Unsupervised Hyperspectral Image Segmentation: A Novel 3-Dimensional Clustering Methodology

Unsupervised learning algorithms prove robust on hyperspectral images (HSI) with unavailable ground truth. We have used unsupervised learning models on the HSIs obtained from the surface reflectance data from the Earth Surface Mineral Dust Source Investigation (EMIT) Imaging Spectrometer. The abundances extracted from the proposed unsupervised method have been illustrated as clusters of where a particular endmember is located in the region of interest. The processing steps utilise a proposed 3-dimensional unsupervised clustering algorithm and generate a map of differently classified areas. Furthermore, various modified silhouette indexes have been used to evaluate clustering, which in this research is 0.91.