Method for evaluating body fatigue under exercise
By collecting and analyzing electrocardiogram and electroencephalogram (EEG) signals, constructing a brain network, and selecting the optimal feature set, a scientific assessment of exercise fatigue was achieved. This solved the problem of the difficulty in accurately assessing exercise fatigue in existing technologies and avoided damage to the body caused by excessive exercise.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV OF POSTS & TELECOMM
- Filing Date
- 2023-04-24
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies cannot accurately assess the body's fatigue state during exercise in real time, which may lead to damage from excessive exercise.
ECG and EEG signals were collected at rest and after exercise. A brain network was constructed through noise reduction and feature extraction. Dual-modal detection was performed by combining ECG and EEG features, and the optimal feature set was selected for fatigue assessment.
It enables scientific assessment of exercise fatigue, avoids damage to the body from excessive exercise, and improves the accuracy and real-time nature of the assessment.
Smart Images
Figure CN116211308B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of human motion measurement technology, specifically relating to a method for assessing fatigue in the body during exercise. Background Technology
[0002] With rapid socio-economic development, modern lifestyles have undergone significant changes. Increasing evidence suggests that most adults and adolescents suffer from severe physical insufficiency due to prolonged sitting, unhealthy dietary habits, and irregular sleep patterns. These factors contribute to a rapid increase in the risk of obesity, cardiovascular disease, and related metabolic disorders worldwide, becoming a pressing concern. Regular, scientific exercise can promote the release of endorphins and other chemicals, reduce the activity of hormones like adrenaline and cortisol, improve cardiopulmonary function, boost metabolism, alleviate anxiety and stress, and promote a positive mental and physical state, thereby reducing the risk of various diseases. While exercise can strengthen the body and bring numerous health benefits, a lack of scientific knowledge about exercise and potential risks mean that excessive, inappropriate exercise can have negative effects, even threatening life.
[0003] Previous assessments of exercise status mostly relied on simple indicators such as self-perception and biochemical parameters. These methods cannot continuously monitor exercise in real time and accurately measure whether the current exercise intensity is appropriate for the body. The cardiovascular system is closely related to the body's functional state and exercise capacity; its real-time functional state reflects the degree of stimulation from exercise. During strenuous exercise, the body's oxygen demand increases, and insufficient supply causes internal environment disturbances, increasing metabolic activity and consequently increasing the workload on the heart. The heart primarily adapts to this increased demand by regulating the autonomic nervous system, specifically through the control of the sympathetic and parasympathetic nervous systems. These two types of neural activity maintain a dynamic balance, rapidly adapting to changes in the external environment and ensuring stable cardiac activity. Therefore, cardiac function is an important indicator for assessing exercise. Furthermore, while exercise increases the cardiac load, it also increases blood flow to the brain, increases neurotransmitter secretion, and alters neuronal activity within the brain. Exercise also has a certain impact on the state and function of the central and peripheral nervous systems. Exercise has a positive effect on the brain, continuously reorganizing and optimizing internal brain pathways, improving neural transmission efficiency and higher cognitive levels. Therefore, by recording and analyzing changes in cardiac and cerebral activity during exercise, we can scientifically assess exercise fatigue, which has important guiding significance for exercise training plans. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a method for assessing bodily fatigue during exercise, which includes:
[0005] S1: Collect ECG and EEG signals in the resting state and the fatigue state after exercise;
[0006] S2: Denoise the ECG signal; locate the R wave of the denoised ECG signal to obtain the RR interval sequence; obtain ECG features based on the RR interval sequence, including heart rate, time-domain features of heart rate variability and frequency-domain features of heart rate variability;
[0007] S3: Denoise the EEG signal and calculate the power spectral density of the denoised EEG signal;
[0008] S4: Construct a brain network, where nodes are EEG electrodes and edge weights are phase-locking values between nodes;
[0009] S5: Calculate brain network parameters, including clustering coefficients, feature path lengths, global efficiency, and small-world properties;
[0010] S6: Screen the ECG features, EEG signal power spectral density, and brain network parameters to obtain the optimal feature set;
[0011] S7: Perform a fatigue assessment of the organism based on the optimal feature set to obtain the fatigue assessment results.
