Unsupervised Geometric Learning of Hyperspectral Images

The problem of unsupervised learning and segmentation of hyperspectral images is a significant challenge in remote sensing. The high dimensionality of hyperspectral data, presence of substantial noise, and overlap of classes all contribute to the difficulty of automatically segmenting and clustering hyperspectral images. In this article, we propose an unsupervised learning technique that combines a density-based estimation of class modes with partial least squares regression (PLSR) on the learned modes. The density-based learning incorporates the geometry of the hyperspectral data by using diffusion distance to promote learning a unique mode from each class. These class modes are then used to generate class cores which approximate training labels. Partial least squares regression using these learned class cores as labeled training points consequently determines a labeling of the entire dataset. The proposed method is shown to perform competitively against state-of-the-art clustering and dimension reduction methods, and often achieves performance comparable to fully supervised PLSR.