Intelligent rehabilitation glove system and method based on noninvasive motor nerve interface

Abstract

The invention belongs to the field of rehabilitation medical treatment, and particularly relates to an intelligent rehabilitation glove system and method based on a noninvasive motor nerve interface, and the method comprises the steps: a wearable electromyographic signal collection unit collects an electromyographic signal at an arm according to a collection instruction sent by a control unit, packages the signals in the same time period, and sends the packaged signals to the control unit; the control unit is used for sending a collection instruction to the wearable electromyographic signal collection unit; meanwhile, the received original electromyographic signals are processed, feature values are obtained, feature vectors are formed, a gesture classifier is trained, the gesture category of the healthy side of the patient is recognized according to the gesture classifier, a motion instruction signal corresponding to the gesture category is output to the flexible pneumatic rehabilitation glove, the gesture identical to the instruction is executed, and then the hand is driven to move. And active rehabilitation training is realized. By optimizing the layout structure, the system is completely integrated on the diseased side arm of the patient, and active and convenient rehabilitation training of various actions of the diseased side hand of the hemiplegic patient is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of rehabilitation medicine, and specifically relates to an intelligent rehabilitation glove system and method based on a non-invasive motor nerve interface. Background Art

[0002] The dexterity and coordination of hand movements play a very important role in people's daily lives and interpersonal interactions. However, neurological diseases such as stroke and neuromuscular atrophy damage the patient's neuronal cells, causing hemiplegia and motor disorders of the upper and lower limbs (including the hands), which seriously affects the patient's life and work. Rehabilitation training is an effective means to restore the patient's motor function by providing compensatory training for the patient's lost functions, promoting disease recovery, and helping damaged neurons to rebuild. Therefore, the rehabilitation training system for hand function recovery has become a hot topic in current research.

[0003] Compared with one-on-one rehabilitation training between patients and therapists, rehabilitation training with equipment to assist the hand not only saves medical expenses, but also ensures a high degree of consistency in the training process and enhances the training effect. Compared with rigid structures, hand rehabilitation equipment with flexible structure design has a simple structure, is lighter and more economical. In addition, electromyographic signals contain rich motion information and have low time lag. They are widely used as external signal sources for rehabilitation training systems. Rehabilitation equipment driven by electromyographic signals can achieve active rehabilitation training with better functional recovery effects.

[0004] However, the existing flexible hand function rehabilitation devices driven by electromyographic signals have the following main shortcomings:

[0005] (1) The control system and energy supply element of the existing hand rehabilitation training device for hemiplegic patients are placed outside the body, and the energy supply element and the drive device are connected by wire, resulting in the rehabilitation process being limited by space and unable to achieve convenient and flexible hand rehabilitation training.

[0006] (2) The existing electromyographic signal acquisition electrodes have poor stretchability and need to be designed with electrodes of different sizes according to the hand sizes of different patients, resulting in poor versatility of the acquisition electrodes.

[0007] (3) The surface electromyographic signal of the affected side of hemiplegic patients is significantly weaker than that of normal people. Most of the existing gesture recognition algorithms use the time-frequency domain information of low-density surface electromyographic signals as features. The classification model based on this type of features cannot accurately identify the patient's movement intention. Summary of the invention

[0008] In view of the above problems, the present invention is committed to developing an intelligent rehabilitation glove system based on a non-invasive motor neural interface. The included distributed flexible high-density surface electromyography acquisition system adopts a stretchable modular high-density electrode structure design to meet the non-invasive electromyography signal acquisition needs of different patient-specific muscles; the high-density surface electromyography electrodes are combined with fabrics to make the device easy to wear; the electromyography decomposition algorithm is used to decompose the high-density surface electromyography signals of the affected side into motor unit signals, so that the distributed flexible high-density surface electromyography acquisition system becomes a non-invasive motor neural interface for the affected side of hemiplegic patients, and the movement intention is accurately identified from the neuromuscular control level by analyzing the characteristics of the motor units. An electromyography signal pattern recognition model is established based on the transfer learning method to achieve rapid classifier training and accurate, rapid and continuous gesture recognition; the design of flexible pneumatic rehabilitation gloves is completed to achieve refined gesture rehabilitation training.

[0009] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: an intelligent rehabilitation glove system based on a non-invasive motor nerve interface, characterized in that it includes: a wearable electromyographic signal acquisition unit, a flexible pneumatic rehabilitation glove and a control unit;

[0010] The wearable electromyographic signal acquisition unit is worn on the patient's hand and is used to collect electromyographic signals at the arm according to the acquisition instructions sent by the control unit, convert the electromyographic signals into digital signals with a resolution of 24 bits, and package the signals within the same time period and send them to the control unit;

[0011] The control unit is used to send a collection instruction to the wearable electromyographic signal collection unit to control it to collect electromyographic signals; at the same time, receive the signal sent by the wearable electromyographic signal collection unit; and process the obtained original electromyographic signal to obtain the characteristic value and form a characteristic vector to train the gesture classifier, identify the patient's healthy side gesture category according to the gesture classifier, and output the motion command signal corresponding to the gesture category to the flexible pneumatic rehabilitation glove;

[0012] Flexible pneumatic rehabilitation gloves are used to execute gestures identical to the motion command signals sent by the control unit, thereby driving hand movements and achieving active rehabilitation training.

[0013] The wearable electromyographic signal acquisition unit comprises: a distributed flexible high-density surface electromyographic electrode module, a signal transmission module and an elastic cuff;

[0014] The elastic sleeve is placed on the patient's forearm; the distributed flexible high-density surface electromyography electrode module is fixed to the inner wall of the elastic sleeve by fabric glue. When the patient wears it, the reference point of the elastic sleeve is aligned with the corresponding joint of the upper limb to ensure that the electrode contact is located at the accurate electromyography signal collection position;

[0015] The signal transmission module is ring-shaped and is tied to the outside of the elastic cuff;

[0016] The signal transmission module is connected to the distributed flexible high-density surface electromyography electrode module, and is used to receive and collect electromyographic signals at the arm, convert the electromyographic signals into digital signals with a 24-bit resolution, and package the signals within the same time period and send them to the control unit.

