Surface electromyogram image-based multi-muscle synergistic abnormal contraction pattern detection method

Abstract

The present application relates to the technical field of electromyography processing, and in particular to a method for detecting abnormal muscle contraction mode based on surface electromyography image. According to the characteristics of the intensity and position of different pixel points in the electromyography image, and the equivalent impedance of subcutaneous fat of each electrode at the latest time, the present application obtains the electromyography compensation intensity of each pixel point. According to the three-dimensional instantaneous torque vector at the latest time, the present application dilates the initial reference electromyography image to obtain a dynamic dilated image. In combination with the distribution characteristics of different pixel points in the electromyography image, the preset initial dynamic attenuation factor and the frequency domain characteristics of the electromyography voltage amplitude sequence, the present application obtains the electromyography correction intensity of each pixel point. According to the difference between the electromyography compensation intensity and the electromyography correction intensity of different pixel points of the antagonistic muscle area and the active muscle area, the present application detects the abnormal muscle contraction mode. The present application improves the accuracy of the abnormal muscle contraction mode detection by dynamically eliminating the influence of multi-dimensional physiological interference factors.

Description

Technical Field

[0001] This invention relates to the field of electromyography (EMG) processing technology, and specifically to a method for detecting multi-muscle coordinated abnormal contraction patterns based on surface EMG images. Background Technology

[0002] Complex movements of the human body are accomplished by the coordinated activation and inhibition of multiple muscles in time and space. By acquiring surface electromyography images through a high-density electrode array attached to the body surface, the spatial topology and temporal evolution characteristics of the motor drive commands issued by the central nervous system in local muscle groups can be intuitively reflected.

[0003] In existing technologies, abnormal contraction patterns are detected based on electromyography (EMG) images, relying on a single, fixed benchmark. However, in actual rehabilitation training, EMG signals are affected by various normal physiological changes. Different patient body shapes result in blurred and weak electrical signals detected through subcutaneous fat; different levels of force exerted by patients lead to an expansion of the normal muscle contraction range; and prolonged training causes muscle fatigue, which can cause normal distortion of the signal waveform. Therefore, existing technologies cannot eliminate complex and dynamic normal physiological changes, and are prone to misjudging normal compensatory force exertion or fatigue as abnormal contraction, resulting in low accuracy in detecting abnormal contraction patterns. Summary of the Invention

[0004] To address the problem that existing technologies cannot eliminate complex and dynamic normal physiological changes, and are prone to misjudging normal compensatory exertion or fatigue as abnormal contraction, the present invention aims to provide a method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography (EMG) images. The specific technical solution adopted is as follows: This invention proposes a method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography (EMG) images, the method comprising: The electromyographic voltage amplitude sequence of each electrode in the target muscle region during a single movement is obtained to obtain the corresponding electromyographic image; the subcutaneous fat equivalent impedance and three-dimensional instantaneous torque vector of each electrode at different times are obtained; the target muscle region includes antagonistic muscle regions and agonist muscle regions; Based on the electromyographic intensity and location characteristics of different pixels in the electromyographic image, the muscle activation center point and the lateral diffusion penalty factor of each pixel are obtained; based on the subcutaneous fat equivalent impedance of each electrode at the latest time, the lateral diffusion penalty factor of each pixel and the electromyographic intensity, the electromyographic compensation intensity of each pixel is obtained. Based on the latest three-dimensional instantaneous torque vector, the initial baseline electromyography (EMG) image is dilated to obtain a dynamic dilated image. Based on the latest three-dimensional instantaneous torque vector, the distribution characteristics of different pixels in the EMG image relative to the muscle activation center point, and the preset initial dynamic attenuation factor, the spatial orientation weight of each pixel is obtained. Based on the spatial orientation weight of each pixel and the frequency domain characteristics of the EMG voltage amplitude sequence at the muscle activation center point, the EMG intensity of each pixel in the dynamic dilated image is adjusted to obtain the corrected EMG intensity of each pixel. Based on the differences in electromyographic compensation intensity and electromyographic correction intensity between antagonist and agonist muscle regions at different pixels, abnormal feature classification coefficients are obtained to detect abnormal contraction patterns.

[0005] Furthermore, the method for obtaining the muscle activation center point includes: Select the pixel with the maximum electromyographic intensity value in the electromyographic image, and use the average coordinate of the corresponding pixel as the muscle activation center point; Obtain the relative distance between each pixel and the muscle activation center point, and select the minimum value between the relative distance and the preset distance threshold as the diffusion distance; obtain the product between the preset diffusion constraint coefficient and the diffusion distance, and perform exponential mapping as the lateral diffusion penalty multiple for each pixel.

[0006] Furthermore, the method for obtaining the electromyographic compensation intensity includes: An interpolation algorithm was used to map the subcutaneous fat equivalent impedance of each electrode at the latest time point into a two-dimensional impedance distribution map aligned with the electromyography image, so as to obtain the impedance mapping value of each pixel. The electromyographic compensation intensity of each pixel is obtained based on the impedance mapping value, lateral diffusion penalty factor, and electromyographic intensity. The impedance mapping value, lateral diffusion penalty factor, and electromyographic intensity are all positively correlated with the electromyographic compensation intensity.

[0007] Furthermore, the method for obtaining the dynamically dilated image includes: The dynamic expansion radius is obtained based on the three-dimensional instantaneous torque vector at the latest moment; Two-dimensional morphological dilation processing is performed on the initial baseline electromyography image based on the dynamic dilation radius to obtain a dynamically dilated image.

[0008] Furthermore, the method for obtaining the dynamic expansion radius includes: Obtain the magnitude of the three-dimensional instantaneous torque vector at the latest moment and normalize it to obtain the force load rate; Obtain the first product between the power load rate and the preset recruitment expansion coefficient; obtain the difference between the power load rate and the preset co-contraction activation threshold, select the maximum value between the difference and the preset activation lower limit, and calculate the second product between the square of the maximum value and the preset co-contraction compensation expansion coefficient; The sum of the preset safety tolerance radius, the first product, and the second product is obtained and used as the dynamic expansion radius.

