Electromyographic signal fatigue detection method based on time window analysis

Abstract

The invention provides an electromyographic signal fatigue detection method based on time window analysis, and the method comprises the steps: carrying out the preprocessing of collected electromyographic signals, and carrying out the self-adaptive time window segmentation based on a sliding overlapping mechanism; performing spectral analysis on each time window signal by using a Wilch method, and extracting frequency domain features such as median frequency, average power frequency and power spectral density ratio; a normalized composite fatigue index is constructed, and smoothing processing is carried out through multi-time scale moving average; finally, a self-adaptive multi-level threshold judgment mechanism is introduced, and fatigue state level recognition and dynamic updating are achieved. The fatigue evolution process can be continuously and quantitatively reflected, and the practicability and the intelligent level of electromyographic signal analysis under the scenes of training monitoring, man-machine work efficiency, rehabilitation evaluation and the like are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of myoelectric fatigue detection, and in particular relates to a myoelectric signal fatigue detection method based on time window analysis. Background Art

[0002] Electromyographic signals are bioelectric signals that reflect muscle contraction activity. Analyzing them can provide insights into muscle activation levels and fatigue levels. Traditional electromyographic fatigue research primarily relies on time-domain and frequency-domain metrics, such as root mean square (RMS) and median frequency (MF). These methods lack the ability to describe the dynamics of fatigue and suffer from the following deficiencies:

[0003] (1) The commonly used fixed window or overall signal processing method lacks the ability to capture the dynamic changes of the signal continuously, resulting in the inability to quantify the fatigue change process in real time;

[0004] (2) Most existing methods use a single indicator such as MF or RMS to evaluate muscle fatigue, ignoring the overall morphological changes of the spectrum, resulting in the evaluation model being susceptible to local noise and having poor stability and sensitivity;

[0005] (3) Fixed threshold fatigue level judgment is often used, which cannot adapt to individuals with different physical signs or training status. In addition, fatigue indicators fluctuate, which is prone to false positives or false negatives.

[0006] Therefore, we need to develop a fatigue detection method for electromyographic signals based on time window analysis, which can restore the fatigue change process through a sliding time window, improve the expression of fatigue characteristics using a multi-frequency domain feature fusion index, and achieve personalized and accurate judgment of fatigue levels. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for detecting fatigue of electromyographic signals based on time window analysis, so as to solve the problems mentioned in the above background technology, such as the lack of real-time and continuous capture of dynamic changes of electromyographic signals, the single dimension of fatigue assessment, and the inability to individualize fatigue level judgment.

[0008] To achieve the above objectives, the present invention provides a method for detecting fatigue of electromyographic signals based on time window analysis, the method being as follows:

[0009] Step S1, collecting electromyographic signals and preprocessing them to obtain a time domain signal sequence, and performing adaptive time window segmentation on the time domain signal sequence to obtain a time window signal sequence;

[0010] Step S2: Use the Welch method to perform spectrum analysis on each time window signal in the time window signal sequence to obtain the power spectrum density function and the corresponding frequency sequence of the time window signal, and establish a time window sequence number. The mapping relationship between the power spectrum density and the frequency sequence is used to obtain the time-frequency spectrum database;

[0011] Step S3, extracting fatigue characteristic parameters based on the time-frequency spectrum database, including median frequency, average power frequency and power spectrum density ratio;

[0012] Step S4, constructing a standardized fatigue index by weighted fusion of the fatigue characteristic parameters, normalizing the fatigue index, and obtaining a smoothed value of the fatigue index using a multi-time scale smoothing technique;

[0013] The calculation formula of the fatigue index is: ,in 、 、 are the median frequency, average power frequency and power spectrum density ratio of the nth time window, 、 、 are the corresponding weights respectively;

[0014] Step S5: setting an adaptive multi-level threshold based on the fatigue index smoothing value, dynamically determining the fatigue state level currently corresponding to the electromyographic signal, and providing corresponding intervention suggestions based on the fatigue state level.

[0015] Based on the above scheme, the adaptive time window segmentation adopts a sliding overlapping mechanism, which is to perform time window segmentation on the time domain signal sequence to analyze the changing characteristics of the electromyographic signal over time. It is necessary to first set the time window length and the window moving step. There is partial overlap between two adjacent time windows to improve the continuity recognition ability of the fatigue evolution process.