[0017] The distributed flexible high-density surface electromyography electrode module comprises: a flexible PCB substrate, electrode contacts, wires and connectors;

[0018] The electrode contacts include: 16 electromyographic signal acquisition modules, 1 reference electrode and 1 bias electrode connector connected to the connector via wires;

[0019] Each of the electromyographic signal acquisition modules is composed of 4 point electrodes forming a high-density electrode array; the length of the wire connected to each electromyographic signal acquisition module is different to ensure that comprehensive forearm muscle group activation information is obtained;

[0020] The wire is a spring-shaped structure wire to achieve a stretching effect; the electrode contacts and the wire are made of copper foil, and the surface of the electrode contact is coated with 2uinch electroless nickel immersion gold to enhance its conductivity and oxidation resistance;

[0021] The connector is located at the end of each electrode connected by the wire, and is used to connect the distributed flexible high-density surface electromyography electrode module and the signal transmission module;

[0022] The electrode contacts and the back of the connector are reinforced with glass fibers to increase the rigidity of the electrode and the connector;

[0023] The elastic sleeve is fixedly connected to the glass fiber, and when worn, the sleeve is tightened to ensure that the electrode contacts are in full contact with the patient's skin.

[0024] The signal transmission module includes: a flexible electrode connector, an electromyographic signal processing module, a main control module, and a battery management module;

[0025] The flexible electrode connector is connected to the electrode contact through a copper cable to reduce the influence of the cable impedance on the signal;

[0026] The electromyographic signal processing module is used to sample, measure, amplify, and perform analog-to-digital conversion on the electromyographic signal, and has the functions of electrode detachment detection and bias drive circuit output, and transmits the processed electromyographic signal to the main control module;

[0027] An ESD protection circuit and an RC low-pass filter circuit are designed at the front end of the electromyographic signal processing module to protect the working circuit and implement 20kHz low-pass filtering of the input signal.

[0028] There are four electromyographic signal processing modules, which are connected in a daisy chain manner, and a 1:2 clock buffer is used to output a synchronous clock to the four electromyographic signal processing modules to achieve strict synchronous acquisition of multi-channel electromyographic signals;

[0029] The main control module is used to control the electromyographic signal processing module, read and transmit data, and transmit the collected data to the control unit by wireless transmission and wired transmission; when it is wireless transmission, the main control module sends data to the control system in real time via WIFI, and when it is wired transmission, data transmission is realized by USB serial port communication;

[0030] The power management module is used to realize the power supply, charging management and power detection operations of the signal transmission module through the control of the main control module.

[0031] The control unit comprises: a signal processing module, a classification recognition module and an instruction output module connected in sequence;

[0032] The signal processing module is connected to the main control module of the signal transmission module, and is used to perform bandpass filtering, notch filtering, sliding window segmentation, short-time energy value extraction, time domain and spectrum feature extraction operations on the original electromyographic signal transmitted by the main control module, and pass the extracted feature values ​​to the classification and recognition module;

[0033] A classification and recognition module is used to transfer the patient's electromyographic signal feature values ​​to obtain a dedicated classifier, and identify the patient's healthy side hand movements through the dedicated classifier;

[0034] The command output module is used to generate motion commands based on the recognized hand movements and transmit them to the flexible pneumatic rehabilitation gloves.

[0035] The flexible pneumatic rehabilitation glove comprises: a glove body and a drive module;

[0036] The glove body comprises: a glove, an independent air chamber and an air guide duct;

[0037] An independent air chamber is fixed behind each finger of the glove to achieve bending and stretching of the finger;

[0038] The driving mechanism and control unit are both located in the wearable device on the patient's upper arm and driven through the air duct;

[0039] The driving mechanism comprises: a control panel, a relay, an air pump and a reversing valve;

[0040] The control panel receives the motion command sent by the control unit, sends a corresponding signal to the relay, controls the operation of the reversing valve and the air pump, controls the bending and extension of the air chamber by changing the air pressure inside the air chamber, acts on the glove through the air duct, and drives the bending and extension of the patient's fingers on the affected side.

[0041] A control method for an intelligent rehabilitation glove system based on a non-invasive motor nerve interface comprises the following steps:

[0042] 1) Wear the elastic cuff on the patient's healthy arm to ensure that the distributed flexible high-density surface electromyography electrode module and the reference electrode are in close contact with the patient's skin, and connect the electrode contacts to the signal transmission module through a connector to complete the wearable electromyography signal acquisition system; wear the flexible pneumatic rehabilitation glove on the hand to be trained on the affected side;

[0043] 2) The control unit issues an acquisition instruction to control the wearable electromyographic signal acquisition unit to start working; the wearable electromyographic signal acquisition unit converts the acquired electromyographic signals into digital signals with a 24-bit resolution, and packages the signals in the same time period and sends them to the control unit; the control unit integrates the signals transmitted by multiple front-end acquisition control units, and transmits the data to the file under the specified path at a regular time;

[0044] 3) The signal processing module pre-processes the original electromyographic signal, and then decomposes the electromyographic signal by module, decomposing the surface electromyographic signal collected by each module into motor unit spike sequence and motor unit action potential, calculating the eigenvalues ​​and forming the eigenvector respectively, which is used to train the gesture classifier;

[0045] 4) The classification recognition module identifies the patient's healthy side gesture category according to the gesture classifier, and outputs the motion command signal corresponding to the gesture category to the flexible pneumatic rehabilitation glove control module and sends it to the command output module;

[0046] 5) The command output module outputs the motion command to the driving module, driving the glove body to make the same gesture, and the glove body moves the hand to achieve active rehabilitation training.

[0047] The step 3) is specifically:

[0048] 3-1) After the signal processing module obtains the collected original electromyographic signal, it removes the data of the first 3s and the last 1s of the signal to ensure the quality of the signal, and then performs Butterworth bandpass filtering of 30-500Hz and notch filtering of 50Hz and its multiple frequency in sequence to reduce noise interference;

[0049] 3-2) performing data expansion and whitening on the signal to obtain an expanded decorrelated electromyographic signal;

[0050] 3-3) In the process of EMG decomposition, the FastICA algorithm is used to estimate the original independent components of the EMG signal in modules, the EMG signal of each module is decomposed into motor unit spike sequences, and the action potential waveforms of each motor unit of the affected arm of hemiplegic patients are extracted respectively;

[0051] 3-4) Extract features from the decomposed motor unit information; calculate the peak-to-peak value of each motor unit action potential waveform; calculate the firing rate F of each motor unit based on its spike sequence, that is:

[0052]

[0053] Where L is the signal length, st j is the number of firings per unit time of the jth motor unit;

[0054] 3-5) The feature values ​​extracted by each module are combined into feature vectors and sent to the classification and recognition module for classification tasks.

[0055] The step 4) is specifically:

[0056] 4-1) Build a CNN model for training the source network:

[0057] CNN model, including: 3 convolutional blocks and 1 fully connected layer;

[0058] Each convolution block contains 1 convolution layer, 1 normalization layer, and 1 Relu activation layer. In addition, the first two convolution blocks also have 1 pooling layer.

[0059] Each convolutional layer consists of several filters with a stride of 3×1, and uses L2R-egularization and a dropout layer with a dropout rate of 0.03 in the fully connected layer to avoid overfitting;

[0060] 4-2) The feature vector is used as the input data of the model. The initial learning rate during training is 0.002, and the Adam gradient descent algorithm is used as the learning rate optimization method. Through the minimum batch training method, the maximum number of model training times is 300, and the learning rate is adjusted to 0.0002 after 250 training times; finally, the output of the model is flattened into a 1-dimensional vector, and the classification result is obtained through a softmax layer with multiple units;

[0061] 4-3) Through the fixed value migration strategy, the fully connected network is trained for migration, and part of the network is frozen during the migration process so that the network structure and weight of this part remain unchanged;

[0062] 4-4) After the training is completed, the classification and recognition module outputs the patient's gesture recognition classifier, which is used for gesture classification and recognition, identifies the gesture category of the patient's healthy side, and outputs the motion command signal corresponding to the gesture category through the command output module, which is used to control the flexible pneumatic rehabilitation gloves to make the same gesture to achieve active rehabilitation training.

[0063] The step 5) is specifically:

[0064] The control board receives the motion command sent by the control unit, sends a corresponding signal to the relay, controls the operation of the reversing valve and the air pump, controls the bending and extension of the air chamber by changing the air pressure inside the air chamber, acts on the glove through the air duct, and drives the bending and extension of the patient's fingers on the affected side.

[0065] The present invention has the following beneficial effects and advantages:

[0066] 1. The intelligent rehabilitation glove system based on non-invasive motor nerve interface established by the present invention adopts flexible pneumatic gloves with independent fingers, flexible wearable high-density surface electromyography electrodes, and gesture recognition strategies based on transfer learning and motor nerve decoding. By optimizing the layout structure, the system is fully integrated into the patient's affected arm, realizing active and convenient rehabilitation training of multiple movements of the affected hand of hemiplegic patients.

[0067] 2. The wearable EMG signal acquisition system established by the present invention optimizes the electrode structure, especially the electrode wire, the high-density array structure of the electrode contacts, the spatial position distribution of the electrode contacts, and the combination design of the electrode and the fabric. Each EMG signal acquisition module contains 4 square-arranged electrodes, which can realize the extraction of human motor nerve signals. At the same time, it also effectively improves the comfort and convenience of wearing the electrodes, and provides a universal high-density surface EMG signal acquisition solution for the upper limb specific muscles of different hemiplegic patients.

[0068] 3. The control system established by the present invention uses an electromyography decomposition algorithm to decompose the high-density surface electromyography signal of the affected side into motor nerve signals, constructs a non-invasive motor nerve interface for the affected side of hemiplegic patients, and identifies the patient's movement intention from the neuromuscular control level by analyzing the characteristics of the motor units.

[0069] 4. The classification and recognition module established by the present invention can realize accurate recognition of a variety of refined gestures of different individuals based on a small amount of data by constructing an electromyographic signal pattern recognition model and a transfer learning model, which greatly reduces the time required to train a personalized classifier for a new individual.

[0070] 5. The flexible pneumatic rehabilitation gloves of the present invention adopt independent positive pressure and vacuum pneumatic circuit control for each finger, and each finger can independently achieve three states of bending, stretching and relaxation, thereby realizing refined gesture movement rehabilitation training. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a structural diagram of the flexible hand function rehabilitation system of the present invention;

[0072] Figure 2 This is a diagram showing the structure of a distributed flexible high-density surface electromyography electrode module of the present invention;

[0073] Figure 3 It is a structural diagram of the signal transmission module of the present invention;

[0074] Figure 4 It is a flow chart of the method for realizing the functions of the present invention;

[0075] Figure 5 This is a diagram showing the main body of the flexible pneumatic rehabilitation gloves of the present invention;

[0076] Figure 6 This is a diagram showing the structure of the driving module of the flexible pneumatic rehabilitation gloves of the present invention;

[0077] Among them, 0100 is a distributed flexible high-density surface electromyography electrode module, 0101 is a reference electrode, 0102 is an electrode contact, 0103 is a bias electrode, 0104 is a connector, 0105 is a wire, 0200 is a signal transmission module, 0201 is a signal indicator light, 0202 is an electromyography signal processing module, 0203 is a main control module, 0204 is a charging interface, 0205 is a power switch, 0206 is a flexible electrode connector, 0300 is a control unit, 0400 is an elastic cuff, 0500 is a glove body, 0501 is a glove, 0502 is an independent air chamber, 0503 is an air duct, 0504 is a control board, 0505 is a relay, and 0506 is a reversing valve. DETAILED DESCRIPTION

[0078] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0079] like Figure 1 As shown, it is a diagram of the structure of the flexible hand function rehabilitation system of the present invention. The flexible hand function rehabilitation training system of the present invention includes three parts: a wearable electromyographic signal acquisition system, a flexible pneumatic rehabilitation glove and a control system;

[0080] The wearable EMG signal acquisition system consists of three parts: a distributed flexible high-density surface EMG electrode module, a signal transmission module, and an elastic cuff;

[0081] The flexible pneumatic rehabilitation gloves include a glove body and a drive module;

[0082] The control unit is equipped with a microprocessor and has three built-in modules: signal processing, classification recognition, and command output.

[0083] The signal processing module analyzes the electromyographic signals of the healthy side of the patient obtained from the wearable electromyographic signal acquisition system, identifies the movement intention and sends movement instructions to the rehabilitation glove in real time, prompting the diseased side to make corresponding gestures.

[0084] Wearable electromyographic signal acquisition system, including: distributed flexible high-density surface electromyographic electrode module, elastic cuff and signal transmission module;

[0085] like Figure 2 As shown, it is a diagram of the structure of the distributed flexible high-density surface electromyography electrode module of the invention. The distributed flexible high-density surface electromyography electrode module includes a flexible PCB substrate, electrode contacts 0102, wires 0105, and connectors 0104.

