A GA-Based Feature Selection Algorithm for Remote Sensing Images

We present a GA–based feature selection algorithm in which feature subsets are evaluated by means of a separability index. This index is based on a filter method, which allows to estimate statistical properties of the data, independently of the classifier used. More specifically, the defined index uses covariance matrices for evaluating how spread out the probability distributions of data are in a given n−dimensional space. The effectiveness of the approach has been tested on two satellite images and the results have been compared with those obtained without feature selection and with those obtained by using a previously developed GA– based feature selection algorithm.