[0012] Preferably, the formula for calculating the edge weights of a brain network is:
[0013]
[0014] in, Indicates the phase lock value. This represents the total number of nodes in the brain network. Indicates the first electrode lead signal The instantaneous phase, Indicates the second electrode lead signal The instantaneous phase, It represents the imaginary unit.
[0015] Preferably, the clustering coefficient of the brain network is the average of the clustering coefficients of all nodes; the formula for calculating the node clustering coefficient is:
[0016]
[0017] in, This represents the clustering coefficient of the i-th node. This represents the number of adjacent nodes of the i-th node. This represents the number of edges connecting the i-th node to its neighboring nodes.
[0018] Preferably, the formula for calculating the length of the feature path is:
[0019]
[0020] in, Indicates the characteristic path length of a brain network. This represents the total number of nodes in the brain network. This represents the shortest path length between two connected nodes i and j. It represents the set of all nodes in the network.
[0021] Preferably, the global efficiency is the reciprocal of the shortest path length.
[0022] Preferably, the formula for calculating small-world properties is:
[0023]
[0024] in, This represents the small-world properties of brain networks. , These represent the clustering coefficients of the current brain network and the clustering coefficients of the random network corresponding to the current network, respectively. , These represent the feature path length of the current brain network and the feature path length of the random network corresponding to the current network, respectively.
[0025] Preferably, the process of obtaining the optimal feature set includes: constructing a first feature set by combining electrocardiogram features, power spectral density of electroencephalogram signals, and brain network parameters; performing a significance analysis on the first feature set to screen out features under fatigue state that differ greatly from the resting state to form a second feature set; calculating the screening index for all features in the second feature set; constructing a third feature set by selecting the N features with the highest screening index in the second feature set; and taking the intersection of the second feature set and the third feature set as the optimal feature set.
[0026] Furthermore, the process of calculating the feature selection criteria includes: calculating the information content of the feature; calculating the correlation between features, and calculating the independence of the feature based on the correlation between the feature and other features; and using the product of the feature information content and the feature independence as the feature selection criterion.
[0027] The beneficial effects of this invention are as follows: This invention collects users' electroencephalogram (EEG) signals for brain network analysis, spatially measuring the coupling changes between channel signals. Compared with traditional time-frequency analysis, this invention can more comprehensively explore the impact of exercise on the brain. This invention combines electrocardiogram (ECG) and electroencephalogram (EEG), extracting multiple parameter features from ECG and EEG to explore the impact of exercise on the heart and brain state. By fusing ECG and EEG signals for dual-modal detection and assessment of exercise fatigue, the results are more accurate than fatigue assessment based on a single physiological signal, enabling more scientific monitoring of exercise intensity and avoiding damage to the body from excessive exercise. Attached Figure Description
[0028] Figure 1 This is a flowchart of the method for assessing body fatigue during exercise in this invention. Detailed Implementation
[0029] 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.
[0030] This invention proposes a method for assessing fatigue in the body during exercise, such as... Figure 1 As shown, the method includes the following:
[0031] S1: Collect ECG and EEG signals in the resting state and in the fatigue state after exercise.
[0032] The system collects ECG and EEG signals from users in two states: resting state (i.e., no movement) and post-exercise fatigue state. For the post-exercise fatigue state, ECG and EEG signals from multiple fatigue states need to be collected during training data collection.
[0033] S2: Denoise the ECG signal; locate the R wave of the denoised ECG signal to obtain the RR interval sequence; based on the ECG characteristics of the RR interval sequence, including heart rate, time-domain characteristics of heart rate variability and frequency-domain characteristics of heart rate variability.
[0034] The ECG signal is denoised using the following methods: a Butterworth bandpass filter is used to filter out ECG signals with frequencies between 3 and 25 Hz; median filtering is used for baseline correction of the ECG signal, using the following formula:
[0035]
[0036] in, To remove baseline drift noise from the electrocardiogram signal, The ECG signal to be processed; The output sequence, composed of the median selected after sorting the electrocardiogram data in the moving window, is the baseline drift data of the signal.
[0037] Locate the R wave in the denoised ECG signal to obtain the RR interval sequence, specifically:
[0038] Signal differential processing: The changes in the rising and falling slopes of the R-wave are the steepest and most obvious. The signal is first processed using first-order forward differential processing, and the difference equation is expressed as:
[0039]
[0040] in, Represents a difference sequence; The sequence is the signal to be detected.
