PIoT-oriented multi-target recognition of substation infrared images driven by deep learning

Abstract

Substation infrared imaging plays a crucial role in condition monitoring and fault detection of Power Internet of Things (PIoT). However, the accurate and efficient recognition of multiple targets in substation infrared images remains a challenging task. This paper proposes a deep learning-based multi-target recognition framework for substation infrared images in PIoT. This paper presents a method for recognizing various electrical equipment in infrared images of substations using a faster region-based convolutional neural network (Faster RCNN). The optimization of Faster RCNN includes class rectification inspired by non-maximum suppression (NMS), enabling the correction of misclassified equipment parts and enhancing recognition accuracy. The approach combines NMS and class rectification to retain region proposals with optimal recognition performance. Experimental results demonstrate the effectiveness of the proposed method in improving the recognition accuracy of electrical equipment in infrared images.