CN111696645A - Hand exoskeleton rehabilitation training device and method based on surface electromyographic signals - Google Patents

Abstract

The invention discloses a hand exoskeleton rehabilitation training device and method based on surface electromyographic signals, which comprises a mechanical glove, a surface electromyographic signal acquisition system and a drive control system, wherein the surface electromyographic signal acquisition system comprises two electromyographic signal acquisition instruments, and each electromyographic signal acquisition instrument transmits the electromyographic signals to the drive control system through a Bluetooth module; the drive control system comprises an FPGA chip, a first rechargeable lithium battery, a Bluetooth module and a digital filter, wherein the FPGA chip carries out secondary filtering processing on the electromyographic signals after receiving the electromyographic signals of the electromyographic signal acquisition instrument; the method comprises the steps of obtaining an active segment signal by using a standard deviation threshold detection method, extracting time domain characteristics of the signal, inputting time domain characteristic parameters into an optimized BP neural network, identifying gesture actions of various fingers through the BP neural network, adding labels to the different gesture actions by an FPGA chip, converting surface electromyographic signals into control signals, and sending the control signals to mechanical gloves through a Bluetooth module so as to drive the fingers to move.

Description

Hand exoskeleton rehabilitation training device and method based on surface electromyographic signals

Technical Field

The invention relates to the field of rehabilitation training equipment, in particular to a hand exoskeleton rehabilitation training device and method based on surface electromyographic signals.

Background

Most of stroke patients can leave sequelae after operation, which is characterized in that: the walking is inconvenient, the hands and the limbs are stiff, and the like, but the hands play an indispensable role in daily life, so the rehabilitation of the stroke patient after the operation is very important. At present, the main postoperative rehabilitation treatment modes are as follows: the physical therapy engineer is used for massaging the limbs of the patient manually to perform passive exercise stimulation, but the method consumes a large amount of manpower, requires a great amount of experience of the physical therapy engineer, and brings great inconvenience to the patient and family life due to the long treatment period after operation.

Therefore, the postoperative rehabilitation equipment based on the mechanical gloves is gradually accepted by people. However, most of the traditional rehabilitation mechanical gloves adopt simple mechanical structures, and fixed actions are set through devices such as sliders and connecting rods to assist patients in performing rehabilitation exercises.

The intelligent exoskeleton rehabilitation manipulator and the method thereof based on the healthy-side biological electric control are disclosed by Chinese patent publication numbers, a driving mechanism is arranged on a palm back platform and is used for independently driving each mechanical unit of an exoskeleton manipulator body to move, and a control system receives position signals fed back by the driving mechanism and controls the exoskeleton manipulator to move and/or stop; the surface electromyography sensor of the surface electromyography system is arranged at the muscle belly of relevant muscles of hands and forearms of a healthy side upper limb of a patient, surface electromyography signals of the corresponding muscles are collected and subjected to feature extraction, the signal recognition module receives surface electromyography feature parameters, the feature parameters are sent to a trained classifier to recognize different action types, the bioelectricity evaluation system receives the surface electromyography feature parameters, individual features and real-time states of the patient are evaluated to obtain an evaluation result, and the recognition result and the evaluation result are sent to the control system to control the movement of the driving mechanism, so that the movement of the hand at the affected side is driven.

Then, it has the following disadvantages:

the finger stall is used for connecting the finger of the patient with the mechanical arm, according to an observation matching picture, the mechanical arm finger stall in the step A is in a semi-circular arc shape, so that the finger and the finger stall cannot be well and tightly connected together, and the use inconvenience is increased; the myoelectric sensors are placed on the arms and the palm in a scattered manner, the positions of the muscles need to be accurately found when the myoelectric sensors are used each time, and how to transmit signals to the signal identification module is not described; multiple characteristics such as time domain characteristics, frequency domain characteristics, time-frequency domain characteristics, nonlinear analysis and the like are selected and extracted, the complexity of a system in processing the characteristic part is increased, the processing time of the system is prolonged, meanwhile, the redundancy among the characteristics is increased, and the burden is increased on the recognition of actions; the deep neural network needs a large amount of data for training, the data needs to be acquired from a patient, great pain is caused to the patient, the use interest of the patient is reduced, and the time consumption is long in practical application.

Disclosure of Invention

The invention aims to provide a hand exoskeleton rehabilitation training device and method based on surface electromyogram signals, aiming at the defects in the prior art.

