A framework for region based quantitative mapping using hybrid constrained PSO based approach

Abstract

In hyperspectral imaging, spectral unmixing and classification of the pixels are some of the major post-processing operations. The spectral unmixing operation is used to map the pixels quantitatively. In general, it is noticed that the algorithm computes abundance fractions of some endmembers computationally, but does not exist in such part of a real scene. These endmembers may be available in other parts of the real scene. To address this issue, a framework is proposed to do quantitative mapping of the data. First, divide the data into the regions of equal pixels size. Subsequently, hybrid constrained PSO based approach is applied for mapping pixels quantitatively. Combination of Spectral Angle Mapper (SAM) and PSO based approach are used for quantitative mapping respectively. For mapping, fully constrained supervised linear mixing model is considered to estimate the abundance fractions. In this work, hybridization of SAM and PSO is done in order to perform the mapping of pixels quantitatively. The proposed framework is tested over synthetic data and has been performing well.