[0016] Based on the above solution, the Welch method, specifically the Welch average periodogram method, and the spectrum analysis include:

[0017] The time window signal is divided into K segments, each segment length is , the overlap between adjacent segments is 50%;

[0018] Add a Hamming window to each segment to obtain the windowed signal segment;

[0019] Perform FFT transformation on the K segments of the signal after adding the Hamming window to obtain K segments of spectrum;

[0020] Calculate power spectral density estimates;

[0021] Mapping the power spectral density estimate to a physical frequency to obtain a power spectral density function;

[0022] Extract the frequency sequence and corresponding power spectrum density within the effective frequency band of the surface electromyography signal.

[0023] Based on the above scheme, the extraction of the median frequency includes:

[0024] The frequency point where the power spectrum of the electromyographic signal is divided into two equal parts in the frequency domain is defined as the frequency value that meets the following conditions : ,in, is the power spectral density function, and They are the lower and upper limits of the effective frequency band of the surface electromyography signal.

[0025] Based on the above solution, the extraction of the median frequency further includes:

[0026] Assume that the power spectrum density sequence corresponding to the nth time window is ,in is the mth frequency component, M is the total number of frequency points under the spectrum resolution, and the cumulative power spectrum density sequence is defined as ,in ;

[0027] Let the sum of the total energy of the nth time window signal in the frequency domain be: , then the median frequency of the nth time window The following conditions are met: there are frequency components , , such that: , then let .

[0028] Based on the above solution, the extraction of the average power frequency includes:

[0029] Calculate and output the average power frequency of the nth time window: ,in, They represent the mth frequency component and its corresponding power spectrum density value, and M is the total number of spectrum points.

[0030] Based on the above solution, the extraction of the power spectral density ratio includes:

[0031] Calculate and output the power spectral density ratio of the nth time window: ,in, is the preset cutoff frequency, and They are the lower and upper limits of the effective frequency band of the surface electromyography signal.

[0032] Based on the above scheme, the fatigue index is constructed by integrating the three fatigue characteristic parameters of MF, MPF, and PSDR. The construction process includes:

[0033] Based on the relative change rate of fatigue characteristic parameters, the initial fatigue index is calculated to preliminarily evaluate the dynamic changes of fatigue status. The calculation formula is as follows: ,in, 、 They represent the percentage change of MF, MPF, and PSDR values ​​in the nth time window relative to the initial levels, 、 、 is the corresponding weight coefficient, satisfying the normalization condition: .

[0034] Based on the above scheme, the initial fatigue index is normalized based on the weighted combination of the fatigue characteristic parameters, and the normalized fatigue index is calculated: , where n represents the sequence number of the time window, is the total number of time windows, represents the i-th fatigue characteristic parameter extracted in the n-th time window, is the total number of fatigue characteristic parameters; is the weight coefficient of the i-th fatigue characteristic parameter, satisfying the normalization condition: .

[0035] Based on the above scheme, the setting of adaptive multi-level threshold is based on prior physiological knowledge and a large amount of experimental data analysis, and L ordered fatigue state levels are pre-set. , and determine the corresponding fatigue index threshold ,grade Indicates that fatigue gradually increases, and the threshold The cutoff points of different fatigue levels are defined. When the fatigue index smoothing value When the fatigue status level is determined to be level ,in, .

[0036] The present invention has the following advantages and effects compared to the prior art:

[0037] (1) Adaptive time window segmentation is performed on the electromyographic signal, and a sliding mechanism with step size and overlap is used to ensure that every moment is covered, maximizing the continuity of signal evolution, achieving high time resolution analysis of the electromyographic signal timing characteristics, dynamically tracking the fatigue evolution process, and improving the integrity and accuracy of real-time monitoring;

[0038] (2) Extract three representative frequency domain features MF, MPF, and PSDR, and construct a unified fatigue index through normalization and feature weighted fusion. This index takes into account the morphology, center of gravity, and frequency band energy distribution, improves fatigue judgment ability, comprehensively describes the morphological changes of muscle frequency domain signals, and realizes multi-dimensional fusion of fatigue level quantification.

