Research on sEMG Pattern Recognition Algorithm and Implementation of a Gesture Recognition System

Pattern recognition of surface electromyogram (surface EMG, sEMG) signals can obtain human movement information. In recent years, this technology has been widely used in many fields. In the algorithm part, this paper proposes a model based on Convolutional Neural Network (CNN) and Recurrent Neural N...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2023 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD) S. 1 - 6
Hauptverfasser: Tian, Yuepeng, Zhang, Zhimin, Li, Yuwen
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 02.11.2023
Schlagworte:
Attention Mechanisms
CNN
Computational modeling
Data acquisition
Data preprocessing
Embedded Systems
Gesture recognition
Pattern Recognition
Real-time systems
Recurrent neural networks
RNN
sEMG
Training
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Pattern recognition of surface electromyogram (surface EMG, sEMG) signals can obtain human movement information. In recent years, this technology has been widely used in many fields. In the algorithm part, this paper proposes a model based on Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). The effects of different algorithm structures and parameter selections are compared. On this basis, Depthwise separable convolution is introduced to reduce the number of parameters while maintaining high accuracy. In addition, the attention module SElayer is introduced to further improve the performance of the algorithm. The final algorithm achieved an accuracy rate of 93.41% on the NinaPro-DB2 dataset. In addition to the algorithm, this paper also builds a sEMG gesture recognition system with the help of an embedded platform. The system is mainly composed of an 8channel sEMG acquisition board and a computer, and includes four functional modules: data acquisition and annotation, data preprocessing, model training and real-time classification. Finally, the system collected sEMG data from 7 subjects. The model achieved good results on the dataset and completed the real-time classification.
DOI:10.1109/ICSMD60522.2023.10491043