[0086] The electrode contacts consist of 16 EMG signal acquisition modules, a reference electrode, and a bias electrode. Each EMG signal acquisition module consists of 4 point electrodes to form a high-density electrode array. The wire 0105 is designed as a spring-like structure to achieve a stretchable effect. The electrode contact 0102 and the wire material are copper foil, and the surface of the electrode contact 0102 is coated with 2uinch electroless nickel immersion gold to enhance its conductivity and oxidation resistance.

[0087] The connector 0104 is located at the end of the electrode and is used to connect the distributed flexible high-density surface electromyography electrode module 0100 and the signal transmission module 0200 .

[0088] The electrode contact 0102 and the back of the connector 0104 are reinforced with glass fiber to improve the rigidity of the electrode and the connector. The elastic sleeve 0400 is fixedly connected with the glass fiber reinforcement, and the sleeve is tightened when worn to ensure that the electrode contact is fully in contact with the patient's skin.

[0089] The distributed flexible high-density surface electromyography electrode module 0100 is combined with the elastic cuff 0400, which is stretchable and adaptable and suitable for different patient-specific muscles.

[0090] like Figure 3 As shown, it is a structural diagram of the signal transmission module of the present invention, and the signal transmission module 0200 includes a flexible electrode connector 0202, an electromyographic signal processing module 0202, a main control module 0203, and a battery management module.

[0091] The flexible electrode connector 0202 is connected to the electrode contact 0102 via a copper flat cable, thereby reducing the influence of the impedance of the flexible electrode connector 0202 and the flat cable on the signal.

[0092] The EMG signal processing module is equipped with a 16-channel bioelectric acquisition chip to achieve EMG signal sampling, measurement, amplification, analog-to-digital conversion, and has the functions of electrode detachment detection and bias drive circuit output. ESD protection circuit and RC low-pass filter circuit are designed at the front end of the EMG signal processing module to protect the working circuit and realize 20kHz low-pass filtering of the input signal. In order to achieve high-density EMG signal acquisition, the four EMG signal processing modules are connected in a daisy chain manner, and a 1:2 clock buffer is used to output a synchronous clock to the four EMG signal processing modules to achieve strict synchronous acquisition of multi-channel EMG signals. The main control module 0203 is equipped with an ESP32-S3 chip to control the main control module 0203 and read and transmit data.

[0093] When transmitting collected data to the control system, it supports both wireless transmission and wired transmission. In wireless transmission, the main control module 0203 sends data to the control system in real time via WIFI, and wired transmission is achieved by USB serial port communication.

[0094] The power management module includes: charging interface 0204, power switch 0205, voltage stabilizing module, acquisition chip, power supply module, charging management module, and power detection module. It is controlled by the main control chip to realize system power supply, charging management, and power detection functions.

[0095] The signal transmission module 0200 is powered by a lithium battery, and the battery voltage is 3.7V when fully charged. The voltage provided by the battery outputs a 3.3V stable voltage through the voltage stabilizing module, which supplies power to the main control module 0203. The 3.3V voltage output by the voltage stabilizing module outputs a ±2.5V voltage through the acquisition chip power supply module, which supplies power to the electromyographic signal processing module 0202. At the same time, the power detection module monitors the system power in real time; the lithium battery is charged through the charging interface 0204, and the charging management is performed through the charging management module. The power switch 0205 is used to control the opening or closing of the system power; the power management module in the present invention is common knowledge known to those skilled in the art, and the specific structure will not be repeated.

[0096] The control unit 0300 is equipped with a microprocessor, including three modules: a signal processing module, a classification and recognition module, and an instruction output module.

[0097] Among them, the signal processing module can realize bandpass filtering, notch filtering, sliding window segmentation, short-time energy value extraction, time domain and spectrum feature extraction of the original electromyographic signal, and pass the feature value to the classification and recognition module; the classification and recognition module performs transfer learning on the patient's electromyographic signal feature value to obtain a dedicated classifier for identifying the patient's healthy side hand movements; the command output module generates motion commands based on the identified hand movements and passes them to the flexible pneumatic rehabilitation gloves.

[0098] like Figure 5-6As shown, the structural schematic diagrams of the glove body and the driving mechanism of the present invention are respectively shown. The flexible pneumatic rehabilitation glove of the present invention includes: a glove body 0500 and a driving mechanism. Among them, an independent air chamber 0502 is fixed on the surface of the glove 0501 in the glove body 0500, and its bending or stretching can be controlled by controlling the air pressure.

[0099] The driving module includes: a control board 0504 equipped with an Arduino UNO development board, a relay 0505, an air pump, and a reversing valve 0506.

[0100] The control board 0504 receives the motion command sent by the control unit 0300, sends a corresponding signal to the relay 0505, controls the reversing valve 0506 and the air pump to work, controls the bending and extension of the air chamber by changing the air pressure inside the air chamber, and drives the bending and extension of the patient's affected side finger. The driving mechanism and control system are placed in the wearable device on the patient's upper arm and driven by the airway 0503.

[0101] like Figure 4 As shown in FIG. 1 , it is a flow chart of a method for realizing the functions of the present invention. Based on the intelligent rehabilitation glove system of the non-invasive motor nerve interface, the present invention provides a control method of the intelligent rehabilitation glove, comprising the following steps:

[0102] 1) Wear the elastic sleeve 0400 on the healthy side of the patient's forearm to ensure that the distributed flexible high-density surface electromyography electrode module 0100 and the reference electrode 0101 are in close contact with the patient's skin, and connect the electrode contact 0102 to the signal transmission module 0200 through the connector 0104 to complete the wearable electromyography signal acquisition system; wear the flexible pneumatic rehabilitation glove on the hand to be trained on the affected side;

[0103] 2) The control unit 0300 issues an acquisition instruction to control the wearable electromyographic signal acquisition unit to start working; the wearable electromyographic signal acquisition unit converts the acquired electromyographic signals into digital signals with a 24-bit resolution, and packages the signals in the same time period and sends them to the control unit 0300; the control unit 0300 integrates the signals transmitted by multiple front-end acquisition control units, and transmits the data to the file under the specified path at a regular time;

[0104] 3) The signal processing module pre-processes the original electromyographic signal, and then decomposes the electromyographic signal by module, decomposing the surface electromyographic signal collected by each module into motor unit spike sequence and motor unit action potential, calculating the eigenvalues ​​and forming the eigenvector respectively, which is used to train the gesture classifier;

[0105] 4) The classification recognition module identifies the patient's healthy side gesture category according to the gesture classifier, and outputs the motion command signal corresponding to the gesture category to the flexible pneumatic rehabilitation glove control module and sends it to the command output module;

[0106] 5) The command output module outputs the motion command to the driving module, driving the glove body 0500 to make the same gesture. The glove body 0500 drives the hand movement to achieve active rehabilitation training.

