Co-occurrence Object Detection of the Transmission Lines Based on the Cross-Domain Interactive Feature Enhancement

作者：Lijuan Zhao, Chang'an Liu, Hongquan Qu · 发表于：IEEE Transactions on Power Delivery · 年份：2023 · DOI：10.1109/tpwrd.2023.3321867 · 被引用次数：9 · 研究领域：Advanced Neural Network Applications、Power Line Inspection Robots、Vehicle License Plate Recognition

The inspection of the transmission lines is of great significance for the safe and stable transportation of electric power. However, the small-sized and easily confused objects of transmission lines are always difficult to detect accurately. In this article, a new network based on the combination of the cross-domain interactive (CDI) enhanced pyramid network and the knowledge graph relational reasoning (KGRR) network is proposed to improve the detection accuracy. First, the self-attention is applied to the feature layer fusion process, and the object features located on different feature layers are enhanced by calculating the similarity of the features of the adjacent layers. Secondly, constructing the knowledge graph of the fitting and using Graph Convolution Network (GCN) to adjust the output of the raw object detector. The experimental results show that the method proposed in this article has greatly improved the detection accuracy of the small-sized fittings of the transmission line, the mean Average Precision (mAP) of this proposed method is improved by 1.9% compared with the baseline algorithm.