The 3D Modeling Technology for Substation Surface Temperature Distribution Based on Thermal-Geometric Joint

Against the "Digital New Infrastructure" backdrop, traditional substation maintenance fails to detect early equipment defects (e.g., porcelain bushings with submm damage and thermal differences of several °C) using visual methods. This paper presents a dual-branch thermal-geometric 3D modeling method based on infrared images, integrating 3D Gaussian splatting. The thermal branch extracts edge/gradient features via multi-scale analysis and heat constraints, while the geometric branch estimates pseudo-depth/normal vectors using polarization and Transformer-based prediction. Cross-branch supervision ensures alignment through the edge/normal consistency.Experiments on UAV datasets show the method achieves thermal-geometric consistent 3D reconstruction, outperforming Colmap/3DGS in PSNR/SSIM, reducing lidar reliance, and enabling precise thermal anomaly localization (e.g., micro-cracks via gradients) for intelligent inspection and early warning.