阅读说明：本技术 基于游戏和肌电信号的cpm下肢康复训练方法及系统 (CPM lower limb rehabilitation training method and system based on game and electromyographic signals ) 是由 史小华 刘瑞发 李月娟 孙杰智 姚吉路 李雪飞 王襄 于 2019-09-25 设计创作，主要内容包括：本发明公开了一种基于游戏和肌电信号的CPM下肢康复训练方法及系统,涉及生机电一体化技术领域,解决患者康复训练时枯燥、训练行动不便的问题。主要包括以游戏为中介,通过建立的生物人机接口采集患者手势动作的表面肌电信号并进行频时域分析以提取手势特征,然后采用主成分分析法或者时间序列评价特征质量准则对手势特征降维,接着通过人工神经网络识别手势特征所对应的手势类型,最后基于手势类型与CPM机行为动作的对应关系确定该患者手势类型对应的CPM机行为动作,并根据该患者手势类型对应的CPM机行为动作控制CPM机以实现训练患者受损下肢的目的。本发明实现了通过患者上肢游戏方式控制CPM机训练患者受损下肢的效果。(The invention discloses a CPM lower limb rehabilitation training method and system based on games and electromyographic signals, relates to the technical field of raw electro-mechanical integration, and solves the problems of boring and inconvenient training and action of a patient during rehabilitation training. The method mainly comprises the steps of taking a game as a medium, collecting surface electromyographic signals of the gesture actions of a patient through an established biological human-computer interface, carrying out frequency-time domain analysis to extract gesture features, then adopting a principal component analysis method or a time sequence evaluation feature quality criterion to carry out dimensionality reduction on the gesture features, then identifying gesture types corresponding to the gesture features through an artificial neural network, finally determining the behavior actions of the CPM machine corresponding to the gesture types of the patient based on the corresponding relation between the gesture types and the behavior actions of the CPM machine, and controlling the CPM machine according to the behavior actions of the CPM machine corresponding to the gesture types of the patient to achieve the purpose of training the damaged lower limbs of the patient. The invention realizes the effect of controlling the CPM machine to train the damaged lower limbs of the patient in a way of upper limb game of the patient.)

1. A CPM lower limb rehabilitation training method based on games and electromyographic signals is characterized by comprising the following steps:

acquiring a sample gesture surface electromyographic signal of a healthy upper limb of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture and surface electromyographic signals corresponding to a resting gesture;

carrying out frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture characteristics, and constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics;

performing dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method or a time series evaluation feature quality criterion to obtain a low-dimensional feature space matrix; the feature vector in the low-dimensional feature space matrix is a sample main gesture feature;

training a neural network model according to the main gesture features of the sample to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is a main gesture feature, and the output of the gesture recognition neural network model is a gesture type;

acquiring and processing real-time gesture surface electromyographic signals of healthy upper limbs of a patient when the patient plays games;

inputting the processed real-time gesture surface electromyographic signals into the gesture recognition neural network model to obtain real-time gesture types corresponding to the real-time gesture surface electromyographic signals;

and determining the CPM machine behavior action corresponding to the real-time gesture type according to the corresponding relation between the gesture type and the CPM machine behavior action, and training the damaged lower limb of the patient according to the CPM machine behavior action corresponding to the real-time gesture type.

2. The CPM lower limb rehabilitation training method based on games and electromyographic signals according to claim 1, wherein before obtaining the sample gesture surface electromyographic signals of the healthy upper limb of the patient, the CPM lower limb rehabilitation training method further comprises: building a surface electromyogram signal acquisition and processing device; the surface electromyogram signal acquisition processing device comprises an acquisition electrode, a data acquisition hardware circuit and an upper computer, wherein the data acquisition hardware circuit is formed by combining a Duinopeak electromyogram sensor and an Arduino UNO singlechip; the acquisition electrode is connected with the Duinpeak electromyography sensor through an electromyography lead wire; the data acquisition hardware circuit is connected with the upper computer.

3. The CPM lower limb rehabilitation training method based on games and electromyographic signals according to claim 2, wherein the frequency domain analysis and the time domain analysis are performed on the sample gesture surface electromyographic signals to obtain sample gesture features, and a high-dimensional feature space matrix is constructed according to the sample gesture features, specifically comprising:

carrying out frequency domain analysis on the sample gesture surface electromyographic signals according to a high-pass-low-pass filter circuit in the Duinpeak electromyographic sensor; the up-to-down frequency of the high-pass and low-pass filter circuit is 100Hz, and the down-to-down frequency of the high-pass and low-pass filter circuit is 50 Hz;

segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows, and extracting time domain features from each time window; the time domain features are sample gesture features;

and constructing a high-dimensional feature space matrix according to the sample gesture features.

