Answers are grounded in this paper and cite their sources.
For condition monitoring and intelligent inspection of substation equipment, equipment recognition and localization in infrared images provide important spatial information for subsequent inspection and maintenance. To address the detection difficulties caused by complex thermal backgrounds and differences in equipment scale, this study proposes an infrared object detection method based on YOLO11n that combines scale-aware spectral modulation and spatial-frequency residual collaboration. The method decomposes multiscale features under unified normalized frequency coordinates, learns independent frequency-band responses for different detection levels, and integrates frequency-reconstructed information into the original spatial features through residual correction. On the independent test set of a self-constructed dataset containing 6020 images and 5 equipment classes, the proposed method achieves mAP50 and mAP50:95 values of 91.0% and 71.3%, respectively, exceeding the baseline by 3.5 and 2.4 percentage points. Five-fold cross-validation further indicates that the proposed method maintains good detection stability across different data splits. On the experimental platform used in this study, the model achieves a forward inference speed of 136.7 FPS. The results show that the method can improve infrared detection of substation equipment with a small additional parameter overhead, supporting equipment recognition and localization during intelligent inspection.
No takes yet. Share an insight, caveat, or question.
Lu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: