A UAV Thermal Infrared Image Super‐Resolution Method Based on Diffusion Models and Visible Image Texture Transfer

Due to hardware limitations of Thermal Infrared (TIR) cameras, TIR images captured by Unmanned Aerial Vehicles (UAVs) suffer from Low Resolution (LR) and blurred textures. Improving the spatial resolution of TIR images is of great significance for subsequent applications. Existing image Super-Resolution (SR) methods rely on High-Resolution (HR) ground truth for supervised training, resulting in poor general-ization ability. They also lack constraints on the temperature information of TIR imag-es, failing to maintain the consistency of temperature information reconstruction. To address these two issues, this paper proposes a UAV TIR image SR method based on diffusion models and cross-modal texture transfer, which introduces HR information from Visible (VIS) images into TIR images. Firstly, a Multi-Stage Decomposition Latent Low-Rank Representation (MS-DLatLRR) method is adopted to extract multi-scale de-tailed textures from VIS images. Secondly, prior information of object thermal radia-tion is introduced, and combined with the segmentation map of VIS images, a guided coefficient map for VIS multi-scale detailed texture transfer is constructed to provide constraints for temperature consistency during the cross-modal texture transfer pro-cess. Finally, the multi-scale detailed textures and the guided coefficient map are in-troduced into a diffusion model (MP-DDNM) for SR processing of TIR images. Exper-imental results show that compared with existing methods, the proposed method im-proves the resolution of UAV TIR images while maintaining the consistency of tem-perature information as much as possible.