4. The CPM lower limb rehabilitation training method based on games and electromyographic signals according to claim 1, wherein before acquiring and processing the real-time gesture surface electromyographic signals of the healthy upper limb of the patient while playing the game, the CPM lower limb rehabilitation training method further comprises:

constructing a stone throwing game based on threshold control; the stone throwing game is characterized in that the amplitude of the myoelectric signal of the hand gesture surface is collected through a quantitative sensor so as to control the stretching length of a rubber band in the stone throwing game.

5. The CPM lower limb rehabilitation training method based on games and electromyographic signals according to claim 1, wherein the acquiring and processing of real-time gesture surface electromyographic signals of healthy upper limbs of a patient while playing games specifically comprises:

acquiring real-time gesture surface electromyographic signals of healthy upper limbs of a patient when the patient plays games;

determining the amplitude of the real-time gesture surface electromyographic signal, and deleting the real-time gesture surface electromyographic signal corresponding to the amplitude larger than a set threshold;

performing frequency domain analysis and time domain analysis on the reserved real-time gesture surface electromyographic signals to obtain real-time gesture characteristics, and constructing a real-time high-dimensional characteristic space matrix according to the real-time gesture characteristics;

performing dimensionality reduction on the real-time high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a real-time low-dimensional feature space matrix; and the feature vectors in the real-time low-dimensional feature space matrix are real-time main gesture features.

6. A CPM lower limb rehabilitation training system based on games and electromyographic signals, which is characterized by comprising:

the first acquisition module is used for acquiring sample gesture surface electromyographic signals of healthy upper limbs of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture and surface electromyographic signals corresponding to a resting gesture;

the analysis module is used for carrying out frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture characteristics and constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics;

the dimensionality reduction module is used for carrying out dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a low-dimensional feature space matrix; the feature vector in the low-dimensional feature space matrix is a sample main gesture feature;

the gesture recognition neural network model obtaining module is used for training a neural network model according to the main gesture characteristics of the sample to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is a main gesture feature, and the output of the gesture recognition neural network model is a gesture type;

the second acquisition module is used for acquiring and processing real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when the patient plays games;

the recognition module is used for inputting the processed real-time gesture surface electromyographic signals into the gesture recognition neural network model to obtain real-time gesture types corresponding to the real-time gesture surface electromyographic signals;

and the training module is used for determining the CPM machine behavior action corresponding to the real-time gesture type according to the corresponding relation between the gesture type and the CPM machine behavior action, and training the damaged lower limb of the patient according to the CPM machine behavior action corresponding to the real-time gesture type.

7. The CPM lower limb rehabilitation training system based on games and electromyographic signals according to claim 6, wherein the analysis module specifically comprises:

the frequency domain analysis unit is used for carrying out frequency domain analysis on the sample gesture surface electromyographic signals; the cut-up frequency of the high-pass-low-pass filtering in the frequency domain analysis unit is 100Hz, and the cut-down frequency of the high-pass-low-pass filtering is 50 Hz;

the time domain analysis unit is used for segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows and extracting time domain features from each time window; the time domain features are sample gesture features;

and the high-dimensional characteristic space matrix construction unit is used for constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics.

8. The CPM lower limb rehabilitation training system based on games and electromyographic signals of claim 6, further comprising:

the game construction module is used for constructing a threshold control-based stone throwing game; the stone throwing game is characterized in that the amplitude of the myoelectric signal of the hand gesture surface is collected through a quantitative sensor so as to control the stretching length of a rubber band in the stone throwing game.

9. The CPM lower limb rehabilitation training system based on games and electromyographic signals according to claim 6, wherein the second obtaining module specifically comprises:

the acquisition unit is used for acquiring real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when the patient plays games;

the deleting unit is used for determining the amplitude of the real-time gesture surface electromyographic signal and deleting the real-time gesture surface electromyographic signal corresponding to the amplitude larger than a set threshold;

the analysis unit is used for carrying out frequency domain analysis and time domain analysis on the reserved real-time gesture surface electromyographic signals to obtain real-time gesture characteristics, and constructing a real-time high-dimensional characteristic space matrix according to the real-time gesture characteristics;

the dimension reduction unit is used for carrying out dimension reduction processing on the real-time high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a real-time low-dimensional feature space matrix; and the feature vectors in the real-time low-dimensional feature space matrix are real-time main gesture features.

