A novel sEMG-based hand gesture prediction method using a new motion detection algorithm and an LCNN model.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Surface electromyography (sEMG) signals offer a promising avenue for non-invasive human-computer interaction.
• Accurate real-time prediction of hand gestures from sEMG is crucial for advanced prosthetics and assistive technologies.
• Existing methods often struggle with precise motion detection and effective fusion of temporal and spatial features.

Purpose of the Study:

• To develop a novel and accurate real-time hand gesture prediction method using sEMG signals.
• To improve the detection of hand gesture motion start and end times.
• To enhance the performance of sEMG pattern recognition through an advanced deep learning model.

Main Methods:

• A new time-domain information index combining mean and standard deviation of sEMG signals was defined.
• A novel motion detection algorithm was introduced for precise capture of gesture motion timings.
• A new Long-term Convolutional Neural Network (LCNN) model integrating LSTM was designed for multi-scale feature fusion.

Main Results:

• The proposed method achieved 92.4% average prediction accuracy for 21 gestures on the Zhang et al. dataset.
• An 82.7% average prediction accuracy for six hand gestures was reached on the Krilova et al. dataset.
• The LCNN model demonstrated superior prediction accuracy and real-time performance compared to GRU and LSTM models.

Conclusions:

• The novel motion detection algorithm significantly improves sEMG-based gesture recognition.
• The proposed LCNN model effectively fuses multi-scale features, enhancing prediction accuracy.
• The developed hand gesture prediction method shows significant potential for practical applications in human-computer interaction.