Intelligent dynamic assessment method and system for muscle fatigue of sportswear

Abstract

The invention provides an intelligent sportswear muscle fatigue degree dynamic evaluation method and system, and relates to the technical field of intelligent sportswear, and the method comprises the steps: collecting muscle pressure, surface temperature and electric signal data, extracting a multi-dimensional feature vector through time-frequency domain decomposition, constructing a dynamic connection network through an adaptive neural synaptic algorithm, and carrying out the dynamic evaluation of the muscle fatigue degree of the intelligent sportswear. And establishing a fatigue evaluation function, determining a fatigue value of the muscle group, dynamically updating a personalized fatigue threshold value, and performing real-time monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent sportswear, and particularly relates to a method and system for dynamically evaluating muscle fatigue degree of intelligent sportswear. BACKGROUND

[0002] As a new type of equipment capable of monitoring the human body motion state in real time, intelligent sportswear has gradually attracted the attention of sports enthusiasts and professional athletes. The intelligent sportswear can collect various physiological signals of the human body during the movement, such as muscle pressure, surface temperature and electromyographic signal, and provides an important basis for evaluating the muscle fatigue degree during the movement.

[0003] Muscle fatigue is a common physiological phenomenon in sports. Excessive fatigue may lead to sports injuries, affect the sports effect, and even cause health risks. Therefore, accurate evaluation of muscle fatigue degree is of great significance for scientific arrangement of training intensity and prevention of sports injuries. The existing muscle fatigue evaluation methods still have the problems of lack of comprehensive utilization of multi-source heterogeneous data, difficulty in fully reflecting the complex physiological change process of muscle fatigue, limited suppression ability to environmental noise and motion artifacts, insufficient accuracy and reliability of fatigue evaluation results, and neglect of the synergistic effect between muscle groups and individual differences in the field of intelligent sportswear.

[0004] Therefore, there is an urgent need for a solution to solve the problems in the prior art. SUMMARY

[0005] The embodiments of the present application provide a method and system for dynamically evaluating muscle fatigue degree of intelligent sportswear, which can at least solve some of the problems in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a method for dynamically evaluating muscle fatigue degree of intelligent sportswear, comprising:

[0007] Collecting muscle pressure data, muscle surface temperature data and muscle electrical signal data collected by sensors in the intelligent sportswear as muscle state data;

[0008] Performing time-frequency domain decomposition on the muscle state data, extracting time domain features, frequency domain features and time-frequency joint features of each data, eliminating environmental noise and motion artifacts by combining an adaptive threshold filtering algorithm to obtain standard feature data, decomposing the standard feature data into a plurality of intrinsic mode functions based on an empirical mode decomposition method, extracting the instantaneous frequency and instantaneous amplitude of each intrinsic mode function by Hilbert-Huang transform, and obtaining a multi-dimensional feature vector representing the muscle activity state;

[0009] According to the multi-dimensional feature vector, a dynamic connection network is constructed by using an adaptive neural synapse algorithm, and based on the dynamic connection network, a non-linear feature in a muscle group movement process is analyzed by combining a recursive least square algorithm, and a fatigue degree evaluation function considering the interaction between muscle groups is established;

[0010] Based on the fatigue degree evaluation function and historical movement data, a fatigue degree value of each muscle group is determined, a personalized fatigue threshold is dynamically updated according to the fatigue degree value and is monitored in real time, and a warning signal is sent if a muscle group exceeds the personalized fatigue threshold.

[0011] In an alternative embodiment,

[0012] The muscle state data is decomposed in time and frequency domains, time domain features, frequency domain features and time-frequency joint features of each data are extracted, and standard feature data is obtained by combining an adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts, including:

[0013] Muscle pressure data, muscle surface temperature data and electromyographic signal data in a movement process are obtained and combined to form multi-source movement state data, continuous wavelet transform is performed on the multi-source movement state data based on a mother wavelet function, the mother wavelet function contains a center frequency parameter, time domain feature vectors, frequency domain feature vectors and time-frequency joint feature matrices are obtained;

[0014] Morphological operations are performed on the time domain feature vectors, the frequency domain feature vectors and the time-frequency joint feature matrices, erosion and dilation operations are performed through a structure element, enhanced feature data is obtained, an adaptive threshold function is constructed based on the local standard deviation and the number of sampling points of the enhanced feature data, and the adaptive coefficient of the adaptive threshold function is dynamically adjusted according to the local signal-to-noise ratio;

[0015] The enhanced feature data is decomposed into a plurality of eigenmodes, the instantaneous frequency of each eigenmode is calculated by Hilbert transform, motion artifact components are identified and removed according to the instantaneous frequency, standard feature data is reconstructed, the signal-to-noise ratio, feature coherence and mean square error of the standard feature data are calculated, and a feature quality evaluation index is obtained by weighted calculation combining a preset weight coefficient, the standard feature data is screened based on the feature quality evaluation index, and combined into a standard feature data set.

[0016] In an alternative embodiment,

[0017] The standard feature data is decomposed into a plurality of intrinsic mode functions based on an empirical mode decomposition method, the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform, and a multi-dimensional feature vector representing the muscle activity state is obtained, including:

[0018] The data in the standard feature data set is decomposed into a plurality of intrinsic mode functions and a residual term by an empirical mode decomposition method, to ensure that the intrinsic mode functions satisfy that the number of extreme points and the number of zero-crossing points differ by no more than 1, and the average value of the upper and lower envelope lines formed by the local maximum points and the local minimum points is zero; a cubic spline interpolation method is used to construct the upper and lower envelope lines of the intrinsic mode functions, a mean envelope line is calculated based on the upper and lower envelope lines, and an iteration termination condition is judged according to the standard deviation of the adjacent two iteration results;

[0019] The intrinsic mode functions are subjected to Hilbert-Huang transformation, an analytic signal containing the intrinsic mode functions and the Hilbert-Huang transformation result is constructed, the instantaneous amplitude, the instantaneous phase and the instantaneous frequency of the analytic signal are calculated, and the instantaneous amplitude, the instantaneous phase and the instantaneous frequency are combined into a feature matrix;

[0020] The feature matrix is subjected to Hilbert marginal spectrum analysis, the importance indicators of the feature components in the feature matrix are calculated, and high-importance feature components are screened out, a biomechanics consistency loss function is constructed based on joint torque, muscle force and gravity term, muscle activity patterns and muscle movement link transmission relationships are extracted, and the high-importance feature components are constructed into a multi-dimensional feature vector representing muscle activity states;

[0021] The mutual information entropy between the feature components in the multi-dimensional feature vector is calculated, the feature redundancy is obtained based on the mutual information entropy, the redundant feature components are removed according to a preset threshold, and a multi-dimensional feature vector is obtained.

[0022] In an optional implementation,

[0023] The feature matrix is subjected to Hilbert marginal spectrum analysis, the importance indicators of the feature components in the feature matrix are calculated, and high-importance feature components are screened out, and the high-importance feature components are constructed into a multi-dimensional feature vector representing muscle activity states, including:

[0024] The feature matrix is subjected to Hilbert marginal spectrum analysis, the Hilbert transform of each feature component in the feature matrix is calculated, the analytic signal of each feature component is obtained, the marginal spectrum is calculated based on the analytic signal, the importance indicators of the feature components are obtained by double integration of the marginal spectrum in the time-frequency domain, and the high-importance feature components corresponding to the importance indicators are screened out according to a preset threshold;

[0025] The joint motion parameters collected in advance are acquired, the Jacobian matrix, the muscle force vector, the damping term and the gravity term are calculated based on the joint motion parameters, and the joint torque is obtained by combination;

[0026] construct a biomechanical consistency loss function comprising a kinematic constraint term, a dynamic constraint term and an energy efficiency constraint term, wherein the kinematic constraint term is constructed based on joint angles, the dynamic constraint term is constructed based on joint torques, and the energy efficiency constraint term is constructed based on energy consumption;

[0027] extract a muscle activity pattern matrix based on the biomechanical consistency loss function, optimize and calculate a synergy weight based on the muscle activity pattern matrix and the muscle force vector, obtain a proximal joint angle and a distal joint angle, and construct a kinematic chain transmission relationship based on the proximal joint angle, the distal joint angle and the muscle force vector;

[0028] combine the high-importance feature components, the biomechanical consistency loss function, the synergy weight and the kinematic chain transmission relationship to construct an initial feature vector representing a muscle activity state, calculate mutual information entropy between feature components in the initial feature vector, and filter based on a preset mutual information entropy threshold to obtain a multi-dimensional feature vector.

[0029] In an optional embodiment,

[0030] According to the multi-dimensional feature vector, a dynamic connection network is constructed using an adaptive neural synapse algorithm, and based on the dynamic connection network, a non-linear feature in a muscle group movement process is analyzed using a recursive least squares algorithm, and a fatigue degree evaluation function considering muscle group interaction is established, including:

[0031] An initial dynamic network is established using the multi-dimensional feature vector through an adaptive neural synapse algorithm, a multi-level synapse plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network, a synapse weight update value is obtained through a short-term plasticity dynamic equation, a synapse connection strength update value is obtained through a long-term plasticity time-dependent plasticity rule, and a synapse growth probability value is calculated according to a calcium ion concentration, and the synapse weight update value, the synapse connection strength update value and the synapse growth probability value are input into the initial dynamic network to obtain the dynamic connection network;

[0032] The multi-dimensional feature vector is input into the dynamic connection network, and the output result of the dynamic connection network is optimized through a recursive least squares algorithm, a network output error value is calculated using an error update equation, a network parameter adjustment value is calculated using a gain matrix update equation, and the dynamic connection network is optimized according to the network output error value and the network parameter adjustment value to obtain an optimized output result;

[0033] A neurotransmitter concentration dynamics equation is established, a release term, a degradation term and a diffusion term are set in the neurotransmitter concentration dynamics equation, neurotransmitter concentration distribution values are calculated according to the neurotransmitter concentration dynamics equation, a single muscle fatigue characteristic term and a muscle group interaction term are set based on the optimization output result and the neurotransmitter concentration distribution values, and a fatigue degree evaluation function is constructed.

