An advanced denoising methodology for Martian surface mineral exploration

Hyperspectral imaging of Mars with a high signal-to-noise ratio is crucial for accurate analysis of Martian surface minerals. However, the presence of inevitable noise presents significant challenges in mineral identification. This study introduces E2E-CRISM, an efficient self-supervised denoiser for global CRISM data. More specifically, we project Martian hyperspectral images onto a subspace to remove partial noise and reduce computational reliance. Additionally, we develop an eigenimage-guided neighborhood column sampler to generate training samples from noisy data for learning convolutional neural networks training purposes. E2E-CRISM effectively retrieves accurate mineral information from noisy spectra without excessive smoothing or fabrication of absorption peak features, providing a viable solution for detecting low-abundance minerals that cannot be directly identified by current methods. We demonstrate the superior performance of E2E-CRISM in surface mineral identification on Mars; it offers efficient and rapid processing, enabling easy mineral identification and mapping across the global Martian surface.