A lumbar fatigue state analysis method based on surface electromyography signals

Abstract

This application relates to the fields of neuromuscular physiology and data processing technology, and discloses a method for analyzing lumbar fatigue state based on surface electromyography (EMG) signals, including the following steps: S1, acquiring raw data of the target muscle group in the lumbar region using a multi-channel EMG acquisition device, which can construct a dataset for batch processing; S2, converting the lumbar EMG signals to be analyzed and evaluated using MATLAB software. If the data file format of the lumbar EMG signals is CSV, it is converted to MATLAB format. The FastICA toolbox path is configured, and the default path is tried. If the toolbox path is not found, MATLAB built-in functions are attempted. By acquiring multi-channel lumbar EMG signals and using an improved FastICA algorithm and iterative KDE method, high-precision motor unit decomposition and pulse recognition are achieved, and optimizations are also made for the characteristics of lumbar muscles.

Description

Technical Field

[0001] This application relates to the fields of neuromuscular physiology and data processing technology, specifically a method for analyzing lumbar fatigue state based on surface electromyography signals. Background Technology

[0002] Lower back fatigue is a common health problem in modern society, widely prevalent in scenarios such as prolonged sitting at work, heavy physical labor, driving, and sports training. Persistent lower back fatigue not only reduces work efficiency and affects quality of life, but is also a significant contributing factor to chronic lower back diseases such as lumbar muscle strain and lumbar disc herniation. Surface electromyography (EMG) signals are electrical activity signals of muscles recorded from the skin surface, which can non-invasively reflect muscle activation patterns, contraction intensity, and fatigue levels. Current research only utilizes frequency domain features such as median frequency and average power frequency of surface electromyography (EMG) signals to assess local muscle fatigue. However, lumbar activity is often a complex and dynamic process, and existing methods generally suffer from insufficient feature extraction, weak noise immunity, and low fatigue discrimination accuracy under dynamic tasks. Furthermore, traditional signal processing algorithms such as short-time Fourier transform and wavelet transform struggle to reliably extract effective fatigue features when faced with dynamic lumbar contractions, muscle crosstalk, and non-stationary noise, leading to unreliable assessment results. Therefore, current research cannot accurately assess muscle fatigue states, hindering its value in sports injury prevention, rehabilitation training guidance, and occupational health management.

[0003] Therefore, this application proposes a method for analyzing lumbar fatigue state based on surface electromyography signals to address the shortcomings of the prior art. Summary of the Invention

[0004] The purpose of this application is to provide a method for analyzing lumbar fatigue state based on surface electromyography signals, in order to solve the problems mentioned in the background art.

[0005] The objective of this application can be achieved through the following technical solutions: A method for analyzing lumbar fatigue state based on surface electromyography signals includes the following steps: S1. Raw data of the target muscle group in the waist can be collected by a multi-channel surface electromyography acquisition device, and a dataset can be constructed for batch processing. S2. Use MATLAB software to convert the lumbar surface electromyography (EMG) signals to be analyzed and evaluated. If the data file format of the lumbar EMG signals is CSV, convert it to MATLAB format. Configure the FastICA toolbox path and try the default path. If the toolbox path is not found, try using MATLAB built-in functions. S3. Data loading and parameter setting: Select the data file, set the relevant parameters, and configure the channel labels; S4. Preprocess the collected surface electromyography (EMG) signals from the lumbar region to obtain preprocessed EMG signals. Data preprocessing includes removing DC components, removing motion artifacts, bandpass filtering, and removing power supply noise. Manually implement data standardization (z-score) without relying on the StatisticsToolbox. S5. The signal is decomposed using the improved FastICA algorithm, and its components are sorted and screened based on physiological inspiration. S6. Use iterative kernel density estimation to identify pulses; S7. Visualize the results of the surface electromyography signals of the lumbar region after the above processing. S8. Based on the analysis and assessment of the degree of fatigue in the lumbar muscles, analyze the fatigue status of the target muscle group and conduct a comprehensive evaluation of the results.

