Wavelet support vector machines based on reproducing kernel hilbert space for hyperspectral remote sensing image classification

Some limitations exist in hyperspectral remote sensing image classification by SVM (support vector machine), such as unsatisfactory classification accuracy, difficult kernel parameter selection process and dependence on artificial tricks. In order to solve those problems, the wavelet SVM (WSVM) was proposed based on the investigation to SVM theory, reproducing kernel Hilbert space(RKHS) and the wavelet analysis. The wavelet kernel in RKHS can approximate arbitrary nonlinear functions and effectively handle the impacts of parameter selection. By experimenting the proposed algorithm with hyperspectral image captured by operational modular imaging spectrometer II (OMIS II) data from China and ROSIS data from Italy. The experimental results shown that Coiflet wavelet kernel function was the more effective wavelet in terms of classificiatoin accuracy improvement.