Denoising Lunar Hyperspectral Images Using CNN with Skip Connections

An improved convolutional neural network (CNN) with skip connections and a bottleneck layer is introduced in this study with the aim of hyperspectral image (HSI) denoising. We used Total Variation Chambolle filtering on dataset obtained from the Chandrayaan-2 mission and split the large HSI cube into smaller ones to train our model quickly. By suppressing noise and preserving crucial spectral and spatial information, our methodology can retrieve features from HSIs. We evaluated our proposed methodology using performance evaluation metrics such as Spectral Information Divergence, Peak Signal-to-Noise Ratio, Structural Similarity Index Measure, and Spectral Angle Mapper which performed better in denoising.