In order to achieve the purpose, the invention is implemented according to the following technical scheme:

a hand exoskeleton rehabilitation training device based on surface electromyographic signals comprises mechanical gloves worn on hands and used for driving fingers to move, a surface electromyographic signal acquisition system and a driving control system used for controlling the mechanical gloves and the surface electromyographic signal acquisition system;

the surface electromyographic signal acquisition system comprises two electromyographic signal acquisition instruments worn on any position of an arm, and each electromyographic signal acquisition instrument transmits an electromyographic signal to the drive control system through a Bluetooth module;

the drive control system comprises an FPGA chip, a first rechargeable lithium battery and a Bluetooth module, wherein the first rechargeable lithium battery supplies power for the FPGA chip and the Bluetooth module, the FPGA chip is connected with a digital filter, a signal input end of the FPGA chip is connected with a signal output end of an electromyographic signal acquisition instrument through the Bluetooth module, the FPGA chip carries out secondary filtering processing on the electromyographic signal after receiving the electromyographic signal of the electromyographic signal acquisition instrument, and the electromyographic signal within the frequency range of 20-500Hz is intercepted; the method comprises the steps of obtaining an active segment signal by utilizing a standard deviation threshold detection method in an FPGA chip, extracting time domain characteristics of the signal, inputting time domain characteristic parameters into an optimized BP neural network, identifying gesture actions of various fingers through the BP neural network, adding labels to the different gesture actions by the FPGA chip, converting surface electromyographic signals into control signals, and sending the control signals to mechanical gloves through a Bluetooth module so as to drive the fingers to move.

Furthermore, the electromyographic signal acquisition instrument comprises an annular elastic belt, four groups of surface electromyographic sensors, a second rechargeable lithium battery and a Bluetooth module, wherein the second rechargeable lithium battery supplies power for the electromyographic sensors and the Bluetooth module, each group of surface electromyographic sensors comprises two electromyographic sensors, the two electromyographic sensors in each group are symmetrically distributed, eight electromyographic sensors are fixed on the circumference of the inner surface of the annular elastic belt at equal intervals, each electromyographic sensor comprises three electrodes welded on a circuit board in a straight shape, the electrode in the middle of the electromyographic sensors is used as a reference electrode, the electrodes on two sides acquire the electromyographic signals in a differential structure, and the circuit board is also welded with a signal amplification circuit and a filter circuit which are used for amplifying and filtering the electromyographic signals.

Further, the mechanical glove comprises a palm support plate, a thumb assembly, an index finger assembly, a middle finger assembly, a ring finger assembly, a little finger assembly, a box sleeve, a Bluetooth module, a third rechargeable lithium battery and a nylon buckle; the hand-held glove comprises a thumb assembly, an index finger assembly, a middle finger assembly, a ring finger assembly and a little finger assembly, wherein the thumb assembly, the index finger assembly, the middle finger assembly, the ring finger assembly and the little finger assembly correspond to five fingers of a hand of a person respectively;

the thumb component comprises a first steering engine, a first transmission rod and a first connecting rod which drive a first supporting rod and a second supporting rod, and the rear end of the first supporting rod is connected with the upper end face of the palm supporting plate through the component; the front end of the first supporting rod is rotatably connected with the rear end of the second supporting rod through a pin shaft, the palm supporting plate and the second supporting rod are connected through a first connecting rod, the first steering engine is connected with the first supporting rod through a first transmission rod, the first steering engine is fixed on the other end face of the palm supporting plate, the whole action of the thumb component is controlled through the first steering engine, and a first finger ring for fixing the thumb is arranged on the first supporting rod;

the forefinger subassembly includes the second steering wheel, second transfer line and second connecting rod, the third connecting rod drives the third bracing piece, the fourth bracing piece, the fifth bracing piece, the third bracing piece is connected through the up end of subassembly with the palm backup pad, the fourth bracing piece passes through the second connecting rod and the round pin axle is connected with the third bracing piece, the fifth bracing piece passes through the third connecting rod, round pin axle and third bracing piece, the fourth bracing piece is connected, be connected through the second transfer line between second steering wheel and the third bracing piece, the second steering wheel is fixed in the backup pad, action through second steering wheel control forefinger subassembly, third bracing piece middle part is equipped with the second ring, fourth bracing piece middle part is equipped with the third ring.

In addition, the invention also provides a hand exoskeleton rehabilitation training method based on the surface electromyogram signal, which comprises the following steps:

and S1, wearing the two electromyographic signal acquisition instruments on the arm in a staggered mode, wherein the electromyographic signal acquisition instrument close to the palm is called a near-end acquisition instrument, and the electromyographic signal acquisition instrument far away from the palm is called a far-end acquisition instrument. After the near-end acquisition instrument is worn, the far-end acquisition instrument is worn on the arm in a staggered mode; the method comprises the steps that surface electromyographic signals generated by actions are collected through an electromyographic sensor, the electromyographic signals are preprocessed through a signal amplification circuit and a filter circuit, the surface electromyographic signals are amplified through the signal amplification circuit, 50Hz power frequency interference is filtered through the filter circuit, and then the signals are transmitted to an FPGA chip of a drive control system through a Bluetooth module;

s2, wearing the mechanical gloves on the hands;

s3, after receiving the preprocessed electromyographic signals, the FPGA chip carries out secondary filtering processing on the electromyographic signals through a digital filter, intercepts the electromyographic signals within the frequency range of 20-500Hz, acquires active segment signals by using a standard deviation threshold detection method in the FPGA chip, extracts time domain characteristics of the signals, inputs time domain characteristic parameters into the optimized BP neural network as input parameters of a classifier of the BP neural network, identifies gesture actions of various fingers through the BP neural network, adds labels to the different gesture actions by the FPGA chip, converts surface electromyographic signals into control signals, and sends the control signals to the mechanical gloves through a Bluetooth module so as to drive the fingers to move;

s4, after receiving a control signal of the drive control system, the Bluetooth module in the mechanical glove drives each finger assembly to act through each steering engine, and then drives the fingers to perform rehabilitation training.