[0107] Among them, each module of the control unit in the present invention performs the following steps:

[0108] 1) Data processing module

[0109] After obtaining the original EMG signal collected by the wearable EMG signal acquisition system, the data before 3s and after 1s are removed to ensure the signal quality, and then Butterworth bandpass filtering from 30 to 500Hz and notch filtering at 50Hz and its multiple frequencies are performed to reduce noise interference. In order to meet the requirements of EMG decomposition, the signal is expanded and whitened to obtain an expanded decorrelated EMG signal.

[0110] In the EMG decomposition process, the FastICA algorithm is used to estimate the original independent components of the EMG signal on a module basis, the EMG signal of each module is decomposed into motor unit spike sequences, and the action potential waveforms of each motor unit in the affected arm of hemiplegic patients are extracted respectively.

[0111] The decomposed motor unit information is subjected to feature extraction. The peak-to-peak value of each motor unit action potential waveform is calculated; the firing rate F of each motor unit is calculated based on its spike sequence:

[0112]

[0113] Where L is the signal length, st j is the number of firings per unit time by the jth motor unit.

[0114] The eigenvalues ​​extracted from each module are combined into feature vectors for classification tasks.

[0115] 2) Classification and recognition module

[0116] A CNN model was constructed to train the source network. The model consists of three convolutional blocks and one fully connected layer. Each convolutional block consists of one convolutional layer, one normalization layer, and one Relu activation layer. In addition, the first two convolutional blocks also have a pooling layer. Each convolutional layer consists of several filters with a stride of 3×1, and uses the L2R-egularization method and a Dropout layer with a dropout rate of 0.03 in the fully connected layer to avoid overfitting. The feature vector is used as the input data of the model. The initial learning rate during training is 0.002, and the Adam gradient descent algorithm is selected as the learning rate optimization method. The minimum batch training method is used. The maximum number of model training times is 300, and the learning rate is adjusted to 0.0002 after 250 training times. Finally, the output of the model is flattened into a 1-dimensional vector. The classification result is obtained through a softmax layer with multiple units.

[0117] Use a fixed value migration strategy to perform migration training on the fully connected network. During the migration process, freeze part of the network so that the network structure and weights of that part remain unchanged. This reduces the number of training times and shortens the training time while ensuring high recognition accuracy.

[0118] After training is completed, the classification and recognition module outputs the patient's gesture recognition classifier for subsequent gesture classification and recognition.

[0119] 3) Command output module

[0120] When the patient is undergoing rehabilitation training, the gesture category of the patient's healthy side is identified using a gesture intention recognition classifier based on the feature vector calculated from the surface electromyography signal of the patient's healthy side limb, and the motion command signal corresponding to the gesture category is output to control the flexible pneumatic rehabilitation gloves to make the same gesture to achieve active rehabilitation training.

[0121] In summary, the intelligent rehabilitation glove system based on non-invasive motor nerve interface established by the present invention adopts flexible pneumatic gloves with independent fingers, flexible wearable high-density surface electromyography electrodes, and gesture recognition strategies based on transfer learning and motor nerve decoding. By optimizing the layout structure, the system is fully integrated into the patient's affected arm, realizing active and convenient rehabilitation training of multiple movements of the affected hand of hemiplegic patients.

[0122] The intelligent rehabilitation glove system based on the non-invasive motor nerve interface provided by the present invention is, as an embodiment, a distributed flexible high-density surface electromyography electrode module in the wearable electromyography signal acquisition system is fixed on the inner layer of the elastic cuff and adopts a stretchable wire structure design, which ensures the convenience of wearing and the comfort of the acquisition process, and meets the high-density surface electromyography signal acquisition requirements for the upper limb specific muscles of different hemiplegic patients; the control system is equipped with three modules: data processing, classification and recognition, and command output. The collected high-density surface electromyography signals are decomposed into motor nerve signals on a module basis, and a non-invasive motor nerve interface is constructed. The motor nerve signal characteristics of each module are extracted and a classification model is constructed based on this, so as to realize the accurate recognition of the hand movements on the affected side of the hemiplegic patient, generate movement instructions according to the recognized hand movement type and transmit them to the flexible pneumatic rehabilitation gloves; the flexible pneumatic rehabilitation gloves adopt independent pneumatic circuit control for each finger to complete refined gesture movement rehabilitation training. The overall schematic diagram of the system is shown in the figure below. Figure 1 As shown, there are a distributed flexible high-density surface electromyography electrode module 0100, a signal transmission unit 0200, a control unit 0300, an elastic cuff 0400, and a glove body 0500.

[0123] Embodiment 1:

[0124] This embodiment realizes the corresponding functions of the intelligent rehabilitation glove system through a three-part design. The specific design method and use principle of this embodiment are as follows:

[0125] (1) Design of wearable electromyography signal acquisition system

[0126] In this embodiment, the wearable electromyographic signal acquisition system includes a distributed flexible high-density surface electromyographic electrode module and a signal transmission module. Figure 2 As shown, it includes: a reference electrode 0101, an electrode contact 0102, a bias electrode 0103, a connector 0104, and a wire 0105.

