Deep learning-based infrared thermal image diagnosis method for power equipment and related device
By using improved Mask R-CNN and Swin Transformer models, combined with a temperature gradient channel and an adaptive temperature fusion module, the accuracy and physical interpretability issues of infrared image diagnosis for power equipment in existing technologies are solved, achieving efficient thermal defect identification and diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DATANG HYDROPOWER SCI & TECH RES INST CO LTD
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-19
AI Technical Summary
Existing deep learning-based infrared image diagnostic methods for power equipment have poor accuracy under complex operating conditions, fail to fully utilize the temperature field information in infrared images, and are affected by factors such as occlusion, blurring, and imaging at different distances/angles, resulting in a lack of physical interpretability and fine-grained classification capabilities in the diagnostic results.
An improved Mask R-CNN model is adopted to introduce a temperature gradient channel and an adaptive temperature fusion module. Combined with an improved Swin Transformer model, feature extraction and classification are optimized through a cross-temperature attention mechanism to form an end-to-end infrared thermal imaging diagnostic framework.
It improves the accuracy and robustness of infrared image diagnosis of power equipment, enhances the ability to identify thermal defects, and realizes physically interpretable fine-grained classification and efficient fault identification.
Smart Images
Figure CN122243909A_ABST