Characterization of Forearm Electromyographic Signals for Automatic Classification of Wrist Movements Conference Paper uri icon

abstract

  • In this work, three different classification methods (multi-layer perceptron, support vector machine and decision tree) were used to automatically discern between six wrist movements (palmar flexion, palmar extension, radial deviation, ulnar deviation, supination and pronation of the hand) via time-domain and time-frequency features extracted from electromyographic signals (EMG) of the forearm muscles acquired in a multichannel fashion (eight channels). EMG signals of thirty (%26#x0024;%26#x0024;N=30%26#x0024;%26#x0024; ) healthy volunteers were acquired while they were performing consecutive repetitions of the six different wrist movements. Data processing included filtering, signal segmentation, feature extraction and classification using the above-mentioned methods. Finally, the results obtained with both time-domain and time-frequency features were compared. In the tests carried out with time-domain features up to 98%25 of correct classifications were obtained and up to 95%25 were obtained with the time-frequency features. These results look promising and we are currently working on their implementation in a robotic wrist rehabilitation system. © 2020, Springer Nature Switzerland AG.
  • In this work, three different classification methods (multi-layer perceptron, support vector machine and decision tree) were used to automatically discern between six wrist movements (palmar flexion, palmar extension, radial deviation, ulnar deviation, supination and pronation of the hand) via time-domain and time-frequency features extracted from electromyographic signals (EMG) of the forearm muscles acquired in a multichannel fashion (eight channels). EMG signals of thirty ($$N=30$$ ) healthy volunteers were acquired while they were performing consecutive repetitions of the six different wrist movements. Data processing included filtering, signal segmentation, feature extraction and classification using the above-mentioned methods. Finally, the results obtained with both time-domain and time-frequency features were compared. In the tests carried out with time-domain features up to 98%25 of correct classifications were obtained and up to 95%25 were obtained with the time-frequency features. These results look promising and we are currently working on their implementation in a robotic wrist rehabilitation system. © 2020, Springer Nature Switzerland AG.

publication date

  • 2020-01-01