[0041] Nonlinear mapping: A cubic power function is used to nonlinearly amplify the difference sequence values point by point, while preserving the original positive and negative signs of the difference waveform. At this point, the difference sequence... The expression is:
[0042]
[0043] Set the initial window width and initial threshold: Iterate through the ECG data sequentially using a sliding window. Initial window width Take the value of the sampling frequency. Search window for internal electrical signals Maximum amplitude The expression for determining the initial amplitude threshold is:
[0044]
[0045] in, The ECG amplitude threshold; For the coefficient, preferably, k can take the value of 0.5.
[0046] R-wave detection: Searching for differential sequences within the width of a sliding window. The location point corresponding to the maximum value The position point corresponding to the minimum value of the difference sequence , At this time Calculate the ECG signal to be detected within the interval The maximum value. If If so, this point is initially identified as point R, and the time corresponding to this point is recorded in a new sequence. middle.
[0047] Window width and amplitude threshold update: After determining the positions of the first two R waves, the RR interval width is obtained. At this point, the current window width is updated. Update:
[0048]
[0049] If the current R-wave amplitude is Then the ECG amplitude threshold Updated to:
[0050]
[0051] in, 0.7 is acceptable.
[0052] R-wave false positive / false negative compensation: Sequence obtained from R-wave localization Then, perform first-order differencing to obtain the RR interval sequence. The current RR interval is then... Interval with the previous one Compare. If If this R-wave is missed between the previous R-wave, then the amplitude threshold should be lowered to 90% of the original threshold and the detection should be repeated; if If the interval between the point and the previous R wave is too small, it is considered a false detection and the point is deleted from the sequence.
[0053] The R-wave is detected using the updated window width and amplitude threshold until the differential sequence ends, thus obtaining a complete RR interval sequence.
[0054] The temporal and frequency domain characteristics of heart rate and heart rate variability were obtained from the RR interval sequence.
[0055] Specifically, the heart rate value is obtained by dividing 60 by the RR interval; the heart rate variability (HRV) signal is a sequence composed of multiple RR intervals. The HRV sequence is directly statistically analyzed in a specific time order to calculate time-domain features, including: the standard deviation of the normal sinus RR interval SDNN, the root mean square value of the difference between adjacent RR intervals RMSSD, and the percentage of heartbeats with a difference of more than 50 ms between adjacent RR intervals in the entire RR interval sequence PNN50; the power values of different frequency bands of the HRV signal are extracted by Fourier transform to obtain frequency-domain features, including: total power TP, low-frequency power LF, high-frequency power HF, and the ratio of low-frequency power to high-frequency power LF / HF.
[0056] S3: Denoise the EEG signal and calculate the power spectral density of the denoised EEG signal.
[0057] The EEG signal is denoised by using a bandpass filter to filter out the signal frequency range of 0.5-30Hz, and then using an independent component analysis algorithm to remove artifacts such as electrooculography and electromyography from the EEG signal to obtain the denoised EEG signal.
[0058] The denoised EEG signal was decomposed into four basic rhythm waves: delta, theta, alpha, and beta using a Butterworth filter. The power spectral density of each rhythm wave was then calculated using the Welch method. Specifically, the signal of length... EEG data Divided into paragraph, each paragraph The data point, the first The segment data is represented as:
[0059]
[0060] window function Add it to each data segment, calculate the periodicity chart for each segment, and so on. The periodic chart is as follows:
[0061]
[0062] in, , is the normalization factor.
[0063] Furthermore, since the Welch algorithm allows for signal overlap, this overlap reduces the correlation between different signal segments, thereby reducing errors. Averaging the power spectrum of the overlapping portions yields the power spectral density of the entire segment.
[0064]
[0065] S4: Construct a brain network where nodes are EEG electrodes and edge weights are phase-locked values between nodes.
[0066] EEG electrodes are used as nodes in the network. The edge weights between nodes are obtained by measuring the phase synchronization between channels. The correlation matrix is calculated based on the phase synchronization, and the adjacency matrix is obtained by thresholding. The brain network is then constructed based on the adjacency matrix.
