A Hybrid Cnn-Svm Classifier For Hand Gesture Recognition With Surface Emg Signals

A synthetic approach was proposed to improve the recognition accuracy. Different with the traditional feature extractors, this study used a convolutional neural network (CNN) to automatically extract characteristics from the input of raw EMG image. Then, a Support Vector Machine (SVM) classifier was employed to identify the hand motions. Our experiments showed that the proposed method achieved the accuracy around 2.5% higher than the use of CNN only, and about 9.7% higher than the use of traditional method (i.e. the use of time domain feature and a SVM classifier). Both inter-subject and inter-session tests demonstrated the robustness of the CNN-based feature.