[0009] Furthermore, the method for obtaining the spatial orientation weights includes: Obtain the difference between the coordinates of each pixel and the muscle activation center point as a difference vector; obtain the cosine of the spatial angle between the difference vector and the three-dimensional instantaneous torque vector after projection onto the electromyography image plane; A negative correlation mapping is performed on the power load rate, and the product between the negative correlation mapping result and the preset initial dynamic attenuation factor is used as the power load dynamic attenuation factor. The spatial orientation weight of each pixel is obtained based on the dynamic attenuation factor of the applied load, the difference between the cosine value of the spatial angle and the preset direction containment cosine threshold. The dynamic attenuation factor of the applied load and the difference are both positively correlated with the spatial orientation weight.

[0010] Furthermore, the method for obtaining the electromyographic correction intensity includes: The fatigue coefficient is obtained based on the frequency domain characteristics of the electromyographic voltage amplitude sequence at the muscle activation center point. The corrected electromyography (EMG) intensity of each pixel is obtained based on the fatigue coefficient, the spatial orientation weight of each pixel, and the EMG intensity of the pixels in the dynamically dilated image. Further, the method for obtaining the fatigue coefficient includes: A short-time Fourier transform is performed on the electromyographic voltage amplitude sequence at the muscle activation center point within the neighborhood of the latest time to obtain the transformed short-time median frequency. Obtain the difference between the reference median frequency and the short-time median frequency, calculate the ratio of the difference result to the reference median frequency, and select the maximum value between the ratio result and the preset fatigue lower limit as the fatigue coefficient.

[0011] Furthermore, the method for obtaining the anomaly feature classification coefficient includes: The absolute value of the difference between the electromyography compensation intensity and the electromyography correction intensity of each pixel is obtained as the residual value of each pixel. Based on the ratio of the residual values ​​of pixels in the antagonistic muscle region to the electromyographic correction intensity of pixels in the relative agonistic muscle region, the abnormal feature classification coefficient is obtained, and the residual values ​​are positively correlated with the abnormal feature classification coefficient.

[0012] Furthermore, the method for detecting abnormal contraction patterns includes: The sum of the residual values ​​of all pixels is obtained as the global error. If the global error is less than or equal to the preset safety threshold, it is determined to be the expected normal shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is greater than or equal to the preset classification threshold, it is determined to be the co-shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is less than the preset classification threshold, it is determined to be the inefficient recruitment mode.

[0013] The present invention has the following beneficial effects: This invention obtains the muscle activation center point and the lateral diffusion penalty multiple of each pixel based on the electromyographic intensity and location characteristics of different pixels in the electromyographic image, reflecting the degree of lateral scattering by fat at the pixel's location; it obtains the electromyographic compensation intensity of each pixel based on the subcutaneous fat equivalent impedance of each electrode at the latest time, the lateral diffusion penalty multiple of each pixel, and the electromyographic intensity, reflecting the true muscle force intensity after eliminating the influence of uneven subcutaneous fat thickness; it dilates the initial baseline electromyographic image based on the three-dimensional instantaneous torque vector at the latest time to obtain a dynamically dilated image, effectively simulating the biomechanical phenomenon of peripheral muscles being forcibly awakened and compensated during heavy force exertion, eliminating false positives when force increases; and it also obtains the three-dimensional instantaneous torque vector at the latest time. The invention utilizes the distribution characteristics of different pixels relative to the muscle activation center point in electromyography (EMG) images and a preset initial dynamic attenuation factor to obtain the spatial orientation weight of each pixel. This weight reflects the degree of nonlinear interception and filtering of interfering electrical signals in non-force-generating directions. Based on the spatial orientation weight of each pixel and the frequency domain characteristics of the EMG voltage amplitude sequence at the muscle activation center point, the EMG intensity of each pixel in the dynamic dilation image is adjusted to obtain the corrected EMG intensity of each pixel. This corrected intensity reflects the theoretical muscle force intensity calculated through dilation based on torque magnitude and fatigue level. Based on the differences in EMG compensation intensity and corrected EMG intensity between antagonist and agonist muscle regions at different pixels, an abnormal feature classification coefficient is obtained to detect abnormal contraction patterns. This invention improves the accuracy of abnormal contraction pattern detection by dynamically eliminating the influence of multidimensional physiological interference factors to obtain the abnormal feature classification coefficient. Attached Figure Description

[0014] Figure 1 A flowchart of a method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images, provided as an embodiment of the present invention; Figure 2 A flowchart illustrating a method for obtaining a dynamic expansion radius according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating a method for obtaining spatial orientation weights according to an embodiment of the present invention. Detailed Implementation

[0015] The following description, in conjunction with the accompanying drawings, details the specific scheme of the multi-muscle coordinated abnormal contraction pattern detection method based on surface electromyography images provided by the present invention.

[0016] Please see Figure 1 The diagram illustrates a flowchart of a method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images, according to an embodiment of the present invention. The specific method includes: Step S1: Obtain the electromyographic voltage amplitude sequence of each electrode in the target muscle region during one movement, and obtain the corresponding electromyographic image; obtain the subcutaneous fat equivalent impedance and three-dimensional instantaneous torque vector of each electrode at different times; the target muscle region includes antagonistic muscle regions and agonist muscle regions.

[0017] In an embodiment of the present invention, the skin of the target muscle area of ​​the subject is pretreated, including hair removal, wiping with medical alcohol, and applying an appropriate amount of conductive ointment, in order to minimize the contact resistance between the skin and the electrodes; then, a high-density flexible surface electromyography electrode array is tightly attached to the muscle belly of the target muscle group, and tendons and movement points need to be avoided during the operation.

[0018] An electrode array captures the weak action potentials of motor units conducted along muscle fibers to the skin surface during muscle contraction. These weak analog electrical signals are amplified by a differential amplifier circuit in the front-end acquisition hardware, and then filtered by a bandpass filter and a power frequency notch filter to remove baseline drift and environmental noise interference. Finally, they are digitized by an analog-to-digital converter (ADC) to form frames of raw electromyographic (EMG) image pixel data. The electrodes are attached to the skin in a grid array, creating a uniform two-dimensional matrix. At the same sampling time, the voltage amplitude measured at each physical electrode is filled into the corresponding coordinate cell in the matrix as the pixel data for that location. Each pixel corresponds to the voltage amplitude of one electrode at a given sampling time, and the data from all electrodes at the same time constitute a frame of pixel data. The EMG voltage amplitude sequence of each electrode during a single movement is processed, and the root mean square eigenvalue is calculated as the EMG intensity of the corresponding pixel, generating an EMG image.