[0034] In an alternative embodiment,

[0035] The multi-dimensional feature vector is input into the dynamic connection network, the output result of the dynamic connection network is optimized through a recursive least square algorithm, and a network output error value is calculated by combining an error update equation, which includes:

[0036] A multi-dimensional feature vector is obtained, the multi-dimensional feature vector is input into a dynamic connection network, a three-dimensional nonlinear equation set is used to describe a network state evolution process, network disturbance sensitivity, bifurcation control parameters and periodic characteristic parameters are calculated and integrated, system phase space trajectories are obtained, Lyapunov exponents, correlation dimensions and bifurcation parameters are calculated, and a chaotic feature vector is combined;

[0037] A nonlinear constraint function including a state stability term, an orbit periodicity term and a parameter sensitivity term is constructed based on the chaotic feature vector, the state stability term adopts an exponential function form of the Lyapunov exponent, the orbit periodicity term adopts a sine function form of the correlation dimension, and the parameter sensitivity term adopts a hyperbolic tangent function form of the bifurcation parameter;

[0038] The nonlinear constraint function is integrated into a recursive least square algorithm, a parameter vector, a gain matrix and a regression vector are calculated through the recursive least square algorithm, a parameter search strategy is dynamically adjusted based on the chaotic feature vector, a search step is calculated according to the Lyapunov exponent, a search path is optimized using the correlation dimension, and an update rate is adjusted based on the bifurcation parameter;

[0039] The output result of the dynamic connection network is optimized through the recursive least square algorithm using the search step, the search path and the update rate, the difference between the nonlinear constraint function and the actual output and the predicted output of the network is multiplied, and a network output error value is calculated by combining an error update equation.

[0040] In an alternative embodiment,

[0041] A fatigue degree value of each muscle group is determined based on the fatigue degree evaluation function and historical motion data, and a personalized fatigue threshold is dynamically updated according to the fatigue degree value and is monitored in real time, including:

[0042] obtaining real-time motion data of the muscle groups, the real-time motion data including motion intensity, motion duration and motion frequency of the muscle groups, inputting the real-time motion data into a fatigue evaluation function; combining historical motion data as a reference benchmark, calculating fatigue values of the muscle groups based on the fatigue evaluation function;

[0043] dynamically updating an initial fatigue threshold value by using an adaptive adjustment algorithm based on comparison results of the fatigue values of the muscle groups and the historical motion data, the adaptive adjustment algorithm being used to optimize the initial fatigue threshold value in real time based on motion characteristics and physiological characteristics of different muscle groups, and establishing a personalized fatigue threshold value corresponding to each muscle group;

[0044] taking the personalized fatigue threshold value as a monitoring reference standard, monitoring fatigue states of the muscle groups in real time, and triggering a fatigue warning when the fatigue value exceeds the personalized fatigue threshold value.

[0045] In a second aspect, the embodiment of the present application provides a muscle fatigue dynamic evaluation system of an intelligent sports garment, which comprises:

[0046] a first unit configured to collect muscle pressure data, muscle surface temperature data and muscle electrical signal data collected by sensors in the intelligent sports garment as muscle state data;

[0047] a second unit configured to perform time-frequency domain decomposition on the muscle state data, extract time domain features, frequency domain features and time-frequency joint features of each data, eliminate environmental noise and motion artifacts by using an adaptive threshold filtering algorithm to obtain standard feature data, decompose the standard feature data into a plurality of intrinsic mode functions based on an empirical mode decomposition method, extract instantaneous frequency and instantaneous amplitude of each intrinsic mode function by using a Hilbert-Huang transform, and obtain a multi-dimensional feature vector representing a muscle activity state;

[0048] a third unit configured to construct a dynamic connection network by using an adaptive neural synapse algorithm based on the multi-dimensional feature vector, analyze nonlinear characteristics in a muscle group motion process based on the dynamic connection network and by using a recursive least squares algorithm, and establish a fatigue evaluation function considering interactions between muscle groups;

[0049] a fourth unit configured to determine fatigue values of the muscle groups based on the fatigue evaluation function and historical motion data, dynamically update a personalized fatigue threshold value based on the fatigue values, and monitor in real time, and send a warning signal if a muscle group exceeds the personalized fatigue threshold value.

[0050] In a third aspect, the embodiment of the present application provides an electronic device, which comprises:

[0051] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored by the memory to execute the aforementioned method.

[0052] In a fourth aspect of the embodiments of the present application, a computer readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the aforementioned method.

[0053] In the present application, by collecting the muscle state data collected by the sensors in the intelligent sports clothing, combining the time-frequency domain decomposition technology and the adaptive threshold filtering algorithm, the multi-dimensional feature vector representing the muscle activity state is accurately extracted, the environmental noise and the motion artifacts can be effectively eliminated, the accuracy and reliability of the feature data are improved, the dynamic connection network is constructed by using the adaptive neural synapse algorithm, the nonlinear characteristics in the muscle group movement process are analyzed by combining the recursive least square algorithm, the fatigue degree evaluation function considering the interaction between muscle groups is established, the synergistic relationship between different muscle groups can be comprehensively reflected, the accuracy of fatigue degree evaluation is improved, the muscle group fatigue degree value is determined based on the fatigue degree evaluation function and the historical movement data, the personalized fatigue threshold is dynamically updated and real-time monitoring is performed, and the early warning signal is sent when the muscle group exceeds the personalized fatigue threshold, so that the real-time evaluation and early warning of the muscle fatigue state in the movement process are realized, the risk of sports injury is effectively reduced, and a scientific basis is provided for the formulation of personalized sports programs. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A flowchart of the muscle fatigue degree dynamic evaluation method of the intelligent sports clothing according to the embodiments of the present application is shown in the figure.

[0055] Figure 2 A biomechanics consistency loss function optimization comparison diagram of the muscle fatigue degree dynamic evaluation method of the intelligent sports clothing according to the embodiments of the present application is shown in the figure.

[0056] Figure 3 A network structure dynamic reconstruction visual diagram of the muscle fatigue degree dynamic evaluation method of the intelligent sports clothing according to the embodiments of the present application is shown in the figure.

[0057] Figure 4 An error convergence comparison diagram of the muscle fatigue degree dynamic evaluation method of the intelligent sports clothing according to the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0058] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0059] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments can not be described again for the same or similar concepts or processes.

[0060] Figure 1 The flowchart of the muscle fatigue degree dynamic evaluation method of the intelligent sports clothing in the embodiments of the present application is shown in FIG. 1. Figure 1 The method comprises the following steps.

[0061] The muscle pressure data, muscle surface temperature data and muscle electrical signal data collected by the sensors in the intelligent sports clothing are collected as muscle state data.

[0062] The muscle state data is decomposed in time and frequency domains, the time domain features, frequency domain features and time-frequency joint features of each data are extracted, the standard feature data is obtained by eliminating environmental noise and motion artifacts in combination with an adaptive threshold filtering algorithm, the standard feature data is decomposed into a plurality of intrinsic mode functions based on an empirical mode decomposition method, the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform, and a multi-dimensional feature vector representing the muscle activity state is obtained.

[0063] According to the multi-dimensional feature vector, a dynamic connection network is constructed by using an adaptive neural synapse algorithm, the non-linear features in the muscle group movement process are analyzed based on the dynamic connection network in combination with a recursive least squares algorithm, and a fatigue degree evaluation function considering the interaction between muscle groups is established.

[0064] The fatigue degree values of each muscle group are determined based on the fatigue degree evaluation function and historical movement data, the personalized fatigue threshold is dynamically updated and real-time monitored according to the fatigue degree values, and a warning signal is sent if a muscle group exceeds the personalized fatigue threshold.

[0065] In an optional implementation,

[0066] The muscle state data is decomposed in time and frequency domains, the time domain features, frequency domain features and time-frequency joint features of each data are extracted, and the standard feature data is obtained by eliminating environmental noise and motion artifacts in combination with an adaptive threshold filtering algorithm, which comprises the following steps.

[0067] The muscle pressure data, muscle surface temperature data and electromyographic signal data in the movement process are acquired and combined to form multi-source movement state data, continuous wavelet transform is performed on the multi-source movement state data based on a mother wavelet function, the mother wavelet function contains a center frequency parameter, time domain feature vectors, frequency domain feature vectors and time-frequency joint feature matrices are obtained;

[0068] Morphological operations are performed on the time domain feature vectors, the frequency domain feature vectors and the time-frequency joint feature matrices, erosion and dilation operations are performed through a structure element, enhanced feature data is obtained, an adaptive threshold function is constructed based on the local standard deviation and the number of sampling points of the enhanced feature data, and an adaptive coefficient of the adaptive threshold function is dynamically adjusted according to a local signal-to-noise ratio;

[0069] The enhanced feature data is decomposed into a plurality of eigenmodes, the instantaneous frequency of each eigenmode is calculated through Hilbert transform, motion artifact components are identified and removed according to the instantaneous frequency, standard feature data is reconstructed, the signal-to-noise ratio, feature coherence and mean square error of the standard feature data are calculated and combined with a preset weight coefficient to obtain a feature quality evaluation index, the standard feature data is filtered based on the feature quality evaluation index, and a standard feature data set is combined.

