I consider dimensionality reduction for hyperspectral data. Algorithms and results for kernel methods using graph construction and replacing kernel eigenvalue decomposition are given.