Surface Defect Detection of Hight-speed Railway Hub Based on Improved YOLOv3 Algorithm

摘要

Aiming at the problems of low detection efficiency, cumbersome processing steps, and lack of real-time performance in existing high-speed rail wheel surface defect detection methods. asurface defect detection algorithm based on improved YOLOv3 is proposed. The algorithm uses Darknet-53 to extract the feature maps of each stage of the defect image, Then use the improved feature pyramid structure to increase the bottom-up reverse connection path to fully integrate the feature maps of each stage to obtain a multi-scale feature map with stronger semantic information and positioning information, and finally use non-maximum suppression methods to filter The bounding box with the highest score is obtained; in order to improve the detection effect, the deformed convolution technology is used to enhance the adaptability of the convolution according to the characteristics of the surface of the high-speed rail wheel hub; the experimental results show that the proposed algorithm is compatible with Faster R-CNN and YOLOv3 Compared with this, it has better detection performance, can locate various defects of high-speed rail wheels more accurately, and has better application value in this field.