[0070] Real-time data in the movement process are acquired, including muscle pressure data collected by a pressure sensor, muscle surface temperature data collected by a temperature sensor and electromyographic signal data collected by an electromyographic sensor. The three types of data are aligned and combined according to time sequence to form multi-source movement state data containing a plurality of physiological information. A mother wavelet function suitable for physiological signal analysis is selected, the mother wavelet function has an adjustable center frequency parameter, and continuous wavelet transform is performed on the multi-source movement state data using the mother wavelet function to extract time domain feature vectors reflecting signal time characteristics, frequency domain feature vectors reflecting frequency characteristics and time-frequency joint feature matrices reflecting time-frequency comprehensive characteristics.

[0071] Morphological optimization processing is performed on the extracted feature data, erosion and dilation operations are performed on the time domain feature vectors, the frequency domain feature vectors and the time-frequency joint feature matrices through a designed structure element. The erosion operation can eliminate noise interference in the data, and the dilation operation enhances the effective feature information in the data. The two operations are combined to obtain enhanced feature data. For the enhanced feature data, the standard deviation in the local region is calculated, and an adaptive threshold function is constructed combined with the actual number of sampling points. The adaptive coefficient in the adaptive threshold function is adjusted in real time by analyzing the signal-to-noise ratio of the local data segment, so that the threshold function can adaptively change according to the data characteristics.

[0072] The enhanced feature data is subjected to intrinsic mode decomposition to obtain intrinsic mode functions of different frequency features. The Hilbert transform is performed on each intrinsic mode function to calculate the instantaneous frequency features of the mode functions. By analyzing the change rule of the instantaneous frequency, the motion-induced false difference components are identified, which are removed from the original data, and the remaining effective components are reconstructed to obtain standard feature data. The quality of the standard feature data is evaluated, and the signal-to-noise ratio index, the coherence index between features, and the mean square error index are calculated. The feature quality evaluation index reflecting the data quality is obtained by weighting calculation of the above indexes and the pre-set weight coefficients. The standard feature data set is formed by combining the feature data with high quality according to the feature quality evaluation index.

[0073] Exemplarily, taking the arm flexion and extension movement as an example, the data is collected by the sensor in the sports clothing. The pressure sensor collects the pressure change data of the biceps and triceps, reflecting the muscle contraction and relaxation state; the temperature sensor collects the muscle surface temperature change, reflecting the heat production in the movement process; the electromyographic sensor collects the muscle discharge signal, showing the muscle activity intensity. The collected pressure data (range 0-200 kPa), temperature data (range 30-40℃) and electromyographic signal (range 0-5 mV) are aligned according to 100 sampling points per second. The Mexican hat wavelet is selected as the mother wavelet function, and the center frequency is set to 2 Hz, to obtain the time domain feature vector (reflecting the signal amplitude change), the frequency domain feature vector (reflecting the signal frequency composition) and the time-frequency joint feature matrix (reflecting the signal time-frequency change feature). A 3x3 pixel rectangular structural element is used for morphological processing to enhance the signal features. In the calculated intrinsic mode function, the 2-3 Hz low-frequency false difference caused by arm swinging is identified and deleted to reconstruct the signal. The final standard feature data set contains effective features reflecting the muscle fatigue degree, the signal-to-noise ratio is improved by 40%, the feature coherence reaches 0.85, and the mean square error is reduced to 30% of the original data.

[0074] In this embodiment, the synergistic analysis of muscle pressure, surface temperature and electromyographic signal is realized through the fusion processing mode of multiple data sources, which improves the comprehensiveness and reliability of feature extraction compared with a single data source. The combination of erosion and expansion operations in morphological operation is used to suppress noise while maintaining the integrity of effective features. The combination method based on intrinsic mode decomposition and Hilbert transform realizes accurate identification and removal of motion false difference.

[0075] In an alternative embodiment,

[0076] decomposing the standard feature data into a plurality of intrinsic mode functions based on an empirical mode decomposition method, extracting an instantaneous frequency and an instantaneous amplitude of each intrinsic mode function through a Hilbert-Huang transform, and obtaining a multi-dimensional feature vector representing a muscle activity state including:

[0077] decomposing data in a standard feature data set into a plurality of intrinsic mode functions and a residual term through an empirical mode decomposition method, ensuring that the intrinsic mode functions satisfy a difference between a number of extreme points and a number of zero-crossing points of no more than 1, and an average value of upper and lower envelope lines formed by local maximum points and local minimum points is zero, constructing the upper and lower envelope lines of the intrinsic mode functions using a cubic spline interpolation method, calculating a mean envelope line based on the upper and lower envelope lines, and determining an iteration termination condition according to a standard deviation of adjacent two iteration results;

[0078] performing a Hilbert-Huang transform on the intrinsic mode functions, constructing an analytic signal including the intrinsic mode functions and the Hilbert-Huang transform result, calculating an instantaneous amplitude, an instantaneous phase, and an instantaneous frequency of the analytic signal, and forming a feature matrix including the instantaneous amplitude, the instantaneous phase, and the instantaneous frequency;

[0079] introducing a Hilbert marginal spectrum analysis of the feature matrix, calculating an importance index of each feature component in the feature matrix and screening to obtain selected high importance feature components, constructing a biomechanics consistency loss function based on joint torque, muscle force, and a gravity term, extracting a muscle activity pattern and a muscle movement link transmission relationship, and constructing the high importance feature components as a multi-dimensional feature vector representing a muscle activity state;

[0080] calculating mutual information entropy between feature components in the multi-dimensional feature vector, obtaining a feature redundancy degree based on the mutual information entropy, removing redundant feature components according to a preset threshold, and obtaining a multi-dimensional feature vector.

[0081] Starting from a standard feature data set, an empirical mode decomposition method is used for data processing, local extreme point detection is performed on a data sequence, and all local maximum points and local minimum points are identified. The position of the extreme point is determined by comparing the adjacent data points with the center point. When the center point is greater than the two side points, it is marked as a local maximum point, and when it is less than the two side points, it is marked as a local minimum point. At the same time, the zero-crossing points of the data sequence are detected, and the positions where the signal changes from positive to negative or from negative to positive are recorded. During the decomposition process, the relationship between the number of extreme points and the number of zero-crossing points is continuously monitored to ensure that the difference is no more than 1.

[0082] The envelope line is constructed for the detected local maximum point set and local minimum point set respectively. The discrete extreme points are connected into smooth envelope lines by using cubic spline interpolation method. In the interpolation process, the boundary conditions and node parameters are adjusted to ensure that the generated envelope line can accurately reflect the data fluctuation characteristics and will not produce excessive oscillation. The average value of the upper and lower envelope lines is calculated to obtain the mean envelope line. The original data is subtracted from the mean envelope line to obtain a new data sequence. Repeat the above process for this new sequence until a component that meets the intrinsic mode function condition is obtained. In the iteration process, the standard deviation between the new sequence obtained by each decomposition and the last sequence is calculated to establish the standard deviation sequence. When the standard deviation obtained by continuous multiple iterations is lower than the preset threshold, it is considered that the current component has met the requirements of the intrinsic mode function, and it is extracted. The remaining signal is continuously decomposed, and finally a series of intrinsic mode functions and a residual term are obtained.

[0083] Each intrinsic mode function is input into the Hilbert-Huang transform processing flow. First, a Hilbert transformer is constructed to perform Fourier transform on the input signal, set the negative frequency component in the frequency domain to zero, double the positive frequency component, perform inverse Fourier transform to obtain the imaginary part of the analytic signal. The original signal is taken as the real part, and the imaginary part calculated is used to construct the complex analytic signal together. The analytic signal is subjected to polar coordinate transformation to extract the instantaneous amplitude and instantaneous phase information. The instantaneous frequency is calculated by the time difference of the phase signal, and the difference result is smoothed to eliminate the influence of sudden changes. The extracted instantaneous amplitude, instantaneous phase and instantaneous frequency are aligned by time to form a feature matrix.

[0084] The Hilbert marginal spectrum analysis is performed on the feature matrix to calculate the energy distribution of each time-frequency point. The time-frequency plane is divided into multiple sub-regions, and the energy in each sub-region is accumulated and counted to obtain the frequency marginal spectrum and the time marginal spectrum. Based on the distribution characteristics of the marginal spectrum, the energy aggregation degree, the frequency stability and the time persistence of each feature component are calculated, and these indicators are combined to obtain the feature importance score. According to the score, high importance feature components are screened out. At the same time, a biomechanical consistency loss function is constructed, which includes joint torque item, muscle force item and gravity item. By minimizing the loss function, the muscle contraction-diastolic activity mode and the force transmission relationship between muscle groups are extracted. The high importance feature components screened out are fused with the biomechanical features to construct a multi-dimensional feature vector.

[0085] Perform redundancy analysis on multidimensional feature vectors. Mutual information entropy calculation methods are used to quantify the information correlation between any two components in the feature vector. The probability distribution of each feature component is estimated, and then the joint probability distribution is calculated. Mutual information entropy is calculated based on these distributions. A feature redundancy matrix is ​​constructed, where each element represents the degree of redundancy between the corresponding two feature components. A redundancy threshold is set. When the mutual information entropy of a pair of features exceeds the threshold, the feature with the higher importance score is retained and the other feature is deleted. Redundant features are gradually removed to obtain a streamlined multidimensional feature vector.

[0086] For example, take the activity analysis of the thigh muscle group during running as an example. The collected standard feature data set is subjected to empirical mode decomposition to obtain 8 intrinsic mode functions and 1 residual term. Each intrinsic mode function strictly meets the characteristic requirements. For example, the first mode function contains 120 extreme points and 119 zero-crossing points, and the mean of its upper and lower envelopes always fluctuates around zero. The envelope constructed by cubic spline interpolation can accurately reflect the fluctuation characteristics of the signal. After 15 iterations, the standard deviation of the adjacent iteration results dropped below 0.001, meeting the termination condition.

