, including image resizing and augmentation, to enhance model robustness and generalization. Performance evaluation was conducted using standard metrics such as Precision, Recall, F1-Score, and mean Average Precision (mAP). The results indicate that YOLOv5 excels in real-time detection, making it suitable for field deployment, while RCNN offers superior localization accuracy. The ResNet50-based classification model demonstrated high accuracy in differentiating between weed and crop segments. This integrated approach provides a comprehensive solution for automated weed management, reducing the reliance on manual labor and chemical herbicides.