Abstract: Hyperspectral remote sensing image has the problem of high dimensionahty. A new hyperspectral image terrain classification method, i. c.，direct LDA subspace method, was presented. Firstly, direct linear discriminant analysis(direct LDA) was used to extract features in original high dimensional hyperspectral space, and then shortest distance classifier was used to perform terrain classification in the feature subspace. Recognition results based on airborne visible/infrared imaging spectrometer(AVIRIS) hyperspectral image show that the presented method can remarkably reduce data dimensionality and improve recognition efficiency.

Key words: Terrain classification,Feature subspace,Fcature extraction, Hyperspectral image