UTIDust: A thermal infrared dataset for dust removal in underground coal mine excavation face

UTIDust is a pioneering, large-scale benchmark dataset specifically engineered for thermal infrared image dust removal within the complex and safety-critical environment of underground coal mine excavation faces. To address the physical impossibility of capturing pixel-aligned data in dynamic mining scenes, the dataset employs a CycleGAN-based unsupervised learning framework to synthesize realistic, high-fidelity dust effects onto real-world thermal backgrounds, resulting in 10,075 strictly paired dusty and clear images with a standardized resolution of 512×512 pixels. The data encompasses a diverse spectrum of operational scenes and is scientifically partitioned into training (8,034 pairs), validation (801 pairs), and testing (1,240 pairs) subsets to facilitate rigorous model evaluation.