Specifically, the step S3 of acquiring the active segment signal by using the standard deviation threshold detection method includes the specific steps of:

step 1, intercepting a base signal with the length of L from an electromyographic signal;

step 2, calculating the mean value alpha and the standard deviation beta of the substrate signal:

wherein x (j) is collected electromyographic signal data;

step 3, selecting a sliding window to calculate the standard deviation of the electromyographic signals, setting the window length as W,

wherein u is L + W, L + W +1, …;

step 4, calculating an activity threshold Th:

Th＝θ·β (4)

wherein theta is a magnification factor;

and 5, comparing a threshold value with a standard deviation, setting the electromyographic signal data of the part below the threshold value to zero, calculating the interval length D between the first data of the part above the threshold value and the last data of the part above the threshold value, and considering the signal data of the section as an active section signal when D is more than W, otherwise, discarding the signal data.

Specifically, the time domain characteristic parameters in step S3 include an absolute mean MAV, a root mean square value RMS, a waveform length WL, a zero crossing point ZC, and a slope sign change SSC, and their calculation formulas are respectively:

(1) absolute mean value MAV:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(2) root mean square value RMS:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(3) waveform length WL:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(4) zero crossing number ZC:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(5) slope sign change SSC:

where N is the data segment length and x (k) is the current data sequence.

Specifically, in step S3, a simulated annealing algorithm and a whale optimization algorithm with a fast convergence rate are used to optimize the BP neural network, and the characteristic parameter is recorded as D₀D is₀Divided into training sets D₁Test set D₂The method comprises the following two steps:

step 1, initializing a BP neural network threshold and a connection weight, and establishing an initial model;

step 2, training set data D₁Is imported into the model and is paired with D₁Carrying out normalization;

and 3, optimizing the threshold and the weight of the BP neural network by using an SA-WOA hybrid algorithm:

(1) in the first stage, the weight and the threshold are optimized by using WOA:

a. encircling predation: searching the position of each whale individual, regarding the parameter as a prey, positioning the position of the prey through echo, performing hunting, and updating the position of the whale individual;

b. soaking and net predation: the whale individual contracts and surrounds a prey through a convergence factor, swims to the prey in a spiral motion mode, and continuously updates the position of the whale individual;

c. searching for predation: and (4) carrying out optimizing search on the whale colony according to the coefficient vector parameters and the mutual positions of the individuals to obtain the accurate position of the prey.

(2) In order to prevent the whale optimization algorithm from being trapped in local optimization, parameters obtained by the whale optimization algorithm are further optimized through the simulated annealing algorithm in the second stage:

d. using the optimized parameters obtained by WOA as the current solution S of the simulated annealing algorithm₁Generating a new solution S using the state function₂；

e. Calculating S₁And S₂The increment delta S between the two and whether a new solution S can be accepted or not is judged according to the Metropolis rule₂；

f. Judging whether the obtained result meets the SA termination condition, if so, performing the step 4, otherwise, returning to the part a of the step 3, and performing parameter optimization again;

step 4, taking the optimized parameters obtained in the step 3 as the optimal weight and threshold of the BP neural network;

step 5, retraining the BP neural network, and updating errors until an optimal result is achieved;

step 6, testing data D₂Normalizing, and then bringing the normalized result into a trained neural network, verifying the effectiveness of the classifier, and adding corresponding labels to each gesture action;

and 7, converting different action labels into corresponding control signals according to the result of the step 6, sending the control signals to the mechanical gloves through the Bluetooth module, and driving the fingers to act through the mechanical gloves so as to achieve the aim of rehabilitation training.

Compared with the prior art, the invention has the following beneficial effects:

1. the finger ring is selected to connect the finger and the rehabilitation hand together, and the thumb part of the mechanical hand is provided with the finger ring for fixing the thumb; the other four fingers are fixed with two finger rings which respectively correspond to 3 finger joints of the fingers. And the ring can be adjusted according to the size that the patient pointed, has increased the practicality of manipulator, and the area of contact of ring and finger is less, and the heat dissipation of the finger of being convenient for has promoted user experience. Meanwhile, the Bluetooth module and the rechargeable lithium battery are additionally arranged at the back of the hand, so that the manipulator part can receive a control signal sent by the control module conveniently;

2. adopt two flesh electrical signal collection appearance to gather flesh electrical signal, every flesh electrical signal collection appearance includes 8 flesh electrical sensors, utilizes the elastic cord to enclose 8 sensors together, conveniently wears on the forearm, for gathering more flesh electrical signal, uses 2 collection appearance crisscross each other. Meanwhile, the Bluetooth module and the rechargeable lithium battery are additionally arranged on the sensor, so that the use convenience of the sensor is improved; meanwhile, the collected electromyographic signals are preprocessed, including filtering, amplification and action section signal extraction, so that the gesture action recognition rate is improved, the system processing time is reduced, and the real-time performance is enhanced;

