Dung beetle optimization with composite population initialization and multi-strategy learning for multi-level threshold image segmentation

Abstract As the number of thresholds increases in multi-threshold segmentation of digital images, the com- plexity of determining the ideal thresholds rises sharply, posing significant challenges for conventional approaches. Dung Beetle Optimization (DBO) is a metaheuristic algorithm that mimics the behav- iors of dung beetles, including rolling dung balls, female beetles laying eggs, small beetles searching for food, and thief beetles stealing. However, the original DBO suffers from slow convergence rate and suboptimal solutions. This paper proposes an improved DBO algorithm, named DBO with composite population initialization and multi-strategy learning (CMDBO), to address the issues. The improve- ments include initializing the population using chaotic mapping and oppositional learning, enabling weaker individuals to learn from better ones, and applying quasi-center oppositional-based learning to enhance convergence rate and solution accuracy. To verify its search performance, CMDBO was tested on CEC2017 function set and compared with several algorithms. Furthermore, CMDBO was applied to multi-threshold image segmentation. Experimental results indicate that the proposed CMDBO achieved the best overall performance in terms of convergence speed and solution accuracy.