Fractal modelling for pattern recognition via artificial neural networks

In this study, it is aimed to use fractal modelling parameters for electromyography (EMG) pattern recognition, and to compare EMG pattern recognition performances of the fractal and autoregressive (AR) models. For this purpose, EMG signals from biceps brachii and forearm muscles during ten different movements of the arm and the hand are acquired from a twenty four years old male subject. These signals are modelled with both AR and fractal modelling approaches, and these models are studied for pattern recognition purposes via artificial neural networks. The results indicate that the contraction factor in fractal modelling can be used as a criterion for pattern recognition. However, the AR model parameters yield better results. As a further approach, the contraction factors obtained from fractal modelling are modelled as the outcome of an AR process and these AR model coefficients are also studied for EMG pattern recognition