To accurately detect abnormal heating regions of power equipment in complex substation scenarios, this paper proposes a method for detecting abnormal heating of power equipment by fusing heterogeneous features from infrared and visible images. First, a semantic segmentation network for power equipment is constructed, enhanced through semantic edge information collaboration. A multi-level cross-modal feature fusion module is designed, adopting a hierarchical fusion strategy to integrate cross-modal features, thereby improving the model's ability to comprehend images. During the feature decoding stage, an edge information supervision module is introduced to enhance the clarity and continuity of segmentation boundaries, effectively removing background regions from the images. On this basis, an abnormal heating detection method optimized by attention mechanisms and residual information is proposed. A multi-level feature fusion and attention enhancement module is designed to analyze infrared and visible light foreground features, and residual information from the images is incorporated to further improve the detection capability and significantly enhance detection accuracy for abnormal heating regions. Using field-collected and manually annotated data, a self-built dataset was compared against multiple state-of-the-art algorithms. The proposed method achieved mAcc and mIoU scores of 97.2% and 88.9%, respectively, for semantic segmentation, and the E-measure for abnormal heating detection improved by 2.6% compared to the next-best algorithm. Experimental results demonstrate that the proposed method can efficiently and accurately detect abnormal heating regions of power equipment in complex scenarios.