1 of 1

Title

10th Annual COE Graduate Poster Presentation Competition

Student: Koundinya Challa (PhD)

Advisor(s): Balakrishna Gokaraju, Anish Turlapaty, Raymond Tesiero

Cross-Disciplinary Research Area: xx

Abstract

EMG (electromyography) signals provide data about neuromuscular activity, and they could be used as a feed to myoelectric control system applications (an Exoskeleton Arm). This research describes about a novel machine learning model for efficient classification of EMG signals. A crucial step in this process is to withdraw useful information from EMG signals, and in this research, we extracted time domain features. These features consist of mean absolute value (MAV), waveform length (WL), zero crossings (ZC), and slope sign changes (SSC). Time domain features show how an EMG signal changes with time. Then we change the time domain features into their frequency domain, Fourier transformations are performed to change the time domain features into their frequency domain features. The frequency domain representation of EMG signals allows us to observe a plethora of characteristics of the signal that are not easy to notice when you look at the EMG signal in its time domain. Next these set of features are fed to a novel machine learning model. The goal of the research is to apply the novel machine learning model to the Exoskeleton arm (Exorn), A exorn has 10 degrees of freedom to control joints starting from shoulder to wrist to give better portability and flexibility to the human arm. Using SVM (Support Vector Machines) classifier an accuracy of 87.5% has been achieved which will later be applied to Exoskeleton arm.

Approach

Results

  • We use the EMG dataset that we collected to train and test our model.
  • When we use Random Forest, the accuracy is 89%
  • When we use SVM as our model the accuracy is 87%

Conclusion / Future work

The experimental results show that:

  • Machine learning can effectively learn the EMG signals with sufficient training data.
  • The quantity and quality if the training data have a great influence on the training
  • In this experiment Random Forest was better than SVM.

Next steps in the research:

  • Collect more data
  • Use more ML algorithms o compare the results
  • Apply the Model on a Exo Arm

References

  1. A MACHINE LEARNING SYSTEM FOR CLASSIFICATION OF EMG SIGNALS TO ASSIST EXOSKELETON PERFORMANCE
  2. Surface Electromyography Signal Processing and Classification Techniques
  3. Latent Factors Limiting the Performance of sEMG-Interfaces
  4. Feature Analysis for Classification of Physical Actions Using Surface EMG Data
  5. SVM based Classification Of sEMG Signals using Time Domain Features for the Applications towards Arm Exoskeletons

​

​