[0067] Phase lock value (PLV) is used to measure phase synchronization, which minimizes the influence of instantaneous amplitude and thus allows for better study of the relationships between multi-channel EEG signals. The phase lock value is calculated as follows:
[0068]
[0069] in, Indicates the phase lock value. This represents the total number of nodes in the brain network. Indicates the first electrode lead signal That is, the instantaneous phase of the EEG signal. Indicates the second electrode lead signal The instantaneous phase, It represents the imaginary unit.
[0070] EEG signals The instantaneous phase is calculated using the Hilbert transform:
[0071] The signal undergoes Hilbert transform:
[0072]
[0073] structure The analytic signal is:
[0074]
[0075] in, For signal The amplitude.
[0076] Therefore, the instantaneous phase is:
[0077]
[0078] S5: Calculate brain network parameters, including clustering coefficients, feature path lengths, global efficiency, and small-world properties.
[0079] Clustering coefficients represent the probability that adjacent nodes of a node are also adjacent nodes. Clustering coefficient The calculation formula is:
[0080]
[0081] in, This represents the clustering coefficient of the i-th node. This represents the number of adjacent nodes of the i-th node. This represents the number of edges connecting the i-th node to its neighboring nodes.
[0082] Clustering coefficient of the entire brain network for:
[0083]
[0084] in, This represents the total number of nodes in the brain network.
[0085] Two connected nodes in a network The path with the fewest edges between two nodes is the shortest path between them. The characteristic path length can be obtained by averaging the shortest paths between all nodes in the network. , is represented as:
[0086]
[0087] in, This represents the shortest path length between two connected nodes i and j. It represents the set of all nodes in the network.
[0088] Overall efficiency The reciprocal of the shortest path length is expressed as:
[0089]
[0090] Small-world networks are a network model that lies between regular and random networks. They combine the high clustering of regular networks with the short-distance nature of random networks, offering the topological advantages of both. This makes them an economical structure. Brain networks possess small-world properties, defined as follows:
[0091]
[0092]
[0093]
[0094] in, This represents the small-world properties of brain networks. This represents the clustering coefficient of the current brain network. This represents the clustering coefficient of the random network corresponding to the current network. This indicates the length of the characteristic path in the current brain network. This represents the length of the characteristic path of the random network corresponding to the current network.
[0095] S6: Screen the ECG features, EEG signal power spectral density, and brain network parameter features to obtain the optimal feature set.
[0096] ECG characteristics Power spectral density of EEG signals and brain network parameters Constitute the first feature set , .
[0097] The first feature set is determined based on the resting state and the fatigue state after exercise. A significance analysis was conducted to select features from the fatigue state that differed significantly from the resting state, thus forming a second feature set. Specifically: A significance level less than 0.05 indicates a significant difference, based on the first feature set. Features under fatigue states with a significance greater than 0.05 were selected to form the second feature set. .
[0098] Calculate the second feature set The selection criteria for all features are calculated as follows:
[0099] Measure features using the product of variance and information entropy. The amount of information contained :
[0100]
[0101]
[0102]
[0103] in, The variance of the feature; Indicates the first The eigenvalues of each feature; The length of the feature; For the first The mean of each feature; Information entropy of features; The marginal probability density function is the feature.
[0104] The correlation of features before using the Pearson correlation coefficient is expressed as:
[0105]
[0106] in, and They represent the first The mean of the first feature and the second feature The sample at the th The values of each feature, and They represent the first The mean of the first feature and the second feature One sample in The values of each feature; The number of samples represents the number of times fatigue state data was collected.
[0107] The independence of a feature is calculated based on its correlation with other features, and is expressed as:
[0108]
[0109] in, Indicates the independence of feature i. This represents the correlation between feature i and feature k. This indicates the number of features in the second feature set.
[0110] The product of feature information content and feature independence is used as the feature selection criterion. , represented as:
[0111]
[0112] Based on comprehensive screening criteria The second feature set is selected based on the value of the feature. The N features with the highest screening index constitute the third feature set. N can be selected as an appropriate value according to actual needs. Preferably, N is 50% of the total number of features; the second feature set is taken. With the third feature set The intersection of these features is used as the optimal feature set. .
[0113] The large number of unfiltered features leads to the curse of dimensionality, affecting the subsequent detection and evaluation of exercise fatigue. The above method optimizes the feature set by filtering, which is simple in principle, has low computational complexity, and the filtered features are highly correlated with the body's state.
