Neural network gesture action classification algorithm based on multi-channel combination

Abstract

The invention designs a neural network gesture action classification algorithm based on multi-channel combination, and a four-channel gesture action recognition algorithm is built by combining wavelet packet coefficient and time domain combination features on the premise of processing acquired signals by using an LMS algorithm. Wavelet packet coefficient maximum value features and coefficient energy features in the time domain features, the frequency domain features and the time-frequency domain features are comprehensively compared; each channel is independently analyzed; taking the maximum value of the wavelet packet coefficient of the channel I, the average absolute value of the channel II and the channel III, and the maximum value and the coefficient energy value of the wavelet packet coefficient of the channel IV as combined features to carry out gesture classification on seven different gesture actions of bending a thumb, bending the thumb and an index finger, bending the index finger, bending four fingers, clenching a fist, bending the index finger and a middle finger and bending three middle fingers; a BP neural network is used as a classifier, dimension reduction processing is carried out through a principal component analysis method, and the average recognition rate of 91.1% is obtained.

Description

technical field [0001] The invention relates to the technical field of gesture action pattern recognition and artificial intelligence, in particular to a multi-channel combination-based neural network gesture action classification algorithm. Background technique [0002] Surface EMG signals are widely used in rehabilitation medicine and clinical diagnosis, and gesture action classification based on surface EMG signals is often used in the treatment of hand hemiplegia. Among them, the extraction of surface EMG features is the key to gesture action classification. Because of the crosstalk and weakness of surface EMG signals, good filtering is a prerequisite for obtaining high-quality surface EMG signals. [0003] At present, the main EMG signal filtering method is to use traditional filters to process in the acquisition system, including Butterworth filter, Chebyshev filter, Bessel filter and so on. But a single filter is often worried about the problem of poor filtering effe...

AI Technical Summary

Problems solved by technology

[0005] The object of the present invention is to provide a neural network gesture action classification algorithm based on multi-channel combination, to solve the problem of low surface electromyography signal classification recognition rate due to noise interference and poor feature classification effect

Images