Technical Field

The invention relates to the technical field of bio-electro-mechanical integration, in particular to a CPM lower limb rehabilitation training method and system based on games and electromyographic signals.

Background

And displaying the latest data: the elderly population is expected to reach 2.48 billion by 2020, the threat of limb disorder diseases gradually emerges in society, and frequent occurrence of natural disasters and traffic accidents also causes a large number of patients with damaged limb joints. This type of patient need train with the help of the instrument when carrying out the rehabilitation, nevertheless because the weakening of patient's self motion ability to and the boring during training for the patient has certain degree of difficulty when carrying out the rehabilitation training, hardly insists on. Moreover, the patients of the type need close-fitting care of family members, the normal life of the patients and even the whole family is seriously influenced, the serious life pressure is caused to the patients of the type, and even the patients of the type are violently abandoned and lose the confidence of life.

With the proposal of the concept of man-machine integration in recent years, the rehabilitation training technology meets the important development opportunity. The biological man-machine interface is an important way for human beings to communicate with machine equipment, and aims to take various physiological signals of human bodies as control signal sources for controlling external machine equipment so as to replace the traditional control mode. The collection and analysis of surface electromyographic signals are relatively simple and convenient, and gradually become one of the key research objects of biological human-computer interfaces. The surface electromyogram signal of the patient can be used as an action source during rehabilitation training, so that the training efficiency is greatly improved, but the research result in the aspect is few.

Disclosure of Invention

The invention aims to provide a CPM lower limb rehabilitation training method and system based on games and electromyographic signals.

In order to achieve the purpose, the invention provides the following scheme:

a CPM lower limb rehabilitation training method based on games and electromyographic signals comprises the following steps:

acquiring a sample gesture surface electromyographic signal of a healthy upper limb of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture and surface electromyographic signals corresponding to a resting gesture;

carrying out frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture characteristics, and constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics;

performing dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method or a time series evaluation feature quality criterion to obtain a low-dimensional feature space matrix; the feature vector in the low-dimensional feature space matrix is a sample main gesture feature;

training a neural network model according to the main gesture features of the sample to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is a main gesture feature, and the output of the gesture recognition neural network model is a gesture type;

acquiring and processing real-time gesture surface electromyographic signals of healthy upper limbs of a patient when the patient plays games;

inputting the processed real-time gesture surface electromyographic signals into the gesture recognition neural network model to obtain real-time gesture types corresponding to the real-time gesture surface electromyographic signals;

and determining the CPM machine behavior action corresponding to the real-time gesture type according to the corresponding relation between the gesture type and the CPM machine behavior action, and training the damaged lower limb of the patient according to the CPM machine behavior action corresponding to the real-time gesture type.

Optionally, before obtaining the sample gesture surface electromyogram signal of the healthy upper limb of the patient, the CPM lower limb rehabilitation training method further includes: building a surface electromyogram signal acquisition and processing device; the surface electromyogram signal acquisition processing device comprises an acquisition electrode, a data acquisition hardware circuit and an upper computer, wherein the data acquisition hardware circuit is formed by combining a Duinopeak electromyogram sensor and an Arduino UNO singlechip; the acquisition electrode is connected with the Duinpeak electromyography sensor through an electromyography lead wire; the data acquisition hardware circuit is connected with the upper computer.

Optionally, the frequency domain analysis and the time domain analysis are performed on the sample gesture surface electromyographic signal to obtain a sample gesture feature, and a high-dimensional feature space matrix is constructed according to the sample gesture feature, which specifically includes:

carrying out frequency domain analysis on the sample gesture surface electromyographic signals according to a high-pass-low-pass filter circuit in the Duinpeak electromyographic sensor; the up-to-down frequency of the high-pass and low-pass filter circuit is 100Hz, and the down-to-down frequency of the high-pass and low-pass filter circuit is 50 Hz;

segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows, and extracting time domain features from each time window; the time domain features are sample gesture features;

and constructing a high-dimensional feature space matrix according to the sample gesture features.

Optionally, before obtaining and processing a real-time gesture surface electromyographic signal of a healthy upper limb of a patient when playing a game, the CPM lower limb rehabilitation training method further includes:

constructing a stone throwing game based on threshold control; the stone throwing game is characterized in that the amplitude of the myoelectric signal of the hand gesture surface is collected through a quantitative sensor so as to control the stretching length of a rubber band in the stone throwing game.

