Scholarlink

<p>Due to aggregating the advantages of rough and fuzzy clustering algorithms, rough fuzzy clustering algorithms have gained increasing attention and are widely applied in practical areas, such as data mining, pattern recognition and image processing. However, most existing rough fuzzy clustering algorithms are designed for single-view clustering and evaluate results based solely on a single clustering criterion. Furthermore, these algorithms are typically sensitive to the initialized cluster centers and prone to local optima, which limits the ability to satisfy diverse practical clustering requirements and restricts the overall performance. To solve the above problems, this paper proposes a multi-view information transfer driven surrogate-assisted multi-objective evolutionary rough fuzzy clustering (MIT-SMERFC) algorithm for image segmentation, which combines multi-view learning and transfer learning into rough fuzzy clustering. First, a multi-view information transfer driven rough fuzzy clustering framework is proposed. This framework employs multiple distance metrics as distinct views to explore various data structures, and utilizes extracted consensus membership degrees as transfer knowledge, thus enhancing the segmentation performance and image information mining capabilities of the algorithm. Second, to overcome the drawback of the transfer factor value depending on manual adjustment experience, a difference image-based transfer factor adaptive determination mechanism is designed in MIT-SMERFC. Subsequently, three complementary rough fuzzy clustering objective functions are constructed based on multi-view learning and transfer learning, and incorporated into an adaptive hybrid crossover operator-based surrogate assisted multi-objective evolutionary framework to enhance the search capability of cluster centers. Finally, a rough fuzzy clustering validity index that integrates multi-view consensus information is constructed to select the optimal trade-off solution form the final non-dominated solution set. The effectiveness of the components of MIT-SMERFC have been validated in the experimental section. Moreover, experimental results on synthetic images, Berkeley and Weizmann images demonstrate that MIT-SMERFC exhibits outstanding performance in segmentation performance and computational cost.</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 multi-view information transfer driven surrogate-assisted multi-objective evolutionary rough fuzzy clustering algorithm

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Yujie Yang,
  • Feng Zhao,
  • Hanqiang Liu

摘要

Due to aggregating the advantages of rough and fuzzy clustering algorithms, rough fuzzy clustering algorithms have gained increasing attention and are widely applied in practical areas, such as data mining, pattern recognition and image processing. However, most existing rough fuzzy clustering algorithms are designed for single-view clustering and evaluate results based solely on a single clustering criterion. Furthermore, these algorithms are typically sensitive to the initialized cluster centers and prone to local optima, which limits the ability to satisfy diverse practical clustering requirements and restricts the overall performance. To solve the above problems, this paper proposes a multi-view information transfer driven surrogate-assisted multi-objective evolutionary rough fuzzy clustering (MIT-SMERFC) algorithm for image segmentation, which combines multi-view learning and transfer learning into rough fuzzy clustering. First, a multi-view information transfer driven rough fuzzy clustering framework is proposed. This framework employs multiple distance metrics as distinct views to explore various data structures, and utilizes extracted consensus membership degrees as transfer knowledge, thus enhancing the segmentation performance and image information mining capabilities of the algorithm. Second, to overcome the drawback of the transfer factor value depending on manual adjustment experience, a difference image-based transfer factor adaptive determination mechanism is designed in MIT-SMERFC. Subsequently, three complementary rough fuzzy clustering objective functions are constructed based on multi-view learning and transfer learning, and incorporated into an adaptive hybrid crossover operator-based surrogate assisted multi-objective evolutionary framework to enhance the search capability of cluster centers. Finally, a rough fuzzy clustering validity index that integrates multi-view consensus information is constructed to select the optimal trade-off solution form the final non-dominated solution set. The effectiveness of the components of MIT-SMERFC have been validated in the experimental section. Moreover, experimental results on synthetic images, Berkeley and Weizmann images demonstrate that MIT-SMERFC exhibits outstanding performance in segmentation performance and computational cost.