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] This invention relates to the field of intelligent sportswear technology, and in particular to a method and system for dynamic assessment of muscle fatigue in intelligent sportswear. Background Technology

[0002] As a new type of equipment that can monitor the human body's movement status in real time, smart sportswear has gradually attracted the attention of sports enthusiasts and professional athletes. Smart sportswear can collect a variety of physiological signals of the human body during exercise, such as muscle pressure, surface temperature and electromyography signals, providing an important basis for assessing muscle fatigue during exercise. Muscle fatigue is a common physiological phenomenon during exercise. Excessive fatigue may lead to sports injuries, affect exercise performance, and even cause health risks. Therefore, accurate assessment of muscle fatigue is of great significance for scientifically arranging training intensity and preventing sports injuries. Existing muscle fatigue assessment methods in the field of smart sportswear still have problems such as lack of comprehensive utilization of multi-source heterogeneous data, difficulty in fully reflecting the complex physiological changes of muscle fatigue, limited ability to suppress environmental noise and motion artifacts, resulting in insufficient accuracy and reliability of fatigue assessment results, and neglect of the synergistic effect between muscle groups and individual differences. Therefore, a solution is urgently needed to address the problems existing in the current technology. Summary of the Invention

[0003] This invention provides a method and system for dynamic assessment of muscle fatigue in intelligent sportswear, which can at least solve some of the problems existing in the prior art.

[0004] A first aspect of this invention provides a method for dynamically assessing muscle fatigue in smart sportswear, comprising: The muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in smart sportswear are used as muscle state data. 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 point. An adaptive threshold filtering algorithm is used to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions. The instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. Based on the multidimensional feature vector, an adaptive neural synapse 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 by combining the recursive least squares algorithm, and a fatigue evaluation function considering the interaction between muscle groups is established. The fatigue level of each muscle group is determined based on the fatigue assessment function and historical exercise data. The personalized fatigue threshold is dynamically updated and monitored in real time based on the fatigue level value. If any muscle group exceeds the personalized fatigue threshold, an early warning signal is issued.

[0005] In one alternative implementation, The muscle state data is decomposed in the time-frequency domain to extract the time-domain features, frequency-domain features, and joint time-frequency features of each data point. An adaptive threshold filtering algorithm is then used to eliminate environmental noise and motion artifacts to obtain standard feature data, including: Muscle pressure data, muscle surface temperature data, and electromyographic signal data during exercise are acquired and combined to form multi-source motion state data. Continuous wavelet transform is performed on the multi-source motion state data based on a mother wavelet function, which includes a center frequency parameter, to obtain time-domain feature vectors, frequency-domain feature vectors, and a time-frequency joint feature matrix. Morphological operations are performed on the time-domain feature vector, the frequency-domain feature vector, and the time-frequency joint feature matrix. Erosion and dilation operations are performed through the structuring element to obtain enhanced feature data. An adaptive threshold function is constructed based on the local standard deviation and the number of sampling points of the enhanced feature data. The adaptive coefficient of the adaptive threshold function is dynamically adjusted according to the local signal-to-noise ratio. The enhanced feature data is decomposed into multiple intrinsic mode functions (IMFs). The instantaneous frequency of each IMF is calculated using Hilbert transform. Motion artifacts are identified and removed based on the instantaneous frequencies to reconstruct standard feature data. The signal-to-noise ratio, feature coherence, and mean square error of the standard feature data are calculated and weighted by a preset weighting coefficient to obtain a feature quality evaluation index. The standard feature data is then selected based on the feature quality evaluation index and combined to form a standard feature dataset.

[0006] In one alternative implementation, The standard feature data is decomposed into multiple intrinsic mode functions (IMFs) based on the empirical mode decomposition method. The instantaneous frequency and instantaneous amplitude of each IMF are extracted using the Hilbert-Huang transform to obtain a multidimensional feature vector representing the muscle activity state, including: The data in the standard feature dataset are processed by empirical mode decomposition to obtain multiple intrinsic mode functions and residual terms. It is ensured that the intrinsic mode functions satisfy the following conditions: 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 maxima and local minima is zero. The upper and lower envelopes of the intrinsic mode functions are constructed using cubic spline interpolation. The mean envelope is calculated based on the upper and lower envelopes. The iteration termination condition is determined based on the standard deviation of the results of two adjacent iterations. Perform a Hilbert-Huang transform on the intrinsic mode function to construct an analytic signal containing the intrinsic mode function and the Hilbert-Huang transform result. Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the analytic signal, and combine the instantaneous amplitude, instantaneous phase, and instantaneous frequency to form a feature matrix. Hilbert marginal spectrum analysis is introduced into the feature matrix to calculate the importance index of each feature component in the feature matrix and screen out the high importance feature components. A biomechanical consistency loss function is constructed based on joint torque, muscle force and gravity terms to extract muscle activity patterns and muscle movement link transmission relationships and construct the high importance feature components into a multidimensional feature vector characterizing the muscle activity state. Calculate the mutual information entropy between the feature components in the multidimensional feature vector, obtain the feature redundancy based on the mutual information entropy, remove redundant feature components according to a preset threshold, and obtain the multidimensional feature vector.

[0007] In one alternative implementation, The feature matrix is ​​analyzed using Hilbert marginal spectrum analysis. Importance indices are calculated for each feature component in the feature matrix. High-importance feature components are selected based on these importance indices, and these high-importance feature components are used to construct a multi-dimensional feature vector representing muscle activity state, including: The feature matrix is ​​analyzed by introducing Hilbert marginal spectrum analysis, the Hilbert transform of each feature component in the feature matrix is ​​calculated, the analytical signal of each feature component is obtained, the marginal spectrum is calculated based on the analytical signal, the marginal spectrum is double integrated in the time-frequency domain to obtain the importance index of each feature component, and the high importance feature components corresponding to the importance index are selected according to a preset threshold. Acquire pre-collected joint motion parameters, calculate the Jacobian matrix, muscle force vector, damping term, and gravity term based on the joint motion parameters, and combine them to obtain the joint torque; A biomechanical consistency loss function is constructed, which includes kinematic constraints, dynamic constraints, and energy efficiency constraints, wherein the kinematic constraints are constructed based on joint angles, the dynamic constraints are constructed based on joint torques, and the energy efficiency constraints are constructed based on energy consumption. Based on the biomechanical consistency loss function, a muscle activity pattern matrix is ​​extracted. Based on the muscle activity pattern matrix and the muscle force vector, a collaborative weight is optimized and calculated to obtain the proximal joint angle and the distal joint angle. The proximal joint angle, the distal joint angle, and the muscle force vector are used to construct a kinetic chain transmission relationship. The high-importance feature components, the biomechanical consistency loss function, the synergistic weights, and the kinetic chain transmission 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 filtering based on a preset mutual information entropy threshold.

[0008] In one alternative implementation, Based on the multidimensional feature vectors, an adaptive neural synapse algorithm is used to construct a dynamic connectivity network. Based on this dynamic connectivity network, a recursive least squares algorithm is used to analyze the nonlinear characteristics of muscle group movement, and a fatigue assessment function considering the interaction between muscle groups is established, including: Using the multidimensional feature vectors, an initial dynamic network is established through an adaptive neural synapse algorithm. A multi-level synaptic plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network. Synaptic weight update values ​​are obtained through short-term plasticity dynamic equations, and synaptic connection strength update values ​​are obtained through long-term plasticity time-dependent plasticity rules. Synaptic growth probability values ​​are calculated based on calcium ion concentration. The synaptic weight update values, synaptic connection strength update values, and synaptic growth probability values ​​are input into the initial dynamic network to obtain the dynamic connection network. The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized by the recursive least squares algorithm. The network output error value is calculated by combining the error update equation, and the network parameter adjustment value is calculated by using the gain matrix update equation. The dynamic connection network is optimized based on the network output error value and the network parameter adjustment value to obtain the optimized output result. A neurotransmitter concentration kinetic equation is established, in which release, degradation, and diffusion terms are set. The neurotransmitter concentration distribution value is calculated based on the neurotransmitter concentration kinetic equation. Based on the optimized output result and the neurotransmitter concentration distribution value, a single muscle fatigue characteristic term and a muscle group interaction term are set to construct a fatigue evaluation function.

