Hyperspectral Image Clustering with Spatially-Regularized Ultrametrics

We propose a method for the unsupervised clustering of hyperspectral images based on spatially regularized spectral with ultrametric path distances. The proposed efficiently combines data density and spectral-spatial geometry to distinguish between material classes in data, without need training labels. is efficient, quasilinear scaling number points, enjoys robust theoretical performance guarantees. Extensive experiments synthetic real HSI demonstrate its strong compared benchmark state-of-the-art methods. Indeed, not only achieves excellent labeling accuracy, but also estimates clusters. Thus, unlike almost all existing methods, algorithm essentially parameter-free.