Sensor Fusion for Myoelectric Control Based on Deep Learning With Recurrent Convolutional Neural Networks.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Electromyogram (EMG) signal decoding is crucial for myoelectric control systems.
• Traditional machine learning methods struggle with the complexity and robustness required for real-life EMG signal processing.
• Existing systems lack sufficient adaptability to dynamic conditions like arm movements.

Purpose of the Study:

• To propose a novel Recurrent Convolutional Neural Network (RCNN) model for enhanced EMG signal decoding and hand movement classification.
• To evaluate the performance of the RCNN model against traditional methods like Support Vector Machines (SVM) and Convolutional Neural Networks (CNNs).
• To investigate the impact of sensor fusion (EMG and acceleration data) on improving system adaptability and robustness.

Main Methods:

• Development and application of a deep architecture RCNN model for processing complex time-series EMG data.
• Utilizing transfer learning and parameter transferring techniques to optimize model training and prevent overfitting.
• Implementing sensor fusion by combining EMG signals with acceleration data as multimodal input.
• Comparative analysis of RCNN performance against SVM and CNN on a non-invasive EMG dataset.

Main Results:

• The RCNN model demonstrated superior classification accuracy compared to SVM and CNNs.
• Time-domain input and 1-dimensional convolution yielded higher accuracy within the RCNN architecture.
• Sensor fusion significantly improved model performance, particularly under conditions involving arm movements.
• Parameter transferring accelerated training and mitigated overfitting issues.

Conclusions:

• The proposed RCNN model is a highly effective and promising approach for EMG signal decoding in myoelectric control.
• Sensor fusion is a key strategy for enhancing the accuracy and robustness of myoelectric control systems.
• The RCNN model offers significant advantages in handling complex EMG data for advanced human-machine interfaces.