Efficient Graph Formulation and Latent Space Integration for Lunar Hyperspectral Image Classification.

This paper presents pioneering research that compares different graph formulation techniques for mapping minerals on the lunar surface from hyperspectral images (HSIs) obtained from the Imaging Infrared Spectrometer (IIRS) sensor on-board the Indian Space Research Organisation's (ISRO) Chandrayaan-2 mission. The main focus of the work is on Graph Convolutional Network (GCN). Additionally, we explore the effectiveness of latent space projection for feature extraction and generalization purposes. By leveraging efficient graph formulation techniques and incorporating latent space projection, we aim to improve the interpretability and performance of the GCN in the context of lunar surface mineral mapping. Comparative analyses including fixed adjacency matrix and spectral clustering are used to identify the most efficient approach.