5-Use of Principal Components and Canonical Analysis in Remote Sensing for Classification of Data.

Abstract

Several classification techniques have been used in remote sensing for pattern recognition. These techniques are basically divided into two categories, viz., supervised and unsupervised classification. The classification algorithm identifies the important features of a class, i.e., determines what each class looks like and given the data for a pixel of the imagery, it compares the features of the pixel with the features of each class and then assign the pixel to one of the classes. When multispectral band of data are available, the variances and correlation of spectral responses pattern of different classes are determined from the training data set to identify salient features of the class, to be used in classification. Two classification techniques have been analysed and compared in this study. These techniques are the principal components analyses and the canonical analysis. Both of them are preprocessing techniques. The study has revealed that the internal correlation is of the order of 0.77 to 0.97 and only two orthogonal components can explain 97/0 to 98% of the variance of the original sample. They hence, constitute a data reduction or condensation technique for preprocessing of multi-spectra! and TM data.