Multiple Characteristics Similarity Metric Method for Hyperspectral Image Classification

Abstract

The traditional classification algorithms have been widely applied for hyperspectral imagery (HSI), but many methods exploit spectral or spatial information in HSI data that doesn't make full use of the information of HSI data. To solve this problem, a new classification method, termed multiple characteristics similarity metric (MCSM), was proposed in this paper for HSI classification. In MCSM, a spatial similarity probability relationship, a sparse similarity relationship and a collaborative similarity relationship are integrated to use the multiple information of HSI. At first, a spatial similarity probability is designed by the multiple features fusion, which can reveal the spatial information of HSI. Then, it utilizes sparse representation and collaborative representation to represent the sparse and collaborative properties of HSI. Finally, the class label can be determined by combining spatial similarity probability, sparse similarity and collaborative similarity. To demonstrate the effectiveness of the proposed method, experiments have been conducted on the Indian Pines and Pavia University data sets. The experimental results show that MCSM achieves better classification performance compared with some state-of-the-art classification methods, which indicates that MCSM can make full use of the multiple information of HSI to form the complementarity of different characteristics and enhance the discriminant performance for HSI classification.

Keywords

• Hyperspectral image
• classification
• multiple characteristics
• sparse representation
• collaborative representation