Modified Principal Component Analysis (MPCA) for feature selection of hyperspectral imagery

Abstract

Principal Component Analysis (PCA) is a classical multivariate data analysis method that is useful in linear feature extraction and data compression. It can compress the most information in the original data space into a few features. Generally, remote sensing image contains (is composed of) many different objects such as land cover classes, but for a specific purpose of remote sensing application, only a few classes may be relevant. In this paper, a new method called Modified Principal Component Analysis (MPCA) is proposed and applied to a DAIS (Digital Airborne Imaging Spectrometer) image acquired in Venice, Italy. The results show that the features form MPCA is more effective in information compression, classes separablity and classification accuracy than those form PCA.