[0087] A Hilbert-Huang transform was performed on these intrinsic mode functions to obtain analytical signals reflecting the characteristics of thigh muscle contraction. The extracted instantaneous amplitude shows that the intensity of muscle contraction reaches a peak at the start of the run and then exhibits periodic changes. The instantaneous phase reflects the coordination characteristics when the left and right legs alternate. The instantaneous frequency shows the dynamic changes in the cadence. After Hilbert marginal spectrum analysis, 12 high-importance feature components were screened from the original feature matrix. Combining the hip joint torque (peak value approximately three times body weight), the muscle force generated by the thigh muscles, and the influence of gravity, a complete muscle activity chain was extracted, showing the force transmission process from the hip to the knee joint. Finally, through mutual information entropy analysis, three pairs of redundant features (mutual information entropy values ​​exceeding 0.9) were identified. After removing them, a 9-dimensional feature vector was obtained, which fully retained the key information of the muscle activity state.

[0088] In this embodiment, the accuracy of the intrinsic mode function decomposition is ensured through a strict extreme point and zero-crossing point detection mechanism, and the precise extraction of the instantaneous characteristics of the signal is achieved through the Hilbert-Huang transform, which not only obtains the amplitude information but also accurately captures the dynamic changes of phase and frequency. By constructing a biomechanical consistency loss function, the mechanical characteristics are organically combined with the physiological characteristics, realizing a multi-dimensional characterization of muscle activity.

[0089] In an optional embodiment,

[0090] Introducing the Hilbert marginal spectrum to analyze the feature matrix, calculating the importance index of each feature component in the feature matrix, screening high-importance feature components based on the importance index, and constructing the high-importance feature components into a multidimensional feature vector representing the muscle activity state includes:

[0091] Introducing the Hilbert marginal spectrum to analyze the characteristic matrix, calculating the Hilbert transform of each characteristic component in the characteristic matrix to obtain an analytical signal of each characteristic component, calculating the marginal spectrum based on the analytical signal, performing a double integration of the marginal spectrum in the time-frequency domain to obtain an importance index of each characteristic component, and screening high-importance characteristic components corresponding to the importance index according to a preset threshold;

[0092] Acquire pre-collected joint motion parameters, calculate the Jacobian matrix, muscle force vector, damping term and gravity term according to the joint motion parameters, and combine them to obtain the joint torque;

[0093] Constructing a biomechanical consistency loss function including a kinematic constraint term, a dynamic constraint term, and an energy efficiency constraint term, wherein the kinematic constraint term is constructed based on a joint angle, the dynamic constraint term is constructed based on the joint torque, and the energy efficiency constraint term is constructed based on energy consumption;

[0094] Extracting a muscle activity pattern matrix based on the biomechanical consistency loss function, optimizing and calculating a synergy weight based on the muscle activity pattern matrix and the muscle force vector, obtaining a proximal joint angle and a distal joint angle, and constructing a kinematic chain transmission relationship between the proximal joint angle, the distal joint angle, and the muscle force vector;

[0095] The high-importance feature components, the biomechanical consistency loss function, the synergy weight and the motion chain transfer relationship are combined to construct an initial feature vector representing the muscle activity state, the mutual information entropy between the feature components in the initial feature vector is calculated, and a multidimensional feature vector is obtained by screening based on a preset mutual information entropy threshold.

[0096] The Hilbert marginal spectrum analysis method is introduced to process the characteristic matrix. First, each characteristic component in the matrix is ​​subjected to a Hilbert transform. During the transformation process, the original signal is mapped to the complex plane by constructing an orthogonal transformation operator to obtain the analytical signal corresponding to each characteristic component. Time-frequency analysis is performed on the analytical signal to calculate the instantaneous frequency and instantaneous amplitude, and to construct a time-frequency energy distribution diagram. Marginal spectrum calculation is performed on the time-frequency plane, and the time-frequency energy distribution diagram is integrated in the time dimension and frequency dimension respectively to obtain the time marginal spectrum and frequency marginal spectrum. The two marginal spectra are double-integrated in the entire time-frequency domain to obtain a quantitative indicator that characterizes the importance of each characteristic component. A feature importance threshold is set, and feature components with importance indicators higher than the threshold are screened out as high-importance feature components.

[0097] The joint motion parameters, including joint position, velocity and acceleration information, are obtained from a pre-established motion database. The Jacobian matrix, which describes the mapping relationship between the joint space and the Cartesian space, is calculated based on the motion parameters. Meanwhile, the force vectors generated by each muscle are calculated, taking into account the muscle contraction characteristics and the force arm changes. Combining the damping effect during joint motion, the velocity-related damping term is calculated. Considering the influence of gravity on each joint, the gravity term is calculated. The Jacobian matrix, muscle force vector, damping term and gravity term are combined to obtain the complete joint torque expression.

[0098] A biomechanical consistency loss function is constructed, which includes three main constraint terms. The kinematic constraint term is based on joint angles, which ensures the accuracy of the motion trajectory by calculating the deviation between the predicted angle and the actual angle. The dynamic constraint term is based on joint torque, which ensures the balance of force by comparing the calculated torque and the measured torque. The energy efficiency constraint term is based on the energy consumption of muscle activity, which optimizes the motion efficiency by minimizing the total energy consumption. Each constraint term is combined through a weight coefficient to form a unified loss function.

[0099] The muscle activity pattern is extracted using the biomechanical consistency loss function. By minimizing the loss function, a series of characteristic matrices describing the muscle synergistic contraction pattern are obtained. The matrix is combined with the previously calculated muscle force vector, and an optimization algorithm is used to calculate the synergistic weights between different muscles. These weights reflect the coordination relationship between muscle groups. Based on the principle of kinematic chain, the motion angles of proximal joints (such as the hip joint) and distal joints (such as the ankle joint) are analyzed, and the motion transmission relationship between joints is established. The proximal joint angle, distal joint angle and muscle force vector are analyzed, and a complete kinematic chain transmission relationship model is constructed.

[0100] The high importance feature components selected are fused with biomechanical features. The calculation results of the biomechanical consistency loss function, muscle synergistic weights and kinematic chain transmission relationship are integrated to construct an initial feature vector. The correlation between each component in the feature vector is analyzed, and the mutual information entropy method is used to quantify the information redundancy degree between features. By calculating the mutual information entropy between each pair of features, a feature correlation matrix is established. Set the mutual information entropy threshold, when the mutual information entropy between a pair of features exceeds the threshold, the feature with larger information quantity is retained and the redundant feature is deleted. After such a screening process, an optimized multi-dimensional feature vector is obtained.

[0101] Exemplarily, taking the analysis of the shooting action in basketball as an example. The upper limb joint motion data is obtained by the motion capture system, including the motion parameters of the shoulder joint, elbow joint and wrist joint. The Hilbert marginal spectrum analysis is performed on the collected feature matrix, and 12 high importance feature components are screened out from the original 30 feature components, and these features are mainly concentrated in the shoulder joint abduction and elbow joint extension stage. The joint torque calculated based on the motion data shows that the maximum torque of the shoulder joint reaches 25 Nm, and the maximum torque of the elbow joint is 15 Nm. The constructed biomechanical consistency loss function includes three constraint terms with weights of 0.4, 0.4 and 0.2 respectively. By minimizing the loss function, four main muscle synergies are extracted, in which the synergistic weight of the deltoid and biceps brachii is the highest, reaching 0.8. By analyzing the movement chain of the shooting action, it is found that the transmission efficiency of the shoulder joint driving the elbow joint motion reaches 85%. Through mutual information entropy analysis, 12 feature components are optimized to 8, forming the final multi-dimensional feature vector, which can accurately represent the key muscle activity characteristics in the shooting action.

[0102] In the embodiment, the muscle activity characteristics are comprehensively captured by systematic time-frequency analysis of the feature matrix, a unified optimization framework is established by organically combining the kinematic constraint, the dynamic constraint and the energy efficiency constraint, the muscle activity pattern is extracted by the biomechanical consistency loss function, the complete movement chain transmission relationship is established based on the muscle force vector optimization calculation of the synergistic weight, and the feature screening mechanism based on mutual information entropy is introduced;

[0103] In the prior art, the analysis of the muscle activity state usually adopts a single signal processing method, only pays attention to the time domain or frequency domain features, and it is difficult to comprehensively capture the dynamic characteristics of the muscle activity. The biomechanical characteristics and the signal characteristics are analyzed separately, and there is a lack of effective feature fusion mechanism. In the feature extraction process, a fixed threshold screening method is generally used, which cannot adapt to the feature changes under different motion states, resulting in unstable quality of the extracted features;

[0104] The embodiment can not only extract the instantaneous characteristics of the signal, but also accurately evaluate the importance of the characteristics through marginal spectrum analysis, significantly improve the accuracy and reliability of the feature extraction, consider not only the accuracy of the motion, but also the energy efficiency, so that the extracted features are more in line with the actual law of human motion. By setting an adaptive mutual information entropy threshold, the intelligent simplification of the feature set is realized, the feature dimension is significantly reduced while maintaining the integrity of the information, and more reliable technical support is provided for the accurate analysis of the muscle activity state.

[0105] Figure 2The biomechanical consistency loss function optimization contrast chart of the intelligent sports clothing muscle fatigue degree dynamic evaluation method of the embodiment of the present application shows the biomechanical consistency loss function optimization process of three different methods. The technical solution adopts a multi-constraint biomechanical loss function, and combines kinematic constraints, dynamic constraints and energy efficiency constraints in a weighted manner, and the weights are 0.4, 0.4 and 0.2 respectively. Compared with the single dynamic constraint method and the Euclidean distance method, the technical solution shows different convergence characteristics in the optimization process. As can be observed from the figure, the loss function value of the technical solution rapidly decreases to 0.72 in the first 10 iterations, and then the convergence speed slows down, and stabilizes at about 0.50 after 50 iterations. The single dynamic constraint method converges to 0.31 after 50 iterations, and the Euclidean distance method converges to 0.17.