[0006] Preferably, step S1 includes the following steps: S11. Configure a multi-channel surface electromyography acquisition device to acquire raw data of the target muscle group in electrode input mode; S12. The multi-channel surface electromyography (EMG) acquisition device can realize the synchronous acquisition of EMG signals from multiple sites. It acquires electrical signals generated by muscle activity through surface electrodes. Its sampling rate, PGA gain, channel settings, etc. can be selected according to the actual situation. S13. The muscle groups for collecting lumbar electromyographic signals mainly include the erector spinae, multifidus, and quadratus lumborum, which are distributed on both sides of the lower back.

[0007] Preferably, step S3 includes the following steps: S31. To add a single .mat file, select the corresponding data file name. If there are multiple .mat files, create an sEMG_results folder in the current directory, put all the .mat data files to be analyzed into it, and modify the data file path in the code. The first file is selected by default. S32. Load the data file, determine if it is loaded correctly, and check the data dimensions and size; S33. Centrally manage parameters such as signal, preprocessing, FastICA, and KDE, and screen and extract parameters based on the physiological characteristics of the lumbar muscles; S34. Display basic signal information, including channels, sampling points and sampling rates, and configure channel labels for detailed descriptions to ensure that the number of channel labels matches the actual number of channels.

[0008] Preferably, step S31 includes the following steps: S331. Reduce the frequency band occupancy threshold to improve IC detection rate: ; Among them, the total The quantity refers to the total number of independent components obtained from the decomposition of FastICA, which is effective. Quantity refers to the number of independent components retained after screening using physiologically inspired criteria; S332. Static postures require a reduction in the minimum discharge rate, while dynamic activities require an increase in the maximum discharge rate and a reduction in the signal-to-noise ratio to adapt to the characteristics of surface electromyography signals.

[0009] Preferably, the preprocessing of the acquired lumbar surface electromyography signals is as follows: S411, Remove DC component; S412. Use median filtering to remove low-frequency motion artifacts:

[0010] The window size is 2k+1 (an odd number), median represents the median of all values ​​within the window (i.e., the middle value after sorting), and the window size (in seconds) is generally set to 0.2 seconds by default. The sampling rate (Hz) is 1000Hz by default; S413, using a 4th-order Butterworth filter, 20-450Hz frequency band; S414. Use a notch filter to remove 50Hz power supply noise.

[0011] Preferably, step S5 includes the following steps: S51. Decomposition is performed using the improved FastICA algorithm, which mainly includes automatic determination of the number of components, deterministic initialization, and independent component extraction. S52. Added a fault tolerance mechanism, that is, when the FastICA toolbox fails, it can automatically switch to the principal component analysis PCA method implemented by SVD; S53. Based on physiological inspiration, sort and screen the components.

[0012] Preferably, step S53 includes the following steps: S531. Manually calculate the kurtosis and power spectrum characteristics of the impulse response index; S532. Perform preliminary detection on the pulses and use the median absolute deviation (MAD) as a more robust threshold. S533, Signal-to-Noise Ratio (SNR) is used to evaluate the signal quality of each independent component, and its decibel representation based on the root mean square (RMS) amplitude is as follows:

[0013] in, This is the estimated signal-to-noise ratio, expressed in decibels (dB). It is an independent component The root mean square (RMS) value is calculated using the following formula: ,in, The number of sampling points. For the first The amplitude at each sampling point reflects the effective amplitude of the signal. The estimated value of the noise standard deviation is obtained by dividing the median of the median filtered residuals by 0.6745: ,in, This is the result after median filtering of the independent components, with a filter window length of 0.01× That is, 10ms. At 1000Hz, the window has 10 sampling points. It is the median of the absolute values ​​of the residuals, i.e., the median absolute deviation (MAD). 0.6745 is the Gaussian distribution constant, used to convert MAD into an unbiased estimate of the standard deviation. S534. Perform a comprehensive score and ranking on the signals processed above. The comprehensive score formula is as follows: Screening criteria were set based on the characteristics of the lumbar muscles: the main energy of lumbar electromyography is in the range of 10-450Hz, the frequency band ratio is greater than 0.45%, the discharge rate is 1-50Hz, the SNR is greater than 5dB, and the peak intensity is relatively high, so as to extract effective independent components.

