Dual-Channel Branch Mechanism-basedVisual Defect Detection

Recently, defect detection has received widespread attention in the fields of graphic images and multimedia vision. Although existing visual detection methods have achieved promising performance, they are mainly targeted at object detection tasks under ideal conditions (sufficient lighting, clean background, etc.). When faced with complex and harsh conditions in reality (such as excessive line of sight and noisy background), the performance drops sharply. To address the potential hazards posed by abnormal interferences in power transmission lines due to natural environmental factors and human activities (such as bird nests, bird damage, foreign object coverage, and construction work), this paper proposes a defect detection algorithm for distribution lines based on a dual-branch mechanism. The proposed method first effectively locates and extracts the features of distribution line defects (bird nests, bird damage, foreign object coverage, and construction work). It then integrates a Feature Pyramid Network (FPN) to achieve multi-scale feature fusion, enhancing the prominence of defect areas to improve the accuracy and robustness of the model in target detection tasks. Finally, utilizing the dual-branch mechanism facilitates multi-task learning for both classification and regression training of defects. Experimental results demonstrate that our proposed method outperforms the widely adopted standard architecture deep models in terms of effectiveness, better meeting the practical requirements of industrial inspections and the visual applications for graphic images. Source code can be found at: https://github.com/jxsiaj/DCBM.git.