摘要
爬行动物搜索算法(Reptile Search Algorithm,RSA)存在易陷入局部最优、收敛精度不足、收敛速度慢等问题.针对该问题,提出一个改进爬行动物搜索算法(Improved Reptile Search Algorithm IRSA),并将其应用于多阈值图像分割领域.IRSA采用拉丁超立方抽样优化种群初始化以增强多样性,引入莱维飞行利用其长步长与随机方向特性有助于跳出局部最优,同时结合正余弦算法的更新策略来平衡全局探索与局部开发.实验选取Berkeley图像库中的3幅图片,将IRSA与其他四个优化算法进行对比.结果表明,IRSA在收敛速度、收敛精度方面优于对比算法,在峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)、结构相似性指数(Structural Similarity Index Measure,SSIM)和特征相似性指数(Feature Similarity Index Measure,FSiM)三个图像质量评价指标上也优势明显,验证了其在多阈值分割图像任务中的有效性与优越性.
The Reptile Search Algorithm(RSA)has problems such as being prone to getting stuck in local optima,insufficient convergence accuracy,and slow convergence speed.To address this issue,an improved Reptile Search Algorithm(IRSA)is proposed and applied to the field of multithreshold image segmentation.IRSA uses Latin hypercube sampling to optimize population initialization to enhance diversity,and introduces Levy flight to utilize its long step size and random directional characteristics to help jump out of local optima.At the same time,it combines the update strategy of sine cosine algorithm to balance global exploration and local development.Three images from the Berkeley image library were selected for the experiment,and IRSA was compared with four other optimization algorithms.The results show that IRSA outperforms the compared algorithms in terms of convergence speed and accuracy.It also has significant advantages in three image quality evaluation metrics:Peak Signal to Noise Ratio(PSNR),Structural Similarity Index Measure(SSIM),and Feature Similarity Index Measure(FSIM),which verifies its effectiveness and superiority in multi-threshold image segmentation tasks.
作者
张小萍
ZHANG Xiaoping(School of Computer,Electronics and Information,Guangxi University,Nanning Guangxi 530004,China)
出处
《太原师范学院学报(自然科学版)》
2025年第4期9-17,共9页
Journal of Taiyuan Normal University(Natural Science Edition)
基金
广西重点研发计划(2024AB33478).
关键词
爬行动物搜索算法
多阈值分割
拉丁超立方抽样
莱维飞行
正余弦算法
reptile search algorithm
multi-threshold segmentation
Latin hypercube sampling
Levy flight
sine cosine algorithm
