A NEW SPECTRAL-SPATIAL SUBSPACE CLUSTERING ALGORITHM FOR HYPERSPECTRAL IMAGE ANALYSIS
In the past decade, hyperspectral imaging techniques have been widely used in various applications to acquire high spectral-spatial resolution images from different objects and materials. Although hyperspectral images (HSIs) are useful tools to obtain valuable information from different materials, the processing of such data is challenging due to several reasons such as the high dimensionality and redundancy of the feature space. Therefore, advanced machine learning algorithms have been developed to analyse HSIs. Among the developed algorithms, unsupervised learning techniques have become popular since they are capable of processing HSIs without having prior knowledge. Generally, unsupervised learning algorithms analyse HSIs based on spectral information. However, in many applications, spatial information plays an eminent role, in particular when the input data is of high spatial resolution. In this study, we propose a new clustering approach by utilizing the sparse subspace-based concept within the hidden Markov random field (HMRF) structure to process HSIs in an unsupervised manner. The qualitative analyses of the obtained clustering results show that the proposed spectral-spatial clustering algorithm outperforms the sparse subspace-based clustering algorithm that only uses spectral information.