Scholarlink

<p>In image segmentation, multi-level threshold segmentation (ML-TS) method is an effective image segmentation method. With the increase of multilevel thresholds, the computation amount of the ML-TS method is higher and higher, and the selection of the best multilevel thresholds is more and more difficult. When using existing intelligent group optimization algorithms to optimize the ML-TS method and select the optimal multi-level thresholds, there are problems such as insufficient optimization efficiency of the intelligent group optimization algorithms, susceptibility to local optima, and difficulty in obtaining the optimal multi-level thresholds. In order to effectively overcome the above problems, the paper proposes a new ML-TS method using Q-Learning based on hybrid whale optimization algorithm and gray wolf optimizer (QL-WOA-GWO). Firstly, the paper conducts a modular analysis of the Q-Learning, treating its the state and action as the state module and action module, respectively. Secondly, in the iteration process of the QL-WOA-GWO, the whale optimization algorithm (WOA) is introduced in the state module and the gray wolf optimizer (GWO) is introduced in the action module to enhance its global optimization efficiency, avoid falling into local optima and obtain the optimal multi-level thresholds. In CEC 2017, compared with the other 16 original intelligent group optimization algorithms, the experimental results show that the QL-WOA-GWO could obtain the minimum mean values (0.00E + 00) of almost all the 24 benchmark functions. And its convergent behavior has significant advantages. Compared with the other eight WOA, GWO and their variant algorithms, the QL-WOA-GWO have better convergence behaviors and can obtain the global minimum mean values (0.00E + 00). From the experimental results of the multi-level threshold segmentation of the color images using the Otsu method and Tsallis’s entropy, it can be seen that the QL-WOA-GWO have stronger competitiveness compared to other intelligent group variant algorithms.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A new multi-level thresholding image segmentation method using Q-Learning based on hybrid whale optimization algorithm and gray wolf optimizer

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Jianxun Liu,
  • Ning Zhang,
  • Zhen Wang,
  • Chao Meng,
  • Shipu Diao,
  • Hongtao Ning,
  • Xiuquan Liu

摘要

In image segmentation, multi-level threshold segmentation (ML-TS) method is an effective image segmentation method. With the increase of multilevel thresholds, the computation amount of the ML-TS method is higher and higher, and the selection of the best multilevel thresholds is more and more difficult. When using existing intelligent group optimization algorithms to optimize the ML-TS method and select the optimal multi-level thresholds, there are problems such as insufficient optimization efficiency of the intelligent group optimization algorithms, susceptibility to local optima, and difficulty in obtaining the optimal multi-level thresholds. In order to effectively overcome the above problems, the paper proposes a new ML-TS method using Q-Learning based on hybrid whale optimization algorithm and gray wolf optimizer (QL-WOA-GWO). Firstly, the paper conducts a modular analysis of the Q-Learning, treating its the state and action as the state module and action module, respectively. Secondly, in the iteration process of the QL-WOA-GWO, the whale optimization algorithm (WOA) is introduced in the state module and the gray wolf optimizer (GWO) is introduced in the action module to enhance its global optimization efficiency, avoid falling into local optima and obtain the optimal multi-level thresholds. In CEC 2017, compared with the other 16 original intelligent group optimization algorithms, the experimental results show that the QL-WOA-GWO could obtain the minimum mean values (0.00E + 00) of almost all the 24 benchmark functions. And its convergent behavior has significant advantages. Compared with the other eight WOA, GWO and their variant algorithms, the QL-WOA-GWO have better convergence behaviors and can obtain the global minimum mean values (0.00E + 00). From the experimental results of the multi-level threshold segmentation of the color images using the Otsu method and Tsallis’s entropy, it can be seen that the QL-WOA-GWO have stronger competitiveness compared to other intelligent group variant algorithms.