[0014] Preferably, step S6 includes the following steps: S61. Dynamically adjust the detection threshold using the standard deviation of a sliding window; S62. Pulse waveform extraction and alignment: Centered on the peak point, take a ±15ms window to align the waveform; S63. Through multiple rounds of iteration and gradual optimization of the KDE template, the pulse recognition accuracy is improved. Iterative optimization of the KDE template: For each time point Calculate the probability density: , ; in, For probability density estimation, For bandwidth, It is a Gaussian kernel function; Automatic bandwidth selection based on MAD's Silverman rules: , Where MAD is the median absolute deviation; Based on the squared correlation coefficient, the weights are iteratively updated, and the Pearson correlation coefficient between each pulse waveform and the current template is calculated. Weights: ,in, For pulse weights, For the Pearson correlation coefficient, outliers with weights less than 0.15 are removed to generate a new template, i.e., the amplitude with the highest probability density at each time point is taken as the template value. S64. Use the KDE template for final pulse classification. Based on the 75th percentile of the correlation coefficient, the pulse classification threshold is: ,in It is a quantile, 0 < <1.

[0015] Preferably, step S7 includes the following steps: S71. Display and compare the raw signals and preprocessed signals of all channels; S72. Display the extracted independent components, pulse sequence, and motor unit action potential MUAP template; S73. Calculate the mean and standard deviation of each channel, use a matrix to calculate the correlation coefficient matrix, and form a channel correlation heatmap.

[0016] Preferably, step S8 includes the following steps: S81. Calculate fatigue indices and display fatigue analysis results, including: comprehensive fatigue score, fatigue level, time domain score, frequency domain score, and motor unit score. S82. Combining time domain, frequency domain, and motion unit discharge characteristics into a multi-domain fatigue index, analysis is performed using a 5-second window and a 50% overlapping sliding window to monitor fatigue development in real time. S83. Establish a comprehensive scoring system based on a comprehensive fatigue score using multi-dimensional indicators; S84. Graded fatigue assessment: The fatigue level is divided into three levels: mild, moderate and severe. The results are saved and a report text is generated.

[0017] The beneficial effects of this application are: 1. This application achieves high-precision motor unit decomposition and pulse recognition by acquiring multi-channel lumbar surface electromyography signals and using an improved FastICA algorithm and iterative KDE method. It also optimizes the method for the characteristics of lumbar muscles. Overcoming the limitations of traditional methods, it can automatically determine the optimal number of components, perform multi-dimensional screening of multiple physiological indicators based on physiological inspiration, and add an adaptive screening strategy that automatically lowers the screening conditions when no effective components are found, thereby improving the detection rate. It combines time-domain, frequency-domain, and motor unit indicators for multi-dimensional fatigue assessment and reduces reliance on the MATLAB toolbox, improving compatibility. Multiple charts display the analysis results, providing a comprehensive decomposition and fatigue analysis report, which helps subjects intuitively understand their lumbar fatigue state. This method not only has significant clinical application value but also provides new technical ideas and methods for the field of biosignal processing.

[0018] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the steps of a method for analyzing lumbar fatigue state based on surface electromyography signals, as described in this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Please see Figure 1 As shown, this application discloses a method for analyzing lumbar fatigue state based on surface electromyography signals. The method includes the following steps: S1. Raw data of the target muscle group in the waist can be collected by a multi-channel surface electromyography acquisition device, and a dataset can be constructed for batch processing. S2. Use MATLAB software to convert the format of the lumbar surface electromyography signals to be analyzed and evaluated, and configure the FastICA toolbox. S3. Data loading and parameter setting: Select the data file, set the relevant parameters, and configure the channel labels; S4. Preprocess the collected surface electromyography (EMG) signals from the lumbar region to obtain preprocessed surface EMG signals. S5. The signal is decomposed using the improved FastICA algorithm, and its components are sorted and screened based on physiological inspiration. S6. Use Iterative Kernel Density Estimation (Iterative KDE) to accurately identify pulses; S7. Visualize the results of the surface electromyography signals of the lumbar region after the above processing. S8. Based on the analysis and assessment of the degree of fatigue in the lumbar muscles, analyze the fatigue status of the target muscle group and conduct a comprehensive evaluation of the results.

