Research On Dropper Location And State Recognition Algorithm Based On Deep Neural Network

Railway is an important form of transportation in our country,and plays a key role in China’s comprehensive transportation system.In order to ensure that the railway can better meet the needs of social and economic development,it is necessary to maintain the safe operation of the railway.As an important transmission line of railway,the normal operation of catenary directly affects the normal power supply of railway system.On the basis of summarizing the existing detection methods based on deep learning and traditional state monitoring,this paper designs the detection algorithm with rotating frame and the state detection algorithm based on deep learning and Hough transform,and applies it to the state detection task of railway hanging string.The main research results are as follows:1)In this paper,dropper location algorithm based on rotating frame detection is proposed.By adding angle branch on the basis of Faster R-CNN,and adjusting boxes presetting to better match the dropper,invalid background in dropper detection boxes is eliminated.The experimental results show that the algorithm has good robustness and can complete dropper location task in multiple scenarios.2)Another state recognition algorithm for droppers based on RefineDet Network and Hough transform is proposed.Combining the traditional line detection task with the target detection task,the algorithm framework of shape recognition of the key parts of the hanging string based on the accurate positioning of the target detection is designed.Combined with the actual inspection experience,the state of the hanging string is further obtained.The experimental results show that the algorithm can effectively improve the accuracy of dropper state detection and realize real-time detection.