[0039] (3) The fatigue index is smoothed by using a sliding average of multiple time scales, and the fatigue threshold sequence is dynamically adjusted in combination with personal historical data through an adaptive learning algorithm to improve the stability and individual adaptability of fatigue level judgment, taking into account both short-term mutations and long-term trends, and meeting the needs of intelligent fatigue alarms in complex usage scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0041] Figure 1 This is a flow chart of a method for detecting fatigue of electromyographic signals based on time window analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to more clearly illustrate the purpose, technical solutions and advantages of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The example implementation methods can be implemented in various forms and should not be understood as being limited to the examples described herein. On the contrary, these implementation methods are provided to make the present invention more comprehensive and complete, and to fully convey the concepts of the example implementation methods to those skilled in the art.

[0043] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, it will be appreciated by those skilled in the art that the technical solutions of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0044] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0045] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0046] The present invention will be described in detail below with reference to specific embodiments:

[0047] As attached Figure 1 As shown, embodiment 1 of the present invention provides a method for detecting fatigue of electromyographic signals based on time window analysis, and the specific steps of the method are as follows:

[0048] Step S1, collecting electromyographic signals and preprocessing them to obtain a time domain signal sequence, and performing adaptive time window segmentation on the time domain signal sequence to obtain a time window signal sequence;

[0049] Step S2: Use the Welch method to perform spectrum analysis on each time window signal in the time window signal sequence to obtain the power spectrum density function and the corresponding frequency sequence of the time window signal, and establish a time window sequence number. The mapping relationship between the power spectrum density and the frequency sequence is used to obtain the time-frequency spectrum database;

[0050] Step S3, extracting fatigue characteristic parameters based on the time-frequency spectrum database, including median frequency, average power frequency and power spectrum density ratio;

[0051] Step S4, constructing a fatigue index by weighted fusion of the fatigue characteristic parameters, normalizing the fatigue index, and obtaining a smoothed value of the fatigue index using a multi-time scale smoothing technique;

[0052] Step S5: Based on the fatigue index smoothing value, an adaptive multi-level threshold is set to dynamically determine the fatigue state level currently corresponding to the electromyographic signal, and corresponding intervention suggestions are given based on the fatigue state level.

[0053] Preferably, the hardware device for collecting electromyographic signals in step S1 is a surface electromyographic acquisition electrode, which is attached to the surface of the target muscle. The continuous analog signal is converted into a discrete digital signal. According to the sampling theorem, the sampling frequency It should be at least twice the highest frequency of the signal. Considering that the effective frequency band of surface electromyography signal is generally 20~450Hz, the embodiment of the present invention sets the sampling frequency The sampling rate is 1000Hz, that is, 1000 sample points are collected per second to ensure that the sampling is not distorted and the digital signal is frequency transmission.

[0054] Furthermore, the pre-processing is based on the collected electromyographic signals, and the electromyographic signals are filtered and denoised. The original collected electromyographic signals are easily affected by noises such as baseline drift and power frequency interference, and necessary pre-processing is required. The embodiment of the present invention adopts a digital filtering method, which specifically includes 20Hz high-pass filtering to remove low-frequency baseline drift in the electromyographic signal; 450Hz low-pass filtering to remove high-frequency noise beyond the electromyographic frequency band; 50Hz notch filtering to suppress power frequency power interference. After a series of filtering processes, a pure original electromyographic signal is obtained, which is recorded as a time domain signal sequence. ,in is a continuous time variable.

[0055] Furthermore, the adaptive time window segmentation adopts a sliding overlap mechanism to perform time window segmentation on the time domain signal sequence S(t) and analyze the time-varying characteristics of the electromyographic signal, specifically including:

[0056] Parameter selection for time window segmentation: setting the time window length , in seconds (s), represents the duration of the signal in each time window; the window moving step parameter , in seconds (s), represents the time interval between two adjacent time windows. In this embodiment, , that is, the step length of each time window is 1 second, and the overlap between two consecutive time windows is 0.5 seconds, which not only ensures the real-time nature of fatigue judgment but also retains sufficient signal sample volume; let n be the sequence number of the time window , then The time window signal is expressed as , each As independent signal segments, the frequency domain analysis described in step S2 is performed in sequence.

[0057] Specifically, based on the sampling frequency , time window length , then each time window signal Contains N sampling points, and .

[0058] Preferably, the spectrum analysis in step S2 is performed by using the Welch average periodogram method to calculate the n-th time window signal The Welch average periodogram method is a non-parametric spectrum estimation technique with excellent performance. It performs segmentation, windowing, and averaging on the time window signal, which improves the spectrum resolution while reducing the estimation variance.