3. time domain characteristics in 5 kinds of slope sign change, including an absolute mean value, a root mean square value, a waveform length, a zero crossing point number and slope sign change are selected and extracted, and through experimental tests, actions can be effectively classified and identified through the extracted 5 kinds of characteristic parameters; meanwhile, the extraction of time domain features is more convenient, the processing time of the system is reduced, and the user experience is increased;

4. and selecting a BP neural network optimized by a Whale Optimization Algorithm (WOA) and a simulated annealing algorithm (SA) to classify the action of the electromyographic signals. Compared with a deep neural network, the BP neural network has the advantages that required training data are less, the FPGA is easy to implement, the practicability of the BP neural network is improved, the recognition rate of the optimized BP neural network on gesture actions is higher, and the practical application experience of a user is improved;

5. according to the invention, the electromyographic signal acquisition system and the drive control system are selectively placed on the healthy side arm, the mechanical rehabilitation hand is placed on the affected side arm, the whole system can be worn on the upper limb of a patient, the patient can conveniently carry out rehabilitation training in different environments, and the convenience of using the device is improved.

Drawings

FIG. 1 is a top view of the structure of a mechanical glove.

FIG. 2 is a structural side view of the thumb assembly of the mechanical glove.

FIG. 3 is a side view of the structure of the index finger assembly of the mechanical glove.

Fig. 4 is a schematic diagram of a surface electromyogram signal acquisition instrument.

Fig. 5 is a schematic diagram of the structure of the electromyographic sensor.

Fig. 6 is a signal amplification circuit diagram.

FIG. 7 is a diagram of a 50Hz trap circuit.

Fig. 8 is a schematic diagram of the operation of the drive control system.

FIG. 9 is a gesture motion recognition workflow diagram.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. The specific embodiments described herein are merely illustrative of the invention and do not limit the invention.

As shown in fig. 1, the present invention relates to a surface electromyogram signal-based hand exoskeleton rehabilitation training device, which takes the right hand as an example, and comprises a mechanical glove worn on the hand for driving fingers to move, a surface electromyogram signal acquisition system and a drive control system (not shown in the figure) for controlling the mechanical glove and the surface electromyogram signal acquisition system;

the surface electromyographic signal acquisition system comprises two electromyographic signal acquisition instruments 13 worn on any position of an arm, and each electromyographic signal acquisition instrument 13 transmits an electromyographic signal to the drive control system through a Bluetooth module; the right hand is taken as an example, namely the right hand is a diseased hand, and mechanical gloves are worn on the right hand, so that the surface electromyogram signal acquisition system and the drive control system are worn on the left arm.

The drive control system comprises an FPGA chip, a first rechargeable lithium battery and a Bluetooth module, wherein the first rechargeable lithium battery supplies power for the FPGA chip and the Bluetooth module, the FPGA chip is connected with a digital filter, a signal input end of the FPGA chip is connected with a signal output end of an electromyographic signal acquisition instrument through the Bluetooth module, the FPGA chip carries out secondary filtering processing on the electromyographic signal after receiving the electromyographic signal of the electromyographic signal acquisition instrument, and the electromyographic signal within the frequency range of 20-500Hz is intercepted; the method comprises the steps of obtaining an active segment signal by utilizing a standard deviation threshold detection method in an FPGA chip, extracting time domain characteristics of the signal, inputting time domain characteristic parameters into an optimized BP neural network, identifying gesture actions of various fingers through the BP neural network, adding labels to the different gesture actions by the FPGA chip, converting surface electromyographic signals into control signals, and sending the control signals to mechanical gloves through a Bluetooth module so as to drive the fingers to move.

As shown in fig. 4, the electromyographic signal acquisition instrument 13 includes an annular elastic band 131, four groups of surface electromyographic sensors, second rechargeable lithium batteries and a bluetooth module, the second rechargeable lithium batteries supply power to the electromyographic sensors and the bluetooth module, each group of surface electromyographic sensors includes two electromyographic sensors 132, the two electromyographic sensors 132 in each group are symmetrically distributed, eight electromyographic sensors 132 are fixed on the circumference of the inner surface of the annular elastic band 131 at equal intervals, each electromyographic sensor 132 includes three electrodes welded in a word on a circuit board 1321, the middle electrode is a reference electrode 1322, the electrodes 1323 on both sides acquire the electromyographic signals in a differential structure, as shown in fig. 5, the two differential electrode pads are both 1cm long and 1cm wide, the reference electrode is 1cm long and 0.5cm wide, and each welding plate is 0.5cm apart; and the circuit board is also welded with a signal amplification circuit and a filter circuit which are used for amplifying and filtering the electromyographic signals. Because the strength of the surface electromyogram signal is extremely low, the surface electromyogram signal is not beneficial to post processing, the electromyogram signal is amplified by about 300 times through an amplifying circuit, the amplifying circuit is composed of an instrumentation amplifier INA333 and a peripheral circuit, and the specific amplifying circuit is shown in fig. 6. Because the surface electromyogram signal is easily interfered by 50Hz power frequency, the invention filters the 50Hz signal by designing a 50Hz trap circuit, the trap circuit is realized by adopting LM324 and peripheral circuits thereof, and the specific circuit is shown in figure 7.

