Infrared Image Segmentation Method of Current Transformer Based on DeepLabv3+ Network

Intelligent infrared image analysis is an effective method for transformer substation equipment fault diagnosis, and target device segmentation is its key technology. In order to solve the problem that it is difficult to partition the current transformer in a complex background, this paper uses the DeepLabv3+ neural network based on ResNet50 to train the semantic segmentation model with the current transformer. Using image transformation to expand the sample dataset for the collected samples, semantic segmentation network training semantic segmentation model is built. 420 current mutual inductor infrared images are tested and compared with other mainstream network models. The results show that the average crossover ratio (MIoU) of this method is 94.2%, which is significantly better than other network models and can accurately segment current mutual inductor devices from the test images. It paves the way for subsequent intelligent fault diagnosis of a current transformer.