[0059] Specifically, when performing the spectrum analysis, the following relevant parameters need to be set:

[0060] (1) FFT length : That is, the fast Fourier transform length, and There are the following relationships: To strike a balance between accuracy and computational efficiency, this embodiment prefers , the frequency resolution is approximately: ;

[0061] (2) Data segment length :For a time window signal with N sampling points, The time window is divided into segments, and 50% overlap is allowed between two adjacent time windows. Experience shows that when When N / 8 to N / 2 is used, a stable spectrum estimation result can be obtained. ;

[0062] (3) Window function type : Used to suppress spectrum leakage and improve the dynamic range of the spectrum. Commonly used window functions include rectangular window, Hanning window, Hamming window, Blackman window, etc. This embodiment specifically uses the Hamming window, which is mathematically described as: , The Hamming window has good sidelobe attenuation characteristics and a moderate mainlobe width, making it an ideal choice for power spectrum estimation.

[0063] Furthermore, based on the relevant parameters, the time window signal The specific steps for spectrum analysis are as follows:

[0064] (1) The time window signal Divided into K segments, each segment length is , the overlap between adjacent segments is 50%. represents the jth sampling point of the kth segment, then: , ,in Indicates the overlapping length of two adjacent segments. ; K is the total number of segments, and ;

[0065] (2) For each paragraph Add a Hamming window to get the windowed signal segment : ;

[0066] (3) Add the K segments of the signal after the Hamming window Do it separately Point FFT transformation to obtain K-segment spectrum:

[0067]

[0068] (4) Calculate the power spectral density estimate :

[0069]

[0070] in is the frequency index The corresponding discrete frequencies are , is the normalization coefficient, .

[0071] (4) The power spectrum density estimate Mapped to physical frequency, we get the power spectral density function : ,in Indicates frequency resolution.

[0072] Furthermore, the frequency sequence within the effective frequency band of the surface electromyography signal is extracted and the corresponding power spectral density : , where the effective frequency band range of the surface electromyography signal is [20, 450] Hz, and M is the total number of frequency components within this range.

[0073] In summary, the time window signal It is transformed into the frequency domain to obtain the power spectrum density function reflecting its spectral characteristics and the corresponding frequency sequence For each time window signal in the time window signal sequence Perform spectrum analysis to obtain the corresponding power spectrum density function and the corresponding frequency sequence , establish the time window sequence number and The mapping relationship between and is used to obtain a complete time-frequency spectrum database, which serves as the basic data structure for subsequent feature extraction.

[0074] Preferably, the extraction of fatigue characteristic parameters in step S3 specifically includes:

[0075] Step S301: extracting the median frequency MF.

[0076] The median frequency MF is the frequency point that divides the power spectrum of the electromyographic signal into two equal parts in the frequency domain and is defined as the frequency value that meets the following conditions: :

[0077]

[0078] in, is the power spectral density function, and They are the lower and upper limits of the effective frequency band of surface electromyography signal, usually MF reflects the center position of the EMG power spectrum. When a muscle transitions from a non-fatigued state to a fatigued state, the low-frequency components of the EMG signal increase while the high-frequency components decrease, causing the MF value to decrease. Therefore, continuously monitoring the MF trend can serve as an important indicator for assessing muscle fatigue.

[0079] It should be noted that represents the general form of the power spectral density function, and Specifically refers to the power spectral density function of the nth time window;

[0080] Furthermore, in the case of a discrete spectrum, the definition of the median frequency MF is converted into the following form to calculate the MF value:

[0081] Assume that the power spectrum density sequence corresponding to the nth time window is ,in is the mth frequency component, M is the total number of frequency points under the spectrum resolution. Define the cumulative power spectrum density sequence ,in:

[0082]

[0083] Furthermore, let the total power, that is, the sum of the total energy of the nth time window signal in the frequency domain, be: , then the median frequency The following conditions are met:

[0084] There are frequency components , such that: ;make ;

[0085] In summary, In the discrete spectrum, the frequency point that divides the total power in half can be obtained by accumulating the power spectrum sequence. Obtain.

[0086] For example, the process of extracting the median frequency MF is as follows:

[0087] Input: M-point power spectrum density sequence of the nth time window and frequency component sequence ;

[0088] Output: median frequency of the nth time window ;

[0089] Steps: Calculate the cumulative power spectral density sequence ; Calculate total power ; Find the frequency components that satisfy the following conditions ;make ; Output .

