Scholarlink

<p>Crop-weed classification and detection are essential tasks in precision agriculture for effective weed management. In this study, we employed the most popular YOLO (You Only Look Once) object detection algorithm to address this challenge. A Weed1000 dataset comprising 1228 annotated images was used for training and evaluation. Various versions of the YOLOv5 algorithm, namely YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5X, were implemented and evaluated with reference to performance evaluation metrics including precision, recall, F1 score and mean Average Precision (mAP) at different Intersection over Union (IoU) thresholds. The hyper-parameter tuning and addition of layers in the modified YOLOv5X demonstrated higher precision of 95.3%, recall of 92%, F1 score of 93% and mAP of 94.5% and 47.8% at 0.5 and 0.95 IoU thresholds, respectively. Furthermore, we developed a web application using Flask to deploy the YOLOv5 model, enabling real-time crop-weed detection accessible via a user-friendly interface. We have also compared our proposed approach with previous studies. Notably, this research introduces novel modifications to the YOLOv5 algorithm to enhance its effectiveness and highlights the practical deployment of such models for agricultural applications. The findings underscore the importance of model selection and deployment strategies in achieving accurate and efficient crop-weed detection systems. Additionally, our modifications to the YOLOv5 algorithm and the development of a user-friendly deployment application represent significant advancements in the field.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance Evaluation of Modified YOLOv5 Object Detectors for Crop-Weed Classification and Detection in Agriculture Images

  • 期刊论文

  • 发表日期：

    2025年1月30日

查看全文
  • Sandip Sonawane,
  • Nitin N. Patil

摘要

Crop-weed classification and detection are essential tasks in precision agriculture for effective weed management. In this study, we employed the most popular YOLO (You Only Look Once) object detection algorithm to address this challenge. A Weed1000 dataset comprising 1228 annotated images was used for training and evaluation. Various versions of the YOLOv5 algorithm, namely YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5X, were implemented and evaluated with reference to performance evaluation metrics including precision, recall, F1 score and mean Average Precision (mAP) at different Intersection over Union (IoU) thresholds. The hyper-parameter tuning and addition of layers in the modified YOLOv5X demonstrated higher precision of 95.3%, recall of 92%, F1 score of 93% and mAP of 94.5% and 47.8% at 0.5 and 0.95 IoU thresholds, respectively. Furthermore, we developed a web application using Flask to deploy the YOLOv5 model, enabling real-time crop-weed detection accessible via a user-friendly interface. We have also compared our proposed approach with previous studies. Notably, this research introduces novel modifications to the YOLOv5 algorithm to enhance its effectiveness and highlights the practical deployment of such models for agricultural applications. The findings underscore the importance of model selection and deployment strategies in achieving accurate and efficient crop-weed detection systems. Additionally, our modifications to the YOLOv5 algorithm and the development of a user-friendly deployment application represent significant advancements in the field.