[0019] Based on the miniature impedance detection circuit integrated on the electrode side, the equivalent impedance of subcutaneous fat at each electrode is read; based on the torque sensor at the external device interaction terminal, the three-dimensional instantaneous torque vector is obtained.

[0020] It should be noted that, in the embodiments of the present invention, the voltage amplitude is sampled once every 1 millisecond; the equivalent impedance and the three-dimensional instantaneous torque vector are sampled once every 10 milliseconds; the antagonist muscle region and the agonist muscle region are two pre-calibrated anatomical regions, the antagonist muscle region reflects the position of the muscle that is theoretically inhibited during the movement; the agonist muscle region reflects the position of the muscle that exerts force.

[0021] Step S2: Based on the electromyographic intensity and location characteristics of different pixels in the electromyographic image, obtain the muscle activation center point and the lateral diffusion penalty factor of each pixel; based on the subcutaneous fat equivalent impedance of each electrode at the latest time, the lateral diffusion penalty factor of each pixel, and the electromyographic intensity, obtain the electromyographic compensation intensity of each pixel.

[0022] In actual monitoring scenarios, the subcutaneous fat layer of the human body has a strong damping and attenuation effect on electrical signals. The fat thickness varies among subjects of different weights or in different parts of the same subject, resulting in local dimness and spatial diffusion in the images acquired by the electrodes. According to the propagation and diffusion law of electromagnetic fields in biological tissues, the farther away from the center of muscle exertion, the greater the degree of blurring and the greater the diffusion penalty. Therefore, based on the electromyographic intensity and positional characteristics of different pixels in the electromyographic image, the muscle activation center point and the lateral diffusion penalty multiple of each pixel can be obtained.

[0023] Preferably, in one embodiment of the present invention, the method for obtaining the muscle activation center point and the lateral diffusion penalty multiple includes: Select the pixel with the maximum electromyographic intensity value in the electromyographic image, and use the average coordinate of the corresponding pixel as the muscle activation center point; It should be noted that the area with the highest electromyographic intensity in the electromyographic image represents the physical core location below the electrode where the motor unit discharge is most concentrated and the muscle contraction is most intense, and therefore serves as the muscle activation center point.

[0024] Obtain the relative distance between each pixel and the muscle activation center point, and select the minimum value between the relative distance and the preset distance threshold as the diffusion distance; obtain the product between the preset diffusion constraint coefficient and the diffusion distance, and perform exponential mapping as the lateral diffusion penalty multiple for each pixel.

[0025] It should be noted that subcutaneous fat in the human body is a high-impedance medium. When the action potential generated by the muscle penetrates the fat layer and diffuses laterally to the surrounding area, the greater the relative distance from the muscle activation center, the more the lateral scattering and blurring by the fat intensifies exponentially. Therefore, an exponential function with the natural constant as the base is used for exponential mapping. The greater the distance, the greater the required diffusion penalty factor.

[0026] In an embodiment of the present invention, the formula for the lateral dispersion penalty factor for any pixel is expressed as: ;in, This represents the lateral dispersion penalty factor for a pixel. This indicates the preset dispersion constraint coefficient; Indicates the diffusion distance of a pixel; Represents the natural constant.

[0027] It should be noted that, in the embodiments of the present invention, relative distances are calculated using Euclidean distance or Manhattan distance, etc.; the preset diffusion constraint coefficient characterizes the spatial attenuation characteristics under the volume conduction effect, that is, the spatial attenuation coefficient of electromyographic signals undergoing transverse scattering and attenuation in biological tissues such as subcutaneous fat, and its dimension is the reciprocal of the diffusion distance unit, which is obtained by experimental calibration of the dielectric properties of human tissues.

[0028] Considering the limitations of physical electrical signal transmission, in order to avoid receiving mostly environmental noise or crosstalk from distant, unrelated muscles due to unrestricted transmission, the method for obtaining the preset distance threshold is as follows: In a benchmark test where the subject contracts the target muscle alone, the physical distance corresponding to the decay of the electromyographic amplitude from the activation center outward to equal the average level of the environmental noise is measured, and this distance is converted into pixel distance in the image as the preset distance threshold; the specific means are well known to those skilled in the art and will not be described in detail here.

[0029] As an example, a subject is performing motion monitoring of lifting heavy objects with their upper limbs, wearing a high-density flexible surface electromyography (EMG) electrode array. The subject's physiological characteristics include a thick subcutaneous fat layer, causing localized dimness and spatial diffusion in the images acquired by the electrodes. A diffusion constraint coefficient, calibrated experimentally, is preset. In the electromyography image, select an edge pixel that is far from the center of muscle activation and measure its diffusion distance. Calculations yielded Therefore, at a physical location 10 pixels away from the center of electromyographic activation, the actual electrical activity energy is diluted by nearly 2.7 times on the image due to the scattering effect of fat. The control system must apply a penalty amplification of 2.718 times at this location to trace back to its original intensity.

[0030] Since subcutaneous fat not only causes lateral spatial diffusion but also reduces the absolute amplitude of vertical depth, the electromyography intensity is compensated based on the subcutaneous fat equivalent impedance and the lateral diffusion penalty factor. The electromyography compensation intensity of each pixel is obtained based on the subcutaneous fat equivalent impedance of each electrode at the latest time, the lateral diffusion penalty factor of each pixel, and the electromyography intensity.

[0031] Preferably, in one embodiment of the present invention, the method for obtaining the electromyographic compensation intensity includes: The first step is to use an interpolation algorithm to map the subcutaneous fat equivalent impedance of each electrode at the latest time point into a two-dimensional impedance distribution map aligned with the electromyography image, thereby obtaining the impedance mapping value of each pixel. It should be noted that, in the embodiments of the present invention, since the subcutaneous fat equivalent impedance data is spatially discrete, while the electromyography (EMG) image is a high-resolution continuous pixel matrix, the two cannot be directly correlated in terms of spatial resolution. In order to achieve accurate compensation pixel by pixel, a two-dimensional bilinear interpolation algorithm is used to spatially smooth the subcutaneous fat equivalent impedance of each electrode to generate a continuous two-dimensional impedance distribution map that is perfectly aligned with the resolution of the EMG image. The specific means are well known to those skilled in the art and will not be described in detail here.

[0032] The second step is to obtain the electromyographic compensation intensity of each pixel based on the impedance mapping value, the lateral diffusion penalty factor, and the electromyographic intensity. The impedance mapping value, the lateral diffusion penalty factor, and the electromyographic intensity are all positively correlated with the electromyographic compensation intensity.

