Spectral-Spatial Hyperspectral Image Classification with Superpixel Pattern and Extreme Learning Machine

Yongshan Zhang,Zhihua Cai,Xinwei Jiang,Xinxin Wang

doi:10.3390/rs11171983

Abstract

Spectral-spatial classification of hyperspectral images (HSIs) has recently attracted great attention in the research domain of remote sensing. It is well-known that, in remote sensing applications, spectral features are the fundamental information and spatial patterns provide the complementary information. With both spectral features and spatial patterns, hyperspectral image (HSI) applications can be fully explored and the classification performance can be greatly improved. In reality, spatial patterns can be extracted to represent a line, a clustering of points or image texture, which denote the local or global spatial characteristic of HSIs. In this paper, we propose a spectral-spatial HSI classification model based on superpixel pattern (SP) and kernel based extreme learning machine (KELM), called SP-KELM, to identify the land covers of pixels in HSIs. In the proposed SP-KELM model, superpixel pattern features are extracted by an advanced principal component analysis (PCA), which is based on superpixel segmentation in HSIs and used to denote spatial information. The KELM method is then employed to be a classifier in the proposed spectral-spatial model with both the original spectral features and the extracted spatial pattern features. Experimental results on three publicly available HSI datasets verify the effectiveness of the proposed SP-KELM model, with the performance improvement of 10% over the spectral approaches.

Highlights

Hyperspectral images (HSIs) are acquired from different spaceborne or airborne sensors, where each pixel contains hundreds of spectral channels from ultraviolet to infrared [1,2] and have been an important tool in many hyperspectral image (HSI) applications [2,3]
In [63], a hierarchical principal component analysis (PCA) approach is presented to reduce the dimensionality of hyperspectral data, where an HSI is partitioned into different spatial domains (i.e., 2 × 2 or 4 × 4 parts of the image)
We propose a new spectral-spatial HSI classification model with superpixel pattern (SP) and kernel based extreme learning machine (KELM), called SP-KELM

Summary

Introduction

Hyperspectral images (HSIs) are acquired from different spaceborne or airborne sensors, where each pixel contains hundreds of spectral channels from ultraviolet to infrared [1,2] and have been an important tool in many HSI applications [2,3]. In [11], composite kernel (CK) machine, called SVM-CK, combines both the spectral and spatial information of HSIs into SVM by using multiple kernels. Afterwards, this framework is extended to ELM and kernel based ELM (KELM), named ELM-CK and KELM-CK, respectively [12]. In [13], local binary pattern (LBP) and KELM are incorporated into a spectral-spatial framework, called LBP-KELM, to exploit the texture spatial information (i.e., edges, corners and spots) for the classification of HSIs, which fully extract local image features. The integrated models are called GMM-MRF [18] and SubMLR-MRF [19], respectively. Researchers work with other spatial features in other manners

Methods

Results

Conclusion