Reinforcement Learning Based Golden Jackal Optimization with A Sine Cosine Algorithm for Multilevel Image Segmentation

Multi-threshold image segmentation is an important but complex step in image processing. Although existing optimization algorithms have made significant progress in solving this problem, they still face challenges in terms of the robustness of segmentation results and the ability of jumping out of the local optimal solutions. To solve these problems, this study designs an improved golden jackal optimization with a sine cosine algorithm(SCGJO) based on reinforcement learning(RLSCGJO) for multi-level thresholding. To evaluate the optimization capability and robustness of RLSCGJO, we use Shannon entropy as a metric and compare it with other classical and advanced optimization algorithms such as particle swarm optimization (PSO), marine predators algorithm (MPA), sine cosine algorithm (SCA), gold rush optimizer (GRO), and SCGJO. The Berkeley Segmentation Dataset (BSDS0500) and several classical images are tested. The experimental results show that the proposed RLSCGJO has better performance than its competitors at different threshold levels, demonstrating its strong robustness, excellent optimization capabilities, and friendly resource consuming, which provides potential applications for multi-level thresholding in a more general scope of images.