Multi-Stream Convolutional Neural Network-Based Wearable, Flexible Bionic Gesture Surface Muscle Feature Extraction...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface electromyographic (sEMG) signals are crucial for analyzing human motion intention but are susceptible to noise, complicating acquisition and processing.
• Traditional feature extraction for sEMG analysis requires specialized domain knowledge and significant experimental effort.
• Deep learning offers a promising avenue for automated feature extraction in sEMG-based gesture recognition.

Purpose of the Study:

• To develop an automated feature extraction method for sEMG-based gesture recognition using deep learning.
• To enhance the characterization of hand actions from sEMG signals through a novel multi-stream convolutional neural network (CNN).
• To investigate and improve noise processing, active segment detection, and feature extraction for sEMG signals.

Main Methods:

• Utilized a wearable, flexible bionic device for capturing sEMG signals.
• Proposed a multi-stream CNN algorithm that virtually augments signal channels by reconstructing sample structures.
• Implemented noise filtering, an improved moving average method for segmentation, and compared KNN, LDA, and multi-stream CNN for classification.

Main Results:

• The multi-stream CNN algorithm demonstrated superior performance in characterizing hand actions for gesture recognition.
• Achieved a high classification accuracy of up to 93.69% for 10 gestures.
• Cross-subject analysis showed an average correct classification rate of 93.18% using a pervasive electrode combination.

Conclusions:

• The proposed multi-stream CNN effectively enhances sEMG-based gesture recognition by automating feature extraction.
• The method offers a robust solution for analyzing human motion intention from noisy sEMG signals.
• High accuracy and cross-subject generalizability indicate the practical applicability of this deep learning approach.