Deep Learning-Based Weed Detection for Precision Agriculture Applications

Weed detection and management is an intricate and formidable challenge, affecting the efficiency of human agricultural practices. Object detection has merged as a key technique in precision-based agriculture for combatting this issue. This paper analyzes the performance of two cutting-edge object detection models namely YOLOv8x and Faster RCNN on a specialized weed-detection dataset. By leveraging transfer learning, these models proved computationally feasible and accurate. The models were built and trained in a Kaggle environment, promoting a more collaborative approach to developing and implementing deep-learning models. The Faster RCNN model showcased a comparatively better mAP as compared to the YOLOv8x model. This paper highlights the significance of advanced models like Faster RCNN and YOLOv8x for precision agriculture while advocating for further research in deep-learning techniques for promoting resource-efficient agricultural and farming practices.