[0090] Step S302: extracting the mean power frequency MPF.

[0091] The mean power frequency (MPF) is another important parameter in the frequency domain that characterizes the center of gravity of the power spectrum. Unlike the median frequency (MF), which depends on the distribution of cumulative power, the MPF considers the weighted average of the power spectrum density and frequency, thus more comprehensively reflecting the overall shape characteristics of the spectrum. The mathematical definition of the MPF of the nth time window is:

[0092]

[0093] in, where represents the mth frequency component and its corresponding power spectral density value, respectively, and M represents the total number of points in the spectrum. As can be seen from the above formula, MPF is actually a weighted average of frequencies, using the power of each frequency component as a weight. It reflects the center of distribution of the total power of the EMG signal. Similar to MF, when muscles fatigue, the energy of the low-frequency components of the EMG signal increases, while the energy of the high-frequency components decreases, resulting in a downward trend in MPF. Therefore, continuously tracking the changing pattern of MPF can serve as an important indicator for evaluating fatigue processes.

[0094] Exemplarily, the steps for extracting MPF are as follows:

[0095] Input: M-point power spectral density sequence of the nth time window and frequency component sequence ;

[0096] Output: Average power frequency of the nth time window ;

[0097] step:

[0098] 1. Calculate the frequency-weighted power spectral density sequence ;

[0099] 2. Calculate the sum of the power spectral density sequence , recorded as ;

[0100] 3. Calculate the sum of the frequency-weighted power spectral density sequence , recorded as ;

[0101] 4. Order ;

[0102] 5. Output ;

[0103] It should be noted that by tracking the amplitude and rate of dynamic changes in MPF, quantitative characterization of fatigue status can be achieved. It is worth noting that although MF and MPF can both indicate fatigue levels to a certain extent, due to their different calculation methods, there may be slight differences in their values ​​and change patterns. MPF is more sensitive to the overall shape of the power spectrum, while MF emphasizes the average point of the cumulative power. The fatigue detection scheme of the present invention makes full use of a variety of frequency domain indicators, and by appropriately weighted fusion of them, a more robust and reliable fatigue evaluation result can be obtained.

[0104] Step S303: extracting the power spectral density ratio (PSDR);

[0105] This embodiment introduces the power spectral density ratio (PSDR) to characterize the shape characteristics of the myoelectric power spectrum. PSDR is defined as the ratio of the power spectral density of the high and low frequency bands. The PSDR expression of the nth time window is:

[0106]

[0107] in, is the preset cutoff frequency, which can be , and The PSDR is the lower and upper limits of the effective frequency band of the surface electromyography signal. As can be seen from the definition, PSDR measures the ratio of the power spectral density of the high frequency band (100-450Hz) to the low frequency band (20-100Hz), reflecting the relative distribution of spectral energy in different frequency bands.

[0108] When muscles are fresh, high-frequency EMG energy is relatively high, resulting in a high PSDR value. As muscle fatigue develops and worsens, high-frequency energy gradually decays, while low-frequency energy increases, causing the PSDR value to decrease. Therefore, the changing trend of the PSDR can serve as another quantitative indicator for assessing the fatigue process.

[0109] Exemplarily, the PSDR extraction process is as follows:

[0110] Input: The M-point power spectral density sequence of a time window , frequency component sequence Band breakpoint ;

[0111] Output: Power spectral density ratio of the nth time window ;

[0112] step:

[0113] 1. Find the frequency Corresponding frequency component number ;

[0114] 2. Calculate the cumulative sum of the low-frequency power spectrum density sequence: , recorded as ;

[0115] 3. Calculate the cumulative sum of the high-frequency power spectrum density sequence: , recorded as ;

[0116] 4. Order

[0117] 5. Output .

[0118] Preferably, in step S4, the fatigue characteristic parameters of MF, MPF, and PSDR are integrated to construct a fatigue index. By comprehensively analyzing the three frequency domain indicators of MF, MPF, and PSDR, the shape characteristics of the EMG power spectrum and its evolution pattern during fatigue can be more comprehensively characterized. This embodiment adopts a weighted fusion strategy to appropriately combine the changing trends of the three indicators to construct a fatigue index (FI).