Optionally, the obtaining and processing of the real-time gesture surface electromyogram signal of the healthy upper limb of the patient during game playing specifically includes:

acquiring real-time gesture surface electromyographic signals of healthy upper limbs of a patient when the patient plays games;

determining the amplitude of the real-time gesture surface electromyographic signal, and deleting the real-time gesture surface electromyographic signal corresponding to the amplitude larger than a set threshold;

performing frequency domain analysis and time domain analysis on the reserved real-time gesture surface electromyographic signals to obtain real-time gesture characteristics, and constructing a real-time high-dimensional characteristic space matrix according to the real-time gesture characteristics;

performing dimensionality reduction on the real-time high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a real-time low-dimensional feature space matrix; and the feature vectors in the real-time low-dimensional feature space matrix are real-time main gesture features.

A CPM lower limb rehabilitation training system based on games and electromyographic signals comprises:

the first acquisition module is used for acquiring sample gesture surface electromyographic signals of healthy upper limbs of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture and surface electromyographic signals corresponding to a resting gesture;

the analysis module is used for carrying out frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture characteristics and constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics;

the dimensionality reduction module is used for carrying out dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a low-dimensional feature space matrix; the feature vector in the low-dimensional feature space matrix is a sample main gesture feature;

the gesture recognition neural network model obtaining module is used for training a neural network model according to the main gesture characteristics of the sample to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is a main gesture feature, and the output of the gesture recognition neural network model is a gesture type;

the second acquisition module is used for acquiring and processing real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when the patient plays games;

the recognition module is used for inputting the processed real-time gesture surface electromyographic signals into the gesture recognition neural network model to obtain real-time gesture types corresponding to the real-time gesture surface electromyographic signals;

and the training module is used for determining the CPM machine behavior action corresponding to the real-time gesture type according to the corresponding relation between the gesture type and the CPM machine behavior action, and training the damaged lower limb of the patient according to the CPM machine behavior action corresponding to the real-time gesture type.

Optionally, the analysis module specifically includes:

the frequency domain analysis unit is used for carrying out frequency domain analysis on the sample gesture surface electromyographic signals; the cut-up frequency of the high-pass-low-pass filtering in the frequency domain analysis unit is 100Hz, and the cut-down frequency of the high-pass-low-pass filtering is 50 Hz;

the time domain analysis unit is used for segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows and extracting time domain features from each time window; the time domain features are sample gesture features;

and the high-dimensional characteristic space matrix construction unit is used for constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics.

Optionally, the CPM lower limb rehabilitation training system further includes:

the game construction module is used for constructing a threshold control-based stone throwing game; the stone throwing game is characterized in that the amplitude of the myoelectric signal of the hand gesture surface is collected through a quantitative sensor so as to control the stretching length of a rubber band in the stone throwing game.

Optionally, the second obtaining module specifically includes:

the acquisition unit is used for acquiring real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when the patient plays games;

the deleting unit is used for determining the amplitude of the real-time gesture surface electromyographic signal and deleting the real-time gesture surface electromyographic signal corresponding to the amplitude larger than a set threshold;

the analysis unit is used for carrying out frequency domain analysis and time domain analysis on the reserved real-time gesture surface electromyographic signals to obtain real-time gesture characteristics, and constructing a real-time high-dimensional characteristic space matrix according to the real-time gesture characteristics;

the dimension reduction unit is used for carrying out dimension reduction processing on the real-time high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a real-time low-dimensional feature space matrix; and the feature vectors in the real-time low-dimensional feature space matrix are real-time main gesture features.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the invention provides a CPM lower limb rehabilitation training method and system based on games and electromyographic signals. The method comprises the steps of taking a game as a medium, collecting surface electromyographic signals of the gesture actions of a patient through an established biological human-computer interface, carrying out frequency-time domain analysis to extract gesture features, then carrying out dimensionality reduction on the gesture features by adopting a principal component analysis method or a time sequence evaluation feature quality criterion, identifying gesture types corresponding to the gesture features through an artificial neural network to complete control over game software, finally determining CPM machine behavior actions corresponding to the gesture types of the patient based on the corresponding relation between the gesture types and the CPM machine behavior actions, and controlling the CPM machine according to the CPM machine behavior actions corresponding to the gesture types of the patient to achieve the purpose of training the damaged lower limbs of the patient.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a schematic flow chart of a CPM lower limb rehabilitation training method based on games and electromyographic signals according to an embodiment of the invention;

FIG. 2 is a schematic structural diagram of a surface electromyographic signal acquisition and processing device according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a hardware circuit structure of a Duinopeak electromyography sensor according to an embodiment of the invention;

FIG. 4 is a schematic diagram of an artificial neuron node model according to an embodiment of the present invention;

FIG. 5 is a flow chart of an interactive game according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a CPM lower limb rehabilitation training system based on games and electromyographic signals according to an embodiment of the invention

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention provides a novel and fashionable CPM lower limb rehabilitation training method and system based on games and electromyographic signals and suitable for the disabled, aiming at the social problem that the disabled faces rehabilitation training and needs to be solved urgently.

