Estimation of Spectral Abundance Fractions using Fixed Acceleration Coefficients PSO Approach

The occurrence of mixed pixels is common in hyperspectral data. It is necessary to analyse mixed pixels for classification, detection, discrimination, and quantification. Spectral unmixing is needed for mixed pixel analysis of the hyperspectral data. It includes endmember extraction and abundance estimation of mixed pixels. In this work, fixed acceleration coefficients based PSO approach is applied and analysed for abundance fractions estimation of endmembers in spectral unmixing. Time varying inertia weight strategy and fixed acceleration coefficient values have been used in this approach. For estimation, supervised linear mixing model is considered, following sum-to-one and non-negative constraints, respectively. A proposed approach is tested over real hyperspectral data i.e., jasper ridge dataset. The performance metrics of the approach are Average Abundance Error (AAE) and Root Mean Square Error (RMSE). AAE and RMSE values have been noted over different number of iterations. It is observed that result of fixed acceleration coefficients based PSO approach is promising,