[0023] Specifically, this invention acquires lumbar surface electromyography (EMG) signals using a multi-channel EMG signal acquisition device. After preprocessing, an improved FastICA algorithm and an iterative KDE method are incorporated for analysis. FastICA enables adaptive component determination and deterministic initialization, while iterative optimization and weighted kernel density estimation achieve refined KDE pulse recognition, thus realizing high-precision motor unit decomposition and pulse recognition. This method of the present invention significantly improves component determination, pulse recognition, and fatigue assessment, reduces toolbox dependence, and improves computational efficiency and robustness. Through multi-dimensional feature analysis and adaptive algorithms, parameters are optimized for the characteristics of lumbar muscles, and a multi-layered fault-tolerance mechanism ensures system stability. It provides comprehensive decomposition and fatigue analysis reports, achieving accurate assessment of lumbar muscle fatigue and providing a powerful tool for related research and applications.

[0024] In S1, this embodiment uses a high-precision, low-noise sensor to build a multi-channel surface electromyography (EMG) acquisition device, which acquires raw EMG signals of the target muscle group in the lumbar region using differential electrode mode. During acquisition, the signal electrode is placed parallel to the muscle fiber direction at the center of the muscle, and the reference electrode is placed in an area without muscle activity. The acquired muscle groups mainly include the erector spinae, multifidus, and quadratus lumborum, distributed on both sides of the lumbar back. The sampling rate is set to 1000Hz, the PGA gain is adjusted according to the signal amplitude, and the number of channels can be configured according to actual needs; this embodiment uses 8 channels as an example. The acquired raw data is saved in .mat format in the sEMG_results folder for easy batch processing.

[0025] S2 includes the following steps: S21. If the original data is in "csv" format, first convert it to "mat" format in MATLAB; S22. Configure the FastICA toolbox path: Try the default paths one by one. If the path exists, add it. If the toolbox is not found, the SVD-based PCA method will be used as an alternative. S3 includes the following steps: S31. A single "mat" file can be selected directly; if there are multiple files, all "mat" files in the sEMG_results folder will be read, and the first file will be selected for decomposition by default. S32. Load the data file, check if the variable sEMG_data exists, and verify the data dimensions (number of channels × number of sampling points) to ensure that the data is not empty; S33. Centralized parameter management, setting key parameters based on the physiological characteristics of the lumbar muscles: Signal parameters: Preprocessing parameters: bandpass filter range 20-450Hz, filter order 4, motion artifact removal window 0.2 seconds; FastICA parameters: maximum number of iterations 1500, symmetric method, nonlinear function tanh, convergence threshold 1e-6; KDE parameters: number of iterations 5, outlier threshold 0.15, initial threshold coefficient 25, Gaussian kernel; filtering parameters: frequency band ratio threshold 0.45, minimum discharge rate 1Hz, maximum discharge rate 50Hz, signal-to-noise ratio threshold 5dB.

[0026] S34. Display basic signal information (number of channels, number of sampling points, sampling rate), configure 8-channel lumbar electromyography tags, including left / right erector spinae L3 / L2, left / right quadratus lumborum, left / right multifidus L4, and note the electrode placement position; It is important to note that static postures require a lower minimum discharge rate, while dynamic activities require a higher maximum discharge rate and a lower signal-to-noise ratio to adapt to the characteristics of surface electromyography signals.

[0027] S4 includes the following steps: S41. Data preprocessing includes removing DC components, removing motion artifacts, bandpass filtering, and removing power supply noise. S41 includes the following steps: S411, Remove DC component: Use detrend(...,'constant') for each channel; S412. Removing motion artifacts: Use median filtering. The output formula for median filtering is:

[0028] The window size is 2k+1 (an odd number), and `median` represents the median of all values ​​within the window, i.e., the middle value after sorting. The window size is: The default window size is 0.2 seconds. The sampling rate (Hz) is 1000Hz by default.

[0029] S413, Bandpass Filter: A fourth-order Butterworth filter is used with cutoff frequencies of 20Hz and 450Hz, and zero-phase filtering is performed using FLTFILT. S414. Remove 50Hz power supply noise: Design an IIR notch filter with a notch frequency of 50Hz, a normalized bandwidth of 0.01, and use FLT FLT filtering. S42. Manually implement data standardization (z-score) without relying on the StatisticsToolbox.

