Active rehabilitation training method, system and equipment for hand dysfunction patient

Abstract

The invention discloses an active rehabilitation training method, system and device for a patient with hand dysfunction, and relates to the technical field of rehabilitation training, and the method comprises the following steps: collecting motion track data of each finger of the patient or an electromyographic signal of an arm; receiving an electromyographic signal or motion trail data of a finger, and performing gesture recognition by adopting deep learning; a gesture recognition result is received, inflation and deflation of the flexible driving device are controlled, a first-order dynamic response two-way rate estimation feedforward adjustment algorithm is adopted to combine various adjustment rules, inflation and deflation residual errors are adjusted and eliminated, and fingers of the patient are driven to move. A first-order dynamic response two-way rate estimation feedforward adjustment technology is adopted, various adjustment rules are combined, inflation and deflation residual errors are adjusted and eliminated, electromyographic signals can be recognized in advance, a response is made rapidly, the actions which a patient wants to do and actually do are highly consistent, the hand on the affected side is driven to move, hand mirror image training is achieved, and the training efficiency is improved. Delay is reduced, and rehabilitation is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation training, and particularly to an active rehabilitation training method, system and device for patients with hand dysfunction. Background Art

[0002] Impaired limb function will lead to a decline in the quality of life of patients and also cause anxiety and depression in patients. In this case, how to improve hand function to promote limb rehabilitation has become the main task for stroke patients to recover.

[0003] Generally, patients can improve hand function through physical training. For example, patients perform some repetitive hand function rehabilitation training under the accompaniment of medical staff, and thereby improve the control ability of the hand and enhance the strength of the hand.

[0004] Due to the high precision and complexity of hand function, during the rehabilitation training process, hand function rehabilitation is both difficult and slow. There are already rehabilitation devices on the market that drive finger movement through pneumatic devices. There is a non-linear hysteresis effect during the inflation and deflation adjustment of the finger movement of the rehabilitation device. Especially when multiple joints move in coordination, cumulative errors are likely to occur, resulting in finger movement delay and it is difficult to achieve a good rehabilitation effect. Summary of the Invention

[0005] The purpose of the present invention is to provide an active rehabilitation training method, system and device for patients with hand dysfunction in view of the above-mentioned deficiencies of the prior art, so as to solve the problems in the prior art.

[0006] The present invention specifically provides the following technical solutions: The present invention provides an active rehabilitation training system for patients with hand dysfunction, including: A gesture acquisition module, configured to obtain the movement trajectory data of each finger on the healthy side of the patient or the myoelectric signal of the arm through an acquisition device arranged on the healthy side of the patient; A gesture analysis and transmission module, configured to perform healthy-side gesture recognition through the movement trajectory data of each finger on the healthy side of the patient or the myoelectric signal of the arm; An action execution module, configured to control the inflation and deflation of a flexible driving device arranged on the affected side of the patient according to the healthy-side gesture recognition result, drive each finger on the affected side of the patient to move synchronously with each finger on the healthy side, and when inflating and deflating, combine the proportional, integral and differential adjustment rules by using a first-order dynamic response two-way rate prediction feedforward adjustment algorithm, perform disturbance compensation on the output inflation and deflation signal and the input healthy-side gesture recognition result, adjust and eliminate the residual error of the inflation and deflation signal through the disturbance compensation result, and control the inflation and deflation of the flexible driving device through the inflation and deflation signal with the residual error eliminated.

[0007] The present invention provides an active rehabilitation training method for patients with hand dysfunction, including the following steps: Obtain the motion trajectory data of each finger on the healthy side of the patient or the electromyogram signal of the arm through the acquisition device arranged on the healthy side of the patient; Perform healthy-side gesture recognition based on the motion trajectory data of each finger on the healthy side of the patient or the electromyogram signal of the arm; According to the healthy-side gesture recognition result, control the flexible driving device arranged on the affected side of the patient to inflate and deflate, drive each finger on the affected side of the patient to move synchronously with each finger on the healthy side, and when inflating and deflating, combine the proportional, integral, and differential adjustment laws by using the first-order dynamic response two-way rate prediction feedforward adjustment algorithm, perform disturbance compensation on the output inflation and deflation signal and the input healthy-side gesture recognition result, adjust and eliminate the residual error of the inflation and deflation signal through the disturbance compensation result, and control the inflation and deflation of the flexible driving device through the inflation and deflation signal with the residual error eliminated.

