Scholarlink

Small objects make up a high proportion of Unmanned aerial vehicle (UAV) images, but the existing detection algorithms have problems such as low detection accuracy and high leakage rate. To improve small object detection, an improved YOLOv5s model is proposed in this paper. In the Neck Network of the YOLOv5s model, a channel attention module attention (CBAM) mechanism is added. By adding CBAM after each CSP2_1 module and before the convolution operation, the model can automatically ignore non-critical information and focus highly on key features in the image. The improved YOLOv5s model was evaluated on the VisDrone2019-DET dataset and its performance was compared with YOLOv5s and YOLOv8s. The YOLOv5s-CBAM model has improved mAP_0.5 by 2.2% and mAP_0.5:0.95 by 1.2%. The loss convergence speed is faster and the stability is better.

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

Improved UAV-to-Ground Small Object Detection Algorithm Based on Yolov5s

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ling Luo,
  • Zhang Wu,
  • Wei Han,
  • Xuefei Sun,
  • Tingting Bai

摘要

Small objects make up a high proportion of Unmanned aerial vehicle (UAV) images, but the existing detection algorithms have problems such as low detection accuracy and high leakage rate. To improve small object detection, an improved YOLOv5s model is proposed in this paper. In the Neck Network of the YOLOv5s model, a channel attention module attention (CBAM) mechanism is added. By adding CBAM after each CSP2_1 module and before the convolution operation, the model can automatically ignore non-critical information and focus highly on key features in the image. The improved YOLOv5s model was evaluated on the VisDrone2019-DET dataset and its performance was compared with YOLOv5s and YOLOv8s. The YOLOv5s-CBAM model has improved mAP_0.5 by 2.2% and mAP_0.5:0.95 by 1.2%. The loss convergence speed is faster and the stability is better.