Structured Priors for Sparse-Representation-Based Hyperspectral Image Classification

Abstract

Abstract-Pixel-wise classification, where each pixel is assigned to a predefined class, is one of the most important procedures in hyperspectral image (HSI) analysis. By representing a test pixel as a linear combination of a small subset of labeled pixels, a sparse representation classifier (SRC) gives rather plausible results compared with that of traditional classifiers such as the support vector machine (SVM). Recently, by incorporating additional structured sparsity priors, the second generation SRCs have appeared in the literature and are reported to further improve the performance of HSI. These priors are based on exploiting the spatial dependencies between the neighboring pixels, the inherent structure of the dictionary, or both. In this paper, we review and compare several structured priors for sparserepresentation-based HSI classification. We also propose a new structured prior called the low rank group prior, which can be considered as a modification of the low rank prior. Furthermore, we will investigate how different structured priors improve the result for the HSI classification.

Unsupervised spectral sub-feature learning for hyperspectral image classification

“…A sparse modelling dictionary-based approach has been applied in previous HSI classification methods (Sun et al 2014;Castrodad et al 2011;Chen, Nasrabadi, and Tran 2011), where different types of sparsity constraints have been included in the corresponding dictionary modelling cost functions. The general idea behind these methods is to learn a separate dictionary for each class from labelled data (Sun et al 2014;Castrodad et al 2011), or use the labelled data per se to form dictionaries (Chen, Nasrabadi, and Tran 2011), and classify unknown pixels by determining which classspecific dictionary best describes the sample in terms of minimum value of the reconstruction error.…”

“…The general idea behind these methods is to learn a separate dictionary for each class from labelled data (Sun et al 2014;Castrodad et al 2011), or use the labelled data per se to form dictionaries (Chen, Nasrabadi, and Tran 2011), and classify unknown pixels by determining which classspecific dictionary best describes the sample in terms of minimum value of the reconstruction error. The aforementioned methods use labelled training samples (or at least assume that the classes present in the data set are known a priori (Castrodad et al 2011)) to obtain class-specific dictionaries, whereas our method uses only unlabelled samples and no prior information to learn a single general dictionary.…”

“…Supervised algorithms have recently been developed to take into account both spectral and spatial information in order to improve the classification of multispectral and hyperspectral images (Sun et al 2014;Castrodad et al 2011;Chen, Nasrabadi, and Tran 2011;Zhong, Lin, and Zhang 2014;Ji et al 2014;Tuia et al 2014;Tao et al 2013;Chen et al 2014;Chen, Zhao, and Jia 2015). Spatial information can also be used to complement our proposed method.…”

“…Various feature extraction methods for the purpose of HSI classification have been proposed in the literature, including unsupervised methods (Hotelling 1933;Chang et al 1999;Wang and Chang 2006;Kaewpijit, Le-Moigne, and El-Ghazawi 2003;Tenenbaum, de Silva, and Langford 2000;Roweis and Saul 2000;Belkin and Niyogi 2001;Zhang and Zha 2005;He et al 2005;He and Niyogi 2004;Zhang et al 2007;Qiao, Chen, and Tan 2010), semisupervised (Cai, He, and Han 2007;Chen and Zhang 2011;Sugiyama et al 2010;Liao et al 2013;Shao and Zhang 2014), and supervised methods (Bandos, Bruzzone, and Camps-Valls. 2009;Li and Qian 2011;Kuo and Landgrebe 2004;Kuo et al 2014;Yang et al 2014;Zhong, Lin, and Zhang 2014;Tuia et al 2014;Tao et al 2013;Sugiyama 2007;Chen et al 2014;Chen, Zhao, and Jia 2015;Sun et al 2014;Castrodad et al 2011). …”

“…Therefore, in this article we propose a hyperspectral classification method based on unsupervised learning of spectral sub-features. Supervised methods (Sun et al 2014;Charles, Olshausen, and Rozell 2011;Castrodad et al 2011) using learned dictionaries of spectral pixels have been proposed for hyperspectral data classification in the past. However, unlike the related supervised methods, our unsupervised approach learns feature representations from subsets of spectral data, which we refer to as spectral sub-feature learning.…”