Surface EMG classification during dynamic contractions for multifunction transradial prostheses.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• High usability myo-controlled devices depend on accurate classification of muscle signals during movement.
• Dynamic contractions present challenges for traditional pattern recognition algorithms developed for static conditions.

Purpose of the Study:

• To investigate the impact of training data set selection on pattern recognition algorithm performance during dynamic contractions.
• To evaluate the effectiveness of different feature classification approaches for myoelectric control.

Main Methods:

• Tested several pattern recognition algorithms using varying training data sets.
• Incorporated a contraction onset detection threshold.
• Compared time-domain features with wavelet features for classification.

Main Results:

• Pattern recognition algorithms, with a contraction onset threshold, achieved high accuracy during dynamic contractions.
• Optimizing the training data set significantly improved algorithm performance.
• Simple time-domain feature classification yielded results comparable to complex wavelet feature methods.

Conclusions:

• Robust classification for myo-controlled devices is achievable during dynamic contractions.
• Training data optimization is crucial for enhancing algorithm performance.
• Accessible and simpler feature extraction methods can be effective for myoelectric signal classification.