Scholarlink

<p>Recently, defect detection has received a great deal of attention in graphic image processing and multimedia vision. Although existing visual detection methods have achieved promising performance, they are mainly designed for object recognition tasks under ideal conditions (sufficient lighting, clean background, etc.). Performance drops sharply when faced with complex and challenging real-world conditions (such as excessive line of sight and noisy backgrounds). To address this problem, this work proposes a visual defect detection algorithm for distribution lines from UAVs using a local dual-branch feature enhancement (LDFE) framework. The proposed framework consists of three main steps: (1) first, it effectively locates and extracts the features of distribution line defects (bird nests, bird damage, foreign object coverage, and construction work); (2) by integrating a feature pyramid network (FPN) to achieve multi-scale feature fusion, the prominence of defect areas is enhanced to improve the accuracy and robustness of the model in target detection tasks; (3) finally, the use of the dual-branch mechanism facilitates multi-task learning for both defect classification and regression training. Experimental results show that our proposed method outperforms the widely used standard architecture deep models in terms of effectiveness, better meeting the practical requirements of industrial inspection and visual applications for graphical images. Source code can be found at <a href="https://github.com/jxsiaj/DCBM.git">https://github.com/jxsiaj/DCBM.git</a>.</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

Local dual-branch attention feature learning framework from UAVs for visual defect detection

  • 期刊论文

  • 发表日期：

    2025年1月22日

查看全文
  • Jianbing Xu,
  • Jiangxin Zhou,
  • Dongxu Xu,
  • Yu Chen

摘要

Recently, defect detection has received a great deal of attention in graphic image processing and multimedia vision. Although existing visual detection methods have achieved promising performance, they are mainly designed for object recognition tasks under ideal conditions (sufficient lighting, clean background, etc.). Performance drops sharply when faced with complex and challenging real-world conditions (such as excessive line of sight and noisy backgrounds). To address this problem, this work proposes a visual defect detection algorithm for distribution lines from UAVs using a local dual-branch feature enhancement (LDFE) framework. The proposed framework consists of three main steps: (1) first, it effectively locates and extracts the features of distribution line defects (bird nests, bird damage, foreign object coverage, and construction work); (2) by integrating a feature pyramid network (FPN) to achieve multi-scale feature fusion, the prominence of defect areas is enhanced to improve the accuracy and robustness of the model in target detection tasks; (3) finally, the use of the dual-branch mechanism facilitates multi-task learning for both defect classification and regression training. Experimental results show that our proposed method outperforms the widely used standard architecture deep models in terms of effectiveness, better meeting the practical requirements of industrial inspection and visual applications for graphical images. Source code can be found at https://github.com/jxsiaj/DCBM.git.