Wrapper based feature selection in hyperspectral image data using self-adaptive differential evolution

Hyperspectral sensors acquire a set of images from hundreds of narrow and contiguous bands of electromagnetic spectrum from visible to infrared regions. The computational complexity is very high for classification of hyperspectral images due to the presence of large number of bands. In such a scenario, feature selection is very essential technique for reducing the dimensionality. In the proposed work, an attempt has been made to develop a feature selection technique based on evolutionary approach. Self-adaptive differential evolution (SADE) is used for searching feature subset. In SADE, the parameter values adapt themselves with generation to generation. Proposed method follows wrapper model for subset evaluation. Fuzzy kNN classifier is incorporated to calculate the classification accuracy which is used as evaluation criterion. The proposed methodology also includes a feature estimating technique, called ReliefF method, for removing the redundant feature. To demonstrate the effectiveness of the proposed method, results are compared with differential evolution based, genetic algorithm based and ant colony optimization based feature selection techniques. This method achieves very promising results compared to others in terms of overall classification accuracy and Kappa coefficient.