[0033] It should be noted that the impedance mapping value reflects the degree of absolute amplitude attenuation of the electrical signal in the vertical direction caused by the fat beneath the electrode. The larger the impedance mapping value, the more severe the vertical attenuation. The lateral diffusion penalty factor reflects the volumetric conductive scattering effect of the signal within biological tissue. The electrical signal not only attenuates vertically but also diffuses in all directions. The farther the pixel is from the activation center, the higher the degree of lateral scattering and blurring of the signal. That is, the larger the impedance mapping value and the larger the lateral diffusion penalty factor, the more severe the lateral diffusion and vertical attenuation, the greater the need for compensation of electromyographic intensity, and the greater the electromyographic compensation intensity. Therefore, the impedance mapping value, the lateral diffusion penalty factor, and the electromyographic intensity are all positively correlated with the electromyographic compensation intensity.

[0034] In one embodiment of the present invention, the product of the impedance mapping value, the lateral diffusion penalty factor and the preset vertical compensation constant of each pixel is obtained as the reverse amplification weight of each pixel; the sum of constant 1 and the reverse amplification weight of each pixel is calculated as the reverse amplification weight of each pixel; the product of the reverse amplification weight of each pixel and the electromyographic intensity is obtained as the electromyographic compensation intensity of each pixel.

[0035] It should be noted that, in the embodiments of the present invention, the attenuation of weak electrical signals by subcutaneous fat is a physical blocking effect. In order to linearly map the impedance mapping value to the amplification ratio of image pixel brightness and ensure that the compensation amplitude conforms to the actual physiological electrical attenuation law, a preset vertical compensation constant is introduced for physical mapping. The preset vertical compensation constant is a static system parameter obtained in advance from offline clinical calibration experiments. The value range is set to 0.01-0.05 based on historical experience, and the dimension is the reciprocal of the dimension of the impedance mapping value. It will not be elaborated here.

[0036] For any pixel, the formula for the electromyographic compensation intensity is: ; ;in, Indicates the electromyographic compensation intensity of a pixel; Indicates the electromyographic intensity of a pixel; This represents the inverse amplification weighting of a pixel; This indicates the preset vertical compensation constant; This represents the impedance mapping value of a pixel; This represents the lateral dispersion penalty factor for a pixel. This represents the reverse amplification weight of a pixel.

[0037] Here's an example with a preset vertical compensation constant. Obtain the impedance mapping value below the electrode. The previous step yielded Acquire the weak raw electromyographic intensity measured at each pixel. Calculate the weighted quantity Calculate the electromyography compensation intensity of each pixel. The system combines vertical reduction compensation and lateral diffusion compensation in series to convert the observed values ​​into a single unit. Weak signals are restored to their original form through nonlinear mapping. This ensures that even patients with thick subcutaneous fat and weak electrical signals can accurately capture the true intensity of muscle exertion, preventing missed detections due to dim signals.

[0038] Step S3: Based on the three-dimensional instantaneous torque vector at the latest moment, dilate the initial baseline electromyography (EMG) image to obtain a dynamic dilated image; based on the three-dimensional instantaneous torque vector at the latest moment, the distribution characteristics of different pixels in the EMG image relative to the muscle activation center point, and the preset initial dynamic attenuation factor, obtain the spatial orientation weight of each pixel; based on the spatial orientation weight of each pixel and the frequency domain characteristics of the EMG voltage amplitude sequence at the muscle activation center point, adjust the EMG intensity of each pixel in the dynamic dilated image to obtain the corrected EMG intensity of each pixel.

[0039] In extreme load scenarios such as lifting heavy objects, in order to maintain the torque balance of the joints under extremely high tension, the central nervous system must force peripheral unrelated auxiliary muscles to participate in co-contraction. If the smaller graphic boundary during relaxation is used for comparison, the heavy load force will inevitably produce a large area of ​​false positives. Therefore, based on the three-dimensional instantaneous torque vector at the latest moment, the initial reference electromyography image is expanded to obtain a dynamic expanded image.

[0040] Preferably, in one embodiment of the present invention, the method for obtaining a dynamically dilated image includes: The first step is to obtain the dynamic expansion radius based on the three-dimensional instantaneous torque vector at the latest moment; Preferably, in one embodiment of the present invention, the method for obtaining the dynamic expansion radius is described in [reference needed]. Figure 2 It shows a flowchart of a method for obtaining the dynamic expansion radius, including: Step S201: Obtain the magnitude of the three-dimensional instantaneous torque vector at the latest moment and normalize it as the force load rate.

[0041] It should be noted that the magnitude of the three-dimensional instantaneous torque vector reflects the absolute intensity of the subject's instantaneous force exertion, i.e., the total mechanical work load. In the embodiments of the present invention, the subject's limiting torque benchmark is obtained in advance based on the standard maximum voluntary contraction test, and the ratio of the magnitude of the three-dimensional instantaneous torque vector to the limiting torque benchmark is obtained, i.e., normalized, as the force load rate, reflecting the percentage of the subject's force exertion to their maximum force exertion capacity, and is used to quantify the relative heavy load of the exercise task.

[0042] Step S202: Obtain the first product between the load factor and the preset fundraising expansion coefficient; Obtain the difference between the load rate and the preset co-contraction activation threshold, select the maximum value between the difference and the preset activation lower limit, and calculate the second product between the square of the maximum value and the preset co-contraction compensation expansion coefficient.

[0043] It should be noted that when the force load rate is less than or equal to the preset co-contraction activation threshold, the force exertion is in the low to medium load stage. At this time, the central nervous system can maintain torque by increasing the recruitment frequency within the target muscle group, and the expansion radius expands outward in a gentle linear manner. When the force load rate is greater than the preset co-contraction activation threshold, the system enters the extreme heavy load force exertion scenario. The single agonist muscle reaches the force exertion bottleneck, and the central nervous system is forcibly awakened to participate in compensation with the surrounding resting muscles, resulting in a rapid increase in the size of the expanded pixel radius. In one embodiment of the present invention, the preset activation lower limit is set to 0.

