The basic premise of hyperspectral data classification is that the spectral features of various materials are different from each other. With the development of sensor technology , hyperspectral sensor can collect in as many as several hundreds spectral bands at once. The hyperspectral data, which features high spectral resolution, also brings great difficult ies to classification. Because of the high dimension of data, great confusion and limited samples, the performance of traditional algorithms in hyperspectral data classification is deteriorated. In this paper a new combined algorithm based on independent component analysis( ICA) and neural network is proposed. Firstly, the independent component analysis algorithm is
used for feature abstraction. The spectral features are assumed to be a linear mixture of constituent spectra from the material types. The independent components are informative for classification, meanwhile the number of independent components is much smaller than the bands of original data. Then the multi-layer feed-forward neural network is used for classification. The spectral features of each pixel are used to be the input vector of neural network. The neural network is trained by Levenberg-Marquardt back-propagation ( LMBP) algorithm. The classification precision is remarkably superior to that of the conventional bayes classsifier. It proves by the experiment upon the 220-band hyperspectral data that the performance of this
combined algorithm is satisfying .