Intention detection using a neuro-fuzzy EMG classifier

One of the most important factors in prosthetic and orthotic controllers is the ability to detect the intention of the person to perform a certain activity such as standing up, quiet standing, walking, and sitting down. For these applications, detecting the intention of the person to perform an activity relieves them from the burden of conscious effort in operating the system. Electromyography (EMG) has been used extensively for intention detection and can be considered a bandlimited stochastic process with Gaussian distribution and zero mean, which has varying spectral characteristics in time. Various EMG features have been used for intention detection including the number of zero crossings, the EMG frequency characteristics, and the mean absolute values. There are a number of drawbacks that have been associated with these methods such as the high electrode sensitivity to electrode displacement, low recognition rate, and a perceivable delay in control. In this article we discuss a technique for EMG applications that decreases global delay time and improves time spectral analysis. The technique is aimed at improving the Gabor matching pursuit (GMP) algorithm through the use of genetic algorithms. The key stage of this design feeds EMG features to a neuro-fuzzy classifier that can be designed to detect the intention of the patient.