Endmember Extraction of Hyperspectral Remote Sensing Images Based on an Improved Discrete Artificial Bee Colony Algorithm and Genetic Algorithm

Aiming at improving the performance of the endmember extraction problem in hyperspectral images, a new extraction method based on discrete hybrid artificial bee colony algorithm and genetic algorithm (DABC_GA) is proposed. By analyzing the characteristic of the problem, each dimension of candidate solution is a discrete and exclusive integer. Then we employ an optimization method with integral coding. By inheriting the strong exploration ability of the traditional artificial bee colony algorithm (ABC), we propose a discrete ABC which could quickly obtain more valuable endmembers combinations in the early stage. Then we select some outstanding results of DABC as the potential solutions of GA, which is adopted as another optimization tool in the later stage of iteration. The concept of complementary sets is proposed in the cross and mutation operators to guarantee the diversity and completeness of solutions. Meanwhile, the greedy strategy is adopted to ensure that the favorable potential solutions are not discarded. Compared with conventional extraction algorithms in simulated and real hyperspectral remote sensing data, the experimental results show the validity of our proposed algorithm.