Construction method of brain electrical physiological indexes of athletes based on two-dimensional motor imagery task

Abstract

The application provides a method for constructing an athlete's electroencephalogram physiological index based on a two-dimensional motor imagery task, and belongs to the technical field of electroencephalogram signal processing. The method comprises the following steps: step 1, designing a dynamic motor imagery and visual motor imagery experiment; step 2, collecting original electroencephalogram signals by using an electroencephalogram device; step 3, electroencephalogram signal preprocessing; step 4, calculating and analyzing time domain, frequency domain and entropy domain electroencephalogram characteristics to construct a multi-domain feature fusion matrix, wherein the time domain adopts AR model parameters, the frequency domain adopts an alpha / beta band power ratio, and the entropy domain proposes a cross-scale cumulative multi-scale sample entropy; and step 5, introducing the multi-domain feature fusion matrix into an SVM classifier for recognition analysis. The method selects channels related to the dynamic and visual cortices, constructs an athlete's electroencephalogram physiological index system through complementary fusion of multi-domain characteristics, and both types of motor imagery tasks have a high classification accuracy, thereby providing an objective basis for sports literacy evaluation and neural mechanism research.

Description

Technical Field

[0001] This application belongs to the field of EEG signal processing technology, especially the field of athlete motor literacy recognition based on motor imagination EEG signals, specifically a method for constructing athlete EEG physiological indicators based on a two-dimensional motor imagination task. Background Technology

[0002] With the development of modern competitive sports, research on athletic performance has gradually expanded from simple physical fitness and technical training to the neuroscience level. Long-term professional training induces neuroplasticity changes in the athlete's brain, forming specific neural activation patterns. The formation of motor skills is closely related to brain neural activity. Electroencephalography (EEG), as a non-invasive neuroimaging technique, can record the electrical activity of the cerebral cortex with high temporal resolution, thus reflecting changes in the electrical activity of brain neuronal populations. It is an important tool for studying motor control and motor cognition. In recent years, analyzing the neural activity characteristics of athletes during exercise based on EEG signals has become an important research direction in the fields of sports science and neural engineering.

[0003] Among numerous methods in motor neuron research, motor imagery is considered an effective cognitive training technique. Motor imagery refers to an individual simulating the movement process through internal brain representations without actually performing the movement. Depending on the method of imagery, motor imagery can generally be divided into kinesthetic motor imagery (KMI) and visual motor imagery (VMI). KMI focuses on the muscle sensations and motor experiences experienced when an individual perceives body movement from a first-person perspective, while VMI relies more on visual representations to recreate the movement scene. Numerous studies have shown that high-quality motor imagery involves the same processes as motor execution, activating similar neural mechanisms and cortical activation patterns to actual movement—a functional equivalence model. Furthermore, motor cognitive models suggest functional similarities between motor imagery and motor execution; therefore, motor imagery has been widely applied in fields such as sports training, brain-computer interfaces, and sports rehabilitation.

[0004] According to cognitive development theory, physical activity can trigger changes in brain structure (such as neurogenesis) through the cascade effect of neurotrophic factors, thereby promoting neuroplasticity and improving cognitive performance. Basketball, as a competitive sport that highly relies on rapid decision-making and complex motor coordination, places high demands on athletes' neurocognitive and motor control abilities. During long-term training, basketball players' brains gradually develop specific neural patterns related to motor skills. Therefore, studying the brain activity characteristics of basketball players during motor imagery using electroencephalography (EEG) is of great significance for revealing the mechanisms of motor skill formation, evaluating the effectiveness of sports training, and optimizing sports training strategies.

[0005] However, most existing EEG studies on motor imagery focus on ordinary subjects, with relatively few studies targeting professional athletes. Furthermore, the effectiveness of traditional sports psychology training often relies on subjective questionnaires or coaches' experience-based judgments, lacking objective, quantitative neurophysiological indicators. Moreover, in terms of motor imagery paradigm design, many experiments focus only on a single type of motor imagery task, lacking systematic comparisons of differences between athletes and non-athletes based on different imagery modes (such as kinesthetic and visual-motor imagery). In addition, regarding EEG signal feature extraction, some studies often analyze EEG signal features from only a single dimension, failing to comprehensively reflect complex neural activity information, thus hindering a deeper understanding of the neural mechanisms of motor skills.