Since there is a certain distance between the electromyographic sensors 132 shown in fig. 4, because the electromyographic signals between the distances cannot be collected, two electromyographic signal collectors 13 are alternatively worn on the arm in a staggered manner, the electromyographic signal collector 13 close to the palm is called a proximal end collector, and the electromyographic signal collector 13 far from the palm is called a distal end collector. After the near-end acquisition instrument is worn, the far-end acquisition instrument is worn on the arm in a staggered mode, namely gaps of the near-end sensors are filled with all the electromyographic sensors of the far-end acquisition instrument, so that more electromyographic signals are acquired.

It should be noted that the specific mechanical structure of the mechanical glove is similar to that of the existing common mechanical glove, as shown in fig. 1, the mechanical glove comprises a palm support plate 1, a thumb component 3, an index finger component 4, a middle finger component 5, a ring finger component 6, a little finger component 7, a box sleeve 2, a bluetooth module, a third rechargeable lithium battery, and a nylon fastener tape 8; thumb subassembly 2, forefinger subassembly 3, middle finger subassembly 4, third finger subassembly 5, little finger subassembly 6 correspond the five fingers of staff respectively, and wherein thumb subassembly 3 installs in palm backup pad 1 side, and forefinger subassembly 4, middle finger subassembly 5, third finger subassembly 6 and little finger subassembly 7 install in backup pad 1 front end, be fixed with box cover 2 on the palm backup pad 1 terminal surface, be fixed with third rechargeable lithium cell (not drawn in the picture) in the box cover 2, be used for receiving the bluetooth module (not drawn in the picture) of the control signal that drive control system sent, third rechargeable lithium cell in the box cover 2 supplies power to thumb subassembly 3, forefinger subassembly 4, middle finger subassembly 5, third finger subassembly 6, little finger subassembly 7 and bluetooth module, be equipped with on the box cover 2 and be used for fixing mechanical gloves at the nylon fastener tape 8 on the palm.

As shown in fig. 2, the thumb component 3 includes a first steering engine 34, a first transmission rod 35 and a first connecting rod 38, which drive a first support rod 36 and a second support rod 37, and the rear end of the first support rod 36 is connected with the upper end surface of the palm support plate 1 through a component 310; the front end of the first supporting rod 36 is rotatably connected with the rear end of the second supporting rod 37 through a pin shaft 39, the palm supporting plate 1 and the second supporting rod 37 are connected through a first connecting rod 38, the first steering gear 34 is connected with the first supporting rod 36 through a first transmission rod 35, the first steering gear 34 is fixed on the other end face of the palm supporting plate 1, the whole action of the thumb component is controlled through the first steering gear 34, and a first finger ring 9 used for fixing the thumb is arranged on the first supporting rod 36.

As shown in fig. 3, the index finger assembly 4 includes a second steering gear 43, a second transmission rod 44, a second connecting rod 412, and a third connecting rod 413 drives a third support rod 45, a fourth support rod 46, and a fifth support rod 47, the third support rod 45 is connected with the upper end surface of the palm support plate 1 through an assembly 414, the fourth support rod 46 is connected with the third support rod 45 through a second connecting rod 412 and a pin 411, the fifth support rod 47 is connected with the third support rod 45 and the fourth support rod 46 through a third connecting rod 413 and a pin, the second steering gear 43 is connected with the third support rod 45 through the second transmission rod 44, the second steering gear is fixed on the support plate 1, the action of the index finger assembly is controlled by a second steering gear 49, a second finger ring 11 is disposed in the middle of the third support rod 45, and a third finger ring 10 is disposed in the middle of the fourth support rod 46.

The structures of the middle finger assembly 5, the ring finger assembly 6 and the little finger assembly 7 are consistent with the structure of the index finger assembly 4, and are not described in detail.

When the hand exoskeleton rehabilitation training device based on the surface electromyographic signals is used for carrying out rehabilitation training on fingers, the specific steps are as follows:

1. the two electromyographic signal acquisition instruments are worn on the arm in a staggered mode, the electromyographic signal acquisition instrument close to the palm is called a near-end acquisition instrument, and the electromyographic signal acquisition instrument far away from the palm is called a far-end acquisition instrument. After the near-end acquisition instrument is worn, the far-end acquisition instrument is worn on the arm in a staggered mode; the method comprises the steps that surface electromyographic signals generated by actions are collected through an electromyographic sensor, the electromyographic signals are preprocessed through a signal amplification circuit and a filter circuit, the surface electromyographic signals are amplified through the signal amplification circuit, 50Hz power frequency interference is filtered through the filter circuit, and then the signals are transmitted to an FPGA chip of a drive control system through a Bluetooth module;