[0009] In one alternative implementation, The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized using a recursive least squares algorithm. The network output error value is calculated by combining the error update equation, including: A multidimensional feature vector is obtained, and the multidimensional feature vector is input into a dynamic connection network. The network state evolution process is described by a three-dimensional nonlinear equation system. The network disturbance sensitivity, bifurcation control parameters and periodic characteristic parameters are calculated and integrated to obtain the system phase space trajectory. The Lyapunov exponent, correlation dimension and bifurcation parameters are calculated and combined to obtain the chaotic feature vector. Based on the chaotic feature vector, a nonlinear constraint function is constructed that includes a state stability term, an orbital periodicity term, and a parameter sensitivity term. The state stability term adopts the exponential function form of the Lyapunov exponent, the orbital periodicity term adopts the sine function form of the correlation dimension, and the parameter sensitivity term adopts the hyperbolic tangent function form of the bifurcation parameter. The nonlinear constraint function is integrated into the recursive least squares algorithm. The parameter vector, gain matrix and regression vector are calculated by the recursive least squares algorithm. The parameter search strategy is dynamically adjusted based on the chaotic feature vector. The search step size is calculated according to the Lyapunov exponent. The search path is optimized using the correlation dimension. The update rate is adjusted based on the bifurcation parameter. The output of the dynamic connection network is optimized using the recursive least squares algorithm based on 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 by combining the error update equation.

[0010] In one alternative implementation, Based on the fatigue assessment function and historical exercise data, the fatigue values ​​of each muscle group are determined, and personalized fatigue thresholds are dynamically updated and monitored in real time according to the fatigue values, including: Real-time motion data of muscle groups is acquired, including the intensity, duration, and frequency of the muscle group's movement. The real-time motion data is then input into a fatigue assessment function. Historical motion data is used as a reference benchmark, and the fatigue value of each muscle group is calculated based on the fatigue assessment function. Based on the comparison between the fatigue value of the muscle group and the historical exercise data, the initial fatigue threshold is dynamically updated using an adaptive adjustment algorithm. The adaptive adjustment algorithm optimizes the initial fatigue threshold in real time based on the movement characteristics and physiological features of different muscle groups, and establishes a personalized fatigue threshold for each muscle group. The personalized fatigue threshold is used as a monitoring reference standard to monitor the fatigue status of the muscle group in real time. When the fatigue value exceeds the personalized fatigue threshold, a fatigue warning is triggered.

[0011] A second aspect of the present invention provides an intelligent sportswear muscle fatigue dynamic assessment system, comprising: The first unit is used to collect muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in 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, and combine the adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted through Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. The third unit is used to construct a dynamic connection network based on the multidimensional feature vector using an adaptive neural synapse algorithm, and based on the dynamic connection network, analyze the nonlinear characteristics of the muscle group movement process using a recursive least squares algorithm to establish a fatigue evaluation function that considers 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 it in real time, and issue an early warning signal if any muscle group exceeds the personalized fatigue threshold.

[0012] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] In this invention, muscle state data collected by sensors in smart sportswear is acquired. Combined with time-frequency domain decomposition technology and an adaptive threshold filtering algorithm, multidimensional feature vectors representing muscle activity are accurately extracted. This effectively eliminates environmental noise and motion artifacts, improving the accuracy and reliability of the feature data. An adaptive neural synapse algorithm is used to construct a dynamic connection network, and a recursive least squares algorithm is combined to analyze the nonlinear characteristics of muscle group movement. A fatigue assessment function considering the interaction between muscle groups is established, comprehensively reflecting the synergistic relationship between different muscle groups and improving the accuracy of fatigue assessment. Based on the fatigue assessment function and historical exercise data, muscle group fatigue values ​​are determined, and personalized fatigue thresholds are dynamically updated and monitored in real time. When a muscle group exceeds the personalized fatigue threshold, an early warning signal is issued. This achieves real-time assessment and early warning of muscle fatigue during exercise, effectively reducing the risk of sports injuries and providing a scientific basis for the development of personalized exercise programs. Attached Figure Description

[0015] Figure 1This is a flowchart illustrating the dynamic assessment method for muscle fatigue in intelligent sportswear according to an embodiment of the present invention. Figure 2 This is a comparison chart showing the optimization of the biomechanical consistency loss function in the dynamic assessment method for muscle fatigue of intelligent sportswear according to an embodiment of the present invention. Figure 3 This is a dynamic reconstruction diagram of the network structure of the intelligent sportswear muscle fatigue dynamic assessment method according to an embodiment of the present invention. Figure 4 This is a comparison chart of error convergence of the dynamic assessment method for muscle fatigue in intelligent sportswear according to an embodiment of the present invention; Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0018] Figure 1 This is a flowchart illustrating the dynamic assessment method for muscle fatigue in intelligent sportswear according to an embodiment of the present invention. Figure 1 As shown, the method includes: The muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in smart sportswear are used as muscle state data. 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 point. An adaptive threshold filtering algorithm is used to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions. The instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. Based on the multidimensional feature vector, an adaptive neural synapse 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 by combining the recursive least squares algorithm, and a fatigue evaluation function considering the interaction between muscle groups is established. The fatigue level of each muscle group is determined based on the fatigue assessment function and historical exercise data. The personalized fatigue threshold is dynamically updated and monitored in real time based on the fatigue level value. If any muscle group exceeds the personalized fatigue threshold, an early warning signal is issued.

[0019] In one alternative implementation, The muscle state data is decomposed in the time-frequency domain to extract the time-domain features, frequency-domain features, and joint time-frequency features of each data point. An adaptive threshold filtering algorithm is then used to eliminate environmental noise and motion artifacts to obtain standard feature data, including: Muscle pressure data, muscle surface temperature data, and electromyographic signal data during exercise are acquired and combined to form multi-source motion state data. Continuous wavelet transform is performed on the multi-source motion state data based on a mother wavelet function, which includes a center frequency parameter, to obtain time-domain feature vectors, frequency-domain feature vectors, and a time-frequency joint feature matrix. Morphological operations are performed on the time-domain feature vector, the frequency-domain feature vector, and the time-frequency joint feature matrix. Erosion and dilation operations are performed through the structuring element to obtain enhanced feature data. An adaptive threshold function is constructed based on the local standard deviation and the number of sampling points of the enhanced feature data. The adaptive coefficient of the adaptive threshold function is dynamically adjusted according to the local signal-to-noise ratio. The enhanced feature data is decomposed into multiple intrinsic mode functions (IMFs). The instantaneous frequency of each IMF is calculated using Hilbert transform. Motion artifacts are identified and removed based on the instantaneous frequencies to reconstruct standard feature data. The signal-to-noise ratio, feature coherence, and mean square error of the standard feature data are calculated and weighted by a preset weighting coefficient to obtain a feature quality evaluation index. The standard feature data is then selected based on the feature quality evaluation index and combined to form a standard feature dataset.

[0020] Real-time data during exercise is 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 electromyography (EMG) sensor. These three types of data are aligned and combined according to time series to form multi-source motion state data containing various physiological information. A suitable mother wavelet function with an adjustable center frequency parameter is selected for physiological signal analysis. This mother wavelet function is used to perform continuous wavelet transform on the multi-source motion state data, extracting time-domain feature vectors reflecting the signal's temporal characteristics, frequency-domain feature vectors reflecting its frequency characteristics, and a time-frequency joint feature matrix reflecting the combined time-frequency characteristics.

[0021] Morphological optimization is performed on the extracted feature data. By designing appropriate structuring elements, erosion and dilation operations are applied to the time-domain feature vector, frequency-domain feature vector, and time-frequency joint feature matrix, respectively. Erosion eliminates noise interference in the data, while dilation enhances the effective feature information. The combination of these two operations yields enhanced feature data. For the enhanced feature data, the standard deviation within its local region is calculated, and an adaptive threshold function is constructed based on the actual number of sampling points. By analyzing the signal-to-noise ratio of local data segments, the adaptive coefficients in the adaptive threshold function are adjusted in real time, enabling the threshold function to adapt to changes in data characteristics.

[0022] The enhanced feature data is subjected to intrinsic mode decomposition (IMD) to obtain IMD functions for different frequency features. A Hilbert transform is performed on each IMD function to calculate the instantaneous frequency characteristics. By analyzing the variation patterns of the instantaneous frequencies, artifacts caused by motion are identified and removed from the original data. The remaining effective components are then reconstructed to obtain standard feature data. The standard feature data is then evaluated for quality by calculating the signal-to-noise ratio (SNR), the coherence between features, and the mean square error (MSE). These indicators are weighted with pre-defined weighting coefficients to obtain feature quality assessment indicators reflecting data quality. Based on these indicator selections, the standard feature data is filtered, and high-quality feature data are combined to form a standard feature dataset.