[0119] Specifically, the fatigue index construction process includes:

[0120] (1) Calculate the initial fatigue index based on the relative change rate of the index:

[0121] The calculation formula is as follows:

[0122] in, 、 They represent the percentage change of MF, MPF, and PSDR values ​​in the nth time window relative to the initial levels, 、 、 is the corresponding weight coefficient, satisfying the normalization condition: This formula is mainly used to preliminarily evaluate the dynamic changes of fatigue status.

[0123] (2) Based on the weighted combination of multiple fatigue characteristic parameters, a standardized fatigue index is calculated:

[0124] In order to facilitate practical application and threshold judgment, the initial fatigue index is further standardized, and the mathematical expression is as follows:

[0125]

[0126] Where n represents the sequence number of the time window, is the total number of time windows; represents the i-th fatigue characteristic parameter extracted in the n-th time window, is the total number of fatigue characteristic parameters; is the weight coefficient of the i-th fatigue characteristic parameter, satisfying the normalization condition:

[0127]

[0128] For the three fatigue characteristic parameters MF, MPF, and PSDR used in this embodiment, the standardized fatigue index calculation formula can be expressed as:

[0129]

[0130] The above formula reveals the physical meaning of fatigue index: It comprehensively reflects the overall change of the frequency domain characteristics of the electromyographic signal in the nth time window. Different characteristic parameters describe the evolution of the electromyographic pattern caused by fatigue from the aspects of spectrum center of gravity, spectrum energy distribution, etc. The fusion of multi-parameter information is achieved through weighted averaging, making fatigue assessment more comprehensive and reliable. As muscle fatigue progresses, The value will show a monotonous change trend, and its size is positively correlated with the degree of fatigue.

[0131] It should be pointed out that in In the construction of The selection of weights is crucial, as it determines the influence of different fatigue characteristic parameters in the comprehensive assessment. Weight allocation should comprehensively consider factors such as the sensitivity, stability, and interpretability of each parameter, and should be continuously optimized through empirical or experimental data in practical applications. Furthermore, weight allocation can be adjusted based on specific sports, exercise intensity, and age of the individual to better suit different test subjects and situations.

[0132] Furthermore, the fatigue index is normalized:

[0133] Since the dimensions and numerical ranges of different fatigue characteristic parameters vary greatly, in order to facilitate comprehensive comparison and threshold determination, it is necessary to normalize the parameters when constructing the fatigue index. The purpose of normalization is to map different parameters to the same scale, eliminate the influence of dimension and order of magnitude differences, and make More interpretable and comparable.

[0134] This embodiment adopts the maximum and minimum value normalization method to calculate the fatigue characteristic parameter The normalized value of Defined as:

[0135]

[0136] in, and Respectively represent The maximum and minimum values ​​of fatigue characteristic parameters in the entire monitoring process. After normalization, The value range of is limited to the interval [0, 1], and its changing trend is consistent with the original parameter.

[0137] The normalized fatigue characteristic parameters Substituting into the calculation formula of fatigue index, the normalized fatigue index can be obtained :

[0138]

[0139] Among them, although the normalization process changes the value of the original parameter, it does not affect The physical meaning and changing laws of . It also reflects the evolution of fatigue status over time, and its value falls within the interval [0, 1], which is convenient for subsequent threshold division and level determination.

[0140] Furthermore, fatigue exponential smoothing is performed at multiple time scales:

[0141] Muscle fatigue is a gradual physiological process with a significant cumulative effect over time. To better characterize the evolution of fatigue over time and reduce the interference of random noise and local fluctuations, this embodiment introduces a fatigue index smoothing mechanism under multiple time scales. By taking a sliding average of the fatigue index within different time spans, a more robust and macroscopically meaningful fatigue assessment indicator can be obtained. The fatigue index smoothing mechanism is specifically as follows:

[0142] In the nth time window, define the fatigue exponential smoothing value with a time span of τ for:

[0143]

[0144] in, Indicates the length of the smoothing window, that is, the number of time windows involved in the sliding average, Reflects the The overall level of fatigue index from the first time window to the nth time window. The size of can flexibly control the time scale of smoothing and obtain fatigue state information at different granularities.