[0008] Preferably, the acquisition device includes various sensors or an eight-channel gesture acquisition armband; wherein, the various sensors include bending sensors, gyroscope sensors, and acceleration sensors, and the various sensors are arranged above the flexor carpi ulnaris tendon on the healthy side to collect the motion trajectory data of each finger; the eight-channel gesture acquisition armband is worn on the healthy-side arm for collecting the electromyogram signal of the patient.

[0009] Preferably, performing healthy-side gesture recognition based on the motion trajectory data of each finger on the healthy side of the patient includes: Perform attitude calculation on the motion trajectory data of each finger on the healthy side of the patient, and calculate the heading angle of each finger , roll angle and pitch angle ; wherein the specific expression of the attitude calculation is: = ; wherein, is the attitude calculation result; Obtain the acceleration of each finger through the attitude calculation result, which is specifically expressed as: = ; |g|= ; wherein, , and are the accelerations in the x, y, and z directions respectively, is the modulus of the acceleration; Use the attitude calculation result and the acceleration of each finger as the gesture recognition result.

[0010] Preferably, the healthy-side gesture recognition is performed through the myoelectric signals of the arm, specifically: Export the myoelectric signals into the required format, and based on the myoelectric signals in the required format, use deep learning for gesture recognition, and transmit the analyzed gesture recognition results to the flexible driving device to generate relevant actions for inflation and deflation through the flexible driving device.

[0011] Preferably, the flexible driving device includes an airbag glove, multiple groups of solenoid valves and a servo motor, and the bending degree of the fingers is changed by inflating and deflating the airbag glove.

[0012] Preferably, according to the healthy-side gesture recognition result, control the flexible driving device arranged on the affected side of the patient to inflate and deflate, driving each finger on the affected side of the patient to move synchronously with each finger on the healthy side, specifically: Set the inflation and deflation of the flexible driving device by controlling multiple groups of solenoid valves, and cooperate with a pulse wave with a constant amplitude and period and an adjustable duty cycle to control the servo motor to drive the airbag glove to inflate and deflate, driving each finger on the affected side of the patient to move synchronously with each finger on the healthy side.

[0013] Preferably, the first-order dynamic response two-way rate prediction feedforward adjustment algorithm is used to combine the proportional, integral and derivative adjustment laws, and the output inflation and deflation signal is subjected to disturbance compensation with the input healthy-side gesture recognition result, and the residual error of the inflation and deflation signal is adjusted and eliminated through the disturbance compensation result, specifically: Set the open-loop transfer function of the continuous system for inflation and deflation, and the specific expression is: ; Use the first-order dynamic response two-way rate prediction feedforward adjustment algorithm to combine the proportional, integral and derivative adjustment laws for feedback, where the feedback control quantity Y(t) is specifically expressed as: ; Among them, the input-output expression of the feedforward controller is specifically: ; Convert to discrete form to obtain the input set value of the feedforward controller, specifically: ; Among them, is the transfer function, is a point on the complex plane, is the real part, j is the imaginary unit, ω is the imaginary part, k is the discretized data, T is the time constant, t is the current time variable, and A and B are the output and input xAdjustment parameters; among them, the feedforward coefficient K p = , the integral gain K i =K p , is the integral time constant, the derivative gain , is the derivative time constant; By adjusting the ratio of B to A, when -Y(t)=0, the residual error of charging and discharging is eliminated. At this time, the product of the transfer function of the feedforward link of the feedforward controller and the transfer function of the continuous system is 1.

[0014] The present invention provides a computer device, including a memory and a processor. A program is stored in the memory. When the program is executed by the processor, the processor executes the steps of the above-mentioned active rehabilitation training method for patients with hand dysfunction.

[0015] Compared with the prior art, the present invention has the following remarkable advantages: For patients with hand dysfunction, the healthy hand wears a collection device, and the affected hand wears a flexible driving device. By collecting and recognizing the movements of the healthy hand, due to the non-linear hysteresis effect existing in the process of adjusting finger movement by charging and discharging of the rehabilitation device, the first-order dynamic response two-way rate prediction feedforward adjustment technology is adopted to combine the three adjustment laws of proportion, integral and derivative existing during charging and discharging, adjust and eliminate the residual error of charging and discharging, can recognize the myoelectric signal or the electrical signal generated by the motion trajectory data in advance and make a rapid response, make the actions that the patient wants to do and the actual actions highly consistent, drive the movement of the affected hand, realize hand mirror training, and reduce delay and promote rehabilitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the overall simplified flowchart of the model of the present invention; Figure 2 is the flowchart of model training and intention recognition of the present invention; Figure 3 is the flowchart of the feedforward compensation idea of the present invention; Figure 4 is the modeling of the flexible driving device of the present invention; Figure 5 is the modeling of the eight-channel myoelectric acquisition arm ring of the present invention; Figure 6 is the flowchart of an active rehabilitation training method for patients with hand dysfunction of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0017] Combined with the accompanying drawings in the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0018] As Figure 1 shown, an active rehabilitation training system for patients with hand dysfunction used in the present invention includes a gesture acquisition module, a gesture analysis and transmission module, and an action execution module.

