Semantic Segmentation of Substation Scenes Using Attention-Based Model

Abstract

Inspection robots are widely used in transformer substation for inspection tasks. In order to complete the task, the robots need to understand and recognize road scenes. However, the transformer substation contains lots of similar and small equipment, which are difficult to recognize. To address this problem, we propose a semantic segmentation model to recognize road scenes. Specifically, our proposed model discriminates the equipment by leveraging multi-view method. We use attention mechanism to strengthen the relationship between pixels that belong to an individual category. Compared with previous methods, our approach achieves a good performance in a new dataset from a transformer substation and shows robustness in recognizing equipment of transformer substation.