Genetic Algorithm Framework for 3D Discrete Wavelet Transform based Hyperspectral Image Classification

Joint spatial-spectral feature extraction process is always playing a vital role in the accurate classification of hyperspectral imagery. Such feature extraction techniques are ever demanded for hyperspectral classification. In this proposed work three dimensional DWT (3D-DWT) is used for the decomposition of the hyperspectral image and 3D gray level cooccurrence matrix (GLCM) features are extracted for obtaining the neighborhood information. A genetic Algorithm is incorporated in this work for the selection of the best features among the extracted features for yielding good classification accuracy. The proposed method is experimented on airborne visible infrared imaging sensor (AVIRIS) data of the Indian pine site and reflective optics system imaging spectrometer (ROSIS) data of the Pavia University site. The results witness the accuracy of 94.62% for the Indian pines dataset and 96.48% for University of Pavia dataset before feature selection while only 5% of the samples in each class were used for training the 3D DWT based GLCM features. After incorporating the Genetic Algorithm for selecting the best features the accuracy is increased up to 97.67% for the Indian pines dataset and 97.99% for the University of Pavia dataset respectively, for the same 5% training samples. The proposed method is compared with the other methods and found to be more efficient.