Baseline noise and model parameters in surface electromyography
University of New Brunswick
The Pairwise Attribute Noise Detection Algorithm has been recommended by others as a way to classify surface electromyography signals as clean or noisy. To train the algorithm, simulation can be used to generate clean examples. Used in this way, the algorithm has been shown to perform well for classifying simulated test signals, but not for in vivo SEMG records. This work investigated the poor performance with in vivo SEMG records in order to improve it, if possible. Impact of introducing instrumentation effects into simulated signals was shown to be negligible. Impacts from judicious selection of simulation parameters, including both embedded and user specified values was shown to significantly improve falsely classifying clean records as noisy. A genetic algorithm was developed to provide support for choosing user specified values and using this technique, false positive rates (i.e. classifying clean signals as noisy) decreased from 95 % to 20 %, without degradation of other classification rates (i.e. classifying noisy signals as noisy, clean signals as clean).