Attribute Profiles of Different Attributes for Spectral-Spatial Classification of Hyperspectral Imagery

Incorporation of spatial information along with spectral features has a great impact on the analysis of hyperspectral images (HSIs). Attribute profiles have been proved to be a very effective state-of-the-art method for fusing spectral and spatial information in HSI. This paper recalls the construction of attribute profiles (APs) for HSI classification and provides a comprehensive analysis of different properties of connected components that can be used as an attribute while constructing APs. Various attributes have been widely and successfully used to construct attribute profiles for spectral-spatial classification of HSIs. In this paper, we investigate several unexplored attributes in addition to the existing ones. An empirical study on the attribute profiles, constructed by using different attributes, is carried out considering three real HSI data sets. Our study reveals that each attribute possesses interesting filtering capabilities among which the attributes namely complexity and diameter of equivalent circle, which are new in the HSI literature, are found most promising to incorporate spatial information for HSI classification. The paper also lists out suitable attributes for recognition of different classes in a remote sensing scene.