2. donning a mechanical glove on a hand;

3. after the FPGA chip receives the preprocessed electromyographic signals, as the main energy frequency spectrum of the surface electromyographic signals is concentrated at 20-500Hz, the electromyographic signals are subjected to secondary filtering processing through a digital filter, the electromyographic signals within the frequency range of 20-500Hz are intercepted to achieve the purposes of filtering redundant noise, reducing the signal data quantity and improving the processing speed, after the FPGA chip finishes signal filtering, in order to improve the accuracy of gesture classification, action section data needs to be accurately intercepted, and in consideration of the real-time performance of system processing, a standard deviation threshold value detection method is utilized in the FPGA chip to obtain the active section signals:

step 1, intercepting a base signal with the length of L from an electromyographic signal;

step 2, calculating the mean value alpha and the standard deviation beta of the substrate signal:

wherein x (j) is collected electromyographic signal data;

step 3, selecting a sliding window to calculate the standard deviation of the electromyographic signals, setting the window length as W,

wherein u is L + W, L + W +1, …;

step 4, calculating an activity threshold Th:

Th＝θ·β (4)

wherein theta is a magnification factor;

and 5, comparing a threshold value with a standard deviation, setting the electromyographic signal data of the part below the threshold value to zero, calculating the interval length D between the first data of the part above the threshold value and the last data of the part above the threshold value, and considering the signal data of the section as an active section signal when D is more than W, otherwise, discarding the signal data.

After the FPGA acquires the action segment data, the characteristic parameters of the signal data are used as input parameters of the classifier. Because the invention relates to the real-time control of the rehabilitation hand, the invention selectively extracts the time domain characteristics of the electromyographic signals: the method comprises the following steps of calculating an absolute mean value MAV, a root mean square value RMS, a waveform length WL, zero crossing points ZC and a slope sign change SSC according to the following calculation formulas:

(1) absolute mean value MAV:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(2) root mean square value RMS:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(3) waveform length WL:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(4) zero crossing number ZC:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(5) slope sign change SSC:

where N is the data segment length and x (k) is the current data sequence.

In order to reduce the processing time of the system and improve the real-time performance of the system operation, the drive control system selects a simpler BP (Back propagation) neural network, and the BP neural network is sensitive to initial parameters and has low convergence speed. The method optimizes the BP neural network by using a simulated annealing algorithm (SA) with stronger global search capability and a Whale Optimization Algorithm (WOA) with high convergence speed, helps the BP neural network to find proper weight and threshold value, and is used for constructing a good gesture action recognition system.

The working process of the gesture motion recognition system is shown in fig. 9, and the characteristic parameters are recorded as D₀D is₀Divided into training sets D₁Test set D₂The method comprises the following two parts:

step 1, initializing a BP neural network threshold and a connection weight, and establishing an initial model;

step 2, training set data D₁Is imported into the model and is paired with D₁Carrying out normalization;

and 3, optimizing the threshold and the weight of the BP neural network by using an SA-WOA hybrid algorithm:

(1) in the first stage, the weight and the threshold are optimized by using WOA:

a. encircling predation: searching the position of each whale individual, regarding the parameter as a prey, positioning the position of the prey through echo, performing hunting, and updating the position of the whale individual;

b. soaking and net predation: the whale individual contracts and surrounds a prey through a convergence factor, swims to the prey in a spiral motion mode, and continuously updates the position of the whale individual;

c. searching for predation: and (4) carrying out optimizing search on the whale colony according to the coefficient vector parameters and the mutual positions of the individuals to obtain the accurate position of the prey.

(2) In order to prevent the whale optimization algorithm from being trapped in local optimization, parameters obtained by the whale optimization algorithm are further optimized through the simulated annealing algorithm in the second stage:

d. using the optimized parameters obtained by WOA as the current solution S of the simulated annealing algorithm₁Generating a new solution S using the state function₂；

e. Calculating S₁And S₂The increment delta S between the two and whether a new solution S can be accepted or not is judged according to the Metropolis rule₂；

f. Judging whether the obtained result meets the SA termination condition, if so, performing the step 4, otherwise, returning to the part a of the step 3, and performing parameter optimization again;

step 4, taking the optimized parameters obtained in the step 3 as the optimal weight and threshold of the BP neural network;

step 5, retraining the BP neural network, and updating errors until an optimal result is achieved;

step 6, testing data D₂Normalizing, and then bringing the normalized result into a trained neural network, verifying the effectiveness of the classifier, and adding corresponding labels to each gesture action;

and 7, converting different action labels into corresponding control signals according to the result of the step 6, sending the control signals to the mechanical gloves through the Bluetooth module, and driving the fingers to act through the mechanical gloves so as to achieve the aim of rehabilitation training.

4. The bluetooth module in the mechanical glove receives the control signal of drive control system, through each steering wheel drive each finger subassembly action, and then drives the finger and carry out the rehabilitation training.

