Improved sequential endmember extraction algorithms

Most of sequential endmember extraction algorithms, such as iterative error analysis (IEA), vertex component analysis (VCA), and simplex growing algorithm (SGA), use sequential forward selection (SFS) searching strategy. The advantage is its low computational complexity. However, it is sensitive to the initial condition. To reduce the “nesting effect”, sequential forward floating selection (SFFS) strategy is investigated in this paper. Experimental results show that SFFS can improve the quality of the extracted endmembers without the initial condition problem. In order to reduce the computational cost of SFFS in endmember extraction, we propose a hybrid searching strategy by combining SFS and SFFS, which can produce a similar or even identical endmember set as the original SFFS.