[Paper Review] Deep Residual Shrinkage Networks for EMG-based Gesture Identification

Yueying Ma, Chengbo Wang+2 more|arXiv (Cornell University)|Feb 7, 2022

Muscle activation and electromyography studies4 citations

TL;DR

This paper proposes Deep Residual Shrinkage Networks (DRSN) for high-accuracy electromyography (EMG)-based gesture recognition. By integrating residual learning with soft thresholding in the network's fully connected layers, DRSN effectively suppresses noise and enhances feature discrimination, achieving superior classification accuracy over traditional neural networks on EMG signal datasets.

ABSTRACT

This work introduces a method for high-accuracy EMG based gesture identification. A newly developed deep learning method, namely, deep residual shrinkage network is applied to perform gesture identification. Based on the feature of EMG signal resulting from gestures, optimizations are made to improve the identification accuracy. Finally, three different algorithms are applied to compare the accuracy of EMG signal recognition with that of DRSN. The result shows that DRSN excel traditional neural networks in terms of EMG recognition accuracy. This paper provides a reliable way to classify EMG signals, as well as exploring possible applications of DRSN.

Motivation & Objective

• Address the challenge of low recognition accuracy in EMG-based gesture identification due to signal noise and variability.
• Improve feature representation in EMG signals by incorporating adaptive noise suppression mechanisms.
• Develop a deep learning architecture tailored for EMG signal classification with enhanced robustness and generalization.
• Demonstrate the superiority of DRSN over conventional neural networks in EMG signal recognition tasks.

Proposed method

• Adapt the residual network (ResNet) architecture to include shrinkage modules in the fully connected layers.
• Introduce learnable thresholds within the shrinkage modules to perform soft thresholding on feature maps.
• Apply residual connections to stabilize training and improve gradient flow in deep networks.
• Optimize the network using backpropagation with adaptive learning rates and batch normalization.
• Use ReLU activation functions and dropout layers to prevent overfitting during training.
• Fine-tune the network on EMG signal data to optimize classification performance.

Experimental results

Research questions

• RQ1 Can a deep residual shrinkage network improve EMG signal classification accuracy compared to standard deep neural networks?
• RQ2 How effective is the integration of soft thresholding in residual networks for suppressing noise in EMG signals?
• RQ3 Does the proposed DRSN architecture generalize well across different gesture classes in EMG data?
• RQ4 What is the performance gain of DRSN over traditional neural networks in terms of recognition accuracy?
• RQ5 How does the inclusion of residual learning and shrinkage modules affect model convergence and robustness?

Key findings

• DRSN achieved significantly higher classification accuracy than traditional neural networks on EMG signal datasets.
• The integration of soft thresholding in the fully connected layers reduced the impact of noisy features, improving model robustness.
• Residual connections enabled stable training of deeper networks, leading to better feature learning.
• The proposed method outperformed three baseline algorithms in terms of recognition accuracy, demonstrating its effectiveness.
• The model showed strong generalization ability across diverse EMG gesture patterns with minimal overfitting.
• The results confirm that combining residual learning with adaptive thresholding is highly effective for EMG-based gesture recognition.