A State Recognition Method of Isolation Switch in Traction Substation Based on Key Components Detection and Geometric Ranging

The isolation switch is one of the most important pieces of electrical equipment in traction substation, and its state directly reflects the operation of the power system. However, the realization of automatic isolation switch monitoring based on computer vision still faces three problems, i.e., intra-class diversity, inter-class polymorphism of components, and complex backgrounds. To overcome the issues of isolation switch positioning and states recognition in railway traction substation, we propose a state recognition method of isolation switch in traction substation based on key components detection and geometric ranging. This method is divided into two parts, i.e., ISD-Net for isolation switch detection and geometric ranging for state recognition. Firstly, ISD-Net is proposed to realize isolation switch detection. Secondly, an algorithm based on geometric ranging is designed, which utilizes the coordinate information of key components to realize automatic recognition of isolation switch state. To train the model and evaluate the effectiveness of our method, we established a dataset of isolation switch (called IVIS2021). On IVIS2021, isolation switch detection and state recognition accuracy reach 90.88% and 91.12%, respectively. The experimental results indicated that the proposed method accurately realizes isolation switch detection and state recognition.