IoT 23(3): e1

Research Article

Weed detection with Improved Yolov 7

Download198 downloads
  • @ARTICLE{10.4108/eetiot.v9i3.3468,
        author={Mingkang Peng and Wuping Zhang and Fuzhong Li and Qiyuan Xue and Jialiang Yuan and Peipu An},
        title={Weed detection with Improved Yolov 7},
        journal={EAI Endorsed Transactions on Internet of Things},
        keywords={weed identification, Deep Learning, attention mechanism, yolov7},
  • Mingkang Peng
    Wuping Zhang
    Fuzhong Li
    Qiyuan Xue
    Jialiang Yuan
    Peipu An
    Year: 2023
    Weed detection with Improved Yolov 7
    DOI: 10.4108/eetiot.v9i3.3468
Mingkang Peng1,*, Wuping Zhang1, Fuzhong Li1, Qiyuan Xue1, Jialiang Yuan1, Peipu An1
  • 1: Shanxi Agricultural University
*Contact email:


INTRODUCTION: An improved Yolo v7 model. OBJECTIVES: To solve the weed detection and  identification in complex field background. METHODS: The dataset was enhanced by online data enhancement, in which the feature extraction, feature fusion and feature point judgment of weed image were carried out by Yolov7 to predict the weed situation corresponding to the prior box. In the enhanced feature extraction part of Yolov7, CBAM, an attention mechanism combining channel and space, is introduced to improve the attention of the algorithm to weeds and strengthen the characteristics of weeds. RESULTS: The mean average precision (mAP ) of the improved algorithm reached 91.15%, which was 2.06% higher than that of the original Yolov7 algorithm. Compared with the current mainstream target detection algorithms Yolox, Yolov5l, Fster RCNN, Yolov4-tiny and Yolov3, the mAP value of the improved algorithm increased by 4.35, 4.51, 5.41, 19.77 and 20.65 percentage points. Weed species can be accurately identified when multiple weeds are adjacent. CONCLUSION: This paper provides a detection model based on Yolov7 for weed detection in the field, which has a good detection effect on weed detection, and lays a research foundation for intelligent weeding robot and spraying robot.