[0030] S42 includes the following steps: S421. Data Standardization (z-score): Manually calculate the mean and standard deviation for each channel, avoiding division by zero. Use eps to avoid floating-point precision issues. Replace all zeros with very small random noise for channel data. The formula for calculating the mean is: ,in, As the mean, there are a total of For each sampling point, the standard deviation σ, also known as the sample standard deviation or unbiased estimate, is calculated using the following formula: ,in, As the denominator, we obtain an unbiased estimate. The above mean is the standardized data. for: ,in, This is the original data. The mean of the original data. The standard deviation of the original data is used. If the standard deviation of a channel is less than the machine precision ESP, then a small amount of random noise (10) is added. -6 (Multiply the number of normally distributed random numbers) to avoid subsequent ICA decomposition failures.

[0031] In S5, the following steps are included: S51. Decomposition is performed using the improved FastICA algorithm, which mainly includes automatic determination of the number of components, deterministic initialization, and independent component extraction. S51 includes the following steps: S511 and FastICA separate independent components by maximizing non-Gaussianity; the negative entropy objective function is: ,in, It is a signal entropy, Is with Entropy of Gaussian signals with the same mean and variance, data whitening process: ,in, It is the original signal matrix. It is a whitening matrix. The data is whitened and meets the requirements. , so that the covariance matrix is ​​the identity matrix; Fixed-point iterative update rules: ,in, , It is a nonlinear function. Its derivative; S512 automatically determines the number of source signals and directly calculates the inter-channel covariance matrix to optimize memory usage. Assuming there are... One channel, each channel collects data. Each sampling point, data matrix for:

[0032] in Indicates the first The first channel in the The value of each sampling point; In the FastICA whitening step, the inter-channel covariance matrix needs to be decomposed using SVD to generate the whitening matrix and the channel covariance matrix. It is A symmetric matrix, if the data matrix After subtracting the mean from each channel, the covariance matrix simplifies to:

[0033] in, yes The data matrix yes The transpose of, for SVD decomposition is performed to obtain eigenvalues ; Automatically determine the number of independent components based on eigenvalue inflection points: ,

[0034] in Represents eigenvalues. Represents the difference operator, used to compute negative differences. :

[0035] Take the index corresponding to the maximum value as Then the number of components:

[0036] Ensure the number of components is between 1 and the number of channels; S513. Determine the number of sources based on the "inflection point" of eigenvalue descent, that is, automatically determine the number of independent components, ensure that the number of sources is within a reasonable range, and use the first N principal components as the initial value of FastICA to improve the convergence speed and decomposition stability. S52. Added a fault tolerance mechanism. If the FastICA toolbox call fails, it can automatically switch to the PCA method implemented by SVD and output the principal components as independent components. S53. Based on physiological inspiration, sort and screen the components.

[0037] S53 includes the following steps: S531. Manually calculate the kurtosis and power spectrum characteristics of the impulse response index (bandwidth 20-450Hz). The formula for calculating kurtosis is as follows: ,in, The fourth-order centrality matrix reflects the thickness and kurtosis of the tails of the data distribution, and its calculation formula is as follows: ,in, For sample size, The sample mean. For the first One data point; S532. Perform preliminary pulse detection and set a threshold using the Median Absolute Deviation (MAD): Peak value detection, minimum interval 20ms; S533, Signal-to-Noise Ratio (SNR) is used to evaluate the signal quality of each independent component, and its decibel representation based on the root mean square (RMS) amplitude is as follows: ,in, This is the estimated signal-to-noise ratio, expressed in decibels (dB). The root mean square (RMS) value of the independent component IC is calculated using the following formula: ,in, The number of sampling points. For the first The amplitude at each sampling point reflects the effective amplitude of the signal. The estimated value of the noise standard deviation is obtained by dividing the median of the median filtered residuals by 0.6745: ,in, This is the result after median filtering of the independent components, with a filter window length of [value missing]. , It is the median of the absolute values ​​of the residuals, i.e., the median absolute deviation (MAD). 0.6745 is the Gaussian distribution constant, used to convert MAD into an unbiased estimate of the standard deviation. S534. Perform a comprehensive score and ranking on the signals processed above. The comprehensive score formula is as follows: The samples were sorted in descending order of scores, and screening criteria were set based on the characteristics of the lumbar muscles: frequency band ratio > 0.45, discharge rate 1-50Hz, SNR > 5dB, and high kurtosis. Finally, effective independent components were extracted.

