Multi-population biogeography-based optimization algorithm and its application to image segmentation

Xinming Zhang,Shaochen Wen,Doudou Wang
DOI: https://doi.org/10.1016/j.asoc.2022.109005
IF: 8.7
2022-01-01
Applied Soft Computing
Abstract:In view of the shortcomings of Worst opposition learning and Random-scaled differential mutation Biogeography-Based Optimization (WRBBO) in solving complex optimization problems, such as insufficient search ability and low search efficiency, an improved WRBBO, Multi-population BBO (MPBBO) is proposed. Firstly, a multi-population strategy is adopted: the whole sorted population is divided into 3 different subgroups (high-level, median-level and low-level) through golden section to get some population diversity. Secondly, an aptitude-based teaching approach is proposed to be beneficial to each subgroup's development: a sinusoidal-scaled differential mutation operator is performed on the high-level subgroup to mostly get stronger exploration and reduce the computing cost, a heuristic crossover with dynamic fine adjustment is hybridized with a horizontal crossover and a vertical one to form a multi-crossover on the median-level subgroup to mainly get stronger exploitation, and a best agent guiding strategy is used on the low-level subgroup to improve the search ability. Finally, an information sharing way is adopted: all the individuals in 3 subgroups share information by merging and sorting them. The experimental results on the complex functions from CEC-2013 and CEC-2017 test sets show that MPBBO obtains stronger search ability and higher efficiency than WRBBO and quite a few other state-of-the-art algorithms. MPBBO is applied to image segmentation with fast and robust fuzzy c-means (FRFCM), and the results show that FRFCM-MPBBO has more significant advantages than its comparison methods.
What problem does this paper attempt to address?