Abstract

This dissertation presents a novel Bi-objective Multi-population Genetic Algorithm (BMPGA) for multimodal optimization problems. BMPGA is distinguished by its use of two separate but complementary fitness objectives designed to enhance the diversity of the overall population and exploration of the search space. This is coupled with a multi-population strategy and a clustering scheme, both of which together focus selection pressure within sub-populations, resulting in improved exploitation of promising optimum areas as well as effective identification and retention of potential optima. The practical value of BMPGA is demonstrated in several applications. In optimization of benchmark multimodal functions, it shows clear superiority over other typical multimodal GAs: Multinational GA [1], Dynamic Niche Clustering [2] and Clearing [3], with respect to overall effectiveness, general applicability and reliability. In the application of imagery ellipses detection, BMPGA is compared with both widely used Randomized Hough Transform (RHT) [4] and Sharing Genetic Algorithm (SGA) [5]. In thorough and fair experimental tests, utilizing both synthetic and real-world images, BMPGA exhibits solid advantages over RHT and SGA in terms of accuracy of recognition - even in the presence of noise or/and multiple imperfect ellipses in an image - and speed of computation. Finally, we successfully extend BMPGA to the segmentation of microscopic cells, which is a necessary first step of many automated biomedical image processing procedures