Towards Clearer Mars Images: Self-Supervised Denoising with Large Vision Model
Jiawei Wang, Hui Tian, Junjie Li, Wentao Hu, Xinyuan Li, Weijie Yue · 2024
Mars images are important scientific data in Mars exploration, whose study can contributes to the understanding of Mars’ geological features, environmental conditions, and meteorological phenomena. However, due to sensor equipment and the imaging environment, the observed Mars images are often inevitably corrupted by various noise. The challenge of denoising Mars images lies in the zero-shot paradigm and the lack of prior knowledge. In this paper, we propose a self-supervised model called SAM-MD. We integrate the prior knowledge of large vision model into the proposed model, so as to denoise single Mars image without any training data or knowledge of the noise distribution. Experimental results on Mars images and Tiangong-2 remotely sensed imagery show the reliability and superiority of SAM-MD.