Fault Detection of Subway Bottom Bolts Based on Improved Faster-RCNN
Youxuan Zhu, Zongyi Xing, Sheng Li · 2022 International Conference on Cyber-Physical Social Intelligence (ICCSI) · 2022
With the acceleration of urbanization, the subway has become the first choice for urban residents to go out. Therefore, it becomes more and more important to guarantee the safety of the subway. Bolts are the most numerous part at the subway bottom. In daily operation, bolts falling off due to unexpected factors will bring great hidden dangers to the operation of the subway. This paper raises a new target detection algorithm based on Faster-RCNN. In view of the problem that the bolt size is small and difficult to detect, the FPN structure is introduced on the basis of the traditional Faster-RCNN, and then we improve the traditional FPN structure to enhance its robustness. Besides, we adjust the anchor size in the RPN structure according to the bolt size in the dataset. Finally, the whale optimization algorithm is introduced to ameliorate the loss function of RPN to make the accuracy of the algorithm better.