A sEMG-Based Hand Gesture Recognition Using Mulit-channel CNN and MLP

In this paper, three different classification methods, including the support vector machine (SVM), the deep learning method based on multi-layer perceptron (MLP) and multichannel convolutional neural network (multi-channel CNN), were used to classify 13 hand gestures. The surface electromyography (sEMG) were extracted from six muscles of hand and forearm. For the SVM and MLP, six features in the time domain, frequency domain and time-frequency domain were extracted. For the multi-channel CNN, a sliding window segment of the original sEMG image was used as the input. Hand gesture recognition based on deep learning had similar performance to traditional machine learning in off-line classification. Considering the high robustness and generalization ability, deep learning is likely a more robust alternative to traditional machine learning in the field of sEMG hand gesture recognition.