Base Vector Selection Method Based on Iterative Weighted Eigenvector Fitting

摘要

The selection of base vectors is important to linear expression of hyperspectral imagery. There exists several techniques for determination of representative vectors, but the selected results of them cannot act as good base vectors in usual. In this paper, a base vector selection method is constructed based on iterative weighted eigenvector fitting (IWEF). Beginning with an initial combination of vectors, the method tries to substitute each vector for each selected vector to reduce the fitting error. This procedure is iterated until no more valid replacements are done. In order to further reduce its computational cost, principal component analysis and kernel trick are used in data preprocessing. Experiments on synthetic data and on truth hyperspectral data prove the efficiency of the proposed method.