[0038] In S6, the following steps are included: S61. Use sliding window standard deviation to dynamically adjust the detection threshold: window size 100ms, calculate local standard deviation, i.e., initial threshold = 25 × median, detect peak value, minimum interval 20ms; S62. Pulse waveform extraction and alignment: Centered on the peak point, take a ±15ms window, with a total of 31 sampling points, and align the waveform; S63, Iterative KDE optimization template (5 iterations): Kernel density estimation: for each time point Calculate the probability density: Using a Gaussian kernel: ,in, For probability density estimation, For bandwidth, It is a Gaussian kernel function.

[0039] Automatic bandwidth selection based on MAD's Silverman rules: , Where MAD is the median absolute deviation, the weights are iteratively updated based on the square of the correlation coefficient, and the Pearson correlation coefficient between each pulse waveform and the current template is calculated. Weight Outliers with a weight less than 0.15 are removed.

[0040] Generate a new template: take the amplitude with the largest probability density at each time point as the template value; S64. Use the KDE template for final pulse classification: Based on the 75th percentile of the correlation coefficient, the impulse classification threshold is: ,in It is a quantile, 0 < <1; Calculate the correlation coefficients between all valid pulses and the final template, and use the 75th percentile as the threshold. Pulses with correlation coefficients higher than the threshold are retained. Record the firing time of each motion unit and calculate characteristics such as discharge frequency and ISI coefficient of variation. In S7, the following steps are included: S71. Display a comparison chart of the raw and preprocessed signals of all 8 channels, with two sub-charts for each channel to facilitate observation of the preprocessing effect; S72. Display the time-domain waveforms of the extracted independent components (effective ICs), each sub-component. Figure 1 One component; S73. Display a Raster plot of the motion unit pulse sequence, with time on the horizontal axis and motion unit number on the vertical axis. Each pulse is represented by a scatter plot. S74. Display the motion potential (MUAP) template of the motor unit, with the horizontal axis representing time (ms) and the vertical axis representing normalized amplitude. S75. Calculate the correlation coefficient matrix between channels and draw a heatmap, with colors representing the intensity of correlation. In S8, the following steps are included: S81. Calculation of fatigue indices: Time-domain indices are calculated using root mean square (RMS), integral electromyography (IEMG), mean amplitude (MA), and variance (VAR) with a 5-second window and a 50% overlapping sliding window. Frequency-domain indices are calculated using mean power frequency (MPF), median frequency (MF), and power spectrum slope (10-100Hz) with a sliding window. The discharge characteristics of the motor unit are obtained using the mean discharge frequency and the coefficient of variation of the discharge interval (CV_ISI). S82. Comprehensive Fatigue Score: Time Domain Score (40 points), Frequency Domain Score (40 points), Motor Unit Score (20 points), Total Score 0-100 points. The scoring formula is as follows: Time-domain score = 10×min(1, relative increase of RMS) + 10×min(1, relative increase of IEMG) + 10×min(1, relative increase of MA) + 10×min(1, relative increase of VAR); Frequency domain score = 15×min(1, relative decrease in MPF) + 15×min(1, relative decrease in MF) + 10×min(1, degree of steepness of slope); Motor unit score = 20 × min(1, mean CV_ISI); S83. Fatigue level classification: less than 20 is mild fatigue, 20 to 50 is moderate fatigue, and more than 50 is severe fatigue. S84. Result Saving: Save the decomposition results (raw data, preprocessed data, effective IC, MU pulse, MU template, characteristic parameters, etc.) as a "*_resultsmat" file and generate a text report, outputting information such as the discharge frequency, average ISI, and CV_ISI of each motion unit.

