Navigating Noise in Chandrayaan-2’s IIRS Hyperspectral Data: Advancing Quality via Deep Learning and SURE Loss

This study centers around the refinement of hyperspectral images obtained from the Chandrayaan-2 Imaging Infrared Spectrometer (IIRS) dataset, which are susceptible to diverse sources of noise, including sensor noise and distortions. The existence of such noise sources presents considerable challenges during data interpretation and analysis. To tackle this concern, we propose an innovative approach grounded in deep learning. This method integrates a Convolutional Neural Network (CNN) architecture with Stein’s Unbiased Risk Estimate (SURE) loss function. Through the utilization of the CNN-SURE framework for model training, our objective is to effectively eliminate noise disturbances and elevate the quality of hyperspectral images. Rigorous experimental assessments have been conducted on both synthetic and real-world datasets. These evaluations illustrate the effectiveness of our approach in mitigating noise artifacts, thereby enhancing data interpretability. The denoised hyperspectral images produced by our methodology hold substantial promise for diverse applications within remote sensing and associated domains.