[0019] Among them, when the healthy hand moves and is processed in a mirror rehabilitation mode, the gesture acquisition module is used to obtain the motion trajectory data of each finger of the patient's healthy hand or the myoelectric signal of the arm through the acquisition device arranged on the healthy side of the patient; the gesture analysis and transmission module is used to recognize the healthy side gesture through the motion trajectory data of each finger of the patient's healthy hand or the myoelectric signal of the arm; the action execution module is used to control the inflation and deflation of the flexible driving device arranged on the affected side of the patient according to the recognition result of the healthy side gesture, drive each finger of the patient's affected side to move synchronously with each finger of the healthy side, and when inflating and deflating, combine the proportional, integral, and differential adjustment laws by using a first-order dynamic response two-way rate prediction feedforward adjustment algorithm, perform disturbance compensation on the output inflation and deflation signal and the input recognition result of the healthy side gesture, adjust and eliminate the residual error of the inflation and deflation signal through the disturbance compensation result, and control the inflation and deflation of the flexible driving device through the inflation and deflation signal with the residual error eliminated. The flexible driving device is the action execution device in the figure.

[0020] When the healthy hand moves and is processed in an active rehabilitation mode, directly collect the myoelectric signal of the affected arm through the acquisition device arranged on the healthy side of the patient, obtain the intended gesture action, and communicate through low-power Bluetooth.

[0021] As Figure 6 shown, the following describes an active rehabilitation training method for patients with hand dysfunction based on this system, including the steps: Step S1: Obtain the motion trajectory data of each finger of the patient's healthy hand or the myoelectric signal of the arm through the acquisition device arranged on the healthy side of the patient.

[0022] The gesture acquisition module provides two methods. That is, the acquisition device includes various sensors or an eight-channel gesture acquisition armband. Among them, in the first method, the various sensors include a bending sensor, a gyroscope sensor, and an acceleration sensor. Each module is integrated on a circuit board, and the various sensors are arranged above the flexor carpi ulnaris tendon to collect the movement trajectories of each finger. In the second method, the eight-channel gesture acquisition armband is worn on the healthy arm to collect the myoelectric signals of the patient, and transmits them to the computer via low-power Bluetooth for data processing, analyzes the patient's movement intention, and sends the data instruction to the action execution module MCU to execute relevant actions.

[0023] The gesture acquisition principle is that the gyroscope, as the most intuitive angle detector, can detect the "angular velocity" of an object rotating around the coordinate axis, so as to detect the angular velocity of finger movement. Just as integrating velocity over time can calculate the distance, integrating angular velocity over time can calculate the "angle" of rotation. Since the gyroscope uses integration when measuring angles, there will be integration errors. The smaller the integration time Dt, the smaller the error. For example, when calculating the distance, assuming the driving time is 1 hour and randomly selecting the speed Vt at a certain moment during the driving process and multiplying it by 1 hour, the calculated distance error is extremely large because the speed is not equal to that moment's speed throughout the driving process. If the vehicle speed is detected every 5 minutes, the vehicle speeds at 12 moments V1, V2, V3 - V12 can be obtained. Multiply the speed at each moment by the time interval of 5 minutes and sum these 12 results to obtain a relatively accurate driving distance. Integrating angular velocity over time to obtain the angle is the same. To obtain an accurate angle, it is necessary to continuously increase the sampling frequency, make the integration time D smaller, and reduce the error. For the problems caused by the device's own errors, for example, the error of a certain gyroscope is 0.1 degrees / second. When the gyroscope is stationary, the ideal angular velocity should be 0. No matter how long it is stationary, the measured rotation angle obtained by integrating it is 0, which is the ideal state. However, due to the existence of an error of 0.1 degrees / second, when the gyroscope is stationary, the sampled angular velocity is always 0.1 degrees / second. If it is stationary for 1 minute, the measured rotation angle obtained by integrating it is 6 degrees. If it is stationary for 1 hour, the rotation angle measured by the gyroscope through integration is 360 degrees, that is, it has rotated a full circle. Only when the positive and negative errors can exactly cancel each other can this cumulative error be eliminated. Therefore, this invention adopts an error compensation method to eliminate the error of the gyroscope itself.

