A Study Of Vegetation Extraction In Remote Sensing Data Based On ICA Algorithm

The classical method of vegetation monitoring is vegetation indexes. Vegetation index is based on pixels. A pixel represent a vast region and is likely to contain several features. As a result the actual amount of vegetation is different from the amount derived from vegetation index.This paper presents ICA algorithm, a pixel decomposition method. ICA is a multi-channel signal decomposition method and is derived from blind source separation. Based on the assumption that the source signals are independent between each other, ICA can minimize the statistical dependence between components. And then the estimated signals are extracted from mixed signals. ICA is often used to find the hidden factors(source signals or features) from observational data. ICA is Based on higher order statistics and uses space transformation to process data. Other space transformation methods, such as PCA, only use low-level statistics and can only remove the second order statistics.In a earlier time some papers used ICA in remote sensing image processing, but most of them just introduced ICA and very few people did further research. FastICA was used mostly.In this paper eight kinds of ICA algorithm are introduced to process TM image, and the accuracy of extracted vegetation components are compared to find the best ICA algorithm which has the better effect. The results show that Efica has the highest accuracy and needs relatively small time. The accuracy of vegetation component can improved by using MNF denoising method first.This paper also processes Hyperion images using FastICA and does some research. The results show that band selection is essential in order to improve the accuracy of extracted vegetation component from Hyperion data. Using band selection first can reduce the running time and improve the extraction accuracy.