Improving gesture recognition by bidirectional temporal convolutional networks

Haoyu Chen, Yue Zhang, Dalin Zhou, Honghai Liu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

214 Downloads (Pure)


Surface electromyography (sEMG) based gesture recognition as an important role in Muscle-Computer interface has been researched for decades. Recently, deep learning based method has had a profound impact on this field. CNN, RNN and RNN-CNN based methods were studied by many researchers. Motivated by Bidirectional Long short-term memory (Bi-LSTM) and Temporal Convolutional Networks (TCN), we propose 1D CNN based networks called Bidirectional Temporal Convolutional Networks (Bi-TCN). The positive order signal and reverse order sEMG signal are feed to our networks to learn the different representation of the same sEMG signal. We evaluate proposed networks on two benchmark datasets, Ninapro DB1 and DB5. Our networks yields 90.74% prediction accuracy on DB1 and 90.06% prediction accuracy on DB5. The results demonstrate our networks is comparable to the state-of-the-art works.
Original languageEnglish
Title of host publicationICRRI 2020: Robotics and Rehabilitation Intelligence
Subtitle of host publicationFirst International Conference, ICRRI 2020, Fushun, China, September 9–11, 2020, Proceedings, Part II
EditorsJianhua Qian, Honghai Liu, Jiangtao Cao, Dalin Zhou
Number of pages12
ISBN (Electronic)978-981-33-4932-2
ISBN (Print)978-981-33-4931-5
Publication statusPublished - 3 Jan 2021
Event1st International Conference on Robotics and Rehabilitation Intelligence - Fushun, China
Duration: 9 Sept 202011 Sept 2020

Publication series

NameCommunications in Computer and Information Science
PublisherSpringer, Singapore
ISSN (Print)1865-0929


Conference1st International Conference on Robotics and Rehabilitation Intelligence
Abbreviated titleICRRI 2020


  • sEMG
  • deep learning
  • CNN
  • TCN
  • LSTM
  • gesture recognition


Dive into the research topics of 'Improving gesture recognition by bidirectional temporal convolutional networks'. Together they form a unique fingerprint.

Cite this