Scholarlink

<p>This paper proposes an enhanced method for cotton and hemp fiber classification using the SE-ResNet50 model, which integrates the Squeeze-and-Excitation (SE) attention mechanism into the ResNet50 architecture. Traditional textile fiber classification methods often suffer from slow speeds and limited accuracy, prompting the adoption of deep learning techniques. By incorporating the SE module, the model’s feature extraction capability is significantly improved, enabling it to focus on critical features. Experimental results demonstrate the superiority of the SE-ResNet50 model, achieving a classification accuracy of 98.5%, substantially outperforming the baseline ResNet50 model. Preprocessing techniques, including noise removal and data augmentation, further enhance the model’s robustness and accuracy. Here we show that the SE-ResNet50 model not only improves classification performance but also reduces overfitting, underscoring its potential for widespread application in textile fiber classification.</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 method for cotton and hemp fiber classification optimized with ResNet and attention mechanism

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Xinyue Fang,
  • Binjie Xin,
  • Zhu Zhan,
  • Feng Yan,
  • Zhenhui Jin

摘要

This paper proposes an enhanced method for cotton and hemp fiber classification using the SE-ResNet50 model, which integrates the Squeeze-and-Excitation (SE) attention mechanism into the ResNet50 architecture. Traditional textile fiber classification methods often suffer from slow speeds and limited accuracy, prompting the adoption of deep learning techniques. By incorporating the SE module, the model’s feature extraction capability is significantly improved, enabling it to focus on critical features. Experimental results demonstrate the superiority of the SE-ResNet50 model, achieving a classification accuracy of 98.5%, substantially outperforming the baseline ResNet50 model. Preprocessing techniques, including noise removal and data augmentation, further enhance the model’s robustness and accuracy. Here we show that the SE-ResNet50 model not only improves classification performance but also reduces overfitting, underscoring its potential for widespread application in textile fiber classification.