Skip to Main Content
This paper presents a novel approach to optimize pattern recognition system using genetic algorithm (GA) to identify the type of hand motion employing artificial neural networks (ANNs) with high performance and accuracy suited for practical implementations. To achieve this approach, electromyographic (EMG) signals were obtained from sixteen locations on the forearm of six subjects in ten hand motion classes. In the first step of feature extraction of forearm EMG signals, WPT is utilized to generate a wavelet decomposition tree from which WPT coefficients are extracted. In the second step, standard deviation of wavelet packet coefficients of EMG signals is considered as the feature vector for training purposes of the ANN. To improve the algorithm, GA was employed to optimize the algorithm in such a way that to determine the best values for “mother wavelet function”, “decomposition level of wavelet packet analysis”, and “number of neurons in hidden layer” concluded in a high-speed, precise two-layer ANN with a particularly small-sized structure. This proposed network with a small size can recognize ten hand motions with recognition accuracy of over 98% and also resulted in improvement of stability and reliability of the system for practical considerations.