Bionic manipulator interaction control method based on electromyographic signal mode recognition and particle swarm optimization

Abstract

The invention discloses a bionic manipulator interaction control method based on electromyographic signal mode recognition and particle swarm optimization. The method comprises the steps of firstly, establishing a pattern recognition system based on surface electromyogram signals for pre-learning; acquiring a surface electromyographic signal through a multi-channel electromyographic instrument; after Butterworth filtering and software filtering of a comb filter, performing feature data extraction work of multiple features, performing pattern recognition on the basis of all-channel offline feature data, and performing optimal feature combination and optimal channel combination optimization search by adopting a particle swarm optimization algorithm; acquiring real-time surface electromyogramsignals on the basis of the optimal feature combination and the optimal channel combination, carrying out real-time gesture recognition by the aid of KNN algorithms after the real-time surface electromyogram signals are subjected to filtering processing and feature extraction, and carrying out real-time robust control on bionic manipulators with various degrees of freedom by the aid of recognizedgesture results.

Description

technical field [0001] The invention belongs to the technical field of myoelectric signal detection and pattern recognition, in particular to a bionic manipulator interactive control method based on myoelectric signal pattern recognition and particle swarm optimization. Background technique [0002] The bionic manipulator is a branch of robotics and plays a vital role in the fields of machinery, medical treatment, and scientific research. By combining the cutting-edge high-tech of artificial intelligence, robots and measurement and control instruments, it is possible for the bionic manipulator to cooperate with the myoelectric signal acquisition device to serve a large number of disabled people caused by accidents or acquired diseases, so that they can use myoelectric control prosthetics , and using myoelectric control prosthetics, not only to quickly identify the myoelectric signal, but also to ensure the accuracy of recognition. [0003] The existing technology has the fo...

AI Technical Summary

Problems solved by technology

[0004] The current status of surface EMG signal pattern recognition is in the exploration stage of optimal EMG signal features and classification methods. The EMG features used in different papers are often different, and it is difficult to judge which one is better and which is worse.

Zhang Qizhong, Xi Xugang and others used the characteristics of approximate entropy and fractal dimension in the document "Wrist Movement Pattern Recognition Based on Surface EMG Signal, Chinese Journal of Biomedical Engineering, 2013, 32(03): 257-265." At the same time, a KNN (K-Nearest Neighbor, K nearest neighbor classification) model incremental learning algorithm with incremental learning ability is used as a classifier for pattern recognition. Although the accuracy rate is slightly improved compared with the traditional KNN algorithm, its feature extraction only uses It cannot prove that this feature is the optimal feature combination, not to mention that the calculation of approximate entropy features and fractal dimension features is very complicated and will take up a lot of computing time. Although there are certain advantages in accuracy, but in real time When controlling, we must consider that gesture recognition should be fast

[0005] Zhang Daohui and others extracted the absolute mean value, root mean square, zero-crossing points, waveform length, slope sign change and AR model coefficients in the document "Research on EMG Control Method of Bionic Manipulator, Shenyang University of Technology, 2013." One kind of time-domain features, average power, average frequency, median frequency-three kinds of frequency-domain features, wavelet transform coefficient-one kind of time-frequency domain features, non-breadth entropy and sample entropy-two kinds of nonlinear entropy features and correlation dimension and Box function—2 kinds of fractal features, a total of 14 kinds of EMG signal features, and then combined these 14 kinds of features to get 4 feature combinations, and then used binary tree structure KD tree, multi-fork tree structure IHDR tree, PCA and LDA These four EMG signal classification methods carry out pattern classification, and use PCA and LDA classifiers to conduct EMG signal control experiments of bionic manipulators; however, the feature combination method is artificial combination, and the combination of features is not comprehensive enough. Combination may not be optimal in terms of accuracy and speed of real-time control

[0006] Xu Yanbin also described his experiment of using myoelectric signal to control the robot online in his master thesis "Robot Teleoperation Control Based on EMG Signal [D]. South China University of Technology, 2018." According to the feature extraction method adopted, it can be Divided into three groups: the first group is the number of zero crossing points (ZC), root mean square (RMS), and waveform length (WL), and the second group is the absolute mean value (MAV), root mean square (RMS), and the number of slope sign changes (SSC), the third group is the absolute mean value (MAV), root mean square (RMS), and zero-crossing points (ZC). Using these three feature combinations, combined with the LDA linear discriminant analysis classification model to realize gesture recognition, select the accuracy rate The highest feature combination is used as the feature combination of online control; but there are some shortcomings in this experiment. First of all, the selection of feature combination is artificial combination, and the number of features available for combination is too small, so it is difficult to find a better one. feature combination, so it is difficult to achieve the accuracy and speed that should be met when the EMG signal is controlled online

Images