TSDM-Fusion: a degradation-robust infrared and visible image fusion model with teacher-student learning

Infrared (IR) and visible (VIS) image fusion plays a critical role in surveillance, night vision, and object detection, yet its performance often degrades due to real-world factors like noise, blur, and compression. This paper proposes a degradation-robust fusion model (TSDM-Fusion) based on a teacher-student framework. By constructing a composite degradation model that incorporates Gaussian noise, blurring, JPEG compression, and luminance-contrast variations, we simulate real-world image degradations and enhance the model’s adaptability through teacher-student learning. Experiments on public datasets (TNO, RoadScene, LLVIP, and M3FD) demonstrate that TSDM-Fusion outperforms state-of-the-art methods in metrics such as SSIM and MS-SSIM while maintaining robustness against diverse degradations.