EMG-based Hand Gesture Recognition by Deep Time-frequency Learning for Assisted Living & Rehabilitation

Abstract

As a user-friendly human-computer interaction approach, EMG is regarded as one of the most promising modalities for hand gesture recognition. Though EMG-based hand gesture recognition has been advanced in recent years, to effective detect the patterns from the noisy EMG signal, more advanced algorithms are still highly necessary. Convolutional neural network (CNN) is a popular deep learning algorithm and its unique architecture has gained a great success in the image processing area. In this study, we propose a new deep learning framework for hand gesture recognition from the multi-session EMG signal. In the data representation stage, we also transform the time domain EMG signal to the time-frequency domain by short-term Fourier transform (STFT) to get more time-varying frequency characteristics. Our experiment shows that the proposed framework can effectively detect hand gestures from the multi-session EMG data. This work will greatly advance the hand gesture recognition.