Jointed endmember extraction algorithm and hyperspectral unmixing analysis

This paper presents a semi-automatic jointed algorithm to extract endmember (pure pixel) and feature spectrum from hyperspectral remotely sensed imagery, and proposes the process of hyperspectral unmixing analysis. First of all, a jointed endmember extracting and interactive algorithm which based on Pixel Purity Index (PPI), scatter plots and Ndimensional visualization were used to extract endmember of identity ground targets; the point which located in the edge and edge node were considered as endmember of identity ground targets by an interactive method. Secondly, the Linear Spectral Mixing Model (LSMM) was used for hyperspectral unmixing analysis based on the extracted endmember information; and the residual error and root mean square error (RMS) was selected for the evaluation model. Thirdly, Spectral Angle Match (SAM) algorithm was introduced to match the endmember with the pixels, at the same time, the matching threshold were adjusted interactively. At last, the proportion of endmember is estimated and the abundance maps of each endmember were derived. From the experiment, it has shown that the proposed jointed endmember extraction and unmixing analysis algorithm performs as well as or even better than the commonly used algorithms.