The Force Classification Based on EMG Signals for Intelligent Wearable Systems

In recent years, the development of flexible and stretchable sensors has shown considerable potential in human information collection for wearable applications. With skin-fitting film electrodes, Electromyography (EMG) signals can be monitored stably and detected for effective control actuation in wearable systems. Traditionally in EMG-based activation, researchers usually focused on the design of EMG features, which is time-costing and difficult to always find the optimal combination. In this work, we have proposed a scheme to use convolutional neural network for complicated EMG feature extraction and accurate force classification. Four force levels were recognized by our model. The experimental result stated that the accuracy in each force level has reached 90.64%, 89.94%, 84.21% and 95.24%, respectively. In addition, the performance of our deep learning model has outperformed the traditional manual-feature-based methods, which utilized mean absolute value (MAV), waveform length (WL) and Willison amplitude (WAMP) for force identification. Actually, this work has verified the excellent effect of intelligent methods in EMG feature learning, and can be further applied in a real-time wearable system to promote its convenience and practicality.