[0024] Step S2: Identify the healthy-side gestures through the movement trajectory data of each finger on the healthy side of the patient or the myoelectric signals of the arm.

[0025] The gesture analysis and transmission module provides two methods. Method 1 includes a signal processing module and a single-chip radio frequency transceiver device, which are connected to the gesture acquisition module on the same circuit board. It occupies a small space, is small and portable, and can be worn on the wrist. The data is analyzed and processed by the single-chip microcomputer system, and the state of the non-electric finger is converted into an output signal, which is sent to the receiving module of the main control board of the action execution module by the wireless radio frequency module. Method 2 includes a signal receiving module, a signal processing module, and a signal sending module, which receive the myoelectric signals from the gesture acquisition armband and can export the data in the required format for subsequent signal analysis and processing. Deep learning is used for gesture recognition, and the analyzed gestures are transmitted to the action execution module via Bluetooth.

[0026] Method 1 of the gesture analysis and transmission module includes a structural signal processing module and a single-chip radio frequency transceiver device. The signal processing module includes a digital low-pass filter circuit module to reduce noise, and a digital motion processor is used for attitude calculation to calculate the heading angle ψ, roll angle Φ, and pitch angle θ of the finger. Assuming that in the initial state, the Z-axis, X-axis, and Y-axis of the finger are parallel to the celestial axis, north axis, and east axis respectively. When the finger rotates around its own "Z" axis, it will cause its own "Y" axis direction to deviate from the "north-south" direction of the local coordinate system by a certain angle, and this angle is called the yaw angle. When the carrier rotates around its own "X" axis, it will cause its own "Z" axis direction to deviate from the "celestial-earth" direction of the local coordinate system by a certain angle, and this angle is called the pitch angle. When the carrier rotates around its own "Y" axis, it will cause its own "X" axis direction to deviate from the "east-west" direction of the local coordinate system by a certain angle, and this angle is called the roll angle. Using the above three angles, the attitude determined by the finger can be determined. The single-chip radio frequency transceiver device includes nrf24l01 as the sending module. The working mode of nrf24l01 is determined by the PWR_UP register, PRIM_RX register, and CE. In the present invention, the sending mode and the receiving mode are set respectively. The sending mode uses PWR_UP = 1, PRIM_RX = 0, CE = 1, and the data is in the TX FIFO register. The receiving mode uses PWR_UP = 1, PRIM_RX = 1, CE = 1 to achieve wireless communication between two single-chip microcomputers.

[0027] For Method 2 of the gesture analysis and transmission module, MATLAB is used to process the eight-channel data to achieve data visualization. The collected signals are placed in the trained model for convolution, and the convolution obtains the motion intention and is transmitted to the action execution module via Bluetooth.

[0028] The first method of the gesture acquisition module includes a structure with five bending sensors respectively arranged at the positions of the flexor pollicis longus tendon and the four flexor digitorum superficialis tendons of the mirror glove; in the single-chip microcomputer system, the gyroscope sensor measures the finger bending angle, and the acceleration sensor measures the finger flexion and extension acceleration, and converts the bending state and degree of each finger into a quantity recognizable by the single-chip microcomputer. The second method of the gesture acquisition module includes a structure of an eight-channel gesture acquisition armband. Eight dry electrode electromyography sensors installed in the armband are in direct contact with the arm to collect eight-channel electromyography signals, and are transmitted to a computer through an internal Bluetooth module for further analysis.

[0029] The attitude fusion principle is that since directly measuring the angle with a gyroscope will produce cumulative errors during long-term measurement, a sensor for detecting the inclination angle is introduced; an acceleration sensor is used to detect the inclination angle. It samples the force data by detecting the deformation of the device in each direction, and according to the conversion of F = ma, the sensor directly outputs the acceleration data. When the attitude of the sensor is different, the gravitational acceleration detected on its own respective coordinate axes is different. Using the measurement results in each direction and according to the principle of force decomposition, the included angle between each coordinate axis and gravity can be obtained. Because the direction of gravity is fixed to the "heaven and earth" axis, the angle of rotation of the carrier can be obtained by measuring the included angle between each axis of the carrier coordinate system and the direction of gravity, so as to know the attitude of the carrier; when using a gyroscope to detect the angle, there are defects in the static state and it is affected by time, while when using an acceleration sensor to detect the angle, there are defects in the moving state and it is not affected by time, just complementary; both of these two sensors are used at the same time, and a filtering algorithm is designed. When the object is in a static state, the weight of the acceleration data is increased, and when the object is in a moving state, the weight of the gyroscope data is increased, so as to obtain more accurate attitude data.

