Linear Spectral Mixing Model-Guided Artificial Bee Colony Method for Endmember Generation

Abstract

Endmember extraction (EE) is one important step in hyperspectral unmixing. However, some EE methods under pure-pixel assumption may work badly for highly mixed data due to the complexity of image data. In this work, we propose a linear spectral mixing model-guided artificial bee colony (LSMM-ABC) method for EE to solve the problem under a highly mixed situation. The main innovative point of this work is that each employed bee in LSMM-ABC searches food source position guided by the LSMM, rather than with a neighbor food source position. What is more, this proposed LSMM-ABC is not confined to the pure-pixel assumption. The LSMM could help employed bees to find a better solution in endmember generation based on the ABC algorithm. Experimental results on both synthetic and real Cuprite data sets show us that the proposed LSMM-ABC method can improve the overall EE accuracy compared with the EE methods for highly mixed data.