Scholarlink

<p>Transfer learning has become popular among the meta-heuristic researchers recently to solve complicated optimization tasks. Transferring high-quality solutions among tasks where difficulties are linked may improve performance. Using this concept the Multi-objective Multi-factorial Evolutionary Algorithm (MOMFEA) has shown remarkable performance in convergence and solution quality. However, it is also reported that in MOMFEA convergence becomes weak in the presence of irrelevant tasks. In this manuscript, a decomposition-based Multi-objective Multi-factorial Evolutionary Algorithm (MOMFEA/d) is proposed. The MOMFEA/d optimizes several scalar optimization sub-problems from a multi-objective problem simultaneously, resulting in lower computational cost per generation than many other comparative multi-objective algorithms since each sub-problem is optimized using data from its neighboring sub-problems. This approach integrates MOMFEA’s local search component into each MOEA/D sub-problem allowing global and local search space exploration. The Chebyshev decomposition technique drives the search towards varied parts of the Pareto front, while the local search mechanism refines solutions inside each sub-problem, improving quality and convergence. The MOMFEA/d is verified on IEEE CEC 2021 multi-objective bi-task test suite (ten problems), and IEEE CEC 2017 complex multi-objective bi-task test suite (nine problems). The proposed algorithm is evaluated against state-of-the-art approaches with respect to convergence, diversity, and run-time complexity, and the results demonstrate its superior performance. The proposed MOMFEA/d is then applied to effectively segment the six hyper-spectral images: Kennedy Space Center, Pavia Center, Pavia University, Salinas Valley, SalinasA Scene and IndianPines. Simulation studies reveal the proposed MOMFEA/d gives better accuracy than existing state-of-the-art algorithms: MOMFEA-II, MOMFEA, EMT-PD and NSGA-II.</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

Hyperspectral image segmentation using multiobjective multifactorial evolutionary algorithms based on chebyshev decomposition

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • Rajat Yadav,
  • Aakansha Agarwal,
  • Satyasai Jagannath Nanda

摘要

Transfer learning has become popular among the meta-heuristic researchers recently to solve complicated optimization tasks. Transferring high-quality solutions among tasks where difficulties are linked may improve performance. Using this concept the Multi-objective Multi-factorial Evolutionary Algorithm (MOMFEA) has shown remarkable performance in convergence and solution quality. However, it is also reported that in MOMFEA convergence becomes weak in the presence of irrelevant tasks. In this manuscript, a decomposition-based Multi-objective Multi-factorial Evolutionary Algorithm (MOMFEA/d) is proposed. The MOMFEA/d optimizes several scalar optimization sub-problems from a multi-objective problem simultaneously, resulting in lower computational cost per generation than many other comparative multi-objective algorithms since each sub-problem is optimized using data from its neighboring sub-problems. This approach integrates MOMFEA’s local search component into each MOEA/D sub-problem allowing global and local search space exploration. The Chebyshev decomposition technique drives the search towards varied parts of the Pareto front, while the local search mechanism refines solutions inside each sub-problem, improving quality and convergence. The MOMFEA/d is verified on IEEE CEC 2021 multi-objective bi-task test suite (ten problems), and IEEE CEC 2017 complex multi-objective bi-task test suite (nine problems). The proposed algorithm is evaluated against state-of-the-art approaches with respect to convergence, diversity, and run-time complexity, and the results demonstrate its superior performance. The proposed MOMFEA/d is then applied to effectively segment the six hyper-spectral images: Kennedy Space Center, Pavia Center, Pavia University, Salinas Valley, SalinasA Scene and IndianPines. Simulation studies reveal the proposed MOMFEA/d gives better accuracy than existing state-of-the-art algorithms: MOMFEA-II, MOMFEA, EMT-PD and NSGA-II.