摘要详情

ID / 提交时间

5 / 2023-08-04 15:34:46

标题

Defect Recognition Method of Digital Twin Power Line Image Based on Improved TridentNet

关键字

Neural Networks,Defect Detection,Power Line,TridentNet,Weights normalization

主题及专题

Special Sessions > SS14 - Digital Twin Empowered Intelligent Operation and Maintenance of Renewable Energy Equipment

状态

终稿

作者

Fan XU / State Grid Information & Telecommunication Group CO，LTD；Beijing Insititute of technology

Yuanyuan LIU / State Grid Information & Telecommunication Group CO，LTD

Fan YANG / Aostar Information Technologies Co.,Ltd

Wenpu LI / State Grid Information & Telecommunication Group CO，LTD

Xiaomeng DI / China Electric Power Research Institute

Xiaodong DU / Hebei Electric Power Research Institute of State Grid

Xingtao WANG / State Grid Information & Telecommunication Group CO，LTD

摘要

The detection of defects in Power Line images plays a vital role in timely and effective identification of equipment flaws, thereby preventing equipment failures. Current methods for transmission image defect detection need further algorithmic research to improve their capability to detect defects at different scales. In this paper, we propose an improved method for the TridentNet network specifically designed for transmission line inspection scenarios. We optimize the scale-based training strategy by adjusting the selection range to enhance the model training effectiveness. We introduce a new dimension in the backbone network to improve feature extraction efficiency. Additionally, we design a weight normalization process in the feature extraction part of the network to accelerate model convergence and improve accuracy when training data batches are small. These enhancements contribute to improved accuracy and recall rates for multi-scale defect detection in the same image. Experimental results demonstrate the effectiveness and accuracy of our proposed method.