Multispectral downhole personnel detection method based on unet and multi-scale attention
By combining UNet with multi-scale attention and fusing features from visible light and infrared images, the detection bottleneck in the complex environment of underground coal mines has been solved, achieving high-precision and robust multispectral personnel detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TIANDI CHANGZHOU AUTOMATION
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing single-spectral image detection methods have poor detection performance in complex environments in coal mines (such as changes in lighting, dust interference, and cluttered backgrounds). They are difficult to effectively integrate the rich details of visible light images with the temperature and contour information of infrared images, resulting in high rates of missed detections and false detections.
A multispectral downhole personnel detection method based on UNet and multi-scale attention is adopted. RGB visible light images and IR infrared images are acquired by a dual-light camera. Personnel features are extracted using contour attention and temperature attention. Multi-scale attention feature maps are generated by scale transformation and fused into the multi-resolution visible light feature map of the UNet encoder. Finally, the personnel detection results are output.
It significantly improves the detection accuracy and robustness in complex downhole environments, effectively suppresses background noise interference, stably detects personnel targets at multiple scales, and meets the needs of real-time monitoring in downhole environments.
Smart Images
Figure CN122416486A_ABST