The invention not only achieves a certain purpose of exercising, but also enjoys the pleasure of exercising, and simultaneously rebuilds the confidence of life, is beneficial to the recovery of the physical function and the confidence of the disabled, and has strong value.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

In order to achieve the purpose, the invention adopts the following technical scheme:

as shown in FIG. 1, the invention provides a CPM lower limb rehabilitation training method based on games and electromyographic signals, which specifically comprises the following steps.

Step 101: acquiring a sample gesture surface electromyographic signal of a healthy upper limb of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture, surface electromyographic signals corresponding to a resting gesture and the like.

Step 102: and carrying out frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture characteristics, and constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics.

Step 103: performing dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method or a time series evaluation feature quality criterion to obtain a low-dimensional feature space matrix; and the feature vectors in the low-dimensional feature space matrix are sample main gesture features.

Step 104: training a neural network model according to the main gesture features of the sample to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is the main gesture characteristics, and the output of the gesture recognition neural network model is the gesture type.

Step 105: and acquiring and processing real-time gesture surface electromyographic signals of healthy upper limbs of the patient when playing games.

Step 106: and inputting the processed real-time gesture surface electromyographic signals into the gesture recognition neural network model to obtain real-time gesture types corresponding to the real-time gesture surface electromyographic signals.

Step 107: and determining the CPM machine behavior action corresponding to the real-time gesture type according to the corresponding relation between the gesture type and the CPM machine behavior action, and training the damaged lower limb of the patient according to the CPM machine behavior action corresponding to the real-time gesture type.

Preferably, step 101 is to collect time domain and frequency domain information of the sample gesture surface electromyographic signals of the healthy limb or the upper limb of the patient.

Before step 101 is executed, a surface electromyogram signal acquisition and processing device is constructed. As shown in fig. 2, the surface electromyogram signal acquisition and processing device comprises an acquisition electrode, a data acquisition hardware circuit formed by combining a Duinoseal electromyogram sensor and an Arduino UNO singlechip, and an upper computer. The acquisition electrode is connected with the Duinpeak electromyography sensor through an electromyography lead wire, the acquired information is input by a data acquisition hardware circuit which combines the Duinpeak electromyography sensor and the Arduino UNO singlechip, and the acquisition electrode is a disposable button type Ag/AgCl electrocardioelectrode. The main process is that after the Duinpeak electromyographic sensor acquires the surface electromyographic signals acquired by the acquisition electrodes, the surface electromyographic signals are output to an Arduino UNO singlechip, and finally the acquired signals are uploaded to an upper computer through a USB interface for subsequent processing.

The hardware circuit structure of the Duinopeak electromyography sensor is shown in FIG. 3, and comprises modules of pre-amplification, band-pass filtering, power frequency notch, gain amplification and the like which are connected in sequence.

Preferably, step 102 essentially comprises: the wrist inward turning gesture (WF), the outward turning gesture (WE), the fist making gesture (FC), the palm stretching gesture (PS) and the rest gesture (RT) are collected by the constructed surface electromyographic signal collecting and processing device to serve as sample gestures, five sample gesture surface electromyographic signals are counted, and frequency domain analysis and time domain analysis are carried out on the surface electromyographic signals.

1) And carrying out frequency domain analysis on the collected sample gesture surface electromyographic signals.

The cut-off frequency of the high-pass-low-pass filter circuit combination of the original Duinoseal electromyography sensor directly destroys the frequency of the electromyography signal on the surface of the original sample gesture, and the improved Duinoseal electromyography sensor is adopted for acquiring the information on the frequency domain of the electromyography signal on the surface of the sample gesture. The main effective information of the electromyographic signals on the surface of the sample gesture is concentrated in a frequency band of 50-100 Hz, so that the cut-off frequencies of the sample gesture are respectively 100Hz and 50 Hz. The low pass filtered RC parameter is calculated from equation (1):

wherein R represents a resistance value in the RC circuit,c represents the capacitance value in the RC circuit. To facilitate the selection and use of the components, take C₁₅0.02 μ F. The parameters of the high-pass filter RC are calculated in the same way.

And then, carrying out frequency domain analysis on the collected sample gesture surface electromyographic signals through a newly constructed high-pass-low-pass filter circuit.

