QWT enhanced SVM for Hyperspectral image classification

Higher and higher accuracy is demanded in the development of Hyperspectral images classification technology, which faces the challenge of increasing amount of data. This paper proposes to combine the standard support vector machine (SVM) classification technique, utilized for land-cover classification studies, with the quaternion wavelet transform (QWT) to enhance the classification accuracy of SVM. This novel algorithm applies QWT to generate additional features prior to SVM, which is selected for classifying the images. Furthermore, two simulation experiments on AVIRIS hyperspectral image are conducted for comparing the performances achieved by the proposed QWT enhanced SVM classification method and the original one respectively. The results demonstrate that the improved SVM classification process, which is derived after the application of QWT, is superior to the raw one in relation to the issue of accuracy.