[0044] It should be noted that, in the embodiments of the present invention, the preset cocontraction activation threshold is set to 60%-70% of the maximum force load rate of the subject; the preset cocontraction compensation expansion coefficient reflects the nonlinear and severe expansion amplification factor caused by the compensation effect under extreme heavy load scenarios, and the preset cocontraction compensation expansion coefficient is set to 20 pixels based on relevant experience; the preset recruitment expansion coefficient reflects the linear recruitment expansion coefficient of the motor units inside the target muscle, and its dimension is pixels, representing the linear expansion amount brought about by the load reaching 100% without compensation. Usually, the muscle activation area increases slowly under medium and low loads, and the value is set to 5 pixels based on relevant historical experience; these are not limited or elaborated here.

[0045] Step S203: Obtain the sum of the preset safety tolerance radius, the first product, and the second product, as the dynamic expansion radius.

[0046] It should be noted that, in the embodiments of the present invention, when the subject is in a resting state, the system collects the envelope radius of the activated region in the surface electromyography image and uses it as the preset safety tolerance radius.

[0047] In one embodiment of the present invention, the formula for the dynamic expansion radius is expressed as: ;in, Indicates the dynamic expansion radius; Indicates the preset safety tolerance radius; This indicates the preset fundraising inflation factor; Indicates the load factor; Indicates the preset co-shrinkage compensatory expansion coefficient; Indicates the preset co-contraction activation threshold; Indicates the preset activation lower limit; This represents the function that takes the maximum value.

[0048] Here's an example, with a preset static safety tolerance radius. Preset fundraising inflation coefficient Preset co-contraction compensating expansion coefficient Preset co-contraction activation threshold The current subjects are at 80% of their maximum workload, that is... Calculate the linear fundraising term ; Calculate the nonlinear compensation term: due to There is an overflow. , The final dynamic expansion radius is... When the exertion load is at a low to medium level, i.e., the exertion load rate is less than 65%, the muscle activation area increases only gradually and linearly with the load; when the exertion load rate exceeds the 65% heavy load boundary, The square mechanism is activated, simulating the nonlinear physiological phenomenon of peripheral muscle groups being violently and forcibly awakened, which enables the algorithm to include the legitimate high-load compensation area and avoid system false alarms.

[0049] The second step involves performing two-dimensional morphological dilation processing on the initial baseline electromyography image based on the dynamic dilation radius to obtain a dynamically dilated image.

[0050] It should be noted that the initial baseline electromyography (EMG) image is a standard EMG template taken when the target muscle contracts correctly under conditions of no fatigue and no compensatory crosstalk in the subject. During the acquisition of the dynamic dilation image, the spatial coordinates of each pixel remain fixed; what changes is the EMG intensity value corresponding to each pixel. Specifically, during the dilation process, an EMG intensity update is performed using a two-dimensional spatial convolution smoothing diffusion algorithm based on a Gaussian kernel. That is, a corresponding two-dimensional Gaussian kernel matrix is ​​constructed based on the dynamic dilation radius, where the window size of the Gaussian kernel is set to twice the dynamic dilation radius plus one, and the standard deviation is proportional to the dynamic dilation radius. The expansion radius is proportionally mapped; a two-dimensional Gaussian kernel matrix is ​​used to perform two-dimensional spatial convolution on the initial reference electromyography image to obtain a smooth and divergent transition image; finally, the transition image is compared pixel by pixel with the initial reference electromyography image, and the maximum value of the electromyography intensity of the two is selected as the updated electromyography intensity value of the corresponding pixel. This makes the electromyography intensity values ​​of adjacent pixels that were originally located on the periphery of the force center and had low electromyography intensity show a smooth increase in accordance with the law of bioelectrical decay, thereby effectively simulating the biomechanical phenomenon that the surrounding muscles are forcibly awakened to compensate during heavy force exertion, and eliminating false positives when the force increases.

[0051] In real human biomechanics, the compensatory calls of the central nervous system to muscles are highly directional, and only synergistic muscle groups along the direction of force application are activated. Furthermore, under heavy load, since a single muscle cannot maintain torque, the central nervous system will produce a spillover effect, resulting in decreased control precision, and widespread compensatory activation of muscle groups in different directions. Therefore, by analyzing the torque vector and the force application direction of pixels, and based on the latest three-dimensional instantaneous torque vector, the distribution characteristics of different pixels relative to the muscle activation center point in the electromyography image, and the preset initial dynamic decay factor, the spatial orientation weight of each pixel is obtained.

[0052] Preferably, in one embodiment of the present invention, the method for obtaining spatial orientation weights is described in [reference needed]. Figure 3 It shows a flowchart of a method for obtaining spatial orientation weights, including: Step S301: Obtain the difference between the coordinates of each pixel and the muscle activation center point as the difference vector; obtain the cosine value of the spatial angle between the difference vector and the three-dimensional instantaneous torque vector after projection onto the electromyography image plane.

[0053] It should be noted that, based on the accelerometer and gyroscope data inside the inertial measurement module of the micro-unit synchronously configured at the flexible surface electromyography electrode array, the real-time rotation transformation matrix of the electromyography image plane in the external coordinate system is obtained through the attitude fusion filtering algorithm. The three-dimensional instantaneous torque vector in the external coordinate system is transformed into the local three-dimensional coordinate system of the electromyography image plane, and its two-dimensional coordinate components located in the electromyography image plane are extracted as projection vectors. Then, the cosine value of the spatial angle between the difference vector and the projection vector is calculated. The cosine value of the spatial angle reflects whether the position of the pixel in the muscle tissue is in line with the actual force direction of the human body. The larger the cosine value, the smaller the spatial angle, and the closer it is to the main force direction. The smaller the cosine value, the larger the spatial angle, and the further it is from the force direction. Based on this, considering that when the pixel is the center point of muscle activation, the difference vector is a zero vector, but since the center point of muscle activation is the strongest and most core force position, its spatial angle cosine value is directly set to 1.

[0054] Step S302: Perform negative correlation mapping on the power load rate, and obtain the product between the negative correlation mapping result and the preset initial dynamic attenuation factor, which is used as the dynamic attenuation factor of the power load.

[0055] It should be noted that under low load, muscle control precision is very high, and electromyographic signals that deviate slightly from the direction will be attenuated and filtered out, resulting in a larger dynamic attenuation factor for force load. As the load increases, muscle control precision decreases, and muscle groups in different directions around the muscle are widely compensated and awakened, resulting in a smaller dynamic attenuation factor for force load. In the embodiments of the present invention, an exponential function with a natural constant as the base is used for negative correlation mapping. The higher the force load rate, the lower the muscle group control precision, the greater the tolerance for directional deviation, and the smaller the dynamic attenuation factor for force load. The specific means are well known to those skilled in the art and will not be described in detail here.