[0145] when When it is lower than the preset minimum threshold (such as ≤1), the smoothing window degenerates into a single time window. Degenerate into instantaneous fatigue index , reflecting the detailed changes in fatigue status. As the time range covered by the smoothing window increases, the short-term fatigue fluctuations are gradually suppressed, and the long-term trend of fatigue evolution is highlighted. = 5 corresponds to a smoothing time of 5 seconds, =30 corresponds to a smoothing time of 30 seconds. In actual applications, different The fatigue index smoothing value sequence is obtained by multi-scale analysis of fatigue state in the time dimension and normalized and multi-scale smoothed. ;

[0146] It should be noted that by comparing different The fatigue index curve below can fully grasp the stage characteristics of fatigue development and provide richer decision-making basis for subsequent classification judgment and status warning.

[0147] Preferably, the setting of the adaptive multi-level threshold in step S5 is based on prior physiological knowledge and a large amount of experimental data analysis, and pre-sets L ordered fatigue state levels. , and determine the corresponding fatigue index threshold .grade Indicates that fatigue gradually increases, and the threshold The dividing points of different fatigue levels are defined, and the mathematical expressions are as follows:

[0148] When the fatigue index smoothing value When the fatigue status level is determined to be level .in,

[0149]

[0150] The above formula gives a threshold division scheme based on fatigue index rigidly. However, considering that muscle fatigue is a gradual physiological process, the boundaries between fatigue levels are not absolute, and individual differences are significant. Fixed thresholds are difficult to adapt to all situations. To overcome this problem, this embodiment further proposes an adaptive threshold adjustment mechanism, which dynamically updates the threshold based on factors such as current fatigue level, historical trends, and personal physical condition. The value of .

[0151] Specifically, at the beginning of fatigue monitoring, the system selects a set of initial thresholds from the threshold knowledge base based on the age, gender, physical condition and other prior information of the subject being tested. As the detection process progresses, the accumulated fatigue index data is used to correct the threshold in real time through the adaptive learning algorithm to obtain a dynamically updated threshold sequence. , adaptive threshold The calculation formula is as follows: ,in: represents the lth level threshold after the nth time window is updated, is the initial threshold; is the first fatigue index obtained based on the statistics of the latest N time windows. Level threshold estimate; is the smoothing factor, which controls the rate at which the threshold is updated. The larger it is, the smoother the threshold changes.

[0152] The physical significance of the adaptive threshold is that it dynamically adjusts the judgment criteria by referring to the historical distribution of the individual's fatigue index, taking into account both prior experience and individual differences, making fatigue classification more objective and accurate. The introduction of gives the threshold a certain "inertia", preventing it from responding too sensitively to short-term fluctuations and improving the stability of the classification results.

[0153] When the fatigue index smoothing value Triggering or approaching a certain threshold When the system detects a change in fatigue status, it automatically sends a warning message to the user, suggesting appropriate adjustment strategies (such as reducing exercise intensity, increasing rest time, etc.). At the same time, the system compares the current fatigue status with the data series of past fatigue status levels to analyze the stage characteristics of fatigue development and provide reference for subsequent training management.

[0154] In this embodiment, surface electromyography acquisition electrodes are used to obtain the original signal of the target muscle, the sampling frequency is set to 1000Hz, and noise reduction processing is performed through high-pass, low-pass and notch filters. After adaptive sliding time window segmentation, the power spectrum of each time window signal is estimated using the Welch method to obtain a power spectrum density function reflecting the spectral morphology. Three typical frequency domain fatigue parameters, namely median frequency, average power frequency and power spectrum density ratio, are further extracted to construct a weighted fusion composite fatigue index, and maximum and minimum value normalization and multi-time scale sliding average processing are performed. The response granularity of the fatigue index evolution is controlled by introducing a dynamic smoothing window to adapt to the balance between short-term fluctuations and long-term trends. Finally, the multi-level fatigue judgment threshold is dynamically updated by combining individual prior information with the real-time indicator evolution results to achieve individualized fatigue level identification and early warning. The overall solution has high resolution, high adaptability and strong generalization ability, and is suitable for real-time fatigue status monitoring and control of muscles during dynamic training.

