An Improved Performance of Deep Learning Based on Convolution Neural Network to Classify the Hand Motion by...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Accurate electromyography (EMG) pattern recognition is crucial for advanced prosthetic hand development.
• Current methods often involve complex feature extraction, hindering real-time application.
• Deep learning offers potential for improved accuracy and simplified processing.

Purpose of the Study:

• To enhance prosthetic hand control by simplifying deep learning preprocessing for EMG signal classification.
• To evaluate the performance of a Convolutional Neural Network (CNN) for classifying ten distinct hand motions using raw EMG data.
• To compare the CNN's efficacy against traditional machine learning classifiers.

Main Methods:

• Utilized a Convolutional Neural Network (CNN) algorithm for classifying ten hand motions from two raw EMG signals.
• Eliminated the need for manual feature extraction, simplifying the preprocessing stage.
• Optimized CNN performance by evaluating hyperparameters and validating on a public dataset from ten subjects.

Main Results:

• The CNN successfully discriminated between ten hand motions using raw EMG signals without handcrafted feature extraction.
• CNN demonstrated superior performance compared to Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), and K-Nearest Neighbors (KNN).
• Achieved an average accuracy ranging from 0.77 to 0.93 across all motions, with no significant difference between two-channel and single-channel EMG input.

Conclusions:

• The proposed CNN-based method offers a simplified yet high-performance approach for EMG pattern recognition in prosthetic hand applications.
• The elimination of feature extraction reduces computational complexity, making it suitable for real-time prosthetic control.
• This technique holds significant promise for improving the effectiveness and usability of prosthetic hands.