2) And performing time domain analysis on the sample gesture surface electromyographic signals subjected to the frequency domain analysis.

And segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows, extracting time domain features from each time window, and first predefining a threshold value when extracting the features. The time domain features are sample gesture features.

3) And constructing a high-dimensional feature space matrix according to the sample gesture features.

4) The five gestures collectively extract features in eleven time domains of absolute value integral IAV, mean absolute value MAV, mean absolute value slope MAVS, waveform length WL, peak value PV, mean MV, variance VAR, standard deviation STD, root mean square RMS, form factor MS, Willison amplitude WAMP. The screened sample gesture features, namely the features in 11 time domains, can form an 11-dimensional high-dimensional feature space matrix.

Wherein x is_dRepresenting a sample gesture feature and d represents the dimension of the high-dimensional feature space matrix.

Preferably, step 103 specifically includes:

1) and performing dimensionality reduction on the high-dimensional feature space matrix by adopting a time sequence evaluation feature quality criterion, introducing a time sequence concept to perform feature selection to select optimal time domain features, screening the extracted time domain features by mainly adopting the time sequence evaluation feature quality criterion to select the optimal time domain features, and determining the optimal time domain features as sample gesture features. The time series evaluation characteristic quality criterion comprises parameters such as zero crossing points (ZC), skewness (skew), kurtosis (kurt), autocorrelation analysis, correlation coefficient analysis and Shannon entropy.

2) And performing dimensionality reduction on the high-dimensional feature space matrix by adopting a principal component analysis method.

The invention utilizes principal component analysis method to reduce dimension of features. The principle is that when a statistical analysis method is used to research a multi-variable problem, the complexity of the problem is increased when the number of variables is too large. It is naturally desirable to obtain a larger amount of information with a smaller number of variables. In many cases, there is a certain correlation between variables, and when there is a certain correlation between two variables, it can be interpreted that there is a certain overlap of information reflecting the subject. The principal component analysis is to eliminate redundant repeated variables (closely related variables) for all the originally proposed variables, and establish new variables as few as possible, so that the new variables are irrelevant pairwise, and the new variables keep original information as much as possible in the aspect of reflecting the information of the subject.

Firstly, calculating the correlation coefficient among all the eigenvectors in the high-dimensional characteristic space matrix, and knowing whether redundancy exists among the characteristics and the degree of the redundancy through the correlation coefficient. The correlation coefficient calculation formula is as follows:

the correlation coefficient between the feature vectors can be obtained by the equation (3), and the closer the correlation coefficient is to 1, the higher the degree of similarity between the two features, and the higher the redundancy, and the closer the correlation coefficient is to 0, the lower the redundancy between the two features. The dimension reduction can be realized by comparing the correlation coefficients among the feature vectors and eliminating the features with the correlation number close to 1.

And secondly, simplifying original complex data through orthogonal transformation, and removing redundancy and noise of the data. The 11 extracted time domain features form an 11-dimensional high-dimensional feature space matrix X, and a covariance matrix C of the high-dimensional feature space matrix X is firstly obtained:

of these, cov (x)_i,x_j)＝E[(x_i-E(x_i))(x_j-E(x_j))^T]。

Then, the eigenvectors are arranged from large to small according to the corresponding eigenvalue size, and the front k (k is less than 11) is taken to form a matrix y_iThen the original 11-dimensional feature vector is converted into a k-dimensional new feature vector.

Cα_i＝λ_iα_i (5)；

Wherein λ is_iIs the eigenvalue of covariance matrix C, and the corresponding eigenvector is alpha_i＝(α_i1,α_i2,…,α_i11)^T(i＝1,2,…,k)。

New variable y obtained by orthogonal transformation_iThe expression (i ═ 1,2, …, k) is:

finally, the signal needs to be reconstructed after the orthogonal transformation. Since the first k eigenvectors are selected, the two sides are left-multiplied by alpha_iAnd summed to obtain an approximate reconstruction matrix:

the approximate reconstruction matrix is a high-dimensional feature space matrix after dimensionality reduction, namely a low-dimensional feature space matrix.

Preferably, step 104 essentially comprises:

1) an artificial neural network model is built using artificial neurons as shown in fig. 4 for computation and transfer of information. The structure of the artificial neuron can be equivalent to a nonlinear logic adder with multiple inputs and single output.