[0030] The healthy-side gesture recognition is carried out through the motion trajectory data of each finger on the healthy side of the patient, including: Using a digital motion processor to perform attitude calculation on the motion trajectory data of each finger on the healthy side of the patient, and calculating the heading angle of each finger , roll angle and pitch angle ; The specific expression of the attitude calculation is: = ; Among them, is the result of the attitude calculation.

[0031] The acceleration of each finger is obtained through the result of the attitude calculation, and the specific expression is: = ; |g| = ; Among them, , and are the accelerations in the x, y, and z directions respectively. is the magnitude of the acceleration. Using a single-chip radio frequency transceiver device as the transmitting module. According to the above two equations, the heading angle ψ, roll angle Φ, and pitch angle θ can be obtained. The present invention uses a specific algorithm to calculate the accurate angles successively through Euler angle to rotation matrix, Euler angle to quaternion, quaternion to rotation matrix, and rotation matrix to Euler angle, so as to determine the accurate position of finger movement and ensure the accuracy of gesture recognition.

[0032] Perform healthy-side gesture recognition on the myoelectric signals of the arm, including: Receive the myoelectric signals from the gesture acquisition armband, export the myoelectric signals in the required format, and perform gesture recognition using deep learning based on the myoelectric signals in the required format. Transmit the analyzed gesture recognition results to the flexible driving device via Bluetooth, and perform inflation and deflation through the relevant actions generated by the flexible driving device.

[0033] When exporting the myoelectric signals in the required format, perform feature extraction and classification extraction on the collected myoelectric signals, and obtain control decisions using the myoelectric signals after feature extraction and classification extraction.

[0034] Step S3: According to the healthy-side gesture recognition result, control the flexible driving device arranged on the affected side of the patient to perform inflation and deflation, drive each finger on the affected side of the patient to move synchronously with each finger on the healthy side, and when inflating and deflating, combine the proportional, integral, and differential adjustment laws using the first-order dynamic response two-way rate prediction feedforward adjustment algorithm, perform disturbance compensation on the output inflation and deflation signal and the input healthy-side gesture recognition result, adjust and eliminate the residual error of the inflation and deflation signal through the disturbance compensation result, and control the inflation and deflation of the flexible driving device through the inflation and deflation signal with the residual error eliminated.

[0035] Such as Figure 4 and Figure 5As shown, in one embodiment, the flexible driving device includes an airbag glove, multiple groups of solenoid valves, and a servo motor. The bending degree of the fingers is changed by inflating and deflating the airbag glove. Specifically, the flexible driving device is a pneumatic device. Different from rigid drivers, the flexible driving device shows a certain degree of flexibility through a mechanical structure, has the characteristics of being thin, bendable, and easy to fold, provides a greater degree of freedom, fits the human skin better, can complete bending and stretching of different fingers with different forces, and the flexion and extension forces are independently adjustable. The actuator is optimized in combination with the human body structure, and flexible driving is adopted to avoid excessive force and other injuries to the human hand. Moreover, a scientific traction mode is adopted to increase the range of hand traction activities, enabling any finger to flex and extend freely. With a specific flexible control algorithm, smoother movement and higher error tolerance are achieved. After receiving the signal, the data is analyzed and processed. Through a specific algorithm and circuit, a first-order dynamic response two-way rate prediction feedforward adjustment is adopted to reduce errors, and the actuator is controlled to accurately execute the corresponding actions, achieving the purpose of mirror rehabilitation training.

[0036] According to the recognition result of the healthy-side gesture, control the flexible driving device arranged on the affected side of the patient to inflate and deflate, and combine the proportional, integral, and differential adjustment laws existing during inflation and deflation by using the first-order dynamic response two-way rate prediction feedforward adjustment algorithm to adjust and eliminate the residual error of inflation and deflation. Specifically: The inflation and deflation settings of the flexible driving device are performed by controlling multiple groups of solenoid valves. The servo motor is controlled by a pulse wave with a constant amplitude and period and an adjustable duty cycle to drive the actuator to operate, driving each finger on the affected side of the patient to move synchronously with each finger on the healthy side. A feedforward compensation circuit is arranged on the actuator to feedback the combination of the proportional, integral, and differential adjustment laws. As long as the intensities of the three effects are properly coordinated, rapid adjustment can be achieved, and the residual error can be eliminated, obtaining a satisfactory control effect, quickly adjusting and eliminating the residual error, and driving the actuator to execute the corresponding actions. In this embodiment, 5 groups of solenoid valves are used.

