Spectral-spatial classification of hyperspectral image with locality discriminative embedding kernel extreme learning machine

Abstract

Previous studies have demonstrated that the classification performance of remote sensing images can be effectively improved by using its spatial characteristics. However, How to effectively use the discriminant information and geometric structure information of hyperspectral image data is still a challenge. In this paper, we propose a joint spectral-spatial hyperspectral image classification (HSI) classification model based on locality discriminative embedding kernel extreme learning machine (LDEKELM). Firstly, in order to make full use of spatial information of HSI, we use hierarchical guidance filtering to extract spatial features of remote sensing images. Secondly, we use LDEKELM to classify remote sensing images with joint spectral and spatial information. LDEKELM introduces the discrimination information and manifold structure of data into KELM, which can reveal the underlying essential geometric structure of data, especially the local geometric structure and local discrimination information of data. It optimizes the projection direction of KELM output weight and enhances the classification performance of KELM in HSI. Experimental results on three benchmark HSI datasets show that LDEKELM can effectively improve the classification performance of remote sensing images.