[0106] The final loss function value of the technical solution is higher than that of the other two methods, because the technical solution comprehensively considers various biomechanical constraints, more comprehensively reflects the biomechanical consistency in the movement process, and does not only focus on the optimization of a single index. The experimental results show that the prediction accuracy of the technical solution for the shoulder joint abduction angle and the elbow joint extension angle reaches 96.5% and 95.2% respectively, which is significantly higher than that of the other two methods.

[0107] In an alternative embodiment,

[0108] According to the multi-dimensional feature vector, a dynamic connection network is constructed by using an adaptive neural synapse algorithm, and based on the dynamic connection network, a non-linear feature in a muscle group movement process is analyzed by combining a recursive least square algorithm, and a fatigue degree evaluation function considering the interaction between muscle groups is established, including:

[0109] An initial dynamic network is established by using the multi-dimensional feature vector through an adaptive neural synapse algorithm, a multi-level synapse plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network, a synapse weight update value is obtained through a short-term plasticity dynamic equation, a synapse connection strength update value is obtained through a long-term plasticity time-dependent plasticity rule, a synapse growth probability value is calculated according to a calcium ion concentration, and the synapse weight update value, the synapse connection strength update value and the synapse growth probability value are input into the initial dynamic network to obtain the dynamic connection network;

[0110] The multi-dimensional feature vector is input into the dynamic connection network, and the output result of the dynamic connection network is optimized through a recursive least square algorithm, a network output error value is calculated by combining an error update equation, a network parameter adjustment value is calculated by using a gain matrix update equation, and the dynamic connection network is optimized according to the network output error value and the network parameter adjustment value to obtain an optimized output result;

[0111] A neurotransmitter concentration dynamics equation is established, in which a release term, a degradation term and a diffusion term are set, a neurotransmitter concentration distribution value is calculated according to the neurotransmitter concentration dynamics equation, a single muscle fatigue characteristic term and a muscle group interaction term are set based on the optimization output result and the neurotransmitter concentration distribution value, and a fatigue degree evaluation function is constructed.

[0112] An initial dynamic network is constructed based on a multi-dimensional feature vector. The topology of the network, including the number of neurons, hierarchical division and connection mode, is dynamically determined by an adaptive neural synapse algorithm. The number of hidden layer neurons is automatically adjusted based on the dimension and distribution characteristics of the input features in an iterative optimization manner. After determining the basic structure of the network, a multi-level synaptic plasticity model is introduced, which includes synaptic plasticity mechanisms at two time scales of short-term and long-term.

[0113] In the implementation process of short-term plasticity, a mapping relationship between presynaptic neuron activity and postsynaptic response is established. The rapid change of synaptic efficacy is described by a dynamic equation, considering the calcium influx of presynaptic terminals, the release probability of neurotransmitters and the sensitivity change of postsynaptic membrane. The instantaneous update value of synaptic weight is calculated based on these factors. At the same time, an adaptive term is introduced to enable the weight update to respond quickly to the frequency and intensity changes of the input signal.

[0114] Long-term plasticity is modeled based on time-dependent plasticity rules, considering the firing timing relationship between presynaptic and postsynaptic neurons. When the activity of presynaptic neurons precedes that of postsynaptic neurons, the synaptic connection strength is enhanced; otherwise, the connection strength is weakened. By setting the time window parameter, the time-dependent characteristics of synaptic strength change are determined. At the same time, a saturation factor is introduced to prevent the synaptic strength from increasing or decreasing indefinitely.

[0115] In the regulation of synaptic plasticity, calcium ion concentration is introduced as a key regulatory factor. A correlation model of calcium ion concentration and membrane potential change is established, considering the kinetic characteristics of calcium ion channels and the calcium ion buffering mechanism. When the local calcium ion concentration exceeds the threshold, the synaptic growth mechanism is triggered to promote the formation of new synapses; when the concentration is below the threshold, the synaptic degeneration mechanism is activated to eliminate redundant synapses. Through this mechanism, the dynamic reconstruction of network structure is realized.

[0116] The constructed dynamic connection network is used as the basic framework, and an improved recursive least squares algorithm is used for network optimization. An error function between the network output and the expected output is established, and the direction of parameter update is determined by calculating the gradient of the error function on the network parameters. In the update process, a forgetting factor mechanism is introduced to gradually fade the influence of historical data, better adapting to the feature changes of the current input.

[0117] The error updating process adopts a dynamic step strategy, which adaptively adjusts the updating step according to the error change trend. When the error rapidly decreases, a larger step is used to accelerate convergence; when the error fluctuates, the step is reduced to improve stability. At the same time, a gain matrix updating mechanism is established, which takes into account the covariance characteristics of the data and can adaptively adjust the updating amplitude of different parameters.

[0118] In the modeling of neurotransmitter dynamics, a nonlinear equation system containing multiple coupling terms is constructed. The modeling of the release term considers calcium ion influx, synaptic vesicle mobilization and secretion process, and describes the kinetic characteristics of different stages by introducing multiple time constants. The degradation term includes enzymatic degradation and reabsorption processes, establishing a nonlinear relationship between neurotransmitter concentration and degradation rate. The diffusion term uses an improved diffusion equation, considering the influence of spatial heterogeneity and boundary conditions.

[0119] A fatigue evaluation function is constructed, using a hierarchical structure design. In the single muscle fatigue characteristic term, multiple feature indicators from the network output are integrated, and the fatigue state of a single muscle is evaluated by nonlinear weighting. The weight coefficient is determined by training data optimization, which can adaptively adjust the importance of different features. The muscle group interaction term is based on the spatial distribution characteristics of neurotransmitters, establishing a coupling relationship model between muscles, considering the anatomical position relationship and functional synergy characteristics of muscles, and describing the transmission law of fatigue between muscle groups through diffusion effect.

[0120] Exemplarily, taking the fatigue monitoring of lower limb muscle groups in long-distance running as an example, an initial dynamic network containing input layer, hidden layer and output layer is constructed, the input layer receives an 8-dimensional feature vector containing information such as electromyographic signal features and muscle mechanical features. A synaptic plasticity model is set in the network, short-term plasticity reflects the synaptic weight change on the scale of hundreds of milliseconds, and long-term plasticity describes the connection strength adjustment on the scale of minutes. By monitoring the synaptic gap calcium ion concentration, when the concentration exceeds the threshold, the synaptic growth mechanism is triggered, and when the concentration is below the threshold, the synaptic weakening is caused. Integrating these plasticity mechanisms into the network, the dynamic adjustment of network structure is realized.

[0121] When optimizing the network, the recursive least squares algorithm is used to track the changes of muscle state in real time. By calculating the error between the network output and the actual fatigue state, the network parameters are dynamically adjusted. During the optimization process, the updating of the gain matrix takes into account the changes of exercise intensity, using a larger updating step in high intensity stage and a smaller step in low intensity stage.

[0122] In the neurotransmitter kinetics equation, the release term mainly considers the influence of exercise intensity on neurotransmitter release, the degradation term reflects the recovery characteristics during rest, and the diffusion term describes the transmission effect of fatigue between muscle groups. The fatigue evaluation function constructed can accurately reflect the fatigue degree of different muscle groups during long-distance running and predict the development trend of fatigue.

[0123] In this embodiment, by establishing a kinetics equation containing three key links of release, degradation and diffusion, the accurate description of the spatio-temporal distribution characteristics of neurotransmitters is realized, the real-time optimization and adjustment of network structure are realized through the multi-level synaptic plasticity model, and the convergence speed and stability of the algorithm are significantly improved by introducing the dynamic step strategy and gain matrix updating mechanism.

[0124] In the prior art, the traditional muscle fatigue evaluation method mainly relies on static neural network structure, which is difficult to adaptively adjust according to the change of input characteristics, resulting in poor adaptability of the network to dynamic changes, and when processing synaptic plasticity, only the change of single time scale is considered, the dynamic adjustment characteristics of synaptic strength on different time scales are ignored, the role of neurotransmitters is not considered, and the collaborative fatigue effect between muscle groups is not analyzed in depth, it is difficult to accurately describe the transmission rule of fatigue between muscle groups.

[0125] The dynamic connection network of the embodiment has stronger adaptive ability, can adjust the network structure in real time according to the change of input characteristics, significantly improves the processing ability of the network to non-stationary signals, the improved recursive least square algorithm has faster convergence speed and better stability, greatly improves the network optimization efficiency, and the evaluation method based on neurotransmitter kinetics can more accurately describe the transmission rule of fatigue between muscle groups, and the evaluation result has better consistency with the actual physiological process.

[0126] Figure 3The network structure dynamic reconstruction visual view of the intelligent sportswear muscle fatigue degree dynamic evaluation method of the embodiment of the application presents the network structure from the initial 4 input nodes (I1-I4), 3 hidden nodes (H1-H3) and 2 output nodes (O1-O2), and is optimized to the structure of adding 3 new hidden nodes (H4-H6) through the adaptive neural synapse algorithm. According to the data in the optimization process, the weight of the new connection is significantly higher than that of the initial connection, for example, the weight of I1 to H4 is 0.87, the weight of I2 to H4 is 0.92, the weight of I3 to H5 is 0.78, and the weight of I4 to H5 is 0.83. In the connection between the hidden layers, the weight of H4 to H1 is 0.65, the weight of H4 to H2 is 0.71, the weight of H5 to H2 is 0.68, and the weight of H5 to H3 is 0.74, indicating the optimization of the cross-layer information flow. In the part from the hidden layer to the output layer, the weight of H1 to H6 is as high as 0.89, the weight of H2 to H6 is 0.91, and the weight of H3 to H6 is 0.82. The connection weight of H6 to the output layer is 0.94 and 0.88 respectively, showing the strengthening of the network to the key output path.