2) And (5) carrying out neural network structure design. The method mainly comprises an input layer design, an output layer design and a hidden layer design. In the design of the input layer, because the characteristic space (TD characteristic) reduced by the time sequence is a 6-dimensional space vector and the new characteristic space (PCA characteristic) obtained by the PCA dimension reduction is a 5-dimensional space vector, the number of nodes of the input layer is designed to be 6 in order to conveniently compare the influence of the TD characteristic and the new characteristic space on the identification performance; in the design of the output layer, the test samples need to be divided into 5 types, namely: inversion (WF), eversion (WE), Fist (FC), palm extension (PS) and Rest (RT), so the target vector is 5-dimensional, i.e. the number of neuron nodes of the output layer of the neural network is set to 5.

3) And (5) designing a neural network algorithm. The output from the input layer to the next layer is represented as:

wherein k is the current iteration frequency, the same as below; x (1), x (2), x (3), x (4), x (5), and x (6) are respectively 6 eigenvectors in the feature sample subspace, that is, the eigenvectors in the low-dimensional feature space matrix.

The input and output of each neuron in the middle layer (hidden layer) of the network are

Wherein, the upper corner marks (1), (2) and (3) respectively represent an input layer, a hidden layer and an output layer;

representing hidden layer weight;

representing a hidden layer threshold; hidden layer neuron activation function f (x) adopts a symmetric Sigmoid function.

Finally, each neuron of the output layer of the neural network inputs and outputs

Wherein the content of the first and second substances,representing the weight of the output layer;

represents an output layer threshold; the output layer neuron activation function g (x) employs a non-negative Sigmoid function.

Because the output of the output layer is 5 dimensions and is respectively 5 gesture action codes needing to be identified and classified, the nodes of the output layer do not only have one node output result for each identification, each node of the output layer has one result for each identification, and each node represents the characteristic of one gesture respectively and is used for dividing the identification and classification results in the subsequent cost function.

Defining a network performance indicator function as

Wherein p is the number of current training samples; n is the total number of training samples; rin_l(k) And outputting the target value for the current node.

And (5) searching the negative gradient direction of the weight and the threshold of the output layer by using a performance index function E (k). It is toAnd

the partial derivatives of (A) can be respectively expressed as

Wherein the content of the first and second substances,

finally, obtaining an adjusting algorithm of the network output layer weight and the threshold:

where η is the learning rate of the neural network.

Similarly, an adjusting algorithm for deriving weight and threshold of hidden layer

4) And training the neural network model to obtain a gesture recognition neural network model, and realizing the classification of the gesture, namely training the neural network model by using the main gesture characteristics of the sample.

The desired output result is subjected to scale transformation, i.e., normalization processing. The formula for the input quantity pre-processing scaling is as follows:

wherein the content of the first and second substances,

denotes the normalized value, x_iRepresenting input data samples, x_minMinimum value, x, representing the range of variation of the input data_maxRepresenting the maximum value of the input data variation range.

The invention selects the expression method of '1 in n' according to the required output type, namely, the output result vector is a row vector of 1 multiplied by 5, and 5 elements in the row vector respectively represent 5 types of gestures required to be identified. When the identification result is a certain type, the corresponding element is 1, and the rest n-1 elements are 0. The classification result is visual and beneficial to observation, and is commonly used in the condition of few classification categories, the encoding form of the obtained gesture classification is shown in table 1, and further the training of the neural network is carried out.

TABLE 1 gesture Classification coding

Preferably, before step 105 is executed, a threshold control-based stone shooting game is constructed; mainly includes program (Processing) environment initialization and the operation process as shown in fig. 5. The initialization mainly comprises stones, slingshots with rubber bands, stone walls and backgrounds. The control principle of the game is based on myoelectricity threshold control, namely amplitude of the myoelectricity signal on the surface of the hand gesture is collected through a quantitative sensor so as to control the stretching length of a rubber band in the game. And meanwhile, an internal recorder is arranged, and the duration from the beginning to the end of the game and the quantity of the destroyed stone walls are recorded when the game runs, so that the score displayed by the player after the game is ended is evaluated.

Preferably, step 105 essentially comprises:

1) and acquiring real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when playing games. The interactive game flow structure diagram is shown in fig. 5, in the game, the length of the rubber band is controlled by a player to emit stones so as to destroy the stone wall, and then the real-time gesture surface myoelectric signals of the healthy upper limbs of the patient in the game process are collected.

2) Processing; and determining the amplitude of the real-time gesture surface electromyographic signal, deleting the real-time gesture surface electromyographic signal corresponding to the amplitude when the amplitude is larger than a set threshold value, and then carrying out step 102-step 103 processing on the reserved real-time gesture surface electromyographic signal. The processed real-time gesture surface electromyographic signals are real-time main gesture characteristics.

