Guided Texture Transfer Network for Mid-Infrared Satellite Image Super Resolution

Mid-infrared (MIR) satellite imagery captures thermal information and supports a wide range of remote sensing applications. However, its inherently low spatial resolution limits its utility for detailed spatial analysis. In this work, we formulate MIR super-resolution as a guided super-resolution task using geometrically aligned high-resolution RGB images. We present a Guided Texture transFormer (GTFormer) that transfers fine textures from RGB to MIR while preserving thermal semantics. Also, we propose a two-stage learning strategy to prevent the model from simply copying guidance values. We evaluate the proposed method on real satellite data from KOMPSAT-3A and KOMPSAT-2. Extensive experiments demonstrate that our method outperforms state-of-the-art techniques in both visual quality and thermal information preservation.