Comparison of Pixels Unmixing Approaches and Application to MODIS

Decomposition of mixture pixels was the keystone and nodus for processing of remote sensing image data. This article compared four kinds of unmixing approaches to surface cover image: linear spectral mixed model (LSMM), fuzzy C-means (FCM) approach, neural network (NN) approach and support vector machines (SVMs) approach. All approaches were applied to Moderate Resolution Imaging Spectroradiometer (MODIS) 1B image and simulative image which is filtered from Enhanced Thematic Mapper (ETM) image with 7×7 windows. In general, end-member and parameter selection to models are sensitive to the precision of unminxg approaches. The neural network approach behaves best performance, and the linear spectral mixed model follows. Lacking experience and prior knowledge of parameter selection, the performance of support vector machines approach does not report finer.