Hyperspect ral data have a high spect ral resolution for the object s of the earth. However , many analysis approaches of hyperspectral data do not provide a promising result because of it s great data volume and strong cor-relation between it s neighboring bands. Consequently , it rest rict s the efficiency and broad application of high reso-lution data. The research indicates that feature extraction is the highly effective theory and method to optimize hy-perspectral data and information. The result of experiment shows that with a given precision of classification , the reduction in dimensionality without loss of information improves the classifier performance , and helps to achieve the aims of optimal process and effective utilization of hyperspect ral remote sensing data.