Research On Feature Reduction Of Hyperspectral Remote Sensing Data

Advances in hyperspectral remote sensing have provided an important means for moinitoring the world. The resulting high-dimensional data collected at hundreds of adjoining and narrow wavelengths benifit better discrimination among similar spectral signatures or fingerprints than the traditional multispectral data with low spectral resolution, and have been widely used in aerospace, earth observing, lunar and mars exploration, biomedical engineering etc. However, the vast amount of data volume presents challenging problems for the subsequent information processing. Task-oriented feature reduction has become one of the most important research tasks and attracted more and more attentions. Feature reduction can generally fall into feature extraction or feature selection. Feature extraction transforms the original data from a high dimension into a lower dimension with most of the desired information content preserved, and feature selection tries to identify a subset of original bands for a given task with the physical meanings of original features preserved. Therefore, technologies of feature extraction and feature selection for the hyperspcetral data are systemically and deeply investigated in this dissertation. The research work and the main results are as follows:(1) Margin-based Feature Extraction algorithm (MFE) for hyperspectral data. In view of the facts that for high-dimensional data, the probability of linear separability will grow in case of small samples and the low-dimensional projection is approximately normal, a new feature extraction algorithm, MFE, is proposed for the small sample size problem. MFE introduces a new definition of the margin, which involves not only the between-class scatter and within-class scatter proposed by LDA criterion, but also the diffences of the class variances. Through maximalizing the margin, we can obtain the optimal projection vector, and avoid the small sample size problem due to singularity of the within-class scatter. The alogrithm is further extended to the multi-class problem. The experiment results show that MFE outperforms several improved versions of LDA in the case of small samples, and achieve a satisfying performance for larger samples.(2) Subset Search algorithm based on Hybrid PSO and GA (HPSOGA) for hyperspectral data. Evolutionary computation for feature selection constitutes a different way of looking for features since it allows a randomized search guided by a certain fitness measure, and thus a new subset search algorithm, HPSOGA, is proposed. The algorithm based on particle swarm optimization (PSO) decomposes the update of a...