[0006] The following is a comparison between this application and existing technologies:

[0007] 1. Technical comparison with patent application CN121354906A, "System, Method and Medium for EEG-based Motor Imagination Ability Assessment and Training Enhancement":

[0008] (1) The two have the following essential differences in implementing technical solutions and solving problems.

[0009] Patent application CN121354906A discloses a method for assessing and enhancing motor imagination ability based on EEG. It uses multi-feature fusion technology to extract time-domain, frequency-domain, and spatial-domain features to achieve quantitative assessment of motor imagination ability. Its core assessment indicator is the lateralization index, and it includes an adaptive training module that dynamically adjusts the training difficulty based on the assessment results.

[0010] This application does not focus on general multi-feature fusion and single-sided assessment, but rather designs a two-dimensional motion imagery experimental task for athletes, encompassing kinesthetic motor imagery (KMI) and visual-motor imagery (VMI), constructing a framework covering the time domain (AR model parameters) and the frequency domain (…). The system integrates complementary features from multiple domains, including the frequency band power ratio and the entropy domain (multi-scale cumulative sample entropy), and finally imports them into an SVM classifier for the recognition of motor literacy.

[0011] (2) The two have the following essential differences in application scenarios and research objectives.

[0012] Patent application CN121354906A mainly provides a quantitative assessment and adaptive difficulty adjustment scheme for rehabilitation training of motor imagery brain-computer interface system. It mainly addresses the question of "how to improve the effect of rehabilitation training" and is aimed at the personalized rehabilitation process of ordinary patients.

[0013] This application is aimed at the research of neural mechanisms and sports literacy assessment of professional athletes in the field of competitive sports. It focuses on the changes in neuroplasticity caused by long-term professional training and aims to establish an objective and quantitative electroencephalographic indicator system to distinguish professional athletes from the general population, in order to replace traditional questionnaires and subjective experience assessments.

[0014] 2. Technical comparison with patent application CN121365240A, "A motion imagery coupling system and method based on emotion prediction":

[0015] (1) The two have the following essential differences in implementing technical solutions and solving problems.

[0016] Patent application CN121365240A discloses a motor imagery coupling system and method based on emotion prediction. By collecting EEG data from emotion-related brain regions and motor perception-related brain regions, the system weights and fuses the emotion index weight vector with the motor feature vector to obtain a weighted and enhanced motor imagery feature vector.

[0017] This application does not focus on the weighting of emotion prediction and emotion feature vectors, but instead specifically designs two-dimensional experimental tasks: kinesthetic imagery (KMI) and visual-motor imagery (VMI), from the time domain (AR model parameters) and frequency domain ( The underlying EEG signal features are extracted from three independent dimensions: frequency band power ratio, entropy domain (multi-scale cumulative sample entropy across scales), thereby constructing a multi-domain feature fusion matrix.

[0018] (2) The two have the following essential differences in application scenarios and research objectives.

[0019] Patent application CN121365240A mainly provides a solution to reduce decoding deviations caused by emotional interference. It mainly addresses the problem of "how to improve the decoding accuracy of motion imagery". Its ultimate goal is to output control signals based on the decoding results to drive the terminal device.

[0020] This application targets the research scenario of sports science and motor neuron mechanisms, focusing on the specific neural activation patterns formed by basketball players during long-term specialized training. It aims to construct an objective and quantitative electroencephalographic indicator system through multi-domain feature complementarity, in order to systematically evaluate the athletes' athletic performance.

[0021] 3. Technical comparison with patent application CN112084879A, "A method for extracting block selection cosmic pattern features from motor imagery EEG":

[0022] (1) The two have the following essential differences in implementing technical solutions and solving problems.

[0023] Patent application CN112084879A discloses a method for block selection and cosmic pattern feature extraction in motor imagery EEG. The method preprocesses the data by dividing each channel into frequency bands to construct data blocks, calculates the Fisher ratio to filter the data blocks, and finally uses cosmic pattern (CSP) and support vector machine (SVM) to extract features and classify the data composed of the optimal blocks.

[0024] This application does not focus solely on CSP spatial filtering and data block selection, but rather deeply integrates the nonlinear dynamics of signals, innovatively combining time-domain AR model parameters and frequency-domain... The physiological characteristics of three different dimensions are identified: frequency band power ratio, entropy domain cross-scale cumulative multi-scale sample entropy, etc.

[0025] (2) The two have the following essential differences in application scenarios and research objectives.

