Infrared Image-Based Fault Warning for Substation High-Voltage Switchgear via Multi-Scale Feature Optimization

To address the risk of fire hazards and equipment damage in substations caused by abnormal temperature rise inside high-voltage switchgear, This work advocates a fault early warning technique based on infrared imaging. First, the internal structure of the switchgear and the mechanism of thermal anomalies are analyzed. An infrared thermal image dataset is constructed and enhanced using multi-directional edge reinforcement, an Efficient Channel Attention (ECP) mechanism, and feature pyramid optimization to improve sample quality. Then, to overcome the tendency of the traditional Sparrow Search Algorithm (SSA) tends to get trapped in local optima optima, an improved SSA-SVM prediction model is proposed, incorporating Logistic chaotic mapping and target-tracking mechanisms. Experimental results demonstrate that the optimized feature detection achieves an accuracy of 98.8%, with a training loss of 0.0096, and improves prediction accuracy by $14-16 \%$ compared to baseline methods.