[0023] For example, taking arm flexion and extension movements as an example, data is collected through sensors in sportswear. Pressure sensors collect pressure change data of the biceps and triceps muscles, reflecting muscle contraction and relaxation; temperature sensors collect changes in muscle surface temperature, reflecting heat generation during exercise; and electromyography (EMG) sensors collect muscle discharge signals, displaying muscle activity intensity. The collected pressure data (range 0-200 kPa), temperature data (range 30-40℃), and EMG signals (range 0-5 mV) are aligned at 100 sampling points per second. The Mexican cap wavelet is selected as the mother wavelet function, with a center frequency set to 2 Hz, resulting in a time-domain feature vector (reflecting signal amplitude changes), a frequency-domain feature vector (reflecting signal frequency composition), and a time-frequency joint feature matrix (reflecting signal time-frequency variation characteristics). Morphological processing is performed using 3×3 pixel rectangular structuring elements to enhance signal features. In the calculated intrinsic mode functions, low-frequency artifacts of 2-3 Hz caused by arm movement are identified, removed, and the signal is reconstructed. The final standard feature dataset contains effective features reflecting the degree of muscle fatigue, with a 40% improvement in signal-to-noise ratio, a feature coherence of 0.85, and a mean squared error reduced to 30% of the original data.

[0024] In this embodiment, a multi-source data fusion processing method is used to achieve collaborative analysis of muscle pressure, surface temperature and electromyography signals. Compared with a single data source, the comprehensiveness and reliability of feature extraction are improved. By combining erosion and dilation operations in morphological operations, the integrity of effective features is maintained while suppressing noise. Based on the combination of intrinsic mode decomposition and Hilbert transform, the accurate identification and removal of motion artifacts are achieved.

[0025] In one alternative implementation, The standard feature data is decomposed into multiple intrinsic mode functions (IMFs) based on the empirical mode decomposition method. The instantaneous frequency and instantaneous amplitude of each IMF are extracted using the Hilbert-Huang transform to obtain a multidimensional feature vector representing the muscle activity state, including: The data in the standard feature dataset are processed by empirical mode decomposition to obtain multiple intrinsic mode functions and residual terms. It is ensured that the intrinsic mode functions satisfy the following conditions: 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 maxima and local minima is zero. The upper and lower envelopes of the intrinsic mode functions are constructed using cubic spline interpolation. The mean envelope is calculated based on the upper and lower envelopes. The iteration termination condition is determined based on the standard deviation of the results of two adjacent iterations. Perform a Hilbert-Huang transform on the intrinsic mode function to construct an analytic signal containing the intrinsic mode function and the Hilbert-Huang transform result. Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the analytic signal, and combine the instantaneous amplitude, instantaneous phase, and instantaneous frequency to form a feature matrix. Hilbert marginal spectrum analysis is introduced into the feature matrix to calculate the importance index of each feature component in the feature matrix and screen out the high importance feature components. A biomechanical consistency loss function is constructed based on joint torque, muscle force and gravity terms to extract muscle activity patterns and muscle movement link transmission relationships and construct the high importance feature components into a multidimensional feature vector characterizing the muscle activity state. Calculate the mutual information entropy between the feature components in the multidimensional feature vector, obtain the feature redundancy based on the mutual information entropy, remove redundant feature components according to a preset threshold, and obtain the multidimensional feature vector.

[0026] Starting with a standard feature dataset, the Empirical Mode Decomposition (EMD) method is used for data processing. Local extrema are detected in the data sequence, identifying all local maxima and local minima. The location of extrema is determined through point-by-point comparison. Adjacent data points are compared to the center point; points where the center point is greater than the points on either side are marked as local maxima, and points where the center point is less than the points on either side are marked as local minima. Simultaneously, zero-crossing points are detected in the data sequence, recording the positions where the signal changes from positive to negative or vice versa. Throughout the decomposition process, the relationship between the number of extrema and the number of zero-crossing points is continuously monitored to ensure that the difference does not exceed 1.

[0027] Envelopes are constructed for the detected sets of local maxima and local minima. Cubic spline interpolation is used to connect the discrete extrema into smooth envelopes. During interpolation, boundary conditions and node parameters are adjusted to ensure that the generated envelopes accurately reflect data fluctuations without causing excessive oscillations. The average of the upper and lower envelopes is calculated to obtain the mean envelope. The mean envelope is subtracted from the original data to obtain a new data sequence. This process is repeated for this new sequence until a component satisfying the intrinsic mode function (IMF) condition is obtained. During the iteration, the standard deviation between the new sequence obtained in each decomposition and the previous sequence is calculated to establish a standard deviation sequence. When the standard deviation obtained after several consecutive iterations is lower than a preset threshold, the current component is considered to satisfy the IMF requirement and is extracted. The remaining signal is then further decomposed to obtain a series of IMFs and a residual term.

[0028] Each intrinsic mode function is input into the Hilbert-Huang transform processing flow. First, a Hilbert transformer is constructed to perform a Fourier transform on the input signal, setting negative frequency components to zero and doubling positive frequency components, followed by an inverse Fourier transform to obtain the imaginary part of the analytic signal. The original signal is used as the real part, and together with the calculated imaginary part, a complex-form analytic signal is constructed. A polar coordinate transformation is performed on the analytic signal to extract the instantaneous amplitude and phase information. The instantaneous frequency is calculated using the time difference of the phase signal, and the difference result is smoothed to eliminate abrupt changes. The extracted instantaneous amplitude, instantaneous phase, and instantaneous frequency are aligned in time to form a feature matrix.

[0029] Hilbert marginal spectrum analysis was performed on the feature matrix to calculate the energy distribution at each time-frequency point. The time-frequency plane was divided into multiple sub-regions, and the energy within each sub-region was accumulated and statistically analyzed to obtain the frequency marginal spectrum and the time marginal spectrum. Based on the distribution characteristics of the marginal spectrum, the energy concentration, frequency stability, and temporal persistence of each feature component were calculated, and these indicators were weighted and combined to obtain a feature importance score. High-importance feature components were selected based on their scores. Simultaneously, a biomechanical consistency loss function was constructed, comprising three parts: joint torque term, muscle force term, and gravity term. By minimizing the loss function, the muscle contraction-relaxation activity pattern and the force transmission relationship between muscle groups were extracted. The selected high-importance feature components were fused with the biomechanical features to construct a multi-dimensional feature vector.

[0030] Redundancy analysis is performed on the multidimensional feature vectors. The mutual information entropy calculation method is 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. Based on these distributions, the mutual information entropy is calculated. A feature redundancy matrix is ​​constructed, where each element represents the degree of redundancy between 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. This process of gradually removing redundant features yields a simplified multidimensional feature vector.

[0031] For example, consider the analysis of thigh muscle group activity during running. Empirical Mode Decomposition (EMD) is performed on the collected standard feature dataset, yielding eight intrinsic mode functions (IMFs) and one residual term. Each IMF strictly satisfies characteristic requirements; for instance, the first IMF contains 120 extreme points and 119 zero-crossing points, and the mean of its upper and lower envelopes consistently fluctuates around zero. The envelope constructed using cubic spline interpolation accurately reflects the signal's fluctuation characteristics. After 15 iterations, the standard deviation of adjacent iterations drops below 0.001, reaching the termination condition.

[0032] Hilbert-Huang transform was performed on these intrinsic mode functions to obtain analytical signals reflecting the contraction characteristics of the thigh muscles. The extracted instantaneous amplitudes show that muscle contraction intensity peaks at the start of the run and then exhibits periodic changes; the instantaneous phase reflects the coordination characteristics during the alternation of left and right legs; and the instantaneous frequency shows the dynamic changes in stride frequency. Through Hilbert marginal spectrum analysis, 12 highly important feature components were selected from the original feature matrix. Combining hip joint torque (peak value approximately three times body weight), 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 these, a 9-dimensional feature vector was obtained, which completely preserves the key information of the muscle activity state.

[0033] In this embodiment, the accuracy of intrinsic mode function decomposition is ensured by a strict extreme point and zero-crossing point detection mechanism. The instantaneous features of the signal are accurately extracted by Hilbert-Huang transform, which not only obtains amplitude information but also accurately captures the dynamic changes of phase and frequency. By constructing a biomechanical consistency loss function, mechanical characteristics and physiological characteristics are organically combined to achieve a multi-dimensional representation of muscle activity.