[0056] It should be noted that, in the embodiments of the present invention, the preset initial dynamic decay factor represents the empirical sensitivity coefficient of the subject's muscle control direction specificity under low load conditions. Low load conditions are those in which the instantaneous force exertion is at 5% to 10% of the subject's limit torque baseline. The preset initial dynamic decay factor can be obtained by fitting the spatial decay gradient of electromyographic signals from the force exertion center to the non-force exertion area in a large number of healthy people under low load conditions, with a reference value range of 10 to 20.

[0057] Step S303: Based on the dynamic attenuation factor of the applied load, and the difference between the cosine value of the spatial angle and the preset direction containment cosine threshold, obtain the spatial orientation weight of each pixel. The dynamic attenuation factor of the applied load and the difference are positively correlated with the spatial orientation weight.

[0058] It should be noted that the dynamic attenuation factor of force load reflects the dynamic tolerance of the central nervous system control precision. The larger the value, the more stringent the directional filtering is. The preset directional containment cosine threshold is a preset containment angle cosine threshold based on the myofascial orientation of the target muscle. The containment angle reflects the boundary of the legal fan-shaped region where muscle electrical activity is allowed to reasonably diffuse in space during normal human force exertion. The difference between the spatial angle cosine value and the preset directional containment cosine threshold reflects the anatomical legality of the muscle activation direction. The larger the difference, the closer the electrical activity is to the expected myofascial force exertion sector. The larger the dynamic attenuation factor of force load, the amplified positive and negative effects of the difference are. When the spatial angle cosine value is greater than the preset directional containment cosine threshold, the system absolutely preserves the signal within the legal direction, and the spatial orientation weight is greater. Therefore, both the dynamic attenuation factor of force load and the difference are positively correlated with the spatial orientation weight.

[0059] In one embodiment of the present invention, the dynamic attenuation factor of the force load at the current moment and the product of the difference between the cosine value of the spatial angle and the preset direction containment cosine threshold are obtained, and the Sigmoid activation function is used for mapping so that the product is perfectly compressed to between 0 and 1; that is, the larger the dynamic attenuation factor of the force load, the larger the difference, and the greater the spatial orientation weight.

[0060] For any pixel, the formula for spatial orientation weights is: ; ;in, Represents the spatial orientation weight of a pixel; This represents the dynamic attenuation factor of the applied load; This represents the preset initial dynamic decay factor; Indicates the load factor; Represents the cosine value of the spatial angle; This indicates a preset direction encompassing the cosine threshold. Represents the natural constant.

[0061] It should be noted that, in the embodiments of the present invention, the preset direction encompassing cosine threshold can be set based on the myofascial anatomical orientation of the target muscle. Since muscle fibers are wrapped in bundles by myofascia, the resistance to action potential conduction along the longitudinal direction of the muscle fiber is extremely small and the speed is extremely fast, while the resistance to conduction across the myofascia laterally is extremely large. When the muscle exerts normal force and increases, the diffusion of the electrical signal on the image should be an elliptical expansion along the muscle orientation.

[0062] As an example, muscle fibers are wrapped in myofascia, and action potentials are easily transmitted longitudinally along the direction of force, while the resistance across the myofascia laterally is extremely high. This logic is used to nonlinearly intercept crosstalk signals that deviate from the normal force axis; an initial dynamic attenuation factor is preset. Preset direction encompassing cosine threshold Allows for spatial dispersion within approximately 45 degrees; current operating load rate is... There exists a pixel that deviates from the main force line, with a spatial angle cosine value of Calculate the dynamic decay factor , Due to the 80% high load, control accuracy decreases, and the dynamic attenuation factor of the applied load increases. With a decay rate of 6.74, the system exhibits physiological fault tolerance, retaining a weight of 0.327 even if a pixel deviates from the direction of force application.

[0063] Spatial orientation weights reflect the degree of nonlinear interception and filtering of interfering electrical signals in directions other than the force exertion direction. As muscles continue to contract, local lactic acid accumulation and changes in ion concentration lead to a slower conduction velocity of action potentials on muscle fibers and a leftward shift in the spectrum. By analyzing the frequency domain characteristics of the electromyographic voltage amplitude sequence at the muscle activation center point, the neuromuscular fatigue level of the target muscle is reflected. Therefore, the electromyographic intensity is adjusted. Based on the spatial orientation weights of each pixel and the frequency domain characteristics of the electromyographic voltage amplitude sequence at the muscle activation center point, the electromyographic intensity of each pixel in the dynamic dilation image is adjusted to obtain the corrected electromyographic intensity of each pixel.

[0064] Preferably, in one embodiment of the present invention, the method for obtaining the electromyographic correction intensity includes: The first step is to obtain the fatigue coefficient based on the frequency domain characteristics of the electromyographic voltage amplitude sequence at the muscle activation center point. It should be noted that, in one embodiment of the present invention, a short-time Fourier transform is performed on the electromyographic voltage amplitude sequence at the muscle activation center point in the neighborhood of the latest moment to obtain the transformed short-time median frequency. Obtain the difference between the reference median frequency and the short-time median frequency, calculate the ratio of the difference result to the reference median frequency, and select the maximum value between the ratio result and the preset fatigue lower limit as the fatigue coefficient.

[0065] It should be noted that, in the embodiments of the present invention, the power spectrum is calculated based on the time spectrum obtained after Fourier transform; the frequency at the bisection of the power spectrum area is determined as the short-time median frequency; the reference median frequency is the median frequency obtained in the initial fatigue-free state; the neighborhood range is obtained by taking the latest time as the reference and the range formed by the historical time, where the historical time is 1000 milliseconds; fatigue has non-negativity, and a value of 0 represents the theoretically absolute fatigue-free state, therefore the preset fatigue lower limit is set to 0; in other embodiments of the present invention, the size of the neighborhood range can be set according to specific circumstances, and is not limited or elaborated here.

[0066] The second step is to obtain the corrected electromyography intensity of each pixel based on the fatigue coefficient, the spatial orientation weight of each pixel, and the electromyography intensity of each pixel in the dynamically dilated image.

