Application of Fast Independent Component Analysis on Extracting the Information of Remote Sensing Imagery

The preprocessing of remote sensing imagery (RSI) has great importance on the results of the classification. In this paper, the algorithm of fast independent component analysis (ICA) and its application to the remote sensing imagery classification are presented, and different parameter has different effect on information extraction with ICA. The remote sensing imagery for experiment is from different areas, in different time by several sensors. In succession, a maximum likelihood estimation (MLE) supervised classification method is used to classify the original images and feature images after ICA. As a result, with distinct characters of original images, choosing varied bands and parameters can get better independent component images. The classification result based on feature image is more credible than on image pixel. As for fast independent component analysis, it can remove the correlation of remote sensing imagery, gain high order statistical independent features however some texture information is lost and have better decorrelation result than PCA, which will make for classification