Eliminating Massive Martian Dust Storms from Images of Tianwen-1 via Deep Learning

Abstract Dust storms may remarkably degrade the imaging quality of Martian orbiters and delay progress mapping global topography geomorphology. To address this issue, paper presents an approach that reuses image dehazing knowledge obtained on Earth to resolve dust-removal problem Mars. In approach, we collect remote-sensing images captured by Tianwen-1 manually select hundreds clean dusty images. Inspired haze formation process Earth, formulate a similar visual degradation synthesize sharing feature distribution with realistic These synthetic pairs are used train deep model inherently encodes irrelevant features dust decodes them into dust-free Qualitative quantitative results show can be effectively eliminated proposed leading obviously improved topographical geomorphological details