[0034] In one alternative implementation, The feature matrix is ​​analyzed using Hilbert marginal spectrum analysis. Importance indices are calculated for each feature component in the feature matrix. High-importance feature components are selected based on these importance indices, and these high-importance feature components are used to construct a multi-dimensional feature vector representing muscle activity state, including: The feature matrix is ​​analyzed by introducing Hilbert marginal spectrum analysis, the Hilbert transform of each feature component in the feature matrix is ​​calculated, the analytical signal of each feature component is obtained, the marginal spectrum is calculated based on the analytical signal, the marginal spectrum is double integrated in the time-frequency domain to obtain the importance index of each feature component, and the high importance feature components corresponding to the importance index are selected according to a preset threshold. Acquire pre-collected joint motion parameters, calculate the Jacobian matrix, muscle force vector, damping term, and gravity term based on the joint motion parameters, and combine them to obtain the joint torque; A biomechanical consistency loss function is constructed, which includes kinematic constraints, dynamic constraints, and energy efficiency constraints, wherein the kinematic constraints are constructed based on joint angles, the dynamic constraints are constructed based on joint torques, and the energy efficiency constraints are constructed based on energy consumption. Based on the biomechanical consistency loss function, a muscle activity pattern matrix is ​​extracted. Based on the muscle activity pattern matrix and the muscle force vector, a collaborative weight is optimized and calculated to obtain the proximal joint angle and the distal joint angle. The proximal joint angle, the distal joint angle, and the muscle force vector are used to construct a kinetic chain transmission relationship. The high-importance feature components, the biomechanical consistency loss function, the synergistic weights, and the kinetic chain transmission 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 filtering based on a preset mutual information entropy threshold.

[0035] This paper introduces the Hilbert marginal spectrum analysis method to process the feature matrix. First, a Hilbert transform is performed on each feature component of the matrix. During the transformation, an orthogonal transform operator is constructed to map the original signal onto the complex plane, obtaining the analytic signal corresponding to each feature component. Time-frequency analysis is then performed on the analytic signal to calculate the instantaneous frequency and instantaneous amplitude, constructing a time-frequency energy distribution map. Marginal spectrum calculation is performed on the time-frequency plane, integrating the time-frequency energy distribution map in both the time and frequency dimensions to obtain the time marginal spectrum and the frequency marginal spectrum. Double integration is then performed on the two marginal spectra over the entire time-frequency domain to obtain a quantitative index characterizing the importance of each feature component. A feature importance threshold is set, and feature components with importance indices higher than the threshold are selected as high-importance feature components.

[0036] Joint motion parameters, including joint position, velocity, and acceleration information, are obtained from a pre-established motion database. Based on these parameters, a Jacobian matrix is ​​calculated, describing the mapping between joint space and Cartesian space. Simultaneously, force vectors generated by each muscle are calculated, considering muscle contraction characteristics and lever arm variations. The damping effect during joint movement is incorporated, and velocity-related damping terms are calculated. The influence of gravity on each joint is considered, and gravity terms are calculated. The Jacobian matrix, muscle force vectors, damping terms, and gravity terms are then combined to obtain a complete expression of joint torques.

[0037] A biomechanical consistency loss function is constructed, comprising three main constraint terms. The kinematic constraint term is based on joint angles, ensuring the accuracy of the motion trajectory by calculating the deviation between predicted and actual angles. The dynamic constraint term is based on joint torques, ensuring force balance by comparing the differences between calculated and measured torques. The energy efficiency constraint term is based on the energy expenditure of muscle activity, optimizing motion efficiency by minimizing overall energy consumption. These constraint terms are combined using weighting coefficients to form a unified loss function.

[0038] Muscle activity patterns are extracted using a biomechanical consistency loss function. By minimizing the loss function, a series of feature matrices describing the coordinated contraction patterns of muscles are obtained. These matrices are combined with previously calculated muscle force vectors, and an optimization algorithm is used to calculate the coordination weights between different muscles. These weights reflect the coordination relationship between muscle groups. Based on the kinetic chain principle, the motion angles of proximal joints (such as the hip joint) and distal joints (such as the ankle joint) are analyzed to establish the motion transmission relationship between joints. The proximal joint angles, distal joint angles, and muscle force vectors are correlated to construct a complete kinetic chain transmission relationship model.

[0039] The selected high-importance feature components are fused with biomechanical features. The calculation results of the biomechanical consistency loss function, muscle synergy weights, and kinetic chain transmission relationships are integrated to construct an initial feature vector. The correlation between the components in this feature vector is analyzed, and the mutual information entropy method is used to quantify the degree of information redundancy between features. A feature correlation matrix is ​​established by calculating the mutual information entropy between each pair of features. A mutual information entropy threshold is set; when the mutual information entropy between a pair of features exceeds the threshold, the feature with higher information content is retained, and redundant features are deleted. Through this selection process, an optimized multidimensional feature vector is obtained.

[0040] For example, consider the analysis of the shooting motion in basketball. Upper limb joint motion data, including motion parameters of the shoulder, elbow, and wrist joints, are acquired using a motion capture system. Hilbert marginal spectrum analysis is performed on the collected feature matrix, selecting 12 high-importance feature components from the original 30 components. These features are mainly concentrated in the shoulder abduction and elbow extension phases. Joint torques calculated based on the motion data show 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. By minimizing the loss function, four main muscle synergy patterns are extracted, with the deltoid and biceps having the highest synergy weight, reaching 0.8. Analysis of the shooting motion's kinetic chain reveals that the transmission efficiency of shoulder joint movement driving elbow joint movement reaches 85%. Through mutual information entropy analysis, the 12 feature components are optimized to 8, forming the final multidimensional feature vector, which accurately represents the key muscle activity features in the shooting motion.

[0041] In this embodiment, a comprehensive capture of muscle activity features is achieved by performing a systematic time-frequency analysis on the feature matrix. A unified optimization framework is established by organically combining kinematic constraints, dynamic constraints, and energy efficiency constraints. Muscle activity patterns are extracted through a biomechanical consistency loss function. A complete kinetic chain transmission relationship is established by optimizing the calculation of collaborative weights based on muscle force vectors. A feature screening mechanism based on mutual information entropy is introduced. In existing technologies, the analysis of muscle activity states usually adopts a single signal processing method, focusing only on time domain or frequency domain features. This makes it difficult to fully capture the dynamic characteristics of muscle activity, separates biomechanical features from signal features for analysis, and lacks an effective feature fusion mechanism. In the feature extraction process, a fixed threshold screening method is generally used, which cannot adapt to feature changes under different motion states, resulting in unstable feature quality. This embodiment not only extracts instantaneous features of signals, but also accurately assesses the importance of features through marginal spectrum analysis, significantly improving the accuracy and reliability of feature extraction. It considers not only the accuracy of motion, but also energy efficiency, making the extracted features more consistent with the actual laws of human movement. By setting an adaptive mutual information entropy threshold, it achieves intelligent simplification of the feature set, significantly reducing the feature dimensionality while maintaining information integrity, and providing more reliable technical support for the accurate analysis of muscle activity states.

[0042] Figure 2 This diagram compares the optimization processes of the biomechanical consistency loss function in the intelligent sportswear muscle fatigue dynamic assessment method of this invention, illustrating the optimization process of three different methods. This technical solution employs a multi-constraint biomechanical loss function, weighting kinematic constraints, dynamic constraints, and energy efficiency constraints with weights of 0.4, 0.4, and 0.2, respectively. Compared to the single dynamic constraint method and the Euclidean distance method, this technical solution exhibits different convergence characteristics during the optimization process. As can be observed from the diagram, the loss function value of this technical solution rapidly decreases to 0.72 in the first 10 iterations, then the convergence speed slows down, stabilizing at around 0.50 after 50 iterations. In contrast, the single dynamic constraint method converges to 0.31 after 50 iterations, and the Euclidean distance method converges to 0.17.

[0043] The final loss function value of this technical solution is higher than that of the other two methods because it comprehensively considers multiple biomechanical constraints, reflecting the biomechanical consistency during movement more fully, rather than focusing solely on the optimization of a single indicator. Experimental results show that the prediction accuracy of this technical solution for shoulder abduction angle and elbow extension angle reaches 96.5% and 95.2% respectively, significantly higher than the other two methods.

[0044] In one alternative implementation, Based on the multidimensional feature vectors, an adaptive neural synapse algorithm is used to construct a dynamic connectivity network. Based on this dynamic connectivity network, a recursive least squares algorithm is used to analyze the nonlinear characteristics of muscle group movement, and a fatigue assessment function considering the interaction between muscle groups is established, including: Using the multidimensional feature vectors, an initial dynamic network is established through an adaptive neural synapse algorithm. A multi-level synaptic plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network. Synaptic weight update values ​​are obtained through short-term plasticity dynamic equations, and synaptic connection strength update values ​​are obtained through long-term plasticity time-dependent plasticity rules. Synaptic growth probability values ​​are calculated based on calcium ion concentration. The synaptic weight update values, synaptic connection strength update values, and synaptic growth probability values ​​are input into the initial dynamic network to obtain the dynamic connection network. The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized by the recursive least squares algorithm. The network output error value is calculated by combining the error update equation, and the network parameter adjustment value is calculated by using the gain matrix update equation. The dynamic connection network is optimized based on the network output error value and the network parameter adjustment value to obtain the optimized output result. A neurotransmitter concentration kinetic equation is established, in which release, degradation, and diffusion terms are set. The neurotransmitter concentration distribution value is calculated based on the neurotransmitter concentration kinetic equation. Based on the optimized output result and the neurotransmitter concentration distribution value, a single muscle fatigue characteristic term and a muscle group interaction term are set to construct a fatigue evaluation function.