[0067] It should be noted that the spatial orientation weight filters out interference electromyographic intensity in non-force direction in the dynamic expansion image, and the fatigue coefficient helps to increase the fatigue amplitude and compensate for spatial divergence of the electromyographic intensity after interference is filtered out. Therefore, the two are combined to correct the electromyographic intensity. In one embodiment of the present invention, the first product of the preset amplitude correction coefficient and the fatigue coefficient is obtained; the sum of the constant 1 and the first product is obtained as the adjustment weight; the product between the adjustment weight, the spatial orientation weight of each pixel and the electromyographic intensity of the pixel is obtained as the corrected electromyographic intensity.

[0068] For any pixel, the formula for the electromyographic correction intensity is expressed as: ;in, Indicates the electromyographic correction intensity of a pixel; This represents the electromyographic intensity of a pixel in a dynamically dilated image; This indicates the preset amplitude correction coefficient; Indicates the fatigue coefficient; This represents the spatial orientation weight of a pixel.

[0069] It should be noted that, in the embodiments of the present invention, in the case of fatigue, in order to maintain mechanical force, the central nervous system is forced to send more pulses to the muscles, resulting in an abnormal surge in the amplitude of electromyographic signals. Therefore, based on historical experience, the amplitude correction coefficient is preset to 0.8.

[0070] As an example, time-frequency analysis revealed a leftward shift in the median frequency, leading to the calculation that the muscle currently exhibits a 20% fatigue shift. ; Obtain the electromyographic intensity of the pixel at the current dynamic dilation image. Combined with the calculated spatial orientation weights Then calculate Considering the issues of directional offset filtering under heavy load and artificially high neuromuscular fatigue levels, the electromyography correction intensity of each pixel is obtained through smoothing adjustments. This provides an extremely rigorous benchmark input for determining inefficient recruitment or abnormal co-contraction at the end of the process.

[0071] Step S4: Based on the differences in electromyographic compensation intensity and electromyographic correction intensity between antagonist and agonist muscle regions at different pixels, obtain the abnormal feature classification coefficient and detect abnormal contraction patterns.

[0072] Electromyographic compensation intensity reflects the actual muscle exertion intensity after eliminating the influence of uneven subcutaneous fat thickness; electromyographic correction intensity reflects the theoretical muscle exertion intensity calculated by expansion based on torque magnitude and fatigue level; by analyzing the difference between the two, the degree of abnormality is reflected, that is, the greater the difference, the more disordered the central nervous system control.

[0073] Preferably, in one embodiment of the present invention, the method for obtaining the anomaly feature classification coefficient includes: The absolute value of the difference between the electromyography compensation intensity and the electromyography correction intensity of each pixel is obtained as the residual value of each pixel. Based on the ratio of the residual values ​​of pixels in the antagonistic muscle region to the electromyographic correction intensity of pixels in the relative agonistic muscle region, the abnormal feature classification coefficient is obtained, and the residual values ​​are positively correlated with the abnormal feature classification coefficient.

[0074] It should be noted that the antagonist muscle region reflects the location of the muscle that is theoretically inhibited during movement; the agonist muscle region reflects the location of the muscle exerting force. The greater the residual value of the antagonist muscle region relative to the electromyographic correction intensity of the agonist muscle, the more force is exerted by the antagonist muscle that should not be exerting force, and the greater the abnormal feature classification coefficient. In one embodiment of the present invention, the sum of the residual values ​​of all pixels in the antagonist muscle region is divided by the sum of the electromyographic correction intensities of all pixels in the agonist muscle region and the sum of the smallest positive number. The resulting value is used as the abnormal feature classification coefficient. In order to avoid the sum of the electromyographic correction intensities being 0 when calculating the ratio, a smallest positive number with the same dimensions as the denominator is added. The value of this number is specifically set according to the range of values ​​of the denominator, which will not be elaborated here.

[0075] It should be noted that, in another embodiment of the present invention, the method for detecting abnormal contraction patterns includes: The sum of the residual values ​​of all pixels is obtained as the global error. If the global error is less than or equal to the preset safety threshold, it is determined to be the expected normal shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is greater than or equal to the preset classification threshold, it is determined to be the co-shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is less than the preset classification threshold, it is determined to be the inefficient recruitment mode.

[0076] It should be noted that, in the embodiments of the present invention, the preset safety threshold is set to 1% of the dynamic safety alarm line. The dynamic safety alarm line is a pre-determined global error benchmark value for safe operation. The acquisition process is as follows: under the benchmark action test of the subject under safe load and without fatigue, multiple sets of standard contraction sample sequences are collected and the corresponding residual values ​​are calculated. The maximum value of the sum of residual values ​​in each set of tests is taken as the dynamic safety alarm line.

[0077] Considering that a significant signal co-contraction is only defined when the unexpected activation energy of the antagonist muscle reaches about 30% of the normal force exertion energy of the agonist muscle, and excluding slight neural control overflow under high load, the preset classification threshold is set to 0.3 based on relevant historical experience. In other embodiments of the present invention, the size of the preset safety threshold and the preset classification threshold can be set according to specific circumstances, and will not be limited or elaborated here.

[0078] Based on this, the expected normal contraction pattern indicates that the current muscle nerve control conforms to the dynamic benchmark expectation and does not trigger an alarm; if the abnormal feature classification coefficient is greater than or equal to the preset classification threshold, the molecule expands rapidly, reflecting that the antagonistic muscle area, which should be in a relaxed state, has experienced a violent burst of excessive electrical activity, which is judged as an unexpected co-contraction caused by the failure of the inhibition mechanism; if the abnormal feature classification coefficient is less than the preset classification threshold, the residual is not in the antagonistic muscle area, but the actual force energy of the agonist muscle is much lower than expected, which is judged as inefficient recruitment caused by the failure of positive activation.

[0079] In summary, this invention obtains the electromyographic compensation intensity of each pixel based on the electromyographic intensity and location characteristics of different pixels in the electromyographic image, as well as the subcutaneous fat equivalent impedance of each electrode at the latest time. It then dilates the initial baseline electromyographic image based on the three-dimensional instantaneous torque vector at the latest time to obtain a dynamically dilated image. Finally, it obtains the electromyographic correction intensity of each pixel by combining the distribution characteristics of different pixels in the electromyographic image, a preset initial dynamic attenuation factor, and the frequency domain characteristics of the electromyographic voltage amplitude sequence. Based on the differences in electromyographic compensation intensity and electromyographic correction intensity between antagonist and agonist muscle regions at different pixels, abnormal contraction patterns are detected. This invention improves the accuracy of abnormal contraction pattern detection by dynamically eliminating the influence of multidimensional physiological interference factors to obtain abnormal feature classification coefficients.

