SDSU Department of Computer Science
  
Home Contact Us Undergraduate Student Info Graduate Student Info Faculty & Staff News & Events
News Calendar Problem of the Fortnight

March 2008 Calendar

CS Masters' Thesis Defense

Title: Real-Time Feature Extraction and Classification of Prehensile EMG Signals
Speaker: Christopher Miller
Date: Monday, March 24, 2008
Time: 9:30 a.m.
Location: GMCS 408
Thesis advisor: Dr Marko Vuskovic

Abstract:
The implementation of a real-time feature extraction and classification program is presented. The program recognizes four basic grasps using only three channels of EMG signals collected from the extensor muscles of the human forearm. The feature extraction method employed is a hybrid approach that relies on the wavelength feature for small training sets and spectral moments for larger training sets. The Mahalanobis distance based classifier achieved a 100% correct classification of offline data using spectral moment based feature vectors that were transformed by the Box-Cox transformation. The results demonstrate that spectral moments are an effective feature for prehensile EMG signal classification in real-time. Recommendations are provided for implementing an advanced real-time control structure as well as an adaptive online learning scheme.
webmaster@cs.sdsu.edu page counter College of Sciences