Improved biogeography-based optimization algorithm and its application to clustering optimization and medical image segmentation

Abstract

In order to improve the optimization efficiency of the biogeography-based optimization (BBO) algorithm, an improved BBO algorithm, that is, worst opposition learning and random-scaled differential mutation BBO (WRBBO), is presented in this paper. First, BBO's mutation operator is deleted to reduce the computational complexity and a more efficient random-scaled differential mutation operator is merged into BBO's migration operator to obtain global search ability. Second, in order to balance exploration and exploitation, the BBO's migration operator is replaced with a dynamic heuristic crossover to enhance the local search ability. Finally, a worst opposition learning is merged into the improved algorithm to avoid trapping into local optima. A large number of experiments are made on 18 various kinds of classic benchmark functions and some complex functions from the CEC-2013 test set. In addition, WRBBO is applied to clustering optimization and medical image segmentation. The experimental results show that WRBBO has better optimization efficiency on benchmark function optimization, clustering optimization, and medical image segmentation than quite a few state-of-the-art BBO variants and other algorithms.