[0045] An initial dynamic network is constructed based on multidimensional feature vectors. The network topology, including the number of neurons, hierarchical division, and connection patterns, is dynamically determined using an adaptive neural synapse algorithm. An iterative optimization approach is employed to automatically adjust the number of neurons in the hidden layers based on the dimensionality and distribution characteristics of the input features. After determining the basic network structure, a multi-level synaptic plasticity model is introduced, which incorporates synaptic plasticity mechanisms at both short-term and long-term timescales.

[0046] In the process of achieving short-term plasticity, a mapping relationship between presynaptic neuronal activity and postsynaptic responses is established. Rapid changes in synaptic efficacy are described using dynamic equations, considering calcium ion influx at the presynaptic terminal, the probability of neurotransmitter release, and changes in postsynaptic membrane sensitivity. Instantaneous updates of synaptic weights are calculated based on these factors. An adaptive term is also introduced to enable rapid weight updates in response to changes in the frequency and intensity of the input signal.

[0047] Long-term plasticity is modeled based on time-dependent plasticity rules, considering the firing sequence relationship between presynaptic and postsynaptic neurons. When the activity of the presynaptic neuron precedes that of the postsynaptic neuron, the synaptic connection strength is enhanced; conversely, the connection strength is weakened. The time-dependent characteristics of synaptic strength changes are determined by setting a time window parameter. A saturation factor is also introduced to prevent synaptic strength from increasing or decreasing indefinitely.

[0048] In the regulation of synaptic plasticity, calcium ion concentration is introduced as a key regulatory factor. A correlation model between calcium ion concentration and membrane potential changes is established, considering the kinetic characteristics of calcium ion channels and the calcium ion buffering mechanism. When the local calcium ion concentration exceeds a threshold, the synaptic growth mechanism is triggered, promoting the formation of new synapses; when the concentration is below the threshold, the synaptic degeneration mechanism is activated, leading to the elimination of redundant synapses. This mechanism enables the dynamic reconstruction of the network structure.

[0049] Using the constructed dynamic connection network as the basic framework, an improved recursive least squares algorithm is employed for network optimization. An error function is established between the network output and the desired output. By calculating the gradient of the error function with respect to the network parameters, the direction of parameter updates is determined. During the update process, a forgetting factor mechanism is introduced, enabling the algorithm to gradually reduce the influence of historical data and better adapt to changes in the features of the current input.

[0050] The error update process employs a dynamic step-size strategy, adaptively adjusting the update step size based on the error change trend. When the error decreases rapidly, a larger step size is used to accelerate convergence; when the error fluctuates, the step size is reduced to improve stability. Simultaneously, a gain matrix update mechanism is established, which considers the covariance characteristics of the data and can adaptively adjust the update magnitude of different parameters.

[0051] In the modeling of neuromodulator kinetics, a system of nonlinear equations with multiple coupled terms is constructed. The release term considers calcium ion influx, synaptic vesicle mobilization, and secretion processes, and introduces multiple time constants to describe the kinetic characteristics of different stages. The degradation term includes two processes: enzymatic degradation and reabsorption, establishing a nonlinear relationship between neuromodulator concentration and degradation rate. The diffusion term employs an improved diffusion equation, considering the effects of spatial inhomogeneity and boundary conditions.

[0052] A fatigue assessment function is constructed using a hierarchical structure. In the single-muscle fatigue feature term, multiple feature indicators from the network output are fused, and the fatigue state of a single muscle is comprehensively evaluated through a non-linear weighted approach. The weight coefficients are determined through optimization using training data, enabling adaptive adjustment of the importance of different features. The muscle group interaction term establishes a coupling relationship model between muscles based on the spatial distribution characteristics of neuromodulation, considering the anatomical location relationships and functional synergy of muscles, and describing the transmission pattern of fatigue between muscle groups through a diffusion effect.

[0053] For example, taking fatigue monitoring of lower limb muscle groups during long-distance running as an example, an initial dynamic network is constructed, comprising an input layer, hidden layers, and an output layer. The input layer receives an 8-dimensional feature vector containing information such as electromyographic signal features and muscle mechanics features. A synaptic plasticity model is set in the network; short-term plasticity reflects synaptic weight changes on a millisecond scale, while long-term plasticity describes connection strength adjustments on a minute scale. By monitoring the calcium ion concentration in the synaptic cleft, synaptic growth is triggered when the concentration exceeds a threshold, and synaptic weakening occurs when it falls below the threshold. Integrating these plasticity mechanisms into the network enables dynamic adjustment of the network structure.

[0054] When optimizing the network, a recursive least squares algorithm is used to track changes in muscle state in real time. The network parameters are dynamically adjusted by calculating the error between the network output and the actual fatigue state. During optimization, the gain matrix update takes into account changes in exercise intensity, using a larger update step size during high-intensity phases and a smaller step size during low-intensity phases.

[0055] In the neurotransmitter dynamics equation, the release term mainly considers the effect 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 constructed fatigue assessment function can accurately reflect the fatigue level of different muscle groups during long-distance running and predict the development trend of fatigue.

[0056] In this embodiment, by establishing a dynamic equation that includes three key steps—release, degradation, and diffusion—a precise description of the spatiotemporal distribution characteristics of neuromodulation is achieved. Through a multi-level synaptic plasticity model, real-time optimization and adjustment of the network structure are realized. By introducing a dynamic step size strategy and a gain matrix update mechanism, the convergence speed and stability of the algorithm are significantly improved. In the existing technology, traditional muscle fatigue assessment methods mainly rely on static neural network structures, which are difficult to adapt to changes in input features. This results in poor adaptability of the network to dynamic changes. When dealing with synaptic plasticity, they often only consider changes on a single time scale, ignoring the dynamic adjustment characteristics of synaptic strength at different time scales. They do not adequately consider the role of neuromodulation, lack in-depth analysis of the synergistic fatigue effect between muscle groups, and cannot accurately describe the transmission law of fatigue between muscle groups. The dynamic connection network in this embodiment has stronger adaptive capabilities, and can adjust the network structure in real time according to changes in input characteristics, which significantly improves the network's ability to process non-stationary signals. The improved recursive least squares algorithm has a faster convergence speed and better stability, which greatly improves the network optimization efficiency. The evaluation method based on neuromodulation dynamics can more accurately describe the transmission law of fatigue between muscle groups, and the evaluation results have better consistency with the actual physiological process.

[0057] Figure 3 This is a dynamic reconstruction diagram of the network structure of the intelligent sportswear muscle fatigue dynamic assessment method according to an embodiment of the present invention. It shows the network structure from an initial 4 input nodes (I1-I4), 3 hidden nodes (H1-H3), and 2 output nodes (O1-O2), optimized by an adaptive neural synapse algorithm to a structure with 3 new hidden nodes (H4-H6). According to the data from the optimization process, the weights of the new connections are significantly higher than the initial connections. For example, the weight from I1 to H4 is 0.87, from I2 to H4 is 0.92, from I3 to H5 is 0.78, and from I4 to H5 is 0.83. In the hidden-to-hidden-layer connections, the weight from H4 to H1 is 0.65, from H4 to H2 is 0.71, from H5 to H2 is 0.68, and from H5 to H3 is 0.74, indicating optimization of cross-layer information flow. In the part from the hidden layer to the output layer, the weights of H1 to H6 are as high as 0.89, H2 to H6 are 0.91, H3 to H6 are 0.82, and the connection weights from H6 to the output layer are 0.94 and 0.88 respectively, showing that the network strengthens the key output path.

[0058] During the optimization process, the total number of network connections decreased from the initial 18 to 14, indicating a more streamlined and efficient network structure. In terms of performance metrics, the network error rate significantly decreased from the initial 21.7% to 8.3% in the intermediate stage, reaching a low error rate of only 2.6%, while the accuracy improved to 94.6%. The number of neurons increased from the initial 9 to 12, but the number of connections decreased by 22.2%, and the average synaptic weight increased from the initial 0.42 to the final 0.81, indicating a significant improvement in network connection quality.