[0041] Through the above steps, a method for analyzing lumbar fatigue status based on surface electromyography (EMG) signals was achieved. This method utilizes an improved FastICA algorithm to automatically determine the number of independent components, combined with a physiologically inspired screening strategy to effectively extract motor unit firing sequences. Iterative kernel density estimation is then used to accurately identify pulses, and finally, a multi-domain fatigue index is used to comprehensively assess the degree of fatigue. Experiments show that this method can dynamically monitor the fatigue development process of lumbar muscles under static posture and dynamic activity, providing an objective basis for sports injury prevention and rehabilitation training. The above description is merely an example and illustration of the concept of this application. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in the claims, they should all fall within the protection scope of this application.

Claims

1. A method for analyzing lumbar fatigue state based on surface electromyography signals, characterized in that, Includes the following steps: S1. Raw data of the target muscle group in the waist can be collected by a multi-channel surface electromyography acquisition device, and a dataset can be constructed for batch processing. S2. Use MATLAB software to convert the lumbar surface electromyography (EMG) signals to be analyzed and evaluated. If the data file format of the lumbar EMG signals is CSV, convert it to MATLAB format. Configure the FastICA toolbox path and try the default path. If the toolbox path is not found, try using MATLAB built-in functions. S3. Data loading and parameter setting: Select the data file, set the relevant parameters, and configure the channel labels; S4. Preprocess the collected surface electromyography (EMG) signals from the lumbar region to obtain preprocessed EMG signals. Data preprocessing includes removing DC components, removing motion artifacts, bandpass filtering, and removing power supply noise. Manually implement data standardization (z-score) without relying on the StatisticsToolbox. S5. The signal is decomposed using the improved FastICA algorithm, and its components are sorted and screened based on physiological inspiration. S6. Use iterative kernel density estimation to identify pulses; S7. Visualize the results of the surface electromyography signals of the lumbar region after the above processing. S8. Based on the analysis and assessment of the degree of fatigue in the lumbar muscles, analyze the fatigue status of the target muscle group and conduct a comprehensive evaluation of the results.

2. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, S1 includes the following steps: S11. Configure a multi-channel surface electromyography acquisition device to acquire raw data of the target muscle group in electrode input mode; S12. The multi-channel surface electromyography (EMG) acquisition device can realize the synchronous acquisition of EMG signals from multiple sites. It acquires electrical signals generated by muscle activity through surface electrodes. Its sampling rate, PGA gain, channel settings, etc. can be selected according to the actual situation. S13. The muscle groups for collecting lumbar electromyographic signals mainly include the erector spinae, multifidus, and quadratus lumborum, which are distributed on both sides of the lower back.

3. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 2, characterized in that, S3 includes the following steps: S31. To add a single .mat file, select the corresponding data file name. If there are multiple .mat files, create an sEMG_results folder in the current directory, put all the .mat data files to be analyzed into it, and modify the data file path in the code. The first file is selected by default. S32. Load the data file, determine if it is loaded correctly, and check the data dimensions and size; S33. Centrally manage parameters such as signal, preprocessing, FastICA, and KDE, and screen and extract parameters based on the physiological characteristics of the lumbar muscles; S34. Display basic signal information, including channel, sampling point and sampling rate, and configure channel labels for detailed description to ensure that the number of channel labels matches the actual number of channels.

4. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 3, characterized in that, S31 includes the following steps: S331. Reduce the frequency band occupancy threshold to improve IC detection rate: ; Among them, the total The quantity refers to the total number of independent components obtained from the decomposition of FastICA, which is effective. Quantity refers to the number of independent components retained after screening using physiologically inspired criteria; S332. Static postures require a reduction in the minimum discharge rate, while dynamic activities require an increase in the maximum discharge rate and a reduction in the signal-to-noise ratio to adapt to the characteristics of surface electromyography signals.

5. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, The preprocessing process for the collected surface electromyography signals of the lumbar region is as follows: S411, Remove DC component; S412. Use median filtering to remove low-frequency motion artifacts: ； The window size is 2k+1 (an odd number), median represents the median of all values ​​within the window (i.e., the middle value after sorting), and the window size (in seconds) is generally set to 0.2 seconds by default. The sampling rate (Hz) is 1000Hz by default; S413, using a 4th-order Butterworth filter, 20-450Hz frequency band; S414. Use a notch filter to remove 50Hz power supply noise.

6. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, S5 includes the following steps: S51. Decomposition is performed using the improved FastICA algorithm, which mainly includes automatic determination of the number of components, deterministic initialization, and independent component extraction. S52. Added a fault tolerance mechanism, that is, when the FastICA toolbox fails, it can automatically switch to the principal component analysis PCA method implemented by SVD; S53. Based on physiological inspiration, sort and screen the components.

7. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 6, characterized in that, S53 includes the following steps: S531. Manually calculate the kurtosis and power spectrum characteristics of the impulse response index; S532. Perform preliminary detection on the pulse and use the median absolute deviation (MAD) as a more robust threshold. S533, Signal-to-Noise Ratio (SNR) is used to evaluate the signal quality of each independent component, and its decibel representation based on the root mean square (RMS) amplitude is as follows: ； in, This is the estimated signal-to-noise ratio, expressed in decibels (dB). It is an independent component The root mean square (RMS) value is calculated using the following formula: ,in, The number of sampling points. For the first The amplitude at each sampling point reflects the effective amplitude of the signal. The estimated value of the noise standard deviation is obtained by dividing the median of the median filtered residuals by 0.6745: ,in, This is the result after median filtering of the independent components, with a filter window length of 0.01× That is, 10ms. At 1000Hz, the window has 10 sampling points. It is the median of the absolute values ​​of the residuals, i.e., the median absolute deviation (MAD). 0.6745 is the Gaussian distribution constant, used to convert MAD into an unbiased estimate of the standard deviation. S534. Perform a comprehensive scoring and ranking on the signals processed above. The comprehensive scoring formula is as follows: Screening criteria were set based on the characteristics of the lumbar muscles: the main energy of lumbar electromyography is in the range of 10-450Hz, the frequency band ratio is greater than 0.45%, the discharge rate is 1-50Hz, the SNR is greater than 5dB, and the peak intensity is relatively high, so as to extract effective independent components.

8. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, S6 includes the following steps: S61. Dynamically adjust the detection threshold using the standard deviation of a sliding window; S62. Pulse waveform extraction and alignment: Centered on the peak point, take a ±15ms window to align the waveform; S63. Through multiple rounds of iteration and gradual optimization of the KDE template, the pulse recognition accuracy is improved. Iterative optimization of the KDE template: For each time point Calculate the probability density: , ; in, For probability density estimation, For bandwidth, It is a Gaussian kernel function; Automatic bandwidth selection based on MAD's Silverman rules: , Where MAD is the median absolute deviation; Based on the squared correlation coefficient, the weights are iteratively updated, and the Pearson correlation coefficient between each pulse waveform and the current template is calculated. Weights: ,in, For pulse weights, For the Pearson correlation coefficient, outliers with weights less than 0.15 are removed to generate a new template, i.e., the amplitude with the highest probability density at each time point is taken as the template value. S64. Use the KDE template for final pulse classification. Based on the 75th percentile of the correlation coefficient, the pulse classification threshold is: ,in It is a quantile, 0 < <1.

9. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, S7 includes the following steps: S71. Display and compare the raw signals and preprocessed signals of all channels; S72. Display the extracted independent components, pulse sequence, and motor unit action potential MUAP template; S73. Calculate the mean and standard deviation of each channel, use a matrix to calculate the correlation coefficient matrix, and form a channel correlation heatmap.

10. The method for analyzing lumbar fatigue state based on surface electromyography signals according to claim 1, characterized in that, S8 includes the following steps: S81. Calculate fatigue indices and display fatigue analysis results, including: comprehensive fatigue score, fatigue level, time domain score, frequency domain score, and motor unit score. S82. Combining time domain, frequency domain, and motion unit discharge characteristics into a multi-domain fatigue index, analysis is performed using a 5-second window and a 50% overlapping sliding window to monitor fatigue development in real time. S83. Establish a comprehensive scoring system based on a comprehensive fatigue score using multi-dimensional indicators; S84. Graded fatigue assessment: The fatigue level is divided into three levels: mild, moderate and severe. The results are saved and a report text is generated.

Images