Gesture recognition of continuous wavelet transform and deep convolution attention network.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Missing data features pose a significant challenge in accurate gesture recognition.
• Surface electromyography (sEMG) signals offer rich information for human-computer interaction but are prone to data loss.
• Deep Convolutional Neural Networks (DCNNs) show promise but require enhancement to handle incomplete datasets.

Purpose of the Study:

• To propose an improved gesture recognition method using a DCNN to overcome missing data features.
• To enhance the feature representation capabilities of DCNNs for sEMG signal analysis.
• To achieve higher accuracy in recognizing gestures from sEMG data.

Main Methods:

• Extracted time-frequency spectrograms of sEMG signals using the continuous wavelet transform.
• Developed a DCNN model integrated with a Spatial Attention Module (SAM) and residual modules (DCNN-SAM).
• Incorporated residual modules to improve feature representation and mitigate missing data issues.

Main Results:

• The proposed DCNN-SAM model achieved a gesture recognition accuracy of 96.1% on 10 different gestures.
• The improved method demonstrated a significant accuracy increase of approximately 6 percentage points compared to the standard DCNN.
• The integration of SAM and residual modules effectively addressed the problem of missing data features.

Conclusions:

• The developed DCNN-SAM method provides a robust solution for gesture recognition from sEMG signals, particularly in the presence of missing data.
• The enhanced feature representation through spatial attention and residual modules leads to superior recognition accuracy.
• This approach holds potential for advancing applications in human-computer interaction and prosthetics control.