Multi‐temporal remote sensing change detection based on independent component analysis

Abstract

A change detection approach based on independence component analysis (ICA) was proposed in this letter. Traditional multivariate change detection schemes such as principal component analysis (PCA) were based on 2nd‐order statistics to remove the correlation among multi‐temporal images. However, for the regions where data did not fit a normal distribution, PCA might not be effective. In this letter, ICA was used to separate change information in independent components by reducing the 2nd‐order and higher order dependences in multi‐temporal images. Firstly, the number of images was expanded to the number of land classes by nonlinear band generation. Then, independent component images were obtained based on ICA. The obtained independent component images corresponded to some kind of land or land variation. At last, different kinds of land variation are located by applying maximum likelihood classification (MLC). The experimental results in synthetic and real multi‐temporal images show the effectiveness of the proposed approach.