Image identification method on high speed railway contact network based on YOLO v3 and SENet
Qiang Chen, Li Liu, Rui Han, Jiaying Qian, Donglian Qi · 2019
In order to timely discover the missing spare nut on the U-shaped hoop of the high-speed railway contact network insulator area and add new nut, reduce the risk of failure of insulator connection, research on U-shaped hoop missing spare nut image recognition methods on high speed railway contact network based on YOLO v3 and SENet was carried out. The identification method is divided into two stages: the first stage detects and locates the U-shaped hoop of the insulator area in the high-speed railway contact network image through YOLO v3, and then the positioned U-shaped hoop is taken out from the original image and scaled to a size of 224×224 In the second stage, the scaled U-shaped hoop image is classified and identified by SENet, and the missing spare nut image of the U-shaped hoop is obtained. To enhance the visualization of the image, class activation map is used to locate the U-shaped hoop image. The method solves the problem of high resource occupancy rate for small object recognition on high-resolution images. The recognition result based on the U-shaped hoop missing spare nut scene of Xinjiang high speed railway contact network showed that the accuracy of the identification method reached 88.24%, which could be applied to U-shaped hoop missing spare nut identification on high speed railway contact network.