[0037] Such as Figure 3As shown in the figure, the idea of feedforward compensation control is that for a system with time delay, when the set target value changes, the controlled variable needs a lag period to change and then achieve the corresponding adjustment effect. The feedforward control system works based on the compensation principle according to the changes in interference factors or target values. Its core feature is that when an interference factor appears and before the controlled variable changes, the system can perform real-time control according to the magnitude of the interference, thereby offsetting the impact of the interference on the controlled variable. If the feedforward control system is used properly, it can eliminate the adverse effects of the interference on the controlled variable when the interference just occurs, and avoid deviations of the controlled variable due to changes in interference or target values. Compared with feedback control, feedforward control can implement control more timely and is not restricted by the system time delay. In the field of high-precision motion control, feedforward control can be used to improve the tracking performance of the system. The feedforward control design in this invention is based on the concept of composite control. When the closed-loop system is a continuous system, the result of multiplying the transfer function of the feedforward link by that of the closed-loop system is 1, thereby achieving accurate reproduction of the output to the input.

[0038] A feedforward compensation circuit is set up. The proportional, integral, and differential adjustment laws are combined using the first-order dynamic response two-way rate prediction feedforward adjustment algorithm. The output charging and discharging signal is subjected to disturbance compensation with the input healthy-side gesture recognition result, and the residual error of the charging and discharging signal is adjusted and eliminated through the disturbance compensation result. Specifically: The open-loop transfer function of the continuous system for charging and discharging is set up, and the specific expression is: ; The proportional, integral, and differential adjustment laws are combined together for feedback using the first-order dynamic response two-way rate prediction feedforward adjustment algorithm. The specific expression of the feedback control quantity Y(t) is: ; Among them, the input-output expression of the feedforward controller is specifically: ; The is discretized to obtain the input set value of the feedforward controller. Specifically: ; Among them, is the transfer function, is a point on the complex plane, is the real part, j is the imaginary unit, ω is the imaginary part, k is the discretized data, T is the time constant, t is the current time variable, which is consistent with in the figure, and A and B are the output and input xAdjustment parameters; among them, the feedforward coefficient K p = , the integral gain K i =K p , is the integral time constant, the derivative gain , is the derivative time constant.

[0039] By adjusting the ratio of B to A, when -Y(t)=0, the residual error of charging and discharging is eliminated. At this time, the product of the transfer function of the feedforward link of the feedforward controller and the transfer function of the continuous system is 1. Minimize the error as much as possible to achieve complete reproduction of the output with the input.

[0040] Therefore, the present invention completes difficult hand movements according to the self-movement intention, realizes active treatment, restores nerve activity, accelerates the rehabilitation process, and integrates the first-order dynamic response two-way rate prediction feedforward adjustment, which can effectively reduce the delay of the action execution module, enable the affected hand to move synchronously with the healthy hand, achieve the unity of knowledge and action, and optimize the rehabilitation effect.

[0041] As Figure 2 shown, the present invention includes an offline training classification model and an online recognition of action intention. Among them, the offline training classification model includes offline collection, preprocessing, feature extraction, feature selection, and model training; among them, the online recognition of action intention includes online collection, preprocessing, feature extraction, classification model, and control decision-making, and the classification model is obtained through model training.

[0042] The principle of data preprocessing is as follows: Data preprocessing includes missing value processing, noise filtering, data integration, data reduction, and data transformation; missing value processing has deletion method and imputation method, and the imputation method can be divided into mean imputation, regression imputation, and maximum likelihood estimation. The present invention adopts the imputation method to impute missing values with the most likely values; noise filtering has regression method, mean smoothing method, outlier analysis method, and wavelet filtering method. The present invention adopts the mean smoothing method, which replaces the original data with the mean of several adjacent data for variables with sequence characteristics; data integration is to logically or physically integrate the data in several scattered data sources into a unified data set; the purpose of data reduction is to obtain a data set that can be approximately equivalent to the original data set or even better but with less data volume; data transformation is to convert the data representation form. Standardization is to scale the data proportionally so that it falls into a small specific interval for easy comparison and weighting, and the 0-1 standardization method is adopted. When new data is added, it needs to be redefined; discretization divides continuous data into several segments, which can effectively overcome the hidden defects in the data and make the model results more stable.