[0127] 1) Distributed flexible high-density surface electromyography electrode module;

[0128] The length of each wire 0105 and the angle between adjacent wires of the distributed flexible high-density surface electromyography electrode module 0100 are determined through experiments to ensure that comprehensive forearm muscle group activation information is obtained;

[0129] Distributed flexible high-density surface electromyography electrode module 0100;

[0130] The distributed flexible high-density surface electromyography electrode module 0100 is fixed to the inner wall of the elastic sleeve 0400 by fabric glue. When the patient wears it, the reference point of the elastic sleeve is aligned with the corresponding joint of the upper limb to ensure that the electrode contact is located at the accurate electromyography signal collection position;

[0131] 2) Signal transmission module

[0132] like Figure 3 As shown, the signal transmission module 0200 is connected to the distributed flexible high-density surface electromyography electrode module 0100 via a connector 0104; it includes: a flexible electrode connector 0206, an electromyography signal processing module 0202, a main control module 0203, and a battery management module;

[0133] The flexible electrode connector 0202 is connected to the electrode contact 0102 via a copper cable to reduce the influence of the cable impedance on the signal;

[0134] The electromyographic signal processing module 0202 is used to sample, measure, amplify, and convert the electromyographic signal into analog-to-digital form, and has the functions of electrode detachment detection and bias drive circuit output, and transmits the processed electromyographic signal to the main control module 0203;

[0135] Design ESD protection circuit and RC low-pass filter circuit at the front end of the electromyographic signal processing module 0202 to protect the working circuit and implement 20kHz low-pass filtering of the input signal;

[0136] There are four EMG signal processing modules 0202, which are connected in a daisy chain manner, and a 1:2 clock buffer is used to output synchronous clocks to the four EMG signal processing modules to achieve strict synchronous acquisition of multi-channel EMG signals;

[0137] The main control module 0203 is used to control the electromyographic signal processing module, read and transmit data, and transmit the collected data to the control unit through wireless transmission and wired transmission. When wireless transmission is used, the main control module sends data to the control system in real time through WIFI, and when wired transmission is used, data transmission is realized by USB serial communication;

[0138] The power management module is used to realize the power supply, charging management and power detection operations of the signal transmission module 0200 through the control of the main control module 0203.

[0139] The surface electromyographic signal of the patient's forearm is collected by the distributed flexible high-density surface electromyographic electrode module 0100, and the electromyographic signal is transmitted to the signal transmission module 0200 via the connector 0104; the incoming electromyographic signal passes through the ESD protection circuit and the RC low-pass filter circuit and enters the ADS1299 bioelectric acquisition chip 0202. The ADS1299 bioelectric acquisition chip 0202 is controlled by the ESP32-S3 chip 0203 carried by the main control module;

[0140] At the beginning of the acquisition, the electromyographic signal processing module 0202 of the ADS1299 bioelectric acquisition chip performs synchronous sampling, measurement, amplification, and analog-to-digital conversion of the electromyographic signal under the action of the external crystal oscillator, realizing strict synchronous acquisition of multi-channel electromyographic signals, and transmitting the sampled digital signals to the main control module 0203 equipped with the ESP32-S3 chip. The main control module 0203 communicates with the control unit in real time, outputs the obtained digital signals to the control system, and receives and executes control instructions from the control system.

[0141] (2) Control system software function design

[0142] In this embodiment, the control unit software function flow chart is as follows: Figure 4 The microprocessor on board the control unit is as follows. Figure 4 As shown;

[0143] The patient's electromyographic data collected by the wearable electromyographic signal acquisition unit is wirelessly transmitted to the control unit. The built-in software displays the acquired data in real time and performs preprocessing, electromyographic signal decomposition, feature extraction, and saves it in a designated folder. The classification and recognition module is used to transfer the features to obtain a dedicated gesture classifier, which estimates the patient's hand movements and generates motion instructions to drive the flexible pneumatic rehabilitation gloves.

[0144] The control system workflow is as follows:

[0145] After starting the system software, search for ports according to the software prompts, and start acquiring data after connecting to the signal transmission module in the wearable electromyographic signal acquisition system. Before officially storing the data, open the window monitoring first, and adjust the wearing device by observing the matching degree between the displayed waveform and the action. When the waveform shows fluctuations with the action and low amplitude in the resting state, you can start recording data.

[0146] Users can manage and browse reports and records through the control system software.

[0147] (3) Design of flexible pneumatic rehabilitation gloves

[0148] In this embodiment, the flexible pneumatic rehabilitation glove includes two parts: a glove body and a driving mechanism. Figure 5 As shown. It includes: gloves 0501, independent air chamber 0502, air guide pipe 0503. Driving mechanism, such as Figure 6 Including, in this embodiment, a control board 0504, a relay 0505, and a reversing valve 0506 of an Arduino UNO development board are used.

[0149] An independent air chamber 0502 is fixed behind each finger of the glove 0501 to achieve the bending and stretching of the finger.

[0150] The control board 0504 and the relay 0505 control the position of the valve core of the reversing valve 0506 and the opening and closing of each air pump, thereby generating a vacuum or pressure state, which acts on the glove body through the air duct 0503, so that each finger can be bent and stretched independently.

[0151] The working principle of this embodiment is as follows:

[0152] When in use, wear the elastic sleeve 0400 on the patient's healthy forearm, ensure that the distributed flexible high-density surface electromyography electrode module and the reference electrode are in close contact with the patient's skin, and connect the flexible electrode to the signal transmission module to complete the wearing of the wearable electromyography signal acquisition unit; wear the flexible pneumatic rehabilitation glove on the affected side of the hand to be rehabilitated, and connect the drive mechanism to the control unit 0300 to complete the wearing of the flexible pneumatic rehabilitation glove.

[0153] The control unit 0300 issues a collection instruction to control the wearable electromyographic signal collection unit to start working. The wearable electromyographic signal collection unit converts the collected electromyographic signals into digital signals with a 24-bit resolution, and packages the signals in the same time period and sends them to the control unit 0300. The control unit 0300 integrates the signals transmitted by multiple front-end collection control units and transmits the data to the file under the specified path at a regular time.

[0154] The control unit 0300 preprocesses the original electromyographic signal obtained to ensure the acquisition quality of the electromyographic signal and make the electromyographic signal suitable for the decomposition algorithm. Then the electromyographic signal is decomposed in modules, and the surface electromyographic signal collected by each module is decomposed into a motor unit spike sequence and a motor unit action potential. The eigenvalues ​​are calculated and composed into a eigenvector for training the gesture classifier. In view of the short-term training characteristics of the system, a transfer learning strategy is adopted to reduce the number of training times and save training time. According to the gesture classifier, the gesture category of the patient's healthy side is identified, and the motion command signal corresponding to the gesture category is output to the flexible pneumatic rehabilitation glove control module. The control module controls the drive module to drive the glove body 0500 to make the same gesture according to the motion command, and the glove body 0500 drives the hand movement to realize active rehabilitation training.

