Scholarlink

<p>Aiming at the problems of low detection accuracy, large parameters, and slow model detection speed in the surface defect detection of electroplated safety buckles, in this paper proposes a MCD-YOLO lightweight model. First, we used the lightweight MobileNetV3 as the backbone network to reduce model parameters and computational complexity. SPPELAN-MDCR replaces the SPPF module, which can better maintain the sensitivity to defect features and improve the feature extraction ability of the model. Second, the DDFPN feature fusion network is designed to enhance the ability of feature fusion at different levels, and the ADDC module and down-sampling module are used to enhance the ability of the network to extract the feature information in the image, reduce the calculation of the network, and then improve the detection speed and accuracy of the model. Finally, we conducted experiments on the self-built DAQK dataset and the public NEU-DET dataset. Compared with the original model, the mAP increased by 6.8% and 3.2%, FPS increased by 14.4%, FLOPs decreased by 9.9% and 15.1%, and parameters decreased by 5.3% and 10.6%, respectively. The experimental results show the model realizes the balance between lightweight and detection accuracy, and has real-time and efficient practical value in surface defect detection.</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

MCD-YOLO: an efficient model for detecting surface defects of electroplated safety buckles

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Hongru Song,
  • Xiang Wu,
  • Hui Dong

摘要

Aiming at the problems of low detection accuracy, large parameters, and slow model detection speed in the surface defect detection of electroplated safety buckles, in this paper proposes a MCD-YOLO lightweight model. First, we used the lightweight MobileNetV3 as the backbone network to reduce model parameters and computational complexity. SPPELAN-MDCR replaces the SPPF module, which can better maintain the sensitivity to defect features and improve the feature extraction ability of the model. Second, the DDFPN feature fusion network is designed to enhance the ability of feature fusion at different levels, and the ADDC module and down-sampling module are used to enhance the ability of the network to extract the feature information in the image, reduce the calculation of the network, and then improve the detection speed and accuracy of the model. Finally, we conducted experiments on the self-built DAQK dataset and the public NEU-DET dataset. Compared with the original model, the mAP increased by 6.8% and 3.2%, FPS increased by 14.4%, FLOPs decreased by 9.9% and 15.1%, and parameters decreased by 5.3% and 10.6%, respectively. The experimental results show the model realizes the balance between lightweight and detection accuracy, and has real-time and efficient practical value in surface defect detection.