[0026] Patent application CN112084879A mainly provides a feature extraction algorithm scheme to eliminate differences between individuals and channels. It mainly addresses the problem of "how to improve the classification performance of general BCI (brain-computer interface) systems" and provides a general approach for EEG signal feature extraction.

[0027] This application focuses on a specific comparison between professional basketball players and non-athletes. To address the shortcomings of a single imagination mode in fully reflecting the neural mechanisms of motor skills, a dual-dimensional motor imagination task combining kinesthetic and visual perception was specifically designed to establish a dedicated EEG physiological indicator system for professional athletes.

[0028] 4. Technical comparison with patent application CN121331365A, "Method and related device for optimizing the experimental paradigm of lower limb motor imagery through dynamic mindfulness training closed loop":

[0029] (1) The two have the following essential differences in implementing technical solutions and solving problems.

[0030] Patent application CN121331365A discloses a method for optimizing the experimental paradigm of lower limb motor imagery through a closed-loop dynamic mindfulness training. Based on the dynamic calculation of electrophysiological indicators through a sliding window, the method generates a comprehensive score by weighting and fusing the clarity of motor imagery, focus, and mindfulness. The method dynamically adjusts the stimulus parameters and uses a two-dimensional temporal convolutional network (TCN) model to decode motor intentions.

[0031] This application does not focus on the dynamic closed-loop adjustment of stimulus parameters or simple lower limb tasks. Instead, it adopts a rigorous subjective and objective cross-validation experimental paradigm of kinesthetic and visual two-dimensional motor imagery (combined with MIQ-R and KVIQ scales for evaluation), and uses a clear multi-domain feature system in the time domain, frequency domain and entropy domain in the feature decoding stage, which is directly imported into the traditional SVM classifier for recognition.

[0032] (2) The two have the following essential differences in application scenarios and research objectives.

[0033] Patent application CN121331365A mainly provides a paradigm optimization scheme for clinical rehabilitation training of patients with stroke, and mainly addresses the problem of "how to improve the evoked rate and stability of EEG features in lower limb motor imagery tasks".

[0034] This application targets healthy professional athletes in the field of competitive sports, addressing the question of "how to objectively quantify the changes in professional athletes' athletic literacy and brain neuroplasticity," and providing objective neuroscientific evidence for evaluating the effectiveness of sports training, optimizing training strategies, and revealing the mechanisms of sports skill formation.

[0035] Therefore, conducting systematic research on the EEG signals of motor imagination in basketball players, and designing subjective and objective experimental paradigms that include kinesthetic and visual-motor imagination, to compare and analyze the differences in EEG characteristics between basketball players and non-athletes, is of great significance for revealing the impact of sports training on brain function and constructing a more scientific method for assessing motor neuron function. Summary of the Invention

[0036] To address the shortcomings of existing EEG research methods for motor imagery in the field of sports science, this application proposes a method for constructing EEG physiological indicators for athletes based on a two-dimensional motor imagery task. This method aims to solve the problem that current research subjects mainly focus on the general population and lack research on the neural activity characteristics of professional athletes. Furthermore, existing motor imagery experimental paradigms often employ single-task designs, lacking comprehensive research on the two different modes of imagery: kinesthetic and visual-motor imagery. Moreover, the evaluation of the effectiveness of sports psychology training relies heavily on subjective experience and judgment, lacking objective EEG indicators. In addition, during EEG signal analysis, existing research methods have limited ability to uncover the complex features of EEG signals, making it difficult to comprehensively reflect the changes in brain neural activity during motor skill training.

[0037] To achieve the above objectives, this application employs the following technical solution: This application designs a subjective and objective motor imagery experimental paradigm that includes kinesthetic and visual-motor imagery, collects EEG signals from basketball players and non-athletes on different motor imagery tasks, and fuses time-domain AR model parameters and frequency-domain parameters. The proposed multi-scale cumulative multi-scale sample entropy multi-domain feature algorithm, based on the frequency band power ratio and entropy domain, establishes an EEG physiological index system to assess the subject's motor literacy, achieving efficient recognition of both kinesthetic and visual-motor imagery EEG.

[0038] The technical solution of this application is: a method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task, specifically including the following steps:

[0039] Step 1, Experimental Paradigm Design: Design experiments on kinesthetic and visual-motor imagery;

[0040] Step 2, EEG signal acquisition: Raw EEG signals are acquired using an EEG device;

[0041] Step 3, EEG signal preprocessing: Preprocessing based on EEGLAB toolbox and denoising verification of VMD-PE-WTD algorithm, where VMD-PE-WTD algorithm verified its excellent denoising performance. Preprocessing is mainly based on the analysis and processing of effective EEG data of motor imagery.

