Spectral Unmixing-Based Clustering of High-Spatial Resolution Hyperspectral Imagery

Abstract

This paper introduces a novel unsupervised spectral unmixing-based clustering method for high-spatial resolution hyperspectral images (HSIs). In contrast to most clustering methods reported so far, which are applied on the spectral signature representations of the image pixels, the idea in the proposed method is to apply clustering on the abundance representations of the pixels. Specifically, the proposed method comprises two main processing stages namely: an unmixing stage (consisting of the endmember extraction and abundance estimation (AE) substages) and a clustering stage. In the former stage, suitable endmembers are selected first as the most representative pure pixels. Then, the spectral signature of each pixel is expressed as a linear combination of the endmembers’ spectral signatures and the pixel itself is represented by the relative abundance vector, which is estimated via an efficient AE algorithm. The resulting abundance vectors associated with the HSI pixels are next fed to the clustering stage. Eventually, the pixels are grouped into clusters, in terms of their associated abundance vectors and not their spectral signatures.

Existing System

Discrete Particle Swarm Optimization Algorithm, Clustering Algorithm.

Proposed System

The key challenge of the proposed method spectral unmixing-based clustering (SUBC) is the identification of spatially homogeneous regions comprising different materials. The method consists of two main stages (unmixing and clustering) and generates three significant (by)products, namely: 1) endmembers; 2) abundance vectors (abundance maps); and 3) clusters (classification maps). The key feature of SUBC is the utilization of the abundance representations of the HSI pixels (as they result from the unmixing stage) in the clustering stage. The advantage of using the abundance representation instead of the basic spectral representation of the pixels is that the former, in contrast to the latter, provides subpixel level information, which in turn favors more detailed classification maps. Moreover, the abundance representation is likely to give rise to more well-discriminated clusters that live on subspaces of the abundance space, due to the fact that only a few materials are expected to contribute to the formation of a HSI pixel (sparsity issue). As a consequence, subspace clustering algorithms could also be considered as an alternative in the final stage of the algorithm, since the abundance representations are likely to lead to clusters that live to subspaces of the abundance space. SUBC is unsupervised and does not require class information knowledge of the dataset under study. Moreover, it is image independent, it alleviates the “curse of dimensionality” issue and enhances localization and accuracy since it operates in the subpixel level of information. However, it is noted again that the correct identification of the endmembers number and their correspondence to physical objects/materials is undoubtedly the most critical step for successful SU and, as a consequence, for the clustering processes.

Architecture

BLOCK DIAGRAM