[0127] In the optimization process, the total number of network connections is reduced from 18 initially to 14, indicating that the network structure is more concise and efficient. In terms of performance indicators, the network error rate is significantly reduced from the initial 21.7% to 8.3% in the intermediate stage, reaching a low error rate of only 2.6%, and the accuracy is improved to 94.6%. The number of neurons increases from 9 initially to 12, but the connections are reduced by 22.2%, and the average synapse weight increases from 0.42 initially to 0.81 finally, indicating that the quality of network connections has been greatly improved.

[0128] Under the condition that the calcium ion concentration threshold is 0.35 μmol / L, the synapse growth probability of the newly connected synapse is 0.63, and the degeneration probability is 0.37, realizing the reasonable balance of the network structure. The entire structure optimization process only takes 17.6 seconds, reflecting the efficiency of the scheme. The dynamic reconstruction mechanism enables the network to automatically adjust its structure according to the complexity of the input features, effectively avoiding the limitations of traditional fixed structure networks when dealing with complex fatigue evaluation tasks.

[0129] In an alternative embodiment,

[0130] The multi-dimensional feature vector is input into the dynamic connection network, and the output result of the dynamic connection network is optimized through a recursive least squares algorithm, and the network output error value is calculated by combining an error update equation, which includes:

[0131] The multi-dimensional feature vector is obtained, the multi-dimensional feature vector is input into a dynamic connection network, a three-dimensional nonlinear equation set is used to describe a network state evolution process, network disturbance sensitivity, bifurcation control parameters and periodic characteristic parameters are calculated and integral operation is performed, system phase space trajectories are obtained and Lyapunov indexes, correlation dimensions and bifurcation parameters are calculated, and a chaotic feature vector is obtained by combination;

[0132] A nonlinear constraint function including a state stability term, an orbit periodicity term and a parameter sensitivity term is constructed based on the chaotic feature vector, the state stability term adopts an exponential function form of the Lyapunov index, the orbit periodicity term adopts a sine function form of the correlation dimension, and the parameter sensitivity term adopts a hyperbolic tangent function form of the bifurcation parameter;

[0133] The nonlinear constraint function is integrated into a recursive least square algorithm, a parameter vector, a gain matrix and a regression vector are calculated by the recursive least square algorithm, a parameter search strategy is dynamically adjusted based on the chaotic feature vector, a search step is calculated according to the Lyapunov index, a search path is optimized by using the correlation dimension, and an update rate is adjusted based on the bifurcation parameter;

[0134] The output result of the dynamic connection network is optimized by the recursive least square algorithm by using the search step, the search path and the update rate, the difference between the nonlinear constraint function and the actual output and the predicted output of the network is multiplied, and network output error values are calculated by combining an error update equation.

[0135] Network state evolution analysis is performed based on an input multi-dimensional feature vector. A three-dimensional nonlinear equation set describing network dynamic characteristics is constructed, and the equation set reflects the evolution law of the network state. Numerical analysis is performed on the equation set, and network disturbance sensitivity is calculated by applying a small disturbance, which reflects the response characteristics of the network to external interference. Meanwhile, bifurcation control parameters are obtained by parameter variation analysis, which describe the stability change of the network under different parameter conditions. Periodic analysis is performed on the network output, and periodic characteristic parameters are extracted, which reflect the periodic variation law of the network output.

[0136] Time integral operation is performed on the disturbance sensitivity, the bifurcation control parameters and the periodic characteristic parameters, and system dynamic trajectories are reconstructed in a phase space. Based on the reconstructed phase space trajectories, three key parameters characterizing the chaotic characteristics of the system are calculated: the Lyapunov index reflects the divergence characteristics of the trajectory, the correlation dimension describes the space filling characteristics of the trajectory, and the bifurcation parameter characterizes the stability change of the system. These three parameters are combined to form a chaotic feature vector, which comprehensively describes the nonlinear dynamic characteristics of the network.

[0137] The nonlinear constraint function is constructed based on the chaotic characteristic vector, and the nonlinear constraint function includes three core constraint terms. The state stability term is constructed in the form of an exponential function of Lyapunov exponent, and the term reflects the stability of the system state. The orbit periodicity term is described in the form of a sine function of correlation dimension, and the term embodies the periodic characteristics of the system orbit. The parameter sensitivity term is expressed in the form of a hyperbolic tangent function of the bifurcation parameter, and the term reflects the sensitivity of the system to parameter changes.

[0138] The constructed nonlinear constraint function is integrated into the recursive least squares algorithm framework. In the algorithm, the parameter vector is calculated, including the weight parameters of each layer of the network. At the same time, the gain matrix is calculated, which controls the direction and amplitude of parameter update. The regression vector is constructed, which reflects the influence of historical data on the current optimization. Based on the characteristics of the chaotic characteristic vector, the dynamic adjustment of the parameter search strategy is realized. The step size of parameter search is determined by the size of Lyapunov exponent. When the exponent is large, a small step size is used to ensure stability, and when the exponent is small, a large step size is used to speed up convergence. The search path is optimized by the change characteristics of the correlation dimension. When the dimension is large, the diversity of the search direction is increased, and when the dimension is small, the search is concentrated on the main direction. Based on the change of the bifurcation parameter, the parameter update rate is adjusted. The update rate is reduced near the bifurcation point to improve accuracy.

[0139] Using the determined search step size, search path and update rate, the network output is optimized by the recursive least squares algorithm. The difference between the nonlinear constraint function and the actual output and the predicted output of the network is multiplied to construct an error function containing dynamic constraints. The network output error value is calculated based on the error update equation, which reflects the prediction accuracy and the satisfaction degree of the dynamic constraints.

[0140] Exemplarily, the analysis of electromyographic signals in continuous flexion and extension movement of the arm is taken as an example. The input multi-dimensional characteristic vector contains eight-dimensional information of time-frequency characteristics, mechanical characteristics and other information of electromyographic signals. After the characteristic vector is input into the dynamic connection network, the network state evolution is described by a three-dimensional nonlinear equation set, and the sensitivity of the network to the change of movement amplitude is calculated to be 0.85, the bifurcation control parameter caused by the change of movement frequency is 0.32, and the signal periodicity parameter is 0.64. The integral operation is performed on these parameters, and the system trajectory is obtained in the phase space. The Lyapunov exponent is calculated to be 0.15, the correlation dimension is 2.3, and the bifurcation parameter is 0.28. The nonlinear constraint function is constructed by combining the chaotic characteristic vector. In the optimization process of the recursive least squares algorithm, when the Lyapunov exponent reaches 0.15, the search step size is set to 0.01; when the correlation dimension is 2.3, search is performed in four main directions; and when the bifurcation parameter is 0.28, the update rate is set to 0.08. The obtained network output error value is 0.05, indicating that the optimized network has good prediction accuracy and dynamic characteristics.

[0141] In this embodiment, the evolution process of network state is described by constructing a three-dimensional nonlinear equation set, the dynamic characteristics of the network are accurately described, the nonlinear constraint function including the state stability term, the orbit periodicity term and the parameter sensitivity term is designed to realize the comprehensive constraint of the network dynamics, the nonlinear constraint function is integrated into the recursive least square algorithm, and the dynamic adjustment of the parameter search strategy is realized based on the chaotic characteristic vector.

[0142] In the prior art, the traditional neural network optimization method mainly focuses on the prediction accuracy of the network and ignores the importance of the network dynamics, often uses a simple gradient descent or back propagation algorithm, cannot effectively capture the chaotic characteristics and dynamic evolution law of the system, adopts a fixed parameter search strategy, lacks adaptive adjustment ability to the system state, is easy to fall into a local optimal solution or leads to unstable optimization process, and when constructing the error function, the dynamic constraint of the system is rarely considered, resulting in that the optimization result has good fitting accuracy but may violate the physical characteristics of the system.

[0143] The dynamic adjustment of the parameter search strategy in this embodiment effectively avoids the local optimal solution and realizes the rapid convergence of the global optimal solution, the adaptive optimization mechanism based on the chaotic characteristics significantly improves the stability of the network under different working conditions, provides strong technical support for modeling and optimization of complex nonlinear systems, and has important application value in the fields of signal processing and pattern recognition.

[0144] Figure 4 The error convergence comparison chart of the intelligent sports clothing muscle fatigue degree dynamic evaluation method of the embodiment of the present application shows the error convergence process comparison of the present technology scheme and three mainstream algorithms (traditional RLS algorithm, neural network method and support vector machine) in nonlinear system prediction. From the figure, the trend of the prediction error (RMSE) of the four algorithms with the increase of the iteration number can be clearly observed.

[0145] The technical scheme (square mark curve) shows significant advantages, not only the convergence speed is the fastest, but also the final prediction error is the lowest. Specifically, the technical scheme reaches a stable convergence state after about 24 iterations, and the final prediction error (RMSE) is reduced to 0.052. In contrast, the support vector machine (diamond mark curve) needs about 42 iterations to converge, and the final error is 0.094; the neural network method (circle mark curve) needs about 56 iterations to converge, and the final error is 0.087; the traditional RLS algorithm (triangle mark curve) performs the worst, needs about 68 iterations to reach convergence, and the final error is still as high as 0.130.

[0146] From the shape of the error reduction curve, the technical scheme shows a sharp error reduction characteristic in the early iteration stage (between 10-20 iterations), which is due to the dynamic adjustment function of the chaotic characteristic vector to the search strategy. In contrast, the error reduction rate of the other three methods is relatively flat, especially the traditional RLS algorithm, whose error reduction curve almost changes linearly, indicating the lack of effective optimization acceleration mechanism.

[0147] Notably, during the iteration process, the technical scheme has a significant error rapid reduction at the 10th-15th iteration, which corresponds to the key stage of the algorithm dynamically adjusting the search step length based on the Lyapunov index. When the Lyapunov index reaches 0.15, the algorithm automatically sets the search step length to 0.014, optimizes the 4 main search directions based on the correlation dimension 2.3, and adjusts the update rate to 0.08 according to the bifurcation parameter 0.28. This combination of parameter settings promotes the algorithm to quickly approach the global optimal solution.

