An adaptive biogeography-based optimization with integrated covariance matrix learning for robust visual object tracking

Biogeography-based optimization (BBO) has gained significant popularity as a population-based metaheuristic optimization algorithm. However, the existing variants of BBO encounter difficulties when tackling complex optimization problems with variable coupling features. This is primarily attributed to the rotational variability of the migration operator, which hampers its ability to accurately capture and utilize variable coupling features among decision variables. To overcome this challenge, we propose a solution called integrated covariance matrix learning (ICML), which utilizes distribution information from subsets of current population to capture variable coupling features. ICML employs local distribution information to guide BBO towards a globally optimal solution by building eigen coordinate systems. This enables individuals to identify a more optimal value based on their nearby distribution. A new class of BBOs, referred to as ICML-BBOs, is presented by embedding ICML into existing BBO variants. The performance of ICML is evaluated by applying it to original BBO and three BBO variants, enabling a comprehensive performance comparison. Experimental results on the CEC2005 and CEC2017, as well as a real-world robust visual object tracking optimization problem, showcase the effectiveness of ICML.

Duke Scholars

Author Feng Tian DKU Faculty