Deep neural network for EMG signal classification of wrist position: Preliminary results

Physically impaired people may use Surface Electromyography (SEMG) signals to control rehabilitation and assistive devices. SEMG is the electrical manifestation of the neuromuscular activation associated with a contracting muscle. SEMG directly reflects the human motion intention; thus, they can be used as input information for human-robot interaction. This paper proposes an EMG-based pattern recognition algorithm for classification of joint wrist angular position during flexion-extension movements from EMG signals. The algorithm uses a feature extraction stage based on a combination of time and frequency domain. The pattern recognition stage uses an artificial neural network (NN) as classifier. Also, using an autoencoder, deep NN architecture was tested. It was carried out a set of experiment with 10 subjects. Experiments included five recorded SEMG channels from forearm executing wrist flexion and extension movements, as well as the use of a commercial electrogoniometer to acquire joint angle. Results show that shallow NN had better performance that architectures with more layers based on autoencoders.