Evaluation of the Hilbert-Huang Transform for myoelectric pattern classification: Towards a method to detect movement intention

Time-frequency (TFR) and time-scales representation techniques like short-time Fourier transform (STFT) and wavelet packet transform (WPT), respectively, have received considerable attention in the analysis of nonstationary signals, and specifically in the detection of movement intention. Surface Electromyography (SEMG) plays a leading role in the study and control of motor activities of people with movement disorders. This paper evaluates a new sEMG-based movement intention detection method at upper limb level. The process of feature extraction based on Hilbert-Huang Transform, a time-frequency domain technique. Selected features were the root mean square, average frequency and the central frequency variance of sEMG signals, applied to the frequency and instantaneous amplitude of Hilbert spectrum. In the pattern recognition stage, the method used is the Linear Discriminant Analysis (LDA). Two postprocessing techniques were incorporated: majority vote and reduction transitions. The efficiency of the algorithm was evaluated with parameters of sensitivity, specificity and error rate under Gaussian white noise and 60 Hz interfering conditions. Obtained results may be considered to use Hilbert-Huang transform in a real time application on the myoelectric pattern classification.