A Hybrid Chaotic Bat Algorithm With Deep Learning Optimization for Multi-Level Image Segmentation

This paper examines the applicability of the Bat Algorithm (BA) and its variants as optimization frameworks for various image segmentation paradigms. Rather than introducing a new segmentation model, the study systematically evaluates how BA-based optimization affects segmentation performance across grayscale, colour, and semantic segmentation tasks. Chaotic maps and inertia weight strategies are explored as mechanisms to enhance the exploration–exploitation balance of BA. Grayscale image segmentation is addressed through multilevel threshold optimization using Otsu’s criterion; colour image segmentation is formulated as a clustering problem with K-means initialization; and semantic segmentation is approached through hyperparameter optimization of a convolutional neural network. Experimental results indicate that inertia-weighted BA consistently improves grayscale segmentation quality, while standard K-means remains competitive for colour image segmentation. For semantic segmentation, BA-based hyperparameter optimization achieves stable performance without manual tuning. The findings highlight both the strengths and limitations of BA-based optimization in image segmentation and demonstrate that its effectiveness is strongly task-dependent. This study provides a unified experimental perspective on BA variants, offering practical insights into when and how such metaheuristic optimization strategies can be beneficial in image segmentation applications.