[0059] Under a calcium ion concentration threshold of 0.35 μmol / L, the probability of new synapse growth is 0.63 and the probability of degradation is 0.37, achieving a reasonable balance in the network structure. The entire structure optimization process takes only 17.6 seconds, demonstrating the high efficiency of this 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 handling complex fatigue assessment tasks.

[0060] In one alternative implementation, The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized using a recursive least squares algorithm. The network output error value is calculated by combining the error update equation, including: A multidimensional feature vector is obtained, and the multidimensional feature vector is input into a dynamic connection network. The network state evolution process is described by a three-dimensional nonlinear equation system. The network disturbance sensitivity, bifurcation control parameters and periodic characteristic parameters are calculated and integrated to obtain the system phase space trajectory. The Lyapunov exponent, correlation dimension and bifurcation parameters are calculated and combined to obtain the chaotic feature vector. Based on the chaotic feature vector, a nonlinear constraint function is constructed that includes a state stability term, an orbital periodicity term, and a parameter sensitivity term. The state stability term adopts the exponential function form of the Lyapunov exponent, the orbital periodicity term adopts the sine function form of the correlation dimension, and the parameter sensitivity term adopts the hyperbolic tangent function form of the bifurcation parameter. The nonlinear constraint function is integrated into the recursive least squares algorithm. The parameter vector, gain matrix and regression vector are calculated by the recursive least squares algorithm. The parameter search strategy is dynamically adjusted based on the chaotic feature vector. The search step size is calculated according to the Lyapunov exponent. The search path is optimized using the correlation dimension. The update rate is adjusted based on the bifurcation parameter. The output of the dynamic connection network is optimized using the recursive least squares algorithm based on 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 by combining the error update equation.

[0061] Network state evolution analysis is performed based on the input multidimensional feature vectors. A three-dimensional nonlinear equation system describing the network's dynamic characteristics is constructed, reflecting the evolutionary law of the network state. Numerical analysis is conducted on the equation system, calculating the network's perturbation sensitivity by applying small perturbations. This sensitivity reflects the network's response characteristics to external disturbances. Simultaneously, bifurcation control parameters are obtained through parameter variation analysis, describing the network's stability changes under different parameter conditions. Periodicity analysis is performed on the network output to extract periodic characteristic parameters, reflecting the periodic variation law of the network output.

[0062] The system's dynamic trajectory is reconstructed in phase space by time-integrating the disturbance sensitivity, bifurcation control parameters, and periodicity parameters. Based on the reconstructed phase space trajectory, three key parameters characterizing the system's chaotic properties are calculated: the Lyapunov exponent reflects the trajectory's divergence, the correlation dimension describes the trajectory's space-filling characteristics, and the bifurcation parameter characterizes the system's stability changes. These three parameters are combined to form a chaotic feature vector, which comprehensively describes the network's nonlinear dynamic characteristics.

[0063] A nonlinear constraint function is constructed based on chaotic eigenvectors, comprising three core constraint terms. The state stability term is constructed using an exponential function of the Lyapunov exponent, reflecting the stability of the system state. The orbital periodicity term is described using a sinusoidal function of the correlation dimension, reflecting the periodicity of the system's orbit. The parameter sensitivity term is expressed using a hyperbolic tangent function of the bifurcation parameters, reflecting the system's sensitivity to parameter changes.

[0064] The constructed nonlinear constraint function is integrated into the recursive least squares algorithm framework. The algorithm calculates a parameter vector containing the weight parameters of each network layer. Simultaneously, a gain matrix is ​​calculated, which controls the direction and magnitude of parameter updates. A regression vector is constructed, reflecting the impact of historical data on the current optimization. Based on the characteristics of chaotic eigenvectors, the parameter search strategy is dynamically adjusted. The step size of the parameter search is determined by the magnitude of the Lyapunov exponent; a smaller step size is used when the exponent is large to ensure stability, and a larger step size is used when the exponent is small to accelerate convergence. The search path is optimized by the variation characteristics of the correlation dimension; a larger dimension increases the diversity of search directions, while a smaller dimension focuses the search on the main directions. The parameter update rate is adjusted based on the changes in bifurcation parameters, reducing the update rate near bifurcation points to improve accuracy.

[0065] Using a defined search step size, search path, and update rate, the network output is optimized using a recursive least squares algorithm. An error function incorporating dynamic constraints is constructed by multiplying the nonlinear constraint function by the difference between the actual and predicted network outputs. The network output error value is calculated based on the error update equation, reflecting the prediction accuracy and the degree to which the dynamic constraints are satisfied.

[0066] For example, consider the analysis of electromyographic (EMG) signals during continuous arm flexion and extension movements. The input multidimensional feature vector contains information in eight dimensions, including the time-frequency and mechanical characteristics of the EMG signal. After inputting the feature vector into a dynamically connected network, the network state evolution is described by a three-dimensional nonlinear equation system. The calculated sensitivity of the network to changes in motion amplitude is 0.85, the bifurcation control parameter caused by changes in motion frequency is 0.32, and the signal periodicity parameter is 0.64. Integrating these parameters yields the system trajectory in phase space, resulting in a Lyapunov exponent of 0.15, a correlation dimension of 2.3, and a bifurcation parameter of 0.28. These parameters are combined into a chaotic feature vector to construct a nonlinear constraint function. During the recursive least squares algorithm optimization, when the Lyapunov exponent reaches 0.15, the search step size is set to 0.01; when the correlation dimension is 2.3, the 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.

[0067] In this embodiment, a three-dimensional nonlinear equation system is constructed to describe the network state evolution process, thereby achieving an accurate characterization of the network dynamics. A nonlinear constraint function containing state stability, orbit periodicity, and parameter sensitivity terms is designed to achieve comprehensive constraints on the network dynamics. The nonlinear constraint function is integrated into the recursive least squares algorithm, and the parameter search strategy is dynamically adjusted based on the chaotic feature vector. In existing technologies, traditional neural network optimization methods mainly focus on the prediction accuracy of the network, neglecting the importance of the network's dynamic characteristics. They often use simple gradient descent or backpropagation algorithms, which cannot effectively capture the chaotic characteristics and dynamic evolution of the system. They adopt fixed parameter search strategies, lack the ability to adaptively adjust to the system state, and are prone to getting trapped in local optima or causing instability in the optimization process. When constructing the error function, they rarely consider the dynamic constraints of the system, so although the optimization results have good fitting accuracy, they may violate the physical characteristics of the system. This embodiment effectively avoids local optima by using a dynamically adjusted parameter search strategy, achieving rapid convergence of the global optimum. The adaptive optimization mechanism based on chaotic features significantly improves the stability of the network under different working conditions, providing strong technical support for the modeling and optimization of complex nonlinear systems. It has important application value in fields such as signal processing and pattern recognition.

[0068] Figure 4 This is a comparison chart of the error convergence of the intelligent sportswear muscle fatigue dynamic assessment method according to an embodiment of the present invention. It shows the error convergence process of the present technical solution and three existing mainstream algorithms (traditional RLS algorithm, neural network method, and support vector machine) in nonlinear system prediction. The trend of the prediction error (RMSE) of the four algorithms with the increase of the number of iterations can be clearly observed from the figure.

[0069] This proposed solution (square marker curve) demonstrates significant advantages, exhibiting not only the fastest convergence speed but also the lowest prediction error. Specifically, it achieves stable convergence after approximately 24 iterations, reducing the final prediction error (RMSE) to 0.052. In contrast, the Support Vector Machine (diamond marker curve) requires approximately 42 iterations to converge, with a final error of 0.094; the Neural Network method (circular marker curve) requires approximately 56 iterations, with a final error of 0.087; and the traditional RLS algorithm (triangle marker curve) performs the worst, requiring approximately 68 iterations to converge, with a final error still as high as 0.130.

[0070] Judging from the shape of the error descent curve, our proposed solution exhibits a sharp error descent in the early iteration stage (between 10-20 iterations), which is attributed to the dynamic adjustment function of the chaotic feature vector on the search strategy. In contrast, the error descent rates of the other three methods are relatively slow, especially the traditional RLS algorithm, whose error descent curve changes almost linearly, indicating a lack of effective optimization acceleration mechanisms.

[0071] It is worth noting that during the iteration process, this technical solution exhibited a significant and rapid decrease in error during the 10th-15th iterations. This corresponds to the critical stage where the algorithm dynamically adjusts the search step size based on the Lyapunov exponent. When the Lyapunov exponent reaches 0.15, the algorithm automatically sets the search step size to 0.014, optimizes it to four main search directions based on the correlation dimension of 2.3, and adjusts the update rate to 0.08 according to the bifurcation parameter of 0.28. This combination of parameter settings enables the algorithm to quickly approach the global optimum.

