Automatic diagnosis of heating in oil-filled terminals of cables

Abstract

Abstract A new automatic diagnosis technique was proposed to address the problem of cable oil-filled terminal heating. The infrared and visible light images collected by DJI drones were registered using the scale-invariant feature transform (SIFT) algorithm. The progressive infrared and visible image fusion network based on the illumination aware (PIA Fusion) network was used to fuse the infrared and visible light images. The fused images were used to train the version 5 of You Only Look Once (YOLOv5) network for object detection. The areas prone to heating were identified and mapped to the infrared images before fusion, and the mapped areas of the infrared images were cropped. The cropped images were fed into the DJI infrared analysis software toolkit Thermal Software Development Kit (TSDK) to obtain the temperature information, and diagnosis was performed according to the relevant standards. The infrared images and fused images were separately used for training. The experimental results showed that the mean average precision calculated when the intersection over union (IoU) threshold was 0.5 (mAP@0.5) was 95.3% when training with fused images and the average detection time was 12 ms per image. This technique could replace traditional manual diagnosis to improve detection efficiency and precision.