[0148] In terms of convergence stability, the prediction error of the technical scheme after convergence fluctuates very little, remaining around 0.052, indicating that the nonlinear constraint function constructed based on the chaotic characteristic vector effectively enhances the stability of the algorithm. In contrast, the other three algorithms still show different degrees of fluctuation even after convergence, especially the support vector machine method, whose prediction error still has a small amplitude oscillation in the last stage, indicating its limited adaptability to nonlinear systems.

[0149] In summary, Figure 4 Intuitively demonstrates the significant advantages of the technical scheme in convergence speed, prediction accuracy and stability. By integrating chaos dynamics theory and recursive least squares algorithm, it has achieved better performance than existing algorithms in nonlinear system prediction tasks, especially in handling complex nonlinear relationships and non-stationary time series data, with stronger adaptability and robustness.

[0150] In an alternative embodiment,

[0151] Based on the fatigue evaluation function and historical motion data, the fatigue values of each muscle group are determined, and the personalized fatigue threshold is dynamically updated according to the fatigue values and real-time monitored, including:

[0152] Real-time motion data of the muscle groups are obtained, including the motion intensity, duration and frequency of the muscle groups, and the real-time motion data are input into the fatigue evaluation function; combining historical motion data as a reference benchmark, the fatigue values of each muscle group are calculated based on the fatigue evaluation function;

[0153] According to the comparison result of the fatigue degree value of the muscle group and the historical exercise data, an adaptive adjustment algorithm is used to dynamically update the initial fatigue threshold value, the adaptive adjustment algorithm optimizes the initial fatigue threshold value in real time based on the exercise characteristics and physiological characteristics of different muscle groups, and establishes a personalized fatigue threshold value corresponding to each muscle group;

[0154] The personalized fatigue threshold value is used as a monitoring reference standard to monitor the fatigue state of the muscle group in real time, and a fatigue warning is triggered when the fatigue degree value exceeds the personalized fatigue threshold value.

[0155] Real-time exercise data of the muscle group is collected through a sensor array. Exercise intensity data is obtained by measuring the amplitude of the electromyographic signal, and is calibrated in combination with force sensor data. Exercise duration is recorded by a high-precision timer, while the start and end time points of the exercise are marked. Exercise frequency is obtained by time-frequency analysis of the electromyographic signal, and signal denoising and feature extraction are performed through wavelet transform. These real-time data are arranged in a predetermined format as input parameters of the fatigue degree evaluation function.

[0156] Historical exercise data is retrieved from the database, and the historical exercise data includes historical performance records under the same exercise mode. Statistical analysis is performed on the historical data to establish a baseline data set. The baseline data set includes standard fatigue curves, typical fatigue development patterns and key time node characteristics under different exercise intensities. The real-time exercise data is compared with the baseline data set, and the fatigue degree value of each muscle group is calculated through the fatigue degree evaluation function.

[0157] During the fatigue degree evaluation process, the real-time data is normalized to make it comparable with the baseline data. Then, the basic fatigue accumulation rate is calculated based on the exercise intensity, and the basic fatigue accumulation rate increases nonlinearly with the exercise duration. At the same time, the modulating effect of exercise frequency on fatigue development is considered, high-frequency exercise accelerates fatigue accumulation, and low-frequency exercise provides recovery opportunities. These factors are comprehensively calculated to obtain the real-time fatigue degree value of each muscle group.

[0158] An adaptive adjustment algorithm is implemented, which can dynamically update the fatigue threshold value according to the characteristics of different muscle groups. A muscle characteristics database is established, which includes physiological characteristics of each muscle group, such as muscle fiber type proportion, energy metabolism characteristics, fatigue recovery rate, etc. Based on these characteristic parameters, a personalized fatigue development model is constructed. This model takes into account the fatigue resistance, recovery ability and work load bearing capacity of the muscle.

[0159] In the threshold optimization process, multiple adjustment factors are introduced. Different threshold adjustment strategies are used for strength muscle groups and endurance muscle groups. For strength muscle groups, the focus is on the impact of instantaneous load; for endurance muscle groups, more attention is paid to the effect of cumulative fatigue. At the same time, the training level of the muscle is considered, and the muscle group with high training level has higher fatigue tolerance, and the threshold is correspondingly increased.

[0160] The adaptive adjustment algorithm also contains a real-time feedback mechanism. When the muscle performance is detected to decline, the trend and rate of performance decline are determined through regression analysis. The threshold parameters are adjusted according to the decline characteristics to realize the dynamic optimization of the threshold. At the same time, a cross-validation mechanism is established, and the accuracy of the adjustment algorithm is continuously optimized by comparing the predicted fatigue degree and the actual fatigue performance.

[0161] An independent fatigue monitoring module is established for each muscle group. Real-time comparison is made between the current fatigue value and the personalized fatigue threshold. A hierarchical warning mechanism is designed, and different levels of warning signals are triggered when the fatigue degree reaches different percentages of the threshold. At the same time, the time, frequency and duration of the warning trigger are recorded, which are used for subsequent threshold optimization and training scheme adjustment.

[0162] Exemplarily, taking the biceps monitoring in weightlifting training as an example. The real-time data collected shows that the exercise intensity is maintained at 70% of the maximum muscle strength, the duration is 45 minutes, and the exercise frequency is 8 times per minute. The training data of the athlete in the past three months is retrieved as a reference benchmark. The fatigue degree value of the biceps is calculated to be 0.75 through the fatigue degree evaluation function. Considering that the athlete's biceps are mainly composed of fast muscle fibers and have a good training foundation, the adaptive adjustment algorithm adjusts the initial fatigue threshold 0.8 to 0.85. In subsequent monitoring, when the fatigue degree value reaches 0.82, the first level of warning is triggered, prompting the athlete to pay attention to adjusting the training intensity.

[0163] In this embodiment, by acquiring real-time exercise data of muscle groups and combining historical exercise data as a reference benchmark, accurate evaluation of muscle fatigue state is realized, dynamic optimization of fatigue threshold is realized based on adaptive adjustment algorithm, and the exercise characteristics and physiological characteristics of different muscle groups are fully considered, so that the threshold setting is more personalized and scientific, and different evaluation standards are used for different types of muscle groups, which better adapts to the physiological characteristics and working mode of different muscles, and can accurately identify the fatigue characteristics of each muscle group, providing more targeted guidance for exercise training and rehabilitation.

[0164] In a second aspect of the embodiment of the present application, an intelligent sportswear muscle fatigue degree dynamic evaluation system is provided, comprising:

[0165] The first unit is used for collecting muscle pressure data, muscle surface temperature data and muscle electrical signal data collected by sensors in the intelligent sports clothing as muscle state data.

[0166] The second unit is used for performing time-frequency domain decomposition on the muscle state data, extracting time domain features, frequency domain features and time-frequency joint features of each data, eliminating environmental noise and motion artifacts by combining an adaptive threshold filtering algorithm to obtain standard feature data, decomposing the standard feature data into a plurality of intrinsic mode functions based on an empirical mode decomposition method, and extracting instantaneous frequency and instantaneous amplitude of each intrinsic mode function by Hilbert-Huang transform to obtain a multi-dimensional feature vector representing a muscle activity state.

[0167] The third unit is used for constructing a dynamic connection network by using an adaptive neural synapse algorithm based on the multi-dimensional feature vector, analyzing non-linear features in the muscle group movement process based on the dynamic connection network and combining a recursive least squares algorithm to establish a fatigue degree evaluation function considering the interaction between muscle groups.

[0168] The fourth unit is used for determining fatigue degree values of each muscle group based on the fatigue degree evaluation function and historical movement data, dynamically updating a personalized fatigue threshold value according to the fatigue degree values and real-time monitoring, and issuing a warning signal if a muscle group exceeds the personalized fatigue threshold value.

[0169] In a third aspect, an electronic device is provided, including:

[0170] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0171] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0172] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0173] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A dynamic evaluation method for muscle fatigue of smart sportswear, characterized in that: include: Collect muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in smart sportswear as muscle status data; Decomposing the muscle state data in the time-frequency domain to extract the time-domain features, frequency-domain features, and joint time-frequency features of each data, eliminating environmental noise and motion artifacts using an adaptive threshold filtering algorithm to obtain standard feature data, decomposing the standard feature data into multiple intrinsic mode functions based on an empirical mode decomposition method, extracting the instantaneous frequency and instantaneous amplitude of each intrinsic mode function through a Hilbert-Huang transform, and obtaining a multidimensional feature vector representing the muscle activity state; Based on the multidimensional feature vector, an adaptive neural synaptic algorithm is used to construct a dynamic connection network. Based on the dynamic connection network, a recursive least squares algorithm is used to analyze the nonlinear characteristics of the muscle group movement process, and a fatigue assessment function that takes into account the interaction between muscle groups is established; The fatigue value of each muscle group is determined based on the fatigue evaluation function and historical exercise data, and the personalized fatigue threshold is dynamically updated according to the fatigue value and monitored in real time. If any muscle group exceeds the personalized fatigue threshold, an early warning signal is issued.

