Wavelet-Based Semantic Feature Representation for Hyperspectral Signature Discrimination

Hyperspectral signature discrimination is a quantitative analysis approach for hyperspectral imagery which performs detection and classification of the constituent materials at the pixel level in the scene. However, the substantial volume of data in hyperspectral images poses challenges for practitioners in terms of computation and storage, making the management of acquired hyperspectral data difficult. One alternative to directly operating on raw data is to perform classification using extracted features from hyperspectral signatures, such as energy or shape information. Therefore, there is a need to design features that capture scientifically meaningful cues of the materials under study from hyperspectral signatures. The wavelet transform is widely recognized for its ability to highlight signal discontinuities, making it suitable for representing spectral fluctuations with semantic value. By appropriately modeling wavelet coefficients, it becomes possible to distinguish between fluctuations that have discriminatory value and those that do not. In this paper, we describe a technique that applies non-homogeneous hidden Markov chain (NHMC) models to hyperspectral signature discrimination. The basic idea is to use statistical models (such as NHMC) to characterize wavelet coefficients which capture the spectrum semantics (i.e., structural information) at multiple levels. Experimental results show that the approach based on NHMC models can outperform existing approaches relevant in classification tasks.