Myoelectric gesture recognition method based on multi-feature fusion CNN

Abstract

The invention discloses a myoelectric gesture recognition method based on a multi-feature fusion CNN, and provides a feature fusion convolutional neural network architecture based on parallelism. In order to simplify the training process and reduce the dependence of a training set on the design of a specific feature set, the method is designed on the basis of a convolutional neural network architecture in a deep learning method; secondly, a parallel neural network is used for training a time domain signal and a frequency domain signal at the same time in architecture, so that the training efficiency is improved, and the training time is shortened; and in the training model, shallow data features at the initial stage of training are extracted, the shallow data features are fused with deep features at the tail end of the network, and the fused features are inputted into a classification layer for classification.

Description

technical field [0001] The invention belongs to the field of physiological signal recognition, and in particular relates to a myoelectric gesture recognition method based on multi-feature fusion CNN. Background technique [0002] Physiological signals are an important data resource that can help in the detection, treatment and rehabilitation of diseases. It mainly includes signals spontaneously generated during normal physiological activities of the human body, such as electrophysiological signals such as ECG, EEG, EMG, and EEG, and non-electrophysiological signals such as respiration, pulse, and body temperature. Among them, electromyography (ElectroMyoGram, EMG) is an important bioelectrical signal generated with muscle activity, which is the superposition of many muscle fiber motor potentials in time and space. This information can reflect information such as the state of the joints and the shape and position of the limbs during the movement, and is an important way to p...

AI Technical Summary

Problems solved by technology

[0004] In general, both methods still have shortcomings. Using machine learning methods for sEMG gesture recognition requires selecting different feature sets according to different application scenarios, and the time for model training is relatively short, but for researchers, it is necessary to Designing corresponding feature extraction methods according to different input data is cumbersome and complicated; while using deep learning methods, only continuous training and iterative extraction of features for the input sEMG signals through the neural network can easily lead to layer-by-layer convolution. After product extraction, the features are too simplified, and part of the feature information in the original signal is lost.

Images