[0155] In summary, the above embodiments achieve the following technical effects:

[0156] 1. Universal and convenient collection of electromyographic signals: The distributed flexible high-density surface electromyographic electrode module of the present invention is fixed to the inner wall of the elastic sleeve by fabric glue, which is convenient to wear; the electrode wire structure design is stretchable, and the elastic sleeve enables the wearable electromyographic signal collection system to adaptively adjust according to the patient's arm size, realizing a universal electromyographic signal collection solution suitable for different patient-specific muscles.

[0157] 2. Accurate identification of gesture types of hemiplegic patients: The present invention uses an electromyography decomposition algorithm to decompose the high-density surface electromyography signals of the affected side into motor unit signals, constructs a non-invasive motor nerve interface for the affected side of hemiplegic patients, and identifies the patient's movement intention from the neuromuscular control level by analyzing the motor unit characteristics, thereby significantly improving the accuracy of gesture recognition on the affected side of hemiplegic patients.

[0158] 3. Refined active hand rehabilitation training: The present invention establishes an intelligent rehabilitation glove system based on a non-invasive motor nerve interface and adopts an active rehabilitation training mode. By designing a flexible and independently separable pneumatic glove driving scheme and constructing a variety of gesture recognition strategies based on CNN model training, active rehabilitation training of various hand movements on the affected side of hemiplegic patients can be achieved.

[0159] 4. Efficient training of personalized gesture recognition classifier: The present invention uses a fixed value migration strategy to perform migration training on a fully connected network, which reduces the number of training times and shortens the training time while ensuring a high recognition accuracy of the gesture recognition classifier.

[0160] Those skilled in the art will appreciate that the above are only preferred embodiments of the present invention, and the various embodiments of the present disclosure and / or the features described in the claims may be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or perform equivalent substitutions on some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0161] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent rehabilitation glove system based on a non-invasive motor nerve interface, characterized in that: include: Wearable electromyographic signal acquisition unit, flexible pneumatic rehabilitation gloves and control unit (0300); The wearable electromyographic signal acquisition unit is worn on the patient's hand and is used to collect electromyographic signals at the arm according to the acquisition instructions sent by the control unit, convert the electromyographic signals into digital signals with a resolution of 24 bits, and package the signals within the same time period and send them to the control unit; The control unit is used to send a collection instruction to the wearable electromyographic signal collection unit to control it to collect electromyographic signals; at the same time, receive the signal sent by the wearable electromyographic signal collection unit; and process the obtained original electromyographic signal to obtain the characteristic value and form a characteristic vector to train the gesture classifier, identify the patient's healthy side gesture category according to the gesture classifier, and output the motion command signal corresponding to the gesture category to the flexible pneumatic rehabilitation glove; Flexible pneumatic rehabilitation gloves are used to execute gestures identical to the motion command signals sent by the control unit, thereby driving hand movements and achieving active rehabilitation training.

2. The intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1 is characterized in that: The wearable electromyographic signal acquisition unit comprises: a distributed flexible high-density surface electromyographic electrode module (0100), a signal transmission module (0200) and an elastic cuff (0400); The elastic sleeve (0400) is sleeved on the patient's forearm; the distributed flexible high-density surface electromyography electrode module (0100) is fixed to the inner wall of the elastic sleeve (0400) by fabric glue. When the patient wears it, the reference point of the elastic sleeve is aligned with the corresponding joint of the upper limb to ensure that the electrode contact is located at the accurate electromyography signal collection position; The signal transmission module (0200) is ring-shaped and is tied to the outside of the elastic sleeve (0400); The signal transmission module (0200) is connected to the distributed flexible high-density surface electromyography electrode module (0100) and is used to receive and collect electromyographic signals at the arm, convert the electromyographic signals into digital signals with a 24-bit resolution, and package the signals within the same time period and send them to the control unit.

3. The intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 2, characterized in that: The distributed flexible high-density surface electromyography electrode module (0100) comprises: a flexible PCB substrate, electrode contacts (0102), wires (0105) and a connector (0104); The electrode contact (0102) comprises: 16 electromyographic signal acquisition modules, 1 reference electrode (0101) and 1 bias electrode connector (0103) connected to the connector (0104) via a wire (0105); Each of the electromyographic signal acquisition modules is composed of 4 point electrodes forming a high-density electrode array; the length of the wire connected to each electromyographic signal acquisition module is different to ensure that comprehensive forearm muscle group activation information is obtained; The wire (0105) is a spring-shaped structure wire to achieve a stretching effect; the electrode contact (0102) and the wire (0105) are made of copper foil, and the surface of the electrode contact (0102) is coated with 2uinch electroless nickel immersion gold to enhance its conductivity and oxidation resistance; The connector (0104) is located at the end of each electrode connected by the wire, and is used to connect the distributed flexible high-density surface electromyography electrode module (0100) and the signal transmission module (0200); The backs of the electrode contacts (0102) and the connector (0104) are reinforced with glass fibers to increase the rigidity of the electrodes and the connector (0104); The elastic sleeve (0400) is fixedly connected to the glass fiber, and when worn, the sleeve is tightened to ensure that the electrode contacts are in full contact with the patient's skin.

4. The intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1, characterized in that: The signal transmission module (0200) comprises: a flexible electrode connector (0206), an electromyographic signal processing module (0202), a main control module (0203), and a battery management module; The flexible electrode connector (0202) is connected to the electrode contact (0102) via a copper wiring harness to reduce the impact of wiring harness impedance on the signal; The electromyographic signal processing module (0202) is used to sample, measure, amplify, and perform analog-to-digital conversion on the electromyographic signal, and has the functions of electrode detachment detection and bias drive circuit output, and transmits the processed electromyographic signal to the main control module (0203); An ESD protection circuit and an RC low-pass filter circuit are designed at the front end of the electromyographic signal processing module (0202) to protect the working circuit and implement 20kHz low-pass filtering of the input signal; There are four electromyographic signal processing modules (0202), which are connected in a daisy chain manner, and a 1:2 clock buffer is used to output a synchronous clock to the four electromyographic signal processing modules, so as to realize strict synchronous acquisition of multi-channel electromyographic signals; The main control module (0203) is used to control the electromyographic signal processing module, read and transmit data, and transmit the collected data to the control unit through wireless transmission and wired transmission; when wireless transmission is used, the main control module sends data to the control system in real time through WIFI, and when wired transmission is used, data transmission is realized by USB serial port communication; The power management module is used to realize the power supply, charging management and power detection operations of the signal transmission module (0200) through the control of the main control module (0203).