[0072] From the perspective of convergence stability, the prediction error of this technical solution fluctuates very little after convergence, remaining around 0.052, indicating that the nonlinear constraint function constructed based on chaotic eigenvectors effectively enhances the stability of the algorithm. In contrast, the other three algorithms still exhibit varying degrees of fluctuation even after convergence, especially the support vector machine method, whose prediction error still oscillates slightly in the final stage, indicating its limited adaptability to nonlinear systems.

[0073] In conclusion, Figure 4 This demonstrates the significant advantages of our proposed solution in terms of convergence speed, prediction accuracy, and stability. By integrating chaotic dynamics theory with recursive least squares algorithm, we achieve performance significantly superior to existing algorithms in nonlinear system prediction tasks, particularly demonstrating stronger adaptability and robustness when handling complex nonlinear relationships and non-stationary time-series data.

[0074] In one alternative implementation, Based on the fatigue assessment function and historical exercise data, the fatigue values ​​of each muscle group are determined, and personalized fatigue thresholds are dynamically updated and monitored in real time according to the fatigue values, including: Real-time motion data of muscle groups is acquired, including the intensity, duration, and frequency of the muscle group's movement. The real-time motion data is then input into a fatigue assessment function. Historical motion data is used as a reference benchmark, and the fatigue value of each muscle group is calculated based on the fatigue assessment function. Based on the comparison between the fatigue value of the muscle group and the historical exercise data, the initial fatigue threshold is dynamically updated using an adaptive adjustment algorithm. The adaptive adjustment algorithm optimizes the initial fatigue threshold in real time based on the movement characteristics and physiological features of different muscle groups, and establishes a personalized fatigue threshold for each muscle group. The personalized fatigue threshold is used as a monitoring reference standard to monitor the fatigue status of the muscle group in real time. When the fatigue value exceeds the personalized fatigue threshold, a fatigue warning is triggered.

[0075] Real-time motion data of muscle groups is acquired through a sensor array. Motion intensity data is obtained by measuring the amplitude of electromyographic (EMG) signals and calibrated in conjunction with force sensor data. Motion duration is recorded using a high-precision timer, with the start and end times marked. Motion frequency is obtained through time-frequency analysis of the EMG signals, followed by wavelet transform for signal denoising and feature extraction. This real-time data is then formatted according to a preset format and used as input parameters for a fatigue assessment function.

[0076] Historical exercise data is retrieved from the database, containing records of past performance under the same exercise patterns. Statistical analysis is performed on the historical data to establish a benchmark dataset. The benchmark dataset includes standard fatigue curves, typical fatigue development patterns, and key time point characteristics under different exercise intensities. Real-time exercise data is compared with the benchmark dataset, and fatigue values ​​for each muscle group are calculated using a fatigue assessment function.

[0077] During fatigue assessment, real-time data is normalized to ensure comparability with baseline data. Then, the baseline fatigue accumulation rate is calculated based on exercise intensity, which increases non-linearly with exercise duration. The modulating effect of exercise frequency on fatigue development is also considered; high-frequency exercise accelerates fatigue accumulation, while low-frequency exercise provides recovery opportunities. These factors are then combined to obtain real-time fatigue values ​​for each muscle group.

[0078] An adaptive adjustment algorithm is implemented, which dynamically updates the fatigue threshold based on the characteristics of different muscle groups. A muscle characteristic database is established, containing the physiological characteristics of each muscle group, such as the proportion of muscle fiber types, energy metabolism characteristics, and fatigue recovery rate. Based on these characteristic parameters, a personalized fatigue development model is constructed. This model considers the muscle's fatigue tolerance, recovery ability, and workload capacity.

[0079] In the threshold optimization process, multiple adjustment factors are introduced. Different threshold adjustment strategies are used for strength-type muscle groups and endurance-type muscle groups. For strength-type muscle groups, the focus is on the impact of instantaneous load; for endurance-type muscle groups, the focus is on the cumulative fatigue effect. The training level of the muscles is also considered; muscle groups with higher training levels have higher fatigue tolerance, and their thresholds are correspondingly higher.

[0080] The adaptive adjustment algorithm also includes a real-time feedback mechanism. When a decline in muscle performance is detected, regression analysis is used to determine the trend and rate of performance decline. Threshold parameters are adjusted based on the decline characteristics to achieve dynamic optimization of the threshold. Simultaneously, a cross-validation mechanism is established to continuously optimize the accuracy of the adjustment algorithm by comparing predicted fatigue levels with actual fatigue performance.

[0081] Establish an independent fatigue monitoring module for each muscle group. Compare the current fatigue level with a personalized fatigue threshold in real time. Design a tiered early warning mechanism that triggers different levels of warning signals when the fatigue level reaches different percentages of the threshold. Simultaneously record the time, frequency, and duration of the warning triggers; this data will be used for subsequent threshold optimization and training program adjustments.

[0082] For example, consider biceps monitoring during weightlifting training. Real-time data shows the exercise intensity was maintained at 70% of maximum muscle strength for 45 minutes, with a frequency of 8 repetitions per minute. Training data from the athlete's past three months was used as a benchmark. The fatigue score for the biceps was calculated to be 0.75 using a fatigue assessment function. Considering the athlete's biceps are predominantly composed of fast-twitch fibers and they have a good training foundation, the adaptive adjustment algorithm raised the initial fatigue threshold from 0.8 to 0.85. In subsequent monitoring, when the fatigue score reaches 0.82, a first-level warning is triggered, prompting the athlete to adjust the training intensity.

[0083] In this embodiment, by acquiring real-time motion data of muscle groups and combining it with historical motion data as a reference benchmark, a precise assessment of muscle fatigue status is achieved. Based on an adaptive adjustment algorithm, the fatigue threshold is dynamically optimized, fully considering the motion characteristics and physiological features of different muscle groups. This makes the threshold setting more personalized and scientific. Differentiated assessment standards are adopted for different types of muscle groups, which better adapts to the physiological characteristics and working modes of different muscles. It can accurately identify the fatigue characteristics of each muscle group and provide more targeted guidance for sports training and rehabilitation.

[0084] A second aspect of the present invention provides an intelligent sportswear muscle fatigue dynamic assessment system, comprising: The first unit is used to collect muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in 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, and combine the adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted through Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. The third unit is used to construct a dynamic connection network based on the multidimensional feature vector using an adaptive neural synapse algorithm, and based on the dynamic connection network, analyze the nonlinear characteristics of the muscle group movement process using a recursive least squares algorithm to establish a fatigue evaluation function that considers 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 it in real time, and issue an early warning signal if any muscle group exceeds the personalized fatigue threshold.

[0085] A third aspect of the present invention provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0086] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0087] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic assessment of muscle fatigue in intelligent sportswear, characterized in that, include: The muscle pressure data, muscle surface temperature data, and muscle electrical signal data collected by sensors in smart sportswear are used as muscle state data. 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 point. An adaptive threshold filtering algorithm is used to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions. The instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted by Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. Based on the multidimensional feature vector, an adaptive neural synapse 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 by combining the recursive least squares algorithm, and a fatigue evaluation function considering the interaction between muscle groups is established. The fatigue level of each muscle group is determined based on the fatigue assessment function and historical exercise data. The personalized fatigue threshold is dynamically updated and monitored in real time based on the fatigue level value. 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 joint time-frequency features of each data point. An adaptive threshold filtering algorithm is then used to eliminate environmental noise and motion artifacts to obtain standard feature data, including: Muscle pressure data, muscle surface temperature data, and electromyographic signal data during exercise are acquired and combined to form multi-source motion state data. Continuous wavelet transform is performed on the multi-source motion state data based on a mother wavelet function, which includes a center frequency parameter, to obtain time-domain feature vectors, frequency-domain feature vectors, and a time-frequency joint feature matrix. Morphological operations are performed on the time-domain feature vector, the frequency-domain feature vector, and the time-frequency joint feature matrix. Erosion and dilation operations are performed through the structuring element to obtain enhanced feature data. An adaptive threshold function is constructed based on the local standard deviation and the number of sampling points of the enhanced feature data. The adaptive coefficient of the adaptive threshold function is dynamically adjusted according to the local signal-to-noise ratio. The enhanced feature data is decomposed into multiple intrinsic mode functions (IMFs). The instantaneous frequency of each IMF is calculated using Hilbert transform. Motion artifacts are identified and removed based on the instantaneous frequencies to reconstruct standard feature data. The signal-to-noise ratio, feature coherence, and mean square error of the standard feature data are calculated and weighted by a preset weighting coefficient to obtain a feature quality evaluation index. The standard feature data is then selected based on the feature quality evaluation index and combined to form a standard feature dataset.

