Multi-State Recognition Method of Substation Switchgear Based on Image Enhancement and Deep Learning

Abstract

s. Since the current substation robot inspection process exists in the high-voltage switchgear status recognition is highly susceptible to the influence of complex environments such as low image contrast, light clutter interference, and blurred reading status details, this paper proposes an image enhancement and deep learning based substation switchgear state recognition method. A multi-scale Retinex-based image enhancement is proposed to enhance the adaptability of outdoor switchgear images to light changes; improve the YOLOx target detection network to introduces a lightweight ECA attention mechanism without dimensionality reduction based on the original YOLOx model’s backbone network CSPDarknet, allowing the model to learn classification features while also focusing on learning spatial features. The experimental results show that the improved network can accurately identify the boundary information of anomalies, and the quality of its prediction results will not be reduced for noise-containing data, and the network shows strong generalization, robustness, accuracy and rapidity, providing certain conditions for realizing substation equipment condition monitoring.