Scholarlink

<p>This paper presents CapsResNet, a novel hybrid image recognition network that integrates Convolutional Neural Networks (CNNs), Residual Blocks, and Capsule Layers. CapsResNet is designed to address the limitations of traditional CNNs in capturing spatial hierarchies and relationships between objects within images. By incorporating residual blocks, the model mitigates the vanishing gradient problem, enabling the training of deeper and more complex networks. The capsule layers, on the other hand, provide a richer vector representation of features, preserving spatial relationships and improving the model’s ability to recognize complex shapes and structures. The proposed model is comprehensively evaluated on five benchmark datasets: Fashion-MNIST, CIFAR-10, CIFAR-100, EMNIST, and SVHN. CapsResNet achieves 94.02% accuracy on CIFAR-10 and 96.85% on SVHN, consistently surpassing baseline CNNs and CapsNet-based models in terms of accuracy, precision, recall, and F1-score. An ablation study further confirms the essential role of both residual blocks and capsule layers in enhancing the model’s robustness and performance. Overall, CapsResNet represents a significant step toward more interpretable and efficient image recognition solutions.</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

Enhancing image recognition with CapsResNet: a hybrid model integrating CNNs, residual blocks, and capsule layers

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Fouad Khattari,
  • Youssef El Hadfi,
  • Aissam Hadri

摘要

This paper presents CapsResNet, a novel hybrid image recognition network that integrates Convolutional Neural Networks (CNNs), Residual Blocks, and Capsule Layers. CapsResNet is designed to address the limitations of traditional CNNs in capturing spatial hierarchies and relationships between objects within images. By incorporating residual blocks, the model mitigates the vanishing gradient problem, enabling the training of deeper and more complex networks. The capsule layers, on the other hand, provide a richer vector representation of features, preserving spatial relationships and improving the model’s ability to recognize complex shapes and structures. The proposed model is comprehensively evaluated on five benchmark datasets: Fashion-MNIST, CIFAR-10, CIFAR-100, EMNIST, and SVHN. CapsResNet achieves 94.02% accuracy on CIFAR-10 and 96.85% on SVHN, consistently surpassing baseline CNNs and CapsNet-based models in terms of accuracy, precision, recall, and F1-score. An ablation study further confirms the essential role of both residual blocks and capsule layers in enhancing the model’s robustness and performance. Overall, CapsResNet represents a significant step toward more interpretable and efficient image recognition solutions.