Claims

1. A method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography (EMG) images, characterized in that, The method includes: The electromyographic voltage amplitude sequence of each electrode in the target muscle region during a single movement is obtained to obtain the corresponding electromyographic image; the subcutaneous fat equivalent impedance and three-dimensional instantaneous torque vector of each electrode at different times are obtained; the target muscle region includes antagonistic muscle regions and agonist muscle regions; Based on the electromyographic intensity and location characteristics of different pixels in the electromyographic image, the muscle activation center point and the lateral diffusion penalty factor of each pixel are obtained; based on the subcutaneous fat equivalent impedance of each electrode at the latest time, the lateral diffusion penalty factor of each pixel and the electromyographic intensity, the electromyographic compensation intensity of each pixel is obtained. Based on the latest three-dimensional instantaneous torque vector, the initial baseline electromyography (EMG) image is dilated to obtain a dynamic dilated image. Based on the latest three-dimensional instantaneous torque vector, the distribution characteristics of different pixels in the EMG image relative to the muscle activation center point, and the preset initial dynamic attenuation factor, the spatial orientation weight of each pixel is obtained. Based on the spatial orientation weight of each pixel and the frequency domain characteristics of the EMG voltage amplitude sequence at the muscle activation center point, the EMG intensity of each pixel in the dynamic dilated image is adjusted to obtain the corrected EMG intensity of each pixel. Based on the differences in electromyographic compensation intensity and electromyographic correction intensity between antagonist and agonist muscle regions at different pixels, abnormal feature classification coefficients are obtained to detect abnormal contraction patterns. The methods for obtaining the muscle activation center point and the lateral diffusion penalty multiple for each pixel include: Select the pixel with the maximum electromyographic intensity value in the electromyographic image, and use the average coordinate of the corresponding pixel as the muscle activation center point; Obtain the relative distance between each pixel and the muscle activation center point, and select the minimum value between the relative distance and the preset distance threshold as the diffusion distance; obtain the product between the preset diffusion constraint coefficient and the diffusion distance, and perform exponential mapping as the lateral diffusion penalty multiple for each pixel; The method for obtaining the spatial orientation weights includes: Obtain the difference between the coordinates of each pixel and the muscle activation center point as a difference vector; obtain the cosine of the spatial angle between the difference vector and the three-dimensional instantaneous torque vector after projection onto the electromyography image plane; A negative correlation mapping is performed on the power load rate, and the product between the negative correlation mapping result and the preset initial dynamic attenuation factor is used as the power load dynamic attenuation factor. Based on the dynamic attenuation factor of the applied load, and the difference between the cosine value of the spatial angle and the preset direction containment cosine threshold, the spatial orientation weight of each pixel is obtained. The dynamic attenuation factor of the applied load and the difference are both positively correlated with the spatial orientation weight. The method for obtaining the force load rate includes: obtaining the magnitude of the three-dimensional instantaneous torque vector at the latest moment and normalizing it as the force load rate.

2. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 1, characterized in that, The method for obtaining the electromyographic compensation intensity includes: An interpolation algorithm was used to map the subcutaneous fat equivalent impedance of each electrode at the latest time point into a two-dimensional impedance distribution map aligned with the electromyography image, so as to obtain the impedance mapping value of each pixel. The electromyographic compensation intensity of each pixel is obtained based on the impedance mapping value, lateral diffusion penalty factor, and electromyographic intensity. The impedance mapping value, lateral diffusion penalty factor, and electromyographic intensity are all positively correlated with the electromyographic compensation intensity.

3. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 1, characterized in that, The method for obtaining the dynamically dilated image includes: The dynamic expansion radius is obtained based on the three-dimensional instantaneous torque vector at the latest moment; Two-dimensional morphological dilation processing is performed on the initial baseline electromyography image based on the dynamic dilation radius to obtain a dynamically dilated image.

4. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 3, characterized in that, The method for obtaining the dynamic expansion radius includes: Obtain the first product between the power load rate and the preset recruitment expansion coefficient; obtain the difference between the power load rate and the preset co-contraction activation threshold, select the maximum value between the difference and the preset activation lower limit, and calculate the second product between the square of the maximum value and the preset co-contraction compensation expansion coefficient; The sum of the preset safety tolerance radius, the first product, and the second product is obtained and used as the dynamic expansion radius.

5. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 1, characterized in that, The method for obtaining the electromyographic correction intensity includes: The fatigue coefficient is obtained based on the frequency domain characteristics of the electromyographic voltage amplitude sequence at the muscle activation center point. The corrected electromyography intensity of each pixel is obtained based on the fatigue coefficient, the spatial orientation weight of each pixel, and the electromyography intensity of each pixel in the dynamically dilated image.

6. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 5, characterized in that, The method for obtaining the fatigue coefficient includes: A short-time Fourier transform is performed on the electromyographic voltage amplitude sequence at the muscle activation center point within the neighborhood of the latest time to obtain the transformed short-time median frequency. Obtain the difference between the reference median frequency and the short-time median frequency, calculate the ratio of the difference result to the reference median frequency, and select the maximum value between the ratio result and the preset fatigue lower limit as the fatigue coefficient.

7. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 1, characterized in that, The method for obtaining the classification coefficient of the abnormal features includes: The absolute value of the difference between the electromyography compensation intensity and the electromyography correction intensity of each pixel is obtained as the residual value of each pixel. Based on the ratio of the residual values ​​of pixels in the antagonistic muscle region to the electromyographic correction intensity of pixels in the relative agonistic muscle region, the abnormal feature classification coefficient is obtained, and the residual values ​​are positively correlated with the abnormal feature classification coefficient.

8. The method for detecting multi-muscle coordinated abnormal contraction patterns based on surface electromyography images according to claim 7, characterized in that, The method for detecting abnormal contraction patterns includes: The sum of the residual values ​​of all pixels is obtained as the global error. If the global error is less than or equal to the preset safety threshold, it is determined to be the expected normal shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is greater than or equal to the preset classification threshold, it is determined to be the co-shrinkage mode. If the global error is greater than the preset safety threshold and the abnormal feature classification coefficient is less than the preset classification threshold, it is determined to be the inefficient recruitment mode.

Images