Effective detection of weeds in sesame crop

Abstract Rapid improvements in Computer vision (CV) and Deep Learning (DL) methods have enabled information extraction from images and videos to apply and interpret them for various applications. Numerous agricultural and farming applications have been using DL approaches for resolving agricultural challenges as well as optimizing farming operations. One such significant task is automated weed identification and categorization to improve crop yield. With the recent research and experiments providing information on the adverse effects of the use of chemical herbicides in farming, it is becoming evident that they are damaging the environment while wasting resources. Differentiating weeds from crops before applying herbicides can help with reducing the environmental effects. Detecting weeds in crops from images and videos is a challenging task due to the similarities between the colors, forms, and textures of weeds and crops throughout the growing period. In this paper, a deep Convolutional Neural Networks (CNN) is implemented for accurate detection of weeds in sesame crops for selective chemical herbicides spraying. Three different pre-trained deep learning algorithms like You Only Look Once (YOLO), specifically YOLOv4, YOLOv5, and Faster R-CNN are applied and the detection models achieved mean average precision (mAP) of 74%, 56%, and 52%, respectively. YOLOv4 performed the best among the three models on a limited dataset. The average IoU value outperformed other object detection-based DL models applied for similar tasks with the score of 80%, showing the efficiency of the proposed system.