Scholarlink

<p>Effectively extracting short- and long-term features information from sEMG signals is critical for accurate gesture identification. Existing deep learning approaches often only capture limited spatial or temporal information from multimodal data. This study proposes Temporal Feature Fusion Network with Attention Strategy (TFFNAS) to enhance hand movement classification accuracy and robustness. Specifically, a dilated temporal convolutional model is designed to capture potential long-term temporal and spatial features of multichannel sEMG signals. Concurrently, a multi-scale convolution module extracts crucial short-term features, and an attention fusion module intelligently prioritizes the importance of these diverse temporal features. Experimental results demonstrate TFFNAS’s superior performance, achieving classification accuracies of 99.26% on Ninapro-DB1 and 96.48% on EMAHA dataset, surpassing existing state-of-the-art methods. Our work addresses the challenge of recognising predefined hand movements from multichannel EMG data by emphasizing the integration of multi-model temporal features. These findings highlight the effectiveness of temporal fusion-based systems in enhancing performance and opens new avenues for prosthetic hand control systems.</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

TFF-Net: A hybrid MSC-TCN-Attention network for enhanced long–short term features extraction in gesture classification

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Gautam Shah,
  • Abhinav Sharma,
  • Deepak Joshi,
  • Ajit Singh Rathor,
  • Sunil Semwal

摘要

Effectively extracting short- and long-term features information from sEMG signals is critical for accurate gesture identification. Existing deep learning approaches often only capture limited spatial or temporal information from multimodal data. This study proposes Temporal Feature Fusion Network with Attention Strategy (TFFNAS) to enhance hand movement classification accuracy and robustness. Specifically, a dilated temporal convolutional model is designed to capture potential long-term temporal and spatial features of multichannel sEMG signals. Concurrently, a multi-scale convolution module extracts crucial short-term features, and an attention fusion module intelligently prioritizes the importance of these diverse temporal features. Experimental results demonstrate TFFNAS’s superior performance, achieving classification accuracies of 99.26% on Ninapro-DB1 and 96.48% on EMAHA dataset, surpassing existing state-of-the-art methods. Our work addresses the challenge of recognising predefined hand movements from multichannel EMG data by emphasizing the integration of multi-model temporal features. These findings highlight the effectiveness of temporal fusion-based systems in enhancing performance and opens new avenues for prosthetic hand control systems.