3D aerial thermal mapping of smouldering coal gangue dump based on GAN super-resolution reconstruction and RGB-T image fusion and detail enhancement

Coal gangue dump fires pose significant threats to ecosystems and human health. Accurately identifying fire boundaries remains challenging, complicating fire suppression and emphasizing the need for precise detection methods. Thermal infrared (TIR) remote sensing is an important tool for detecting coal fires and reconstructing three-dimensional (3D) temperature fields in affected areas. However, TIR imagery is limited by low resolution, poor contrast, weak detail representation, and blurred contours. To overcome these limitations, this study proposes a method based on generative adversarial network (GAN) super-resolution reconstruction and RGB-thermal (RGB-T) image feature fusion to improve 3D thermal imaging. A high-resolution TIR dataset was created, and a GAN-based model was used to optimize the upsampling of low-resolution TIR images. Multi-sensor calibration was then performed to align the RGB and TIR image planes, enabling RGB-derived structural details to be integrated into TIR images. The method was applied at Yuxing Coal Mine, where aerial thermal remote sensing of a smouldering coal gangue dump was used to construct 3D visible-light and TIR models. The results demonstrate that the proposed method effectively increases the number of keypoints, improves keypoint identification, and enhances the detail representation of the TIR model. This method provides a visualized 3D temperature field for risk assessment and analysis, thereby supporting the control of coal gangue dump fires.