2. The method according to claim 1, characterized in that The muscle state data is decomposed in the time-frequency domain to extract the time-domain features, frequency-domain features and time-frequency joint features of each data. The standard feature data is obtained by combining the adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts. Obtain muscle pressure data, muscle surface temperature data, and electromyographic signal data during exercise and combine them to form multi-source motion state data, perform continuous wavelet transform on the multi-source motion state data based on a mother wavelet function, wherein the mother wavelet function includes a center frequency parameter, and obtain a time domain eigenvector, a frequency domain eigenvector, and a time-frequency joint eigenmatrix; performing morphological operations on the time-domain feature vector, the frequency-domain feature vector, and the time-frequency joint feature matrix, performing erosion and dilation operations on structural elements to obtain enhanced feature data, constructing an adaptive threshold function based on a local standard deviation and the number of sampling points of the enhanced feature data, and dynamically adjusting an adaptive coefficient of the adaptive threshold function according to a local signal-to-noise ratio; The enhanced feature data is decomposed into multiple intrinsic mode functions, the instantaneous frequency of each intrinsic mode function is calculated by Hilbert transform, the motion artifact component is identified and removed according to the instantaneous frequency, and the standard feature data is reconstructed. The signal-to-noise ratio, feature coherence and mean square error of the standard feature data are calculated and weighted calculation is performed in combination with preset weight coefficients to obtain a feature quality evaluation index. The standard feature data is screened based on the feature quality evaluation index and combined into a standard feature data set.

3. The method according to claim 1, characterized in that The standard feature data is decomposed into multiple intrinsic mode functions based on the empirical mode decomposition method. The instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform to obtain a multidimensional feature vector representing the muscle activity state, including: The data in the standard feature data set are subjected to the empirical mode decomposition method to obtain multiple intrinsic mode functions and residual terms, ensuring that the intrinsic mode function satisfies the condition that the difference between the number of extreme points and the number of zero-crossing points does not exceed 1, and the average value of the upper and lower envelopes formed by the local maximum points and the local minimum points is zero; the upper and lower envelopes of the intrinsic mode function are constructed using the cubic spline interpolation method, the mean envelope is calculated based on the upper and lower envelopes, and the iteration termination condition is determined according to the standard deviation of the results of two adjacent iterations; Performing a Hilbert-Huang transform on the intrinsic mode function, constructing an analytical signal including the intrinsic mode function and the Hilbert-Huang transform result, calculating the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the analytical signal, and forming a characteristic matrix with the instantaneous amplitude, the instantaneous phase, and the instantaneous frequency; The Hilbert marginal spectrum is introduced to analyze the characteristic matrix, the importance index of each characteristic component in the characteristic matrix is ​​calculated and high-importance characteristic components are selected. A biomechanical consistency loss function is constructed based on joint torque, muscle force and gravity terms. The muscle activity pattern and muscle movement link transmission relationship are extracted and the high-importance characteristic components are constructed into a multidimensional feature vector representing the muscle activity state. The mutual information entropy between the feature components in the multidimensional feature vector is calculated, feature redundancy is obtained based on the mutual information entropy, and redundant feature components are removed according to a preset threshold to obtain a multidimensional feature vector.

4. The method according to claim 3, characterized in that Introducing the Hilbert marginal spectrum to analyze the feature matrix, calculating the importance index of each feature component in the feature matrix, screening high-importance feature components based on the importance index, and constructing the high-importance feature components into a multidimensional feature vector representing the muscle activity state includes: Introducing the Hilbert marginal spectrum to analyze the characteristic matrix, calculating the Hilbert transform of each characteristic component in the characteristic matrix to obtain an analytical signal of each characteristic component, calculating the marginal spectrum based on the analytical signal, performing a double integration of the marginal spectrum in the time-frequency domain to obtain an importance index of each characteristic component, and screening high-importance characteristic components corresponding to the importance index according to a preset threshold; Acquire pre-collected joint motion parameters, calculate the Jacobian matrix, muscle force vector, damping term and gravity term according to the joint motion parameters, and combine them to obtain the joint torque; Constructing a biomechanical consistency loss function including a kinematic constraint term, a dynamic constraint term, and an energy efficiency constraint term, wherein the kinematic constraint term is constructed based on a joint angle, the dynamic constraint term is constructed based on the joint torque, and the energy efficiency constraint term is constructed based on energy consumption; Extracting a muscle activity pattern matrix based on the biomechanical consistency loss function, optimizing and calculating a synergy weight based on the muscle activity pattern matrix and the muscle force vector, obtaining a proximal joint angle and a distal joint angle, and constructing a kinematic chain transmission relationship between the proximal joint angle, the distal joint angle, and the muscle force vector; The high-importance feature components, the biomechanical consistency loss function, the synergy weight and the motion chain transfer relationship are combined to construct an initial feature vector representing the muscle activity state, the mutual information entropy between the feature components in the initial feature vector is calculated, and a multidimensional feature vector is obtained by screening based on a preset mutual information entropy threshold.

5. The method according to claim 1, wherein Based on the multidimensional feature vector, an adaptive neural synaptic algorithm is used to construct a dynamic connection network. Based on the dynamic connection network, the nonlinear characteristics of the muscle group movement process are analyzed in combination with a recursive least squares algorithm. A fatigue evaluation function that considers the interaction between muscle groups is established, including: An initial dynamic network is established using the multidimensional feature vector and an adaptive neural synapse algorithm. A multi-level synaptic plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network. A synaptic weight update value is obtained through a short-term plasticity dynamic equation. A synaptic connection strength update value is obtained through a long-term plasticity time-dependent plasticity rule. A synaptic growth probability value is obtained based on calcium ion concentration calculation. The synaptic weight update value, the synaptic connection strength update value, and the synaptic growth probability value are input into the initial dynamic network to obtain the dynamic connection network. Inputting the multidimensional feature vector into the dynamic connection network, optimizing the output result of the dynamic connection network using a recursive least squares algorithm, calculating a network output error value in combination with an error update equation, calculating a network parameter adjustment value using a gain matrix update equation, and optimizing the dynamic connection network according to the network output error value and the network parameter adjustment value to obtain an optimized output result; A neuromodulator concentration kinetic equation is established, a release term, a degradation term, and a diffusion term are set in the neuromodulator concentration kinetic equation, a neuromodulator concentration distribution value is calculated according to the neuromodulator concentration kinetic equation, a single muscle fatigue characteristic term and a muscle group interaction term are set based on the optimization output result and the neuromodulator concentration distribution value, and a fatigue assessment function is constructed.

6. The method according to claim 5, characterized in that Inputting the multidimensional feature vector into the dynamic connection network, optimizing the output result of the dynamic connection network by a recursive least squares algorithm, and calculating the network output error value by combining the error update equation includes: Obtaining a multidimensional eigenvector, inputting the multidimensional eigenvector into a dynamically connected network, using a three-dimensional nonlinear equation system to describe the network state evolution process, calculating the network disturbance sensitivity, bifurcation control parameters, and periodic characteristic parameters and performing integration operations to obtain the system phase space trajectory and calculate the Lyapunov exponent, correlation dimension, and bifurcation parameter, and combining them to obtain the chaotic eigenvector; Constructing a nonlinear constraint function including a state stability term, an orbital periodicity term, and a parameter sensitivity term based on the chaotic eigenvector, wherein the state stability term adopts an exponential function form of the Lyapunov index, the orbital periodicity term adopts a sinusoidal function form of the correlation dimension, and the parameter sensitivity term adopts a hyperbolic tangent function form of the bifurcation parameter; Integrating the nonlinear constraint function into a recursive least squares algorithm, calculating a parameter vector, a gain matrix, and a regression vector using the recursive least squares algorithm, dynamically adjusting a parameter search strategy based on the chaotic eigenvector, calculating a search step size based on the Lyapunov exponent, optimizing a search path using the correlation dimension, and adjusting an update rate based on the bifurcation parameter; The output result of the dynamically connected network is optimized by the recursive least squares algorithm using the search step size, the search path and the update rate, the nonlinear constraint function is multiplied by the difference between the actual output and the predicted output of the network, and the network output error value is calculated in combination with the error update equation.

7. The method according to claim 1, characterized in that Determining fatigue values ​​for each muscle group based on the fatigue evaluation function and historical exercise data, dynamically updating personalized fatigue thresholds based on the fatigue values, and performing real-time monitoring includes: Acquiring real-time motion data of a muscle group, the real-time motion data including the intensity, duration, and frequency of the muscle group's motion, and inputting the real-time motion data into a fatigue assessment function; calculating fatigue values ​​for each muscle group based on the fatigue assessment function, combined with historical motion data as a reference; Dynamically updating the initial fatigue threshold using an adaptive adjustment algorithm based on a comparison result between the fatigue value of the muscle group and the historical motion data. The adaptive adjustment algorithm optimizes the initial fatigue threshold in real time based on the motion characteristics and physiological features of different muscle groups to establish a personalized fatigue threshold for each muscle group. The personalized fatigue threshold is used as a monitoring reference standard to monitor the fatigue state of the muscle group in real time, and a fatigue warning is triggered when the fatigue value exceeds the personalized fatigue threshold.

8. Intelligent sportswear muscle fatigue dynamic evaluation system, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect muscle pressure data, muscle surface temperature data and muscle electrical signal data collected by sensors in the smart sportswear as muscle state data; The second unit is used to perform time-frequency domain decomposition on the muscle state data, extract the time domain features, frequency domain features and time-frequency joint features of each data, combine the adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts to obtain standard feature data, decompose the standard feature data into multiple intrinsic mode functions based on the empirical mode decomposition method, extract the instantaneous frequency and instantaneous amplitude of each intrinsic mode function through the Hilbert-Huang transform, and obtain a multidimensional feature vector representing the muscle activity state; A third unit is configured to construct a dynamic connection network based on the multidimensional feature vector using an adaptive neural synaptic algorithm, analyze the nonlinear characteristics of the muscle group movement process based on the dynamic connection network in combination with a recursive least squares algorithm, and establish a fatigue assessment function that takes into account the interaction between muscle groups; The fourth unit is used to determine the fatigue value of each muscle group based on the fatigue evaluation function and historical exercise data, dynamically update the personalized fatigue threshold according to the fatigue value and monitor in real time, and issue a warning signal if any muscle group exceeds the personalized fatigue threshold.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images