[0114] S7: Perform a fatigue assessment of the organism based on the optimal feature set to obtain the fatigue assessment results.
[0115] Preferably, machine learning methods are used for exercise fatigue detection and assessment, and the optimal feature set is selected and optimized. The system also incorporates a corresponding fatigue level input support vector machine to detect the body's fatigue state and obtain fatigue assessment results. Real-time acquisition of ECG and EEG data from the user's resting and fatigue states is performed, processed, and the optimal feature set is calculated and input into the trained model. This allows for real-time monitoring of the user's fatigue state, enabling motion monitoring and preventing damage from excessive exercise.
[0116] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for assessing physical fatigue during exercise, characterized in that, include: S1: Collect ECG and EEG signals in the resting state and the post-exercise fatigue state; S2: Denoise the ECG signal; locate the R wave of the denoised ECG signal to obtain the RR interval sequence; obtain ECG features based on the RR interval sequence, including heart rate, time-domain features of heart rate variability and frequency-domain features of heart rate variability; S3: Denoise the EEG signal and calculate the power spectral density of the denoised EEG signal; S4: Construct a brain network, where nodes are EEG electrodes and edge weights are phase-locking values between nodes; S5: Calculate brain network parameters, including clustering coefficients, feature path lengths, global efficiency, and small-world properties; S6: The optimal feature set is obtained by filtering the power spectral density of ECG features, EEG signals, and brain network parameters; The process of finding the optimal feature set includes: constructing a first feature set from ECG features, power spectral density of EEG signals, and brain network parameters; performing a significance analysis on the first feature set to select features from the fatigue state that differ significantly from the resting state to form a second feature set; calculating the selection index for all features in the second feature set; and constructing a third feature set from the N features with the highest selection index in the second feature set; and taking the intersection of the second and third feature sets as the optimal feature set. The process of calculating the selection criteria for features includes: calculating the information content of the features, including: using the product of variance and information entropy to measure the amount of information contained in feature i. ; ； in, The variance of the feature; Indicates the first The eigenvalues of each feature; Information entropy of features; The correlation between features is calculated using the Pearson correlation coefficient, and the independence of features is calculated based on their correlation with other features, expressed as: ； in, Indicates the independence of feature i. This represents the correlation between feature i and feature k. This indicates the number of features in the second feature set; The product of the information content of a feature and the independence of the feature is used as the feature selection criterion. S7: Perform fatigue assessment on the body based on the optimal feature set to obtain fatigue assessment results; including: using machine learning methods to detect and assess exercise fatigue, inputting the selected optimal feature set and corresponding fatigue degree into a support vector machine to detect the fatigue state of the body and obtain fatigue assessment results.
2. The method for assessing physical fatigue during exercise according to claim 1, characterized in that, The formula for calculating the edge weights of a brain network is: ； in, Indicates the phase lock value. This represents the total number of nodes in the brain network. Indicates the first electrode lead signal The instantaneous phase, Indicates the second electrode lead signal The instantaneous phase, It represents the imaginary unit.
3. The method for assessing physical fatigue during exercise according to claim 1, characterized in that, The clustering coefficient of a brain network is the average of the clustering coefficients of all nodes; the formula for calculating the node clustering coefficient is: ； in, This represents the clustering coefficient of the i-th node. This represents the number of adjacent nodes of the i-th node. This represents the number of edges connecting the i-th node to its neighboring nodes.
4. The method for assessing physical fatigue during exercise according to claim 1, characterized in that, The formula for calculating the feature path length is: ； in, Indicates the characteristic path length of a brain network. This represents the total number of nodes in the brain network. This represents the shortest path length between two connected nodes i and j. It represents the set of all nodes in the network.
5. The method for assessing physical fatigue during exercise according to claim 1, characterized in that, Global efficiency is the reciprocal of the shortest path length.
6. The method for assessing physical fatigue during exercise according to claim 1, characterized in that, The formula for calculating small-world properties is: ； in, This represents the small-world properties of brain networks. , These represent the clustering coefficients of the current brain network and the clustering coefficients of the random network corresponding to the current network, respectively. , These represent the feature path length of the current brain network and the feature path length of the random network corresponding to the current network, respectively.