[0042] Step 4, Feature Extraction and Fusion: Calculate and analyze the parameters of the 6th-order AR model in the time domain and the frequency domain respectively. Bandwidth power ratio and entropy domain and The feature values ​​of multi-scale sample entropy accumulated across frequency bands were used to extract EEG feature values ​​of different numbers of electrode channels based on the differences in experimental tasks. Among them, the EEG features of 9 electrode channels were extracted for the kinesthetic imagination experiment, including the frontal lobe F3, F4, and Fz channels, the central region C3, C4, and Cz channels, and the parietal lobe P3, P4, and Pz channels. The EEG features of 5 electrode channels were extracted for the visual-motor imagination experiment, including the parietal lobe P3, P4, and Pz channels and the occipital lobe O1 and O2 channels. Thus, a multi-domain feature fusion matrix was constructed.

[0043] Step 5, SVM recognition: Import the feature matrix calculated and analyzed in step 4 into the Matlab classifier SVM for classification to achieve recognition of motor literacy.

[0044] Furthermore, the requirements for experimental subjects, experimental environment, and experimental equipment during EEG signal acquisition in step 2 are as follows:

[0045] (1) Experimental subjects: 10 first-level basketball players and 10 non-athletes with no competition or training experience were selected from the sports department. All 20 subjects were college students aged 19-23 years old, who were right-handed, had normal or corrected vision, and had no neurological or musculoskeletal diseases.

[0046] (2) Experimental environment: a comfortable, quiet, and electromagnetically free shielded indoor environment;

[0047] (3) Experimental equipment: 19-channel wet electrode EEG acquisition equipment.

[0048] Furthermore, step 3 includes the following steps:

[0049] Step 3.1, EEGLAB Toolbox Preprocessing: First, electrode positioning is performed, and 50Hz power frequency interference is removed. At the same time, a 0.5-40Hz bandpass filter is applied to remove high-frequency signal interference. Next, the sampling rate is reduced to 256Hz, and bad conductors and bad segments are interpolated and removed. Then, A1 / A2 bilateral mastoid electrodes are used for rereference. Finally, independent component analysis (ICA) is performed to remove artifacts and bad segment components related to electrooculography (EOG) and electromyography (EMG).

[0050] Step 3.2, VMD-PE-WTD algorithm denoising: First, the signal processed in Step 3.1 is subjected to variational mode decomposition to obtain IMF components, where the number of modes K is set to 6 and the penalty factor alpha is set to 3000; then, the normalized permutation entropy of each IMF component is calculated, and the normalized permutation entropy threshold is set to 0.8. Based on the set permutation entropy threshold, the IMF components are divided into noise components and effective signal components; finally, wavelet threshold denoising is performed on the noise components. Wavelet threshold denoising includes wavelet decomposition, thresholding, and wavelet reconstruction. The decomposition level is set to 5, and the db4 wavelet is selected for denoising. The threshold calculation formula is as follows: ,in It is the noise standard deviation. It is the signal length. The sign is logarithmic, and the signal is finally reconstructed from the effective signal components to obtain the denoised signal.

[0051] Furthermore, the algorithm steps for calculating the permutation entropy described in step 3.2 are as follows:

[0052] Step 1: Phase space reconstruction;

[0053] Let the original time series be... Given the embedding dimension and time delay The sequence is reconstructed in phase space to obtain a set of reconstruction matrices, where the embedding dimension is... Set to 3, time delay Set to 1;

[0054] The reconstructed first Each component Represented as:

[0055] ;

[0056] in, ,and each They are all one A subsequence of dimension;

[0057] Step 2: Ascending order arrangement and symbolic mapping;

[0058] For each reconstructed component of The elements are sorted in ascending order of their numerical values;

[0059] Assumption The original index position corresponding to the element in the middle is The sorted sequence is:

[0060] ;

[0061] If there are equal values, sort them according to their index numbers, that is, the smaller index is placed first;

[0062] Therefore, each component All of them can be mapped to a unique permutation pattern or symbol sequence. :

[0063] ;

[0064] Step 3: Calculate the probability distribution of the permutation pattern;

[0065] exist In 3D space, there are a total of 100 permutation patterns. Count the possible combinations and statistically analyze each permutation pattern. The number of times it appears in all reconstructed vectors is denoted as . ;

