End-to-End Detection for Key Equipment in Natural Gas Station with DETR

To enhance operational efficiency at natural gas stations, it is essential to establish the functional connection topology of station components. This includes the reliability block diagram and fault tree model during reliability assessment and optimization. The complex technological processes and interactions among various components require accurate and efficient component identification. In this paper, a novel detector for component identification has been established based on the Detection Transformer (DETR). It directly predicts object bounding boxes and categories without post-processing, therefore, enabling end-to-end detection. To test the proposed model, a comprehensive dataset has been generated combining data augmentation. Experimental results showed the well performance of the proposed model in identifying natural gas station equipment. The identification accuracy is 89.94%, which is the state of the art of the existed model. The proposed model can provide effective technical support for safe operation and equipment management, and also shows strong ability to deal with different scenarios and environments.