COTTON WEED DETECTION USING FASTER R-CNN ON COTTONWEEDDET3 DATASET FOR PRECISION AGRICULTURE
Keywords:
Cotton weed detection, Faster R-CNN, im- age processing, computer vision, deep learning, precision agriculture, object detectionAbstract
Weed species can significantly impact crop productivity, and manual weeding is often impractical due to labor intensity and scale. Consequently, many recent studies have focused on automating weed detection using image-based approaches. However, accurately detecting weeds through images remains a challenging task because the texture, color, and shape of weeds and crops are often very similar. In this study, we propose a deep learning-based solution using the Faster Region- Based Convolutional Neural Network (Faster R-CNN) architecture to detect three cotton weed species: carpetweed, morning-glory, and Palmer amaranth. We utilize a publicly available dataset, CottonWeedDet3, which contains 848 RGB images annotated with bounding boxes following the Common Objects in Context (COCO) format. Our proposed model achieved a mean Average Precision (mAP@0.5) of 92.3% at an Intersection Over Union (IoU) threshold of 0.5. The findings demonstrate the effectiveness of Faster R-CNN for accurate and auto- mated cotton weed detection in the context of precision agriculture.