The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims (7)

1. A hand exoskeleton rehabilitation training device based on surface electromyographic signals comprises mechanical gloves worn on hands and used for driving fingers to move, and is characterized by further comprising a surface electromyographic signal acquisition system and a driving control system used for controlling the mechanical gloves and the surface electromyographic signal acquisition system;

the surface electromyographic signal acquisition system comprises two electromyographic signal acquisition instruments worn on any position of an arm, and each electromyographic signal acquisition instrument transmits an electromyographic signal to the drive control system through a Bluetooth module;

the drive control system comprises an FPGA chip, a first rechargeable lithium battery and a Bluetooth module, wherein the first rechargeable lithium battery supplies power for the FPGA chip and the Bluetooth module, the FPGA chip is connected with a digital filter, a signal input end of the FPGA chip is connected with a signal output end of an electromyographic signal acquisition instrument through the Bluetooth module, the FPGA chip carries out secondary filtering processing on the electromyographic signal after receiving the electromyographic signal of the electromyographic signal acquisition instrument, and the electromyographic signal within the frequency range of 20-500Hz is intercepted; the method comprises the steps of obtaining an active segment signal by utilizing a standard deviation threshold detection method in an FPGA chip, extracting time domain characteristics of the signal, inputting time domain characteristic parameters into an optimized BP neural network, identifying gesture actions of various fingers through the BP neural network, adding labels to the different gesture actions by the FPGA chip, converting surface electromyographic signals into control signals, and sending the control signals to mechanical gloves through a Bluetooth module so as to drive the fingers to move.

2. The surface electromyography-based hand exoskeleton rehabilitation training device of claim 1, wherein: the electromyographic signal acquisition instrument comprises an annular elastic belt, four groups of surface electromyographic sensors, a second rechargeable lithium battery and a Bluetooth module, wherein the second rechargeable lithium battery supplies power for the electromyographic sensors and the Bluetooth module, each group of surface electromyographic sensors comprises two electromyographic sensors, the two electromyographic sensors of each group are symmetrically distributed, eight electromyographic sensors are fixed on the circumference of the inner surface of the annular elastic belt at equal intervals, each electromyographic sensor comprises three electrodes welded on a circuit board in a word shape, the electrode in the middle of the electromyographic sensors serves as a reference electrode, the electrodes on two sides acquire the electromyographic signals in a differential structure, and the circuit board is further welded with a signal amplification circuit and a filter circuit which are used for amplifying and filtering the electromyographic signals.

3. The surface electromyography-based hand exoskeleton rehabilitation training device of claim 1, wherein: the mechanical glove comprises a palm support plate, a thumb assembly, an index finger assembly, a middle finger assembly, a ring finger assembly, a little finger assembly, a box sleeve, a Bluetooth module, a third rechargeable lithium battery and a nylon fastener; the hand-held glove comprises a thumb assembly, an index finger assembly, a middle finger assembly, a ring finger assembly and a little finger assembly, wherein the thumb assembly, the index finger assembly, the middle finger assembly, the ring finger assembly and the little finger assembly correspond to five fingers of a hand of a person respectively;

the thumb component comprises a first steering engine, a first transmission rod and a first connecting rod which drive a first supporting rod and a second supporting rod, and the rear end of the first supporting rod is connected with the upper end face of the palm supporting plate through the component; the front end of the first supporting rod is rotatably connected with the rear end of the second supporting rod through a pin shaft, the palm supporting plate and the second supporting rod are connected through a first connecting rod, the first steering engine is connected with the first supporting rod through a first transmission rod, the first steering engine is fixed on the other end face of the palm supporting plate, the whole action of the thumb component is controlled through the first steering engine, and a first finger ring for fixing the thumb is arranged on the first supporting rod;

the forefinger subassembly includes the second steering wheel, second transfer line and second connecting rod, the third connecting rod drives the third bracing piece, the fourth bracing piece, the fifth bracing piece, the third bracing piece is connected through the up end of subassembly with the palm backup pad, the fourth bracing piece passes through the second connecting rod and the round pin axle is connected with the third bracing piece, the fifth bracing piece passes through the third connecting rod, round pin axle and third bracing piece, the fourth bracing piece is connected, be connected through the second transfer line between second steering wheel and the third bracing piece, the second steering wheel is fixed in the backup pad, action through second steering wheel control forefinger subassembly, third bracing piece middle part is equipped with the second ring, fourth bracing piece middle part is equipped with the third ring.

