A multi-scale collaborative optimization approach for infrared target detection of electrical apparatus in switchgear cabinets

Switchgear cabinets are critical components of power-distribution infrastructure, and accurate infrared target detection of their electrical apparatus provides a prerequisite for component-level temperature extraction and overheating assessment. Such capability supports condition monitoring and preventive maintenance, thereby contributing to the reliable and continuous delivery of electrical energy. However, practical infrared images are often affected by complex background thermal interference, long-tailed class distribution, weak small-target responses, and blurred thermal boundaries, which degrade detection and localization performance. To address these challenges, this paper proposes a multi-scale collaborative optimization framework that integrates input enhancement, sample-distribution regulation, feature representation, and bounding-box regression. Specifically, a semantic-guided region-of-interest–background decoupled enhancement strategy is introduced to suppress irrelevant background responses while preserving target-related thermal details. Minority-class-oriented augmentation and class-frequency weighting are employed to alleviate long-tailed training bias. In addition, the C2f module embedded with SimAM attention is incorporated to recalibrate local feature responses, and Minimum Point Distance Intersection over Union loss is adopted to improve localization under blurred thermal boundaries. Experimental results show that the proposed method achieves 0.971 mAP@0.5 and 0.779 mAP@0.5:0.95 with 15.32 M parameters and 58.9 GFLOPs. Compared with the YOLOv10m baseline, mAP@0.5 and mAP@0.5:0.95 are improved by 1.0 and 2.0 percentage points, respectively, without increasing model complexity. These results demonstrate the effectiveness of the proposed framework as a front-end perception method for infrared inspection of switchgear cabinets, providing reliable target localization for component-level thermal monitoring and preventive maintenance in power-distribution infrastructure.