A novel kernel-based nonparametric feature extraction method for remotely sensed hyperspectral image classification

Feature extraction has been an essential technique for enhancing the recognition of patterns. General feature extraction methods are constructed to extract linear features because constructing methods for extracting nonlinear features are not easy. Kernel-based method provides a framework for developing a nonlinear extension of one existing linear feature extraction. A novel kernel-base feature extraction method based on our previously proposed cosine-based nonparametric feature extraction (CNFE) were addressed and named KCNFE. The experimental results show that classifiers 1NN and SVM with KCNFE features can achieves better classification rate than some existing feature extraction methods.