Mixture of experts extra tree-based sEMG hand gesture recognition.

Area of Science:

• Robotics and Machine Learning
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG) is a non-invasive technique for human hand gesture recognition, crucial for robotic hand control.
• Overfitting and biases are significant challenges in developing generalized classifiers for multi-gesture recognition using sEMG data.
• Existing methods often struggle to achieve high accuracy and robustness across diverse gestures and subjects.

Purpose of the Study:

• To introduce and evaluate a Mixture of Experts Extra Tree (MEET) classifier for enhanced human hand gesture recognition using sEMG.
• To address the common issues of overfitting and bias in multi-gesture classification.
• To compare the performance of the MEET classifier against eleven other models.

Main Methods:

• Development of a Mixture of Experts Extra Tree classifier, utilizing individual Extra Tree models as experts and a gating Extra Tree model.
• Collection of sEMG data from four subjects performing six distinct hand gestures.
• Evaluation of the MEET classifier on both the collected dataset and a publicly available dataset with fifteen gesture classes.

Main Results:

• The MEET classifier demonstrated superior performance compared to eleven other evaluated algorithms.
• Accuracies for the MEET classifier on the collected data were 86.8%, 89.2%, 87.9%, and 78.4% across the four subjects.
• The MEET classifier achieved an improved mean accuracy of 1.25% on the public dataset.

Conclusions:

• The Mixture of Experts Extra Tree classifier is highly effective for accurate human hand gesture recognition from sEMG signals.
• This model offers a robust solution for overcoming overfitting and bias in sEMG-based gesture classification.
• The MEET classifier shows significant potential for advancing robotic hand control and human-computer interaction systems.