[0066] Calculate the probability of each permutation pattern occurring. :

[0067] ;

[0068] Step 4: Calculate the permutation entropy;

[0069] According to the definition of Shannon entropy, the embedding dimension of this time series is calculated. The entropy of the arrangement :

[0070] ;

[0071] Step 5: Normalization process;

[0072] because The maximum value is When all permutation patterns occur with equal probability, they are normalized to facilitate comparison under different parameters, resulting in the normalized permutation entropy. :

[0073] ;

[0074] The range of values ​​is :

[0075] The closer to 0, the more regular and simple the sequence;

[0076] The closer to 1, the more random and complex the sequence.

[0077] Permutation entropy is a nonlinear method for measuring the complexity and randomness of a signal. It can quantitatively assess the random noise contained in a signal and has advantages such as fast computation speed and strong noise resistance.

[0078] The permutation entropy is positively correlated with the noise power. The larger the permutation entropy, the more complex the signal and the more noise it contains.

[0079] Furthermore, step 4 includes the following steps:

[0080] Step 4.1, Feature Extraction: Starting from the kinesthetic and visual-motor imagery experimental tasks, electrode positioning was performed according to the 10-20 international standard lead system placement method. To accurately capture the athlete's specific neural activation patterns, the kinesthetic imagery experiment selected nine features: the frontal lobe (left F3, right F4, midline Fz) located in the front of the brain, the central area spanning the vertex motor cortex (left C3, right C4, midline Cz) and the parietal lobe located in the upper back of the brain (left P3, right P4, midline Pz). Electrode channels were used to fully cover the complete cortical mapping from motor intention planning (frontal lobe) to motor execution (central area) and then to proprioceptive feedback (parietal lobe). The visual-motor imagery experiment selected five electrode channels: the parietal lobe (left P3, right P4, midline Pz) located in the upper posterior part of the brain, and the occipital lobe (left O1, right O2) covering the visual center of the posterior brain, to capture visual-spatial depth processing features. Subsequently, the Burg algorithm was used to extract the temporal 6th-order AR model parameter features, and wavelet packet decomposition was used to generate key frequency bands for motor imagery. and ,extract Band power ratio characteristics, and based on and Two-band cross-scale cumulative multi-scale sample entropy features; finally, 81 feature parameters were calculated for kinesthetic motion imagination and 45 feature parameters were calculated for visual motion imagination;

[0081] Step 4.2, Feature Fusion: The EEG feature parameters of basketball players and non-athletes extracted in Step 4.1 are fused to obtain a multi-domain feature fusion vector, and then a multi-domain feature fusion matrix is ​​constructed to generate a kinesthetic motion imagery sample feature dataset and a visual motion imagery sample feature dataset, respectively.

[0082] This application first conducts a design logic analysis based on the physiological characteristics of the athlete population:

[0083] During long-term, high-intensity specialized training, basketball players experience specific neuroplasticity changes in their brain's motor control center and visual-spatial processing center. Compared to the general population, high-level athletes exhibit higher neural network activation efficiency and more precise cortical resource allocation when engaging in motor imagery. To accurately characterize this specificity, this application specifically constructs a multi-domain feature association system encompassing the time domain, frequency domain, and entropy domain.

[0084] (1) At the time domain level, the autoregressive evolution law of the EEG sequence of the moment of motion imagination can be extracted by using AR model parameters, which can serve as an important indicator to characterize the dynamic rhythm of the cortex.

[0085] (2) At the frequency domain level, using The bandwidth power ratio measures the level of cognitive resource consumption and neural activation in the brain during motor imagery tasks, and can serve as an important indicator for assessing brain functional status and motor-related neural activity.

[0086] (3) At the entropy domain level, the complexity, regularity and cross-scale dynamic characteristics of motor imagery EEG signals at multiple cumulative time scales can be characterized by cross-scale cumulative multi-scale sample entropy, which can be used as an important indicator to evaluate the efficiency of motor-related neural information processing and the complexity of brain function.

[0087] The fusion of the above multi-domain feature indicators enables a comprehensive mapping from modeling the dynamic changes over time and frequency band energy regulation to the complexity of brain function, establishing a professional athlete's electrophysiological indicator system that is significantly different from that of the general population.

