RSHaze5K

Haze interference in remote sensing images significantly degrades image quality and subsequent interpretation accuracy. However, most existing datasets consist of synthetically hazed homogeneous images. To address this, we propose RSHaze5K, a novel large-scale remote sensing image dehazing dataset. RSHaze5K contains 5,000 pairs of high-resolution remote sensing images. Each pair includes a synthetically hazed image generated using real cloud and haze layers as masks combined with a professional atmospheric scattering model, alongside its corresponding haze-free clear reference image. Compared to existing remote sensing dehazing datasets, our proposed RSHaze5K offers a larger volume of data and greater realism. It encompasses diverse terrains, with various haze coverage levels for each terrain type. We conducted quantitative evaluations and ablation experiments on this dataset using multiple mainstream dehazing algorithms. The results highlight the significant value of RSHaze5K in advancing remote sensing dehazing technology. This dataset is intended to serve related research fields such as remote sensing image enhancement, atmospheric correction, and computer vision.