5. The intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1, characterized in that: The control unit (0300) comprises: a signal processing module, a classification recognition module and an instruction output module connected in sequence; The signal processing module is connected to the main control module (0203) of the signal transmission module (0200), and is used to perform bandpass filtering, notch filtering, sliding window segmentation, short-time energy value extraction, time domain and spectrum feature extraction operations on the original electromyographic signal transmitted by the main control module (0203), and transmit the extracted feature values ​​to the classification and recognition module; A classification and recognition module is used to transfer the patient's electromyographic signal feature values ​​to obtain a dedicated classifier, and identify the patient's healthy side hand movements through the dedicated classifier; The command output module is used to generate motion commands based on the recognized hand movements and transmit them to the flexible pneumatic rehabilitation gloves.

6. The intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1, characterized in that: The flexible pneumatic rehabilitation glove comprises: a glove body (0500) and a drive module; The glove body (0500) comprises: a glove (0501), an independent air chamber (0502) and an air guide duct (0503); An independent air chamber (0502) is fixed behind each finger of the glove (0501) to achieve bending and stretching of the fingers; The driving mechanism and the control unit are both arranged in the wearable device at the patient's upper arm and driven through the airway (0503); The driving mechanism comprises: a control panel (0504), a relay (0505), an air pump and a reversing valve (0506); The control panel (0504) receives the motion command sent by the control unit (0300), sends a corresponding signal to the relay (0505), controls the operation of the reversing valve (0506) and the air pump, controls the bending and extension of the air chamber by changing the air pressure inside the air chamber, and acts on the glove (0501) through the air duct (0503), thereby driving the bending and extension of the patient's fingers on the affected side.

7. A control method for an intelligent rehabilitation glove system based on a non-invasive motor neural interface according to any one of claims 1 to 6, characterized in that: The following steps are involved: 1) Wearing the elastic sleeve (0400) on the healthy side of the patient's forearm, ensuring that the distributed flexible high-density surface electromyography electrode module (0100) and the reference electrode (0101) are in close contact with the patient's skin, and connecting the electrode contact (0102) to the signal transmission module (0200) through the connector (0104) to complete the wearable electromyography signal acquisition system; wearing the flexible pneumatic rehabilitation glove on the hand to be trained on the affected side; 2) The control unit (0300) issues a collection instruction to control the wearable electromyographic signal collection unit to start working; the wearable electromyographic signal collection unit converts the collected electromyographic signals into digital signals with a resolution of 24 bits, and packages the signals in the same time period and sends them to the control unit (0300); the control unit (0300) integrates the signals transmitted by multiple front-end collection control units, and transmits the data to a file under a specified path at a regular time; 3) The signal processing module pre-processes the original electromyographic signal, and then decomposes the electromyographic signal by module, decomposing the surface electromyographic signal collected by each module into motor unit spike sequence and motor unit action potential, calculating the eigenvalues ​​and forming the eigenvector respectively, which is used to train the gesture classifier; 4) The classification recognition module identifies the patient's healthy side gesture category according to the gesture classifier, and outputs the motion command signal corresponding to the gesture category to the flexible pneumatic rehabilitation glove control module and sends it to the command output module; 5) The command output module outputs the motion command to the driving module, driving the glove body (0500) to make the same gesture, and the glove body (0500) drives the hand to move, thereby realizing active rehabilitation training.

8. The control method of the intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1 is characterized in that: The step 3) is specifically: 3-1) After the signal processing module obtains the collected original electromyographic signal, it removes the data of the first 3s and the last 1s of the signal to ensure the quality of the signal, and then performs Butterworth bandpass filtering of 30-500Hz and notch filtering of 50Hz and its multiple frequency in sequence to reduce noise interference; 3-2) performing data expansion and whitening on the signal to obtain an expanded decorrelated electromyographic signal; 3-3) In the process of EMG decomposition, the FastICA algorithm is used to estimate the original independent components of the EMG signal in modules, the EMG signal of each module is decomposed into motor unit spike sequences, and the action potential waveforms of each motor unit of the affected arm of hemiplegic patients are extracted respectively; 3-4) Extract features from the decomposed motor unit information; calculate the peak-to-peak value of each motor unit action potential waveform; calculate the firing rate F of each motor unit based on its spike sequence, that is: Where L is the signal length, st j is the number of firings per unit time of the jth motor unit; 3-5) The feature values ​​extracted by each module are combined into feature vectors and sent to the classification and recognition module for classification tasks.

9. The control method of the intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1 is characterized in that: The step 4) is specifically: 4-1) Construct a CNN model for training the source network: CNN model, including: 3 convolutional blocks and 1 fully connected layer; Each convolution block contains 1 convolution layer, 1 normalization layer, and 1 Relu activation layer. In addition, the first two convolution blocks also have 1 pooling layer. Each convolutional layer consists of several filters with a stride of 3×1, and uses L2R-egularization and a dropout layer with a dropout rate of 0.03 in the fully connected layer to avoid overfitting; 4-2) The feature vector is used as the input data of the model. The initial learning rate during training is 0.002, and the Adam gradient descent algorithm is used as the learning rate optimization method. Through the minimum batch training method, the maximum number of model training times is 300, and the learning rate is adjusted to 0.0002 after 250 training times; finally, the output of the model is flattened into a 1-dimensional vector, and the classification result is obtained through a softmax layer with multiple units; 4-3) Through the fixed value migration strategy, the fully connected network is trained for migration, and part of the network is frozen during the migration process so that the network structure and weight of this part remain unchanged; 4-4) After the training is completed, the classification and recognition module outputs the patient's gesture recognition classifier, which is used for gesture classification and recognition, identifies the gesture category of the patient's healthy side, and outputs the motion command signal corresponding to the gesture category through the command output module, which is used to control the flexible pneumatic rehabilitation gloves to make the same gesture to achieve active rehabilitation training.

10. The control method of the intelligent rehabilitation glove system based on non-invasive motor neural interface according to claim 1, characterized in that: The step 5) is specifically: The control panel (0504) receives the motion command sent by the control unit (0300), sends a corresponding signal to the relay (0505), controls the operation of the reversing valve (0506) and the air pump, controls the bending and extension of the air chamber by changing the air pressure inside the air chamber, and acts on the glove (0501) through the air duct (0503), thereby driving the bending and extension of the patient's fingers on the affected side.

Images