[0043] The construction principle of the convolutional neural network is as follows: First, it is necessary to import the necessary libraries. In this invention, Pytorch is used as the main library for constructing and training the neural network, and torchvision will be used to process the dataset and perform conversions. Secondly, the architecture of the convolutional neural network is defined, including the convolutional layer and the pooling layer. Then, the hardware settings are made. To be able to use the CPU, in this invention, the device is set to run on the GPU. Hyperparameters are defined, which are configuration settings used to adjust the model training method. The MNIST dataset is downloaded and loaded using the torchvision.datasets module. The neural network is initialized and moved to the GPU. The loss and optimizer are defined. In this invention, cross-entropy loss is used for classification, and the Adam optimizer is used to update the weights of the model. The dataset is looped multiple times and the model weights are updated according to the loss. The model is continuously trained to improve its accuracy so that it can accurately identify the motion intention without error.

[0044] The convolutional layer is the building block of the CNN and consists of a filter (kernal), a stride, and padding. The filter is a small matrix that slides over the input image, performs element-wise multiplication, and then sums the results. Each filter is designed to detect specific features in the input image. For example, the filter can detect horizontal edges, vertical edges, or more complex textures. The output of applying the filter to the input image is called the feature map or activation map, and multiple filters will result in multiple feature maps. The stride is the step size at which the filter moves over the input image. A stride of 1 means the filter moves one pixel at a time, both horizontally and vertically. A larger stride will reduce the size of the feature map because the filter will skip more pixels. For example, a stride of 2 means the filter moves two pixels at a time, effectively downsampling the feature map. Padding involves adding extra pixels around the boundaries of the input image. These extra pixels are usually set to zero (zero padding). Padding ensures that the filter correctly covers the image, especially at the edges. Without padding, the size of the feature map will decrease after each convolutional operation. For example, for a 5x5 input image and a 3x3 filter without padding, the resulting feature map will be 3x3. When the padding is 1, the feature map remains the same size as the input. The pooling layer reduces the spatial dimension of the feature map, which helps improve the computational efficiency of the network and reduce overfitting. There are two main types of pooling: max pooling and average pooling. Max pooling takes the maximum value from each patch of the feature map. For example, in a 2x2 max pooling operation, the maximum value of each 2x2 block of the feature map is taken to create a new, smaller feature map. This operation reduces the size of the feature map by half in both the horizontal and vertical directions while retaining the most prominent features. Average pooling takes the average value of each patch of the feature map, similar to max pooling, but instead of the maximum value, it takes the average value of each block. In this invention, max pooling is adopted.

[0045] A feature map is the output of a convolutional layer after applying filters to an input image. Each feature map corresponds to a different filter and captures different features from the input. Multiple feature maps are stacked together to form a multi-channel output, which is used as the input for the next layer.

[0046] The present invention also provides a computer device, including a memory and a processor. When a program stored in the memory is executed by the processor, the processor is caused to execute the steps of an active rehabilitation training method for patients with hand dysfunction.

[0047] According to the disclosed embodiments, the computer device can communicate with one or more external devices (such as a keyboard, a pointing device, Bluetooth communication, etc.), or communicate with any device (such as a router, a demodulator, etc.) that enables the computing device to communicate with one or more other computing devices.

[0048] In the description of the present application, it should be noted that the terms "installation", "connection", and "coupling" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or a connection through an intermediate medium. The above are only specific embodiments of the present application to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown in the text, but will conform to the broadest scope consistent with the principles and novel features claimed in the text.

[0049] The above content is a further detailed description of the present invention in combination with specific preferred embodiments. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, which should all be regarded as belonging to the protection scope of the present invention.

Claims

1. An active rehabilitation training system for patients with hand dysfunction, characterized in that: include: A gesture acquisition module is used to acquire the motion trajectory data of each finger on the healthy side of the patient or the electromyographic signal of the arm through an acquisition device set on the healthy side of the patient; The gesture analysis and transmission module is used to identify the healthy side gestures through the motion trajectory data of each finger on the healthy side of the patient or the electromyographic signal of the arm; The action execution module is used to control the flexible driving device arranged on the affected side of the patient to inflate and deflate according to the healthy side gesture recognition result, so as to drive each finger on the affected side of the patient to move synchronously with each finger on the healthy side, and when inflating and deflation, a first-order dynamic response bidirectional rate estimation feedforward adjustment algorithm is used to combine the three adjustment laws of proportion, integration and differentiation, and the output inflation and deflation signal is disturbed by the input healthy side gesture recognition result, and the residual of the inflation and deflation signal is adjusted and eliminated through the disturbance compensation result, and the inflation and deflation of the flexible driving device is controlled by the inflation and deflation signal that eliminates the residual.

