Hyperspectral Feature Extraction using Selective PCA based on Genetic Algorithm with Subgroups

Feature selection and extraction are important applications of hyperspectral images. In this paper, a new algorithm, selective principal component analysis based on genetic algorithm with subgroups (SPCA-GAS) is proposed. This algorithm has the distinct characteristic of combining feature selection and extraction together via the introduction of subgroup concept. SPCA-GAS tries to solve the problem of how many and which bands should be selected for feature extraction (principal component analysis, PCA), and then the transformed features are used for classification. Concretely, bands with high separability are first selected and then transformed to remove correlation between them. In this way, useful features with high separability and no correlation are obtained, which help to improve classification accuracies compared with traditional feature selection and extraction methods. Numerical experiments are conducted on AVIRIS data with 100 bands, and satisfying results are reached by SPCA-GAS