[0155] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present invention are indicated by the claims. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and that various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for detecting fatigue of electromyographic signals based on time window analysis, characterized in that: include: Step S1, collecting electromyographic signals and preprocessing them to obtain a time domain signal sequence, and performing adaptive time window segmentation on the time domain signal sequence to obtain a time window signal sequence; Step S2: Use the Welch method to perform spectrum analysis on each time window signal in the time window signal sequence to obtain the power spectrum density function and the corresponding frequency sequence of the time window signal, and establish a time window sequence number. The mapping relationship between the power spectrum density and the frequency sequence is used to obtain the time-frequency spectrum database; Step S3, extracting fatigue characteristic parameters based on the time-frequency spectrum database, including median frequency, average power frequency and power spectrum density ratio; Step S4, constructing a standardized fatigue index by weighted fusion of the fatigue characteristic parameters, normalizing the fatigue index, and obtaining a smoothed value of the fatigue index using a multi-time scale smoothing technique; The calculation formula of the fatigue index is: ,in 、 、 are the median frequency, average power frequency and power spectrum density ratio of the nth time window, 、 、 are the corresponding weights respectively; Step S5: setting an adaptive multi-level threshold based on the fatigue index smoothing value, dynamically determining the fatigue state level currently corresponding to the electromyographic signal, and providing corresponding intervention suggestions based on the fatigue state level.

2. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 1, characterized in that: The adaptive time window segmentation adopts a sliding overlapping mechanism, which performs time window segmentation on the time domain signal sequence and analyzes the time-varying characteristics of the electromyographic signal. The time window length and the window moving step must be set first. There is partial overlap between two adjacent time windows to improve the continuity recognition capability of the fatigue evolution process.

3. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 1, characterized in that: The Welch method, specifically the Welch average periodogram method, and the spectrum analysis include: The time window signal is divided into K segments, each segment length is , the overlap between adjacent segments is 50%; Add a Hamming window to each segment to obtain the windowed signal segment; Perform FFT transformation on the K segments of the signal after adding the Hamming window to obtain K segments of spectrum; Calculate power spectral density estimates; Mapping the power spectral density estimate to a physical frequency to obtain a power spectral density function; Extract the frequency sequence and corresponding power spectrum density within the effective frequency band of the surface electromyography signal.

4. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 1, characterized in that: Extracting the median frequency includes: The frequency point where the power spectrum of the electromyographic signal is divided into two equal parts in the frequency domain is defined as the frequency value that meets the following conditions : ,in, is the power spectral density function, and They are the lower and upper limits of the effective frequency band of the surface electromyography signal.

5. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 4, characterized in that: The extraction of the median frequency further includes: Assume that the power spectrum density sequence corresponding to the nth time window is ,in is the mth frequency component, M is the total number of frequency points under the spectrum resolution, and the cumulative power spectrum density sequence is defined as ,in ; Let the sum of the total energy of the nth time window signal in the frequency domain be: , then the median frequency of the nth time window The following conditions are met: there are frequency components , , such that: , then let .

6. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 5, characterized in that: The extraction of the average power frequency comprises: Calculate and output the average power frequency of the nth time window: ,in, They represent the mth frequency component and its corresponding power spectrum density value, and M is the total number of spectrum points.

7. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 6, characterized in that: The extraction of the power spectral density ratio comprises: Calculate and output the power spectral density ratio of the nth time window: ,in, is the preset cutoff frequency, and They are the lower and upper limits of the effective frequency band of the surface electromyography signal.

8. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 7, characterized in that: The fatigue index is constructed by integrating the three fatigue characteristic parameters of MF, MPF, and PSDR. The construction process includes: Based on the relative change rate of fatigue characteristic parameters, the initial fatigue index is calculated to preliminarily evaluate the dynamic changes of fatigue status. The calculation formula is as follows: ,in, 、 They represent the percentage change of MF, MPF, and PSDR values ​​in the nth time window relative to the initial levels, 、 、 is the corresponding weight coefficient, satisfying the normalization condition: .

9. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 8, characterized in that: Based on the weighted combination of the fatigue characteristic parameters, the initial fatigue index is normalized to calculate the normalized fatigue index: , where n represents the sequence number of the time window, is the total number of time windows, represents the i-th fatigue characteristic parameter extracted in the n-th time window, is the total number of fatigue characteristic parameters; is the weight coefficient of the i-th fatigue characteristic parameter, satisfying the normalization condition: .

10. The method for detecting fatigue of electromyographic signals based on time window analysis according to claim 9, characterized in that: The setting of the adaptive multi-level threshold is based on prior physiological knowledge and a large amount of experimental data analysis, and pre-sets L ordered fatigue state levels. , and determine the corresponding fatigue index threshold ,grade Indicates that fatigue gradually increases, and the threshold The cutoff points of different fatigue levels are defined. When the fatigue index smoothing value When the fatigue status level is determined to be level ,in, .

Images