2. An active rehabilitation training method for patients with hand dysfunction, characterized in that: include: The motion trajectory data of each finger on the healthy side of the patient or the electromyographic signal of the arm is obtained by a collection device set on the healthy side of the patient; The patient's healthy side gestures are recognized through the movement trajectory data of each finger on the healthy side or the electromyographic signal of the arm; According to the healthy-side gesture recognition result, the flexible drive device arranged on the patient's affected side is controlled to inflate and deflate, driving each finger on the affected side of the patient to move synchronously with each finger on the healthy side. When inflating and deflation, a first-order dynamic response bidirectional rate estimation feedforward adjustment algorithm is used to combine the three regulation laws of proportion, integration and differentiation, and the output inflation and deflation signal is disturbed by the input healthy-side gesture recognition result. The residual of the inflation and deflation signal is adjusted and eliminated through the disturbance compensation result, and the inflation and deflation of the flexible drive device is controlled by the inflation and deflation signal that eliminates the residual.

3. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 2, characterized in that: The acquisition device includes various sensors or an eight-channel gesture acquisition armband; wherein the various sensors include bending sensors, gyroscope sensors and acceleration sensors, and the various sensors are arranged above the ulnar wrist flexor tendon of the healthy side to collect the motion trajectory data of each finger; the eight-channel gesture acquisition armband is worn on the healthy arm to collect the patient's electromyographic signals.

4. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 2, characterized in that: The patient's healthy side gesture recognition is performed through the motion trajectory data of each finger on the healthy side, including: Perform posture calculation on the motion trajectory data of each finger on the healthy side of the patient and calculate the heading angle of each finger , Roll Angle and pitch angle ; The specific expression of posture solution is: = ； in, is the result of attitude solution; The acceleration of each finger is obtained through the posture solution result, which is specifically expressed as: = ； |g|= ； in, , and are the accelerations in the x, y and z directions respectively, is the modulus of acceleration; The posture calculation result and the acceleration of each finger are used as the gesture recognition result.

5. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 2, characterized in that: The healthy side gesture recognition is performed through the electromyographic signal of the arm, specifically: The electromyographic signal is exported into a required format, and based on the electromyographic signal in the required format, deep learning is used to perform gesture recognition, and the analyzed gesture recognition result is transmitted to the flexible driving device, and inflation and deflation are performed through relevant actions generated by the flexible driving device.

6. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 2, characterized in that: The flexible driving device comprises an airbag glove, a plurality of solenoid valves and a servo motor, and changes the bending degree of the fingers by inflating and deflating the airbag glove.

7. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 6, characterized in that: According to the healthy side gesture recognition result, the flexible driving device arranged on the patient's affected side is controlled to inflate and deflate, so as to drive each finger on the affected side of the patient to move synchronously with each finger on the healthy side, specifically: The inflation and deflation of the flexible driving device is set by controlling multiple groups of solenoid valves, and the pulse wave with constant amplitude and period and adjustable duty cycle controls the servo motor to drive the airbag glove to inflate and deflate, driving each finger on the affected side of the patient and each finger on the healthy side to move synchronously.

8. The active rehabilitation training method for patients with hand dysfunction as claimed in claim 7, characterized in that: The first-order dynamic response bidirectional rate estimation feedforward adjustment algorithm combines the three regulation rules of proportion, integration and differentiation, performs disturbance compensation on the output inflation and deflation signal and the input healthy side gesture recognition result, and adjusts and eliminates the residual of the inflation and deflation signal through the disturbance compensation result, specifically: Set the open-loop transfer function of the continuous system of charging and discharging, the specific expression is: ； The first-order dynamic response bidirectional rate estimation feedforward adjustment algorithm is used to combine the three regulation laws of proportion, integration and differentiation for feedback. The specific expression of the feedback control quantity Y(t) is: ； Among them, the input and output expressions of the feedforward controller are Specifically: ； Will Discretize and obtain the input setting value of the feedforward controller, which is: ； in, is the transfer function, is a point on the complex plane, is the real part, j is the imaginary unit, ω is the imaginary part, k For discretized data, T is the time constant, t is the current time variable, A and B are output and input x The adjustment parameters of K p = , integral gain K i =K p , is the integral time constant, the differential gain , is the differential time constant; By adjusting the ratio of B and A, -When Y(t)=0, the residual difference of inflation and deflation is eliminated. At this time, the product of the feedforward link transfer function of the feedforward controller and the continuous system transfer function is 1.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of an active rehabilitation training method for a patient with hand dysfunction as described in any one of claims 2 to 8.

Images