Applying composite kernel to kernel-based nonparametric weighted feature extraction

Abstract

In the recent researches show that nonparametric weighted feature extraction (NWFE) is a useful method for extracting hyperspectral image features. Kernel-based NWFE (KNWFE) is applying the kernel method to extend the more effective projected features in the feature space. It had been showed the performance of KNWFE is better than NWFE. In this study, we would apply a composite kernel function with spectral and spatial information to KNWFE, and hope this composite kernel to KNWFE can get a better performance than the spectral-based kernel function to KNWFE. In the experiment results show that the KNWFE with composite kernel, include the spectral and spatial information, outperforms the KNWFE with the only spectral based kernel function.