3. The method according to claim 1, characterized in that, The standard feature data is decomposed into multiple intrinsic mode functions (IMFs) based on the empirical mode decomposition method. The instantaneous frequency and instantaneous amplitude of each IMF are extracted using the Hilbert-Huang transform to obtain a multidimensional feature vector representing the muscle activity state, including: The data in the standard feature dataset are processed by empirical mode decomposition to obtain multiple intrinsic mode functions and residual terms. It is ensured that the intrinsic mode functions satisfy the following conditions: 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 maxima and local minima is zero. The upper and lower envelopes of the intrinsic mode functions are constructed using cubic spline interpolation. The mean envelope is calculated based on the upper and lower envelopes. The iteration termination condition is determined based on the standard deviation of the results of two adjacent iterations. Perform a Hilbert-Huang transform on the intrinsic mode function to construct an analytic signal containing the intrinsic mode function and the Hilbert-Huang transform result. Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the analytic signal, and combine the instantaneous amplitude, instantaneous phase, and instantaneous frequency to form a feature matrix. Hilbert marginal spectrum analysis is introduced into the feature matrix to calculate the importance index of each feature component in the feature matrix and screen out the high importance feature components. A biomechanical consistency loss function is constructed based on joint torque, muscle force and gravity terms to extract muscle activity patterns and muscle movement link transmission relationships and construct the high importance feature components into a multidimensional feature vector characterizing the muscle activity state. Calculate the mutual information entropy between the feature components in the multidimensional feature vector, obtain the feature redundancy based on the mutual information entropy, remove redundant feature components according to a preset threshold, and obtain the multidimensional feature vector.

4. The method according to claim 3, characterized in that, The feature matrix is ​​analyzed using Hilbert marginal spectrum analysis. Importance indices are calculated for each feature component in the feature matrix. High-importance feature components are selected based on these importance indices, and these high-importance feature components are used to construct a multi-dimensional feature vector representing muscle activity state, including: The feature matrix is ​​analyzed by introducing Hilbert marginal spectrum analysis, the Hilbert transform of each feature component in the feature matrix is ​​calculated, the analytical signal of each feature component is obtained, the marginal spectrum is calculated based on the analytical signal, the marginal spectrum is double integrated in the time-frequency domain to obtain the importance index of each feature component, and the high importance feature components corresponding to the importance index are selected according to a preset threshold. Acquire pre-collected joint motion parameters, calculate the Jacobian matrix, muscle force vector, damping term, and gravity term based on the joint motion parameters, and combine them to obtain the joint torque; A biomechanical consistency loss function is constructed, which includes kinematic constraints, dynamic constraints, and energy efficiency constraints, wherein the kinematic constraints are constructed based on joint angles, the dynamic constraints are constructed based on joint torques, and the energy efficiency constraints are constructed based on energy consumption. Based on the biomechanical consistency loss function, a muscle activity pattern matrix is ​​extracted. Based on the muscle activity pattern matrix and the muscle force vector, a collaborative weight is optimized and calculated to obtain the proximal joint angle and the distal joint angle. The proximal joint angle, the distal joint angle, and the muscle force vector are used to construct a kinetic chain transmission relationship. The high-importance feature components, the biomechanical consistency loss function, the synergistic weights, and the kinetic chain transmission 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 filtering based on a preset mutual information entropy threshold.

5. The method according to claim 1, characterized in that, Based on the multidimensional feature vectors, an adaptive neural synapse algorithm is used to construct a dynamic connectivity network. Based on this dynamic connectivity network, a recursive least squares algorithm is used to analyze the nonlinear characteristics of muscle group movement, and a fatigue assessment function considering the interaction between muscle groups is established, including: Using the multidimensional feature vectors, an initial dynamic network is established through an adaptive neural synapse algorithm. A multi-level synaptic plasticity model and a dynamic structure reconstruction mechanism are set in the initial dynamic network. Synaptic weight update values ​​are obtained through short-term plasticity dynamic equations, and synaptic connection strength update values ​​are obtained through long-term plasticity time-dependent plasticity rules. Synaptic growth probability values ​​are calculated based on calcium ion concentration. The synaptic weight update values, synaptic connection strength update values, and synaptic growth probability values ​​are input into the initial dynamic network to obtain the dynamic connection network. The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized by the recursive least squares algorithm. The network output error value is calculated by combining the error update equation, and the network parameter adjustment value is calculated by using the gain matrix update equation. The dynamic connection network is optimized based on the network output error value and the network parameter adjustment value to obtain the optimized output result. A neurotransmitter concentration kinetic equation is established, in which release, degradation, and diffusion terms are set. The neurotransmitter concentration distribution value is calculated based on the neurotransmitter concentration kinetic equation. Based on the optimized output result and the neurotransmitter concentration distribution value, a single muscle fatigue characteristic term and a muscle group interaction term are set to construct a fatigue evaluation function.

6. The method according to claim 5, characterized in that, The multidimensional feature vector is input into the dynamic connection network, and the output of the dynamic connection network is optimized using a recursive least squares algorithm. The network output error value is calculated by combining the error update equation, including: A multidimensional feature vector is obtained, and the multidimensional feature vector is input into a dynamic connection network. The network state evolution process is described by a three-dimensional nonlinear equation system. The network disturbance sensitivity, bifurcation control parameters and periodic characteristic parameters are calculated and integrated to obtain the system phase space trajectory. The Lyapunov exponent, correlation dimension and bifurcation parameters are calculated and combined to obtain the chaotic feature vector. Based on the chaotic feature vector, a nonlinear constraint function is constructed that includes a state stability term, an orbital periodicity term, and a parameter sensitivity term. The state stability term adopts the exponential function form of the Lyapunov exponent, the orbital periodicity term adopts the sine function form of the correlation dimension, and the parameter sensitivity term adopts the hyperbolic tangent function form of the bifurcation parameter. The nonlinear constraint function is integrated into the recursive least squares algorithm. The parameter vector, gain matrix and regression vector are calculated by the recursive least squares algorithm. The parameter search strategy is dynamically adjusted based on the chaotic feature vector. The search step size is calculated according to the Lyapunov exponent. The search path is optimized using the correlation dimension. The update rate is adjusted based on the bifurcation parameter. The output of the dynamic connection network is optimized using the recursive least squares algorithm based on 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 by combining the error update equation.

7. The method according to claim 1, characterized in that, Based on the fatigue assessment function and historical exercise data, the fatigue values ​​of each muscle group are determined, and personalized fatigue thresholds are dynamically updated and monitored in real time according to the fatigue values, including: Real-time motion data of muscle groups is acquired, including the intensity, duration, and frequency of the muscle group's movement. The real-time motion data is then input into a fatigue assessment function. Historical motion data is used as a reference benchmark, and the fatigue value of each muscle group is calculated based on the fatigue assessment function. Based on the comparison between the fatigue value of the muscle group and the historical exercise data, the initial fatigue threshold is dynamically updated using an adaptive adjustment algorithm. The adaptive adjustment algorithm optimizes the initial fatigue threshold in real time based on the movement characteristics and physiological features of different muscle groups, and establishes a personalized fatigue threshold for each muscle group. The personalized fatigue threshold is used as a monitoring reference standard to monitor the fatigue status of the muscle group in real time. When the fatigue value exceeds the personalized fatigue threshold, a fatigue warning is triggered.

8. A dynamic assessment system for muscle fatigue in intelligent sportswear, used to implement the method described in any one of claims 1-7, 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 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, and combine the adaptive threshold filtering algorithm to eliminate environmental noise and motion artifacts to obtain standard feature data. Based on the empirical mode decomposition method, the standard feature data is decomposed into multiple intrinsic mode functions, and the instantaneous frequency and instantaneous amplitude of each intrinsic mode function are extracted through Hilbert-Huang transform to obtain a multidimensional feature vector characterizing the muscle activity state. The third unit is used to construct a dynamic connection network based on the multidimensional feature vector using an adaptive neural synapse algorithm, and based on the dynamic connection network, analyze the nonlinear characteristics of the muscle group movement process using a recursive least squares algorithm to establish a fatigue evaluation function that considers 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 it in real time, and issue an early warning signal if any muscle group exceeds the personalized fatigue threshold.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke 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 the processor, they implement the method described in any one of claims 1 to 7.

Images