Scholarlink

<p>Convolutional neural networks (CNNs) are indispensable in computer vision, especially for image classification, but designing optimal CNN architectures is challenging and often requires extensive domain-specific knowledge. This paper presents a novel method that combines genetic algorithms (GAs) and particle swarm optimization (PSO) to automate CNN architecture design. The proposed algorithm integrates the global search capability of GAs with the local optimization efficiency of PSO, effectively addressing the limitations of each. Tested on the CIFAR-10 benchmark dataset, the algorithm achieved a classification accuracy of 94.47%, outperforming most existing manually designed and evolutionary algorithm-designed architectures. Statistical analysis confirmed the significance of the improvements. This approach not only enhances the accuracy of CNN architectures but also demonstrates the potential of hybrid metaheuristic algorithms in automated design, making it more accessible for researchers across various fields.</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

Automatic design of CNN architecture based on genetic algorithm and particle swarm optimization

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Shuaifei Lin,
  • Wei Zhang,
  • Nannan Xu,
  • Xueli Liu,
  • Jianfeng Wu

摘要

Convolutional neural networks (CNNs) are indispensable in computer vision, especially for image classification, but designing optimal CNN architectures is challenging and often requires extensive domain-specific knowledge. This paper presents a novel method that combines genetic algorithms (GAs) and particle swarm optimization (PSO) to automate CNN architecture design. The proposed algorithm integrates the global search capability of GAs with the local optimization efficiency of PSO, effectively addressing the limitations of each. Tested on the CIFAR-10 benchmark dataset, the algorithm achieved a classification accuracy of 94.47%, outperforming most existing manually designed and evolutionary algorithm-designed architectures. Statistical analysis confirmed the significance of the improvements. This approach not only enhances the accuracy of CNN architectures but also demonstrates the potential of hybrid metaheuristic algorithms in automated design, making it more accessible for researchers across various fields.