Feature Fusion-Based Improved Capsule Network for sEMG Signal Recognition.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Current deep learning models like CNNs often overlook feature correlations in sEMG signal recognition.
• Accurate sEMG-based hand gesture recognition is crucial for prosthetics and human-computer interaction.

Purpose of the Study:

• To introduce a feature fusion-based improved capsule network (FFiCAPS) for enhanced sEMG signal recognition.
• To improve the robustness of hand gesture recognition against electrode displacement and inter-subject variability.

Main Methods:

• FFiCAPS integrates sEMG signal information with feature data for richer input representations.
• The model incorporates a multilayer convolution layer for multiscale feature extraction and an e-Squash function for improved sensitivity.
• Feature fusion is employed to capture correlations among extracted features.

Main Results:

• FFiCAPS achieved 86.58% overall accuracy under electrode displacement conditions.
• The model demonstrated 82.12% accuracy among subjects, outperforming eight other methods.
• Notable improvements were observed in recognizing specific gestures like hand open and radial flexion.

Conclusions:

• The proposed FFiCAPS method significantly enhances sEMG-based hand gesture recognition accuracy and robustness.
• FFiCAPS effectively addresses limitations of traditional deep learning models by considering feature correlations.
• This approach shows promise for advanced applications requiring reliable sEMG signal interpretation.