4. A hand exoskeleton rehabilitation training method based on surface electromyogram signals is characterized by comprising the following steps:

s1, wearing the two electromyographic signal acquisition instruments on the arm in a staggered mode, wherein the electromyographic signal acquisition instrument close to the palm is called a near-end acquisition instrument, and the electromyographic signal acquisition instrument far away from the palm is called a far-end acquisition instrument; after the near-end acquisition instrument is worn, the far-end acquisition instrument is worn on the arm in a staggered mode; the method comprises the steps that surface electromyographic signals generated by actions are collected through an electromyographic sensor, the electromyographic signals are preprocessed through a signal amplification circuit and a filter circuit, the surface electromyographic signals are amplified through the signal amplification circuit, 50Hz power frequency interference is filtered through the filter circuit, and then the signals are transmitted to an FPGA chip of a drive control system through a Bluetooth module;

s2, wearing the mechanical gloves on the hands;

s3, after receiving the preprocessed electromyographic signals, the FPGA chip carries out secondary filtering processing on the electromyographic signals through a digital filter, intercepts the electromyographic signals within the frequency range of 20-500Hz, acquires active segment signals by using a standard deviation threshold detection method in the FPGA chip, extracts time domain characteristics of the signals, inputs time domain characteristic parameters into the optimized BP neural network as input parameters of a classifier of the BP neural network, identifies gesture actions of various fingers through the BP neural network, adds labels to the different gesture actions by the FPGA chip, converts surface electromyographic signals into control signals, and sends the control signals to the mechanical gloves through a Bluetooth module so as to drive the fingers to move;

s4, after receiving a control signal of the drive control system, the Bluetooth module in the mechanical glove drives each finger assembly to act through each steering engine, and then drives the fingers to perform rehabilitation training.

5. The surface electromyography-based hand exoskeleton rehabilitation training device of claim 4, wherein: the specific steps of acquiring the active segment signal by using the standard deviation threshold detection method in step S3 are as follows:

step 1, intercepting a base signal with the length of L from an electromyographic signal;

step 2, calculating the mean value alpha and the standard deviation beta of the substrate signal:

wherein x (j) is collected electromyographic signal data;

step 3, selecting a sliding window to calculate the standard deviation of the electromyographic signals, setting the window length as W,

wherein u is L + W, L + W +1, …;

step 4, calculating an activity threshold Th:

Th＝θ·β (4)

wherein theta is a magnification factor;

and 5, comparing a threshold value with a standard deviation, setting the electromyographic signal data of the part below the threshold value to zero, calculating the interval length D between the first data of the part above the threshold value and the last data of the part above the threshold value, and considering the signal data of the section as an active section signal when D is more than W, otherwise, discarding the signal data.

6. The surface electromyography-based hand exoskeleton rehabilitation training device of claim 4, wherein: the time domain characteristic parameters in the step S3 include an absolute mean value MAV, a root mean square value RMS, a waveform length WL, zero crossing points ZC, and a slope sign change SSC, and their calculation formulas are respectively:

(1) absolute mean value MAV:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(2) root mean square value RMS:

wherein, N is the length of the data segment, and x (i) is the current data sequence;

(3) waveform length WL:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(4) zero crossing number ZC:

wherein, N is the length of the data segment, and x (t) is the current data sequence;

(5) slope sign change SSC:

where N is the data segment length and x (k) is the current data sequence.

7. The surface electromyography signal-based hand exoskeleton rehabilitation training method of claim 4, wherein: in the step S3, a simulated annealing algorithm and a whale optimization algorithm with high convergence speed are adopted to optimize the BP neural network, and characteristic parameters are recorded as D₀D is₀Divided into training sets D₁Test set D₂The method comprises the following two steps:

step 1, initializing a BP neural network threshold and a connection weight, and establishing an initial model;

step 2, training set data D₁Is imported into the model and is paired with D₁Carrying out normalization;

and 3, optimizing the threshold and the weight of the BP neural network by using an SA-WOA hybrid algorithm:

(1) in the first stage, the weight and the threshold are optimized by using WOA:

a. encircling predation: searching the position of each whale individual, regarding the parameter as a prey, positioning the position of the prey through echo, performing hunting, and updating the position of the whale individual;

b. soaking and net predation: the whale individual contracts and surrounds a prey through a convergence factor, swims to the prey in a spiral motion mode, and continuously updates the position of the whale individual;

c. searching for predation: the whale colony carries out optimizing search according to the coefficient vector parameters and the mutual positions of all individuals to obtain the accurate position of a prey;

(2) in order to prevent the whale optimization algorithm from being trapped in local optimization, parameters obtained by the whale optimization algorithm are further optimized through the simulated annealing algorithm in the second stage:

d. using the optimized parameters obtained by WOA as the current solution S of the simulated annealing algorithm₁Generating a new solution S using the state function₂；

e. Calculating S₁And S₂The increment delta S between the two and whether a new solution S can be accepted or not is judged according to the Metropolis rule₂；

f. Judging whether the obtained result meets the SA termination condition, if so, performing the step 4, otherwise, returning to the part a of the step 3, and performing parameter optimization again;

step 4, taking the optimized parameters obtained in the step 3 as the optimal weight and threshold of the BP neural network;

step 5, retraining the BP neural network, and updating errors until an optimal result is achieved;

step 6, testing data D₂Normalizing, and then bringing the normalized result into a trained neural network, verifying the effectiveness of the classifier, and adding corresponding labels to each gesture action;

and 7, converting different action labels into corresponding control signals according to the result of the step 6, sending the control signals to the mechanical gloves through the Bluetooth module, and driving the fingers to act through the mechanical gloves so as to achieve the aim of rehabilitation training.

Images