Scholarlink

<p>The sensor-based intelligent system has drawn the attention of researchers in the domain of smart homes, healthcare, security, and more. Human Activity Recognition (HAR) is identified as a central application area of the intelligent system. Recently, the HAR models have faced the issue of cross-domain subject identification using activity-based learning due to a lack of complete activity data available for individual subjects. Moreover, transfer learning helps to transfer knowledge from the source domain to the target domain in a cross-subject environment. To address this challenge, we have proposed multi-headed convolutional networks, attached in parallel, to extract efficient features and then employ a recurrent network to identify human activity patterns. Moreover, the proposed methodology (subject-based learning) was affiliated with transfer learning to identify the activity in the cross-subject domain. We have used the publicly available mHealth experimental dataset to evaluate the performance of the proposed methodology. The experimental outcomes demonstrated that the proposed methodology outperformed the state-of-the-art methods by achieving a higher accuracy rate of 3.70%, precision of 24.77%, recall of 24.87%, and F1 Score of 25.28% compared to the leading classical classifier. The effectiveness of this research work is demonstrated by its generalizability to unknown subject identification.</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

Multi-headed Convolutional Embedded with Recurrent Networks for Recognizing the Cross-Subject Activity Using Transfer Learning

  • 期刊论文

  • 发表日期：

    2025年10月28日

查看全文
  • Prabhat Kumar,
  • Shashi Bhushan,
  • S. Suresh

摘要

The sensor-based intelligent system has drawn the attention of researchers in the domain of smart homes, healthcare, security, and more. Human Activity Recognition (HAR) is identified as a central application area of the intelligent system. Recently, the HAR models have faced the issue of cross-domain subject identification using activity-based learning due to a lack of complete activity data available for individual subjects. Moreover, transfer learning helps to transfer knowledge from the source domain to the target domain in a cross-subject environment. To address this challenge, we have proposed multi-headed convolutional networks, attached in parallel, to extract efficient features and then employ a recurrent network to identify human activity patterns. Moreover, the proposed methodology (subject-based learning) was affiliated with transfer learning to identify the activity in the cross-subject domain. We have used the publicly available mHealth experimental dataset to evaluate the performance of the proposed methodology. The experimental outcomes demonstrated that the proposed methodology outperformed the state-of-the-art methods by achieving a higher accuracy rate of 3.70%, precision of 24.77%, recall of 24.87%, and F1 Score of 25.28% compared to the leading classical classifier. The effectiveness of this research work is demonstrated by its generalizability to unknown subject identification.