Preferably, step 107 comprises essentially the following:

1) performing action design on the recognized gesture; and forming one-to-one correspondence between various gesture types and different behavior actions of the CPM machine during the game.

2) The method comprises the steps of collecting real-time gesture surface electromyographic signals of a patient during game by using a designed surface electromyographic signal collecting and processing device, and uploading collected data to an upper computer (PC end), wherein Matlab is used as signal characteristic extraction processing software and pattern recognition classification processing software (gesture recognition neural network model operating software) at the PC end, and an obtained recognition result is sent to a CPM (converged Internet protocol) machine through a wireless signal, the CPM machine performs corresponding actions according to a designed mapping relation, so that the CPM machine is controlled based on the real-time gesture surface electromyographic signals, and the corresponding actions are shown in a table 2 by taking a CPM lower limb rehabilitator as an example.

TABLE 2 action design

As shown in fig. 6, the present invention further provides a CPM lower limb rehabilitation training system based on games and electromyographic signals, comprising:

the first acquisition module 100 is used for acquiring a sample gesture surface electromyographic signal of a healthy upper limb of a patient; the sample gesture surface electromyographic signals comprise surface electromyographic signals corresponding to a wrist inward turning gesture, surface electromyographic signals corresponding to a wrist outward turning gesture, surface electromyographic signals corresponding to a fist making gesture, surface electromyographic signals corresponding to a palm stretching gesture and surface electromyographic signals corresponding to a resting gesture.

The analysis module 200 is configured to perform frequency domain analysis and time domain analysis on the sample gesture surface electromyographic signals to obtain sample gesture features, and construct a high-dimensional feature space matrix according to the sample gesture features.

The dimensionality reduction module 300 is configured to perform dimensionality reduction on the high-dimensional feature space matrix by using a principal component analysis method or a time series evaluation feature quality criterion to obtain a low-dimensional feature space matrix; and the feature vectors in the low-dimensional feature space matrix are sample main gesture features.

A gesture recognition neural network model obtaining module 400, configured to train a neural network model according to the sample main gesture features to obtain a gesture recognition neural network model; the input of the gesture recognition neural network model is the main gesture characteristics, and the output of the gesture recognition neural network model is the gesture type.

And the second acquiring module 500 is used for acquiring and processing the real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when playing games.

The recognition module 600 is configured to input the processed real-time gesture surface electromyographic signal into the gesture recognition neural network model, so as to obtain a real-time gesture type corresponding to the real-time gesture surface electromyographic signal.

The training module 700 is configured to determine a CPM machine behavior corresponding to the real-time gesture type according to a correspondence between the gesture type and the CPM machine behavior, and train the damaged lower limb of the patient according to the CPM machine behavior corresponding to the real-time gesture type.

Preferably, the analysis module 200 specifically includes:

the frequency domain analysis unit is used for carrying out frequency domain analysis on the sample gesture surface electromyographic signals; the high-pass-low-pass filtering in the frequency domain analysis unit has a cut-up frequency of 100Hz and the high-pass-low-pass filtering has a cut-down frequency of 50 Hz.

The time domain analysis unit is used for segmenting the sample gesture surface electromyographic signals subjected to frequency domain analysis by using continuous time windows and extracting time domain features from each time window; the time domain features are sample gesture features.

And the high-dimensional characteristic space matrix construction unit is used for constructing a high-dimensional characteristic space matrix according to the sample gesture characteristics.

Preferably, the CPM lower limb rehabilitation training system further includes:

the game construction module is used for constructing a threshold control-based stone throwing game; the stone throwing game is characterized in that the amplitude of the myoelectric signal of the hand gesture surface is collected through a quantitative sensor so as to control the stretching length of a rubber band in the stone throwing game.

Preferably, the second obtaining module 500 specifically includes:

the acquisition unit is used for acquiring real-time gesture surface electromyographic signals of the healthy upper limbs of the patient when the patient plays games.

And the deleting unit is used for determining the amplitude of the real-time gesture surface electromyographic signal and deleting the real-time gesture surface electromyographic signal corresponding to the amplitude larger than a set threshold value.

And the analysis unit is used for carrying out frequency domain analysis and time domain analysis on the reserved real-time gesture surface electromyographic signals to obtain real-time gesture characteristics, and constructing a real-time high-dimensional characteristic space matrix according to the real-time gesture characteristics.

The dimension reduction unit is used for carrying out dimension reduction processing on the real-time high-dimensional feature space matrix by adopting a principal component analysis method or a time sequence evaluation feature quality criterion to obtain a real-time low-dimensional feature space matrix; and the feature vectors in the real-time low-dimensional feature space matrix are real-time main gesture features.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.