Multi-gesture recognition method of single-channel EMG signal based on deep learning

The electrical signals of human upper limb 1 muscles contain a wealth of motion information, and certain progress has been made in decoding biological signals for motion recognition, such as dexterous artificial hands. However, the current research on gesture recognition still has outstanding problems such as inaccurate action classification, poor generalization effect, and complicated data information. In this regard, this paper proposes a single channel and multi-gesture recognition model of comprehensive feature screening and convolutional neural network (CNN) to quickly and accurately classify the data set of eight types of upper limb hand movements. The research objectives include (1) verify the effectiveness of single channel feature screening; (2) explore the accuracy of single channel EMG recognition combined with convolutional networks. The method was to extract the time domain, frequency domain, time-frequency domain and non-linear features by pre-processing and detecting the start and end points of the collected EMG signals of eight gestures of the upper limbs. The Pearson correlation coefficient method was used to screen high correlation features, and single channel EMG features were constructed; combined with the proposed deep learning model, eight upper limb movements were recognized. The results show that using the proposed model, the average recognition rate of single channel EMG signal gestures is as high as 94.4%, which is much higher than machine learning algorithms such as BP, SVM, ELM and FA-ELM, which fully verifies the effectiveness of the scheme.