X-ray Imaging Defect Detection of Transmission Line Strain Clamps Based on a YOLOX Model
Zhibin Qiu, Junxuan Li, Dazhai Shi, Zuwen Lu · 2022 IEEE International Conference on High Voltage Engineering and Applications (ICHVE) · 2022
Effective detection of internal defects of strain clamps is of vital significance to safe operation of transmission lines, and the X-ray radiographic inspection is a useful method to evaluate the hydraulic crimping quality of strain clamps. This paper presents a method to detect defects in X-ray images of strain clamps using YOLOX algorithm. An X-ray image dataset of strain clamps including 4976 images with 6 types of defects was constructed. The images were preprocessed by Gaussian filtering, histogram equalization, and gamma correction, thus to improve the image quality. The model was trained by the training sample X-ray images combining the Mosaic data augmentation method. The trained YOLOX model was applied to detect the defects in the 498 test sample X-ray images, and the mean average precision (mAP) reached 88.96%. The detection results were also compared to those of other object detection algorithms like SSD, YOLOv3, YOLOv4, etc., which indicates that the YOLOX model has higher precision. This study is helpful to automatically detect the defects of X-ray inspection images of transmission line strain clamps.