[0088] Based on this physiological mechanism analysis, the specific extraction process is as follows:

[0089] The brain activity triggered by kinesthetic and visual-motor imagery differs. Kinesthetic activation involves the supplementary motor area (SMA), premotor area (PMC), primary motor cortex (M1), and somatosensory association cortex (S1), while visual-motor activation involves the visual cortex (V1, V2), visual association cortex (V3, V4, V5), and somatosensory association cortex (S1).

[0090] Furthermore, the AR model parameters mentioned in step 4.1 include:

[0091] (1) Construction of AR model;

[0092] Let the time series be The order is Then AR( The model can be represented as:

[0093] ;

[0094] in:

[0095] These are the parameters of the AR model, reflecting the dynamic characteristics of the signal;

[0096] For the traversal order Variables within a range;

[0097] The model order;

[0098] With zero mean and variance The white noise sequence represents the deviation between the AR model's predictions and the actual values, characterizing the part of the signal that is not explained by the model;

[0099] The response of white noise through a linear system, in this case, is an EEG signal, representing the response using past data. Known values The weighted sum predicts the current value;

[0100] The transfer function of the system is then:

[0101] ;

[0102] in:

[0103] For complex variables in the Z-transform;

[0104] EEG signals Z-transform;

[0105] Z-transform of a zero-mean white noise sequence;

[0106] The power spectrum is:

[0107] ;

[0108] in, This refers to the digital angular frequency in signal power spectrum analysis.

[0109] In summary, can be The unique representation is that AR model coefficients can be used as features of EEG signals;

[0110] (2) Burg algorithm calculates AR model parameters;

[0111] The Burg algorithm is described as follows:

[0112] First, introduce the order variable. , define the first Step, The forward prediction error and the backward prediction error are respectively and The specific recursive formula is as follows:

[0113] Initial prediction error, order :

[0114] ;

[0115] From the order of the steps recurrence order , ,right definition:

[0116] ;

[0117] ;

[0118] in, That is, the first The reflection coefficient of the first order is a real number under real signals, equivalent to the first order. The first AR model coefficients ;

[0119] The core principle of the Burg algorithm is to minimize the power of the forward prediction error. Power of backward prediction error The arithmetic mean, i.e., the average prediction error power. Minimum:

[0120] ;

[0121] make right The partial derivative is zero, that is... It can be derived that the power of the average prediction error is... The smallest First-order reflection coefficient formula:

[0122] ;

[0123] Calculate the reflection coefficient Then, the Levinson-Durbin recursive formula can be used to calculate the current nth... All autoregressive coefficients of the model and updating residual variance :

[0124] ;

[0125] ;

[0126] By from Calculate stepwise recursively to the required model order. You can then obtain the complete AR model parameters. and excitation white noise variance ;

[0127] The Burg algorithm is a parametric AR model estimation method. Its core objective is to minimize the sum of forward and backward prediction errors at each order, and to calculate the reflection coefficients (i.e., some coefficients in the Levinson-Durbin model) step by step through a recursive approach. Unlike the Yule-Walker method, the Burg algorithm does not require prior estimation of the autocorrelation matrix. Instead, it directly obtains the parameters by minimizing the forward and backward prediction errors in the time domain, resulting in more accurate estimation results that are closer to the true values.

[0128] (3) Determining the order of the AR model;

[0129] To determine the optimal model order in the AR model The residual variances of each order calculated using the Burg algorithm described above for the AR model parameters are then used. Substitute into the evaluation criteria and introduce the order variable. And use the two major criteria of FPE and AIC to perform traversal optimization;

[0130] The FPE criterion, used to balance prediction error and the number of parameters, measures the magnitude of the error when using the current model to predict future data. Its expression is:

[0131] ;

[0132] in The data length of the signal. The order of the AR model;

[0133] order The residual variance of the time model fit;

[0134] This is a complexity penalty term;

[0135] The value at a certain optimal order The location will reach its minimum;

[0136] The AIC criterion, based on the idea of ​​maximum likelihood estimation, comprehensively considers the fitting error and the number of parameters. It is used to measure the amount of loss information between the fitted model and the true mechanism that generated the data. Its expression is:

[0137] ;

[0138] The fitting error term representing the model, as the order increases... Increase, residual variance Decrease it, and the value of this term will decrease;

[0139] The penalty term, representing the complexity of the model, increases by 2 for each additional order.

[0140] The value at a certain optimal order The minimum will be reached.

[0141] Furthermore, step 4.1 as described The steps for calculating the bandwidth power ratio are as follows:

[0142] (1) Calculate the power spectral density;

[0143] EEG signals With zero mean and variance The response of white noise through a linear system is known as the AR model. The transfer function of the AR model is:

[0144] ;

[0145] in, The system polynomial of the AR model, i.e. The location of its root determines the stability and frequency domain characteristics of the system. The order of the AR model;

[0146] In a linear time-invariant discrete-time system, according to signal processing theory, the input signal at this time... power spectrum Output signal power spectrum and the frequency response function of the model system The following relationship must be satisfied:

[0147] ;

[0148] The power spectrum of the white noise input to the AR model is The parameters of the AR model are obtained using the Burg algorithm. , ;Will Substituting the transfer function into the AR model yields And the relationship between the input and output power spectrum of the aforementioned linear system. By combining these, the power spectrum of the AR model based on the Burg algorithm can be obtained;

[0149] ;

[0150] Then by Therefore, the power spectral density formula for the AR model based on the Burg algorithm is:

[0151] ;

[0152] in, For actual frequency, The sampling frequency;

[0153] (2) Calculate the bandwidth power ratio;

[0154] This application uses 6-layer wavelet packet decomposition to refine the subband bandwidth of the EEG signal to 2Hz, and selects db4 as the wavelet basis, thereby obtaining... wave and Wave;

[0155] EEG signals frequency band and The power spectral density of the frequency band is respectively used and The bandwidth power of the two target frequency bands is shown below:

[0156] ;

[0157] ;

[0158] calculate The bandwidth power ratio is expressed as follows:

[0159] .

[0160] The frequency bands most closely related to motor imagery tasks are mainly reflected in Frequency band and Frequency bands, and their energy or power variations, are the core basis for decoding motor imagery, reflecting the dynamic working state of the cerebral cortex. Increased frequency band energy typically indicates that a local cortex is in a relatively idle state or that the inhibition of irrelevant information is enhanced, reflecting the conservation of mental resources and a decrease in the activation level of the cortical region; while Increased frequency band energy is often accompanied by increased cortical excitability and enhanced active cognitive processing. Therefore, The bandwidth power ratio can serve as an important indicator for measuring the level of resource consumption and neural activation in the brain during the performance of a specific task, offering advantages such as intuitiveness and strong interpretability. A higher bandwidth power ratio... The ratio usually indicates that the brain is maintaining a low level of activity. While having a high frequency band activity level Frequency band activity level refers to the ability to complete tasks with lower cortical activation costs, and to have stronger irrelevant information suppression capabilities and resource utilization efficiency.

[0161] Furthermore, the specific calculation steps for the cross-scale cumulative multi-scale sample entropy described in step 4.1 are as follows:

[0162] Given a length of One-dimensional discrete time series Let the upper limit of the target scale factor be... Embedding dimension is , The similarity tolerance coefficient ranges from 0.1 to 0.25. The specific calculation steps of the algorithm are as follows:

[0163] Step 1: Signal coarsening processing;

[0164] For a given scale factor ,in The original time series Divided into lengths of The scale is obtained by drawing non-overlapping data windows and calculating the average value of the data points within each window. coarse-grained sequence below The calculation formula is as follows:

[0165] ;

[0166] in, Indicates the length of the coarsened sequence. This is a floor function used to remove sequences ending in less than one scale window. Redundant data points;

[0167] Step 2: Phase space reconstruction;

[0168] At each scale Below, based on coarse-grained sequences Reconstructed separately peacekeeping A sequence of phase space vectors of dimension 1. The dimensional vector is constructed as follows:

[0169] ;

[0170] The dimensional vector is constructed as follows:

[0171] ;

[0172] Step 3: Calculate the Chebyshev distance;

[0173] Chebyshev distance is used to measure the similarity between any two vectors in the reconstruction space. Any two vectors in the phase space and The distance between them is defined as the maximum absolute value of the difference between their corresponding elements:

[0174] ;

[0175] Similarly, The Chebyshev distance between the dimensional vectors is:

[0176] ;

[0177] in, Strictly exclude self-matching;

[0178] Step 4: Count the number of similar matching pairs;

[0179] Calculate coarse-grained sequences The standard deviation of is denoted as . Define similarity tolerance Iterate through all possible combinations of vector pairs, satisfying Statistical analysis peacekeeping The number of matching pairs whose subdimensional distance is less than or equal to the threshold is denoted as . For scale Down Total number of matches for a dimensional vector:

[0180] ;

[0181] remember For scale Down Total number of matches for a dimensional vector:

[0182] ;

[0183] in, This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise.

[0184] Step 5: Calculate the cumulative matching metric and the final feature value;

[0185] Regarding the cumulative characteristics of this algorithm, for the current cumulative scale... , Calculate from scale 1 to scale 2 respectively. of peacekeeping The cumulative sum of the matching logarithms of the dimension:

[0186] ;

[0187] ;

[0188] Based on the above cumulative states, the ratio of cumulative matching probabilities is calculated and the negative natural logarithm is taken to obtain the final result reflecting the time series. On the cumulative time scale Metrics of dynamic complexity :

[0189] ;

[0190] If it occurs at a certain cumulative scale In this case, at this scale Defined as 0; cumulative time scale Increasing from 1 to This will give you a complete set of vectors that reflect the characteristics of the original time series.

[0191] To improve the stability of complexity statistics under large-scale conditions and reduce the computational burden of repeated coarsening and template matching operations at each scale in composite multi-scale methods, thus alleviating the problems of high computational complexity and low computational efficiency in batch feature extraction of multi-channel EEG, this application proposes a method called Cross-scale Cumulative Multi-scale Sample Entropy (CCMSE). This method retains the traditional single coarsening framework of multi-scale entropy. After completing template matching statistics at each scale, instead of using the multi-offset coarse-grained sequence accumulation method at the same scale, it introduces a cross-scale accumulation mechanism to progressively fuse matching information at different scales. It uses the statistical results of previous scales to enhance the matching information at the current scale, thereby improving the problem of insufficient matching samples when the scale increases, enhancing the stability of entropy estimation, and improving the ability to represent EEG complexity.

[0192] Compared to traditional multi-scale entropy methods, CCMSE employs a cross-scale statistical fusion strategy to establish continuous correlations between scales, thus creating a progressive cumulative relationship between features at different scales. As the scale increases, the current-scale features not only reflect information from the current scale but also gradually fuse statistical features from previous scales. Therefore, this method can achieve statistical information enhancement while maintaining low computational complexity, making it more suitable for analyzing the complexity of short-term non-stationary signals such as motor imagery EEG.

[0193] Furthermore, based on the FPE and AIC criteria, the optimal order for the AR model is set to 6; this application selects the following parameter values. , , , Calculate the cumulative multi-scale sample entropy features across scales.

[0194] Furthermore, step 5 includes the following steps:

[0195] Step 5.1, Data partitioning: In the two datasets, the basketball player motion imagery samples are labeled as 1, and the non-athlete motion imagery samples are labeled as 2. The two datasets are randomly divided into 5 equal parts.

[0196] Step 5.2, Alternate Validation: Use 1 set as the validation set and the remaining 4 sets as the training set each time, repeating 5 times;

[0197] Step 5.3, Results Integration: Calculate the average of the 5 validation results as the final inter-group performance difference evaluation index.

[0198] This application provides a method for constructing electroencephalographic indicators for athletes based on a two-dimensional motor imagery task. It has the following beneficial effects:

[0199] (1) A dual-dimensional motor imagery experimental paradigm combining kinesthetic and visual imagery: This application breaks through the limitations of existing EEG experiments that mostly employ single-task designs, and designs a dual-dimensional motor imagery evoked experiment combining kinesthetic motor imagery (KMI) and visual-motor imagery (VMI). Subjective assessment results based on the KVIQ scale show that, according to the nonparametric Mann-Whitney U test, there is a significant difference in the vividness of motor imagery between the two groups (p<0.05), with a highly significant difference between the kinesthetic motor imagery groups (p<0.01), verifying the advantages that athletes have in terms of vividness of motor imagery, internal representation of movement, and cognitive regulation. This paradigm can more comprehensively and scientifically capture the specific neuroplasticity changes in the brain from "first-person perspective muscle sensation" to "visual representation reproduction".

[0200] (2) Design of a VMD-PE-WTD composite denoising algorithm to verify the denoising performance of EEG signals: Addressing the highly susceptible nature of EEG signals to interference, this application innovatively combines Variational Mode Decomposition (VMD), Permutation Entropy (PE), and Wavelet Threshold Denoising (WTD). The algorithm sets the normalized permutation entropy threshold to 0.8, effectively distinguishing noise from valid signal components and performing deep wavelet reconstruction on the noise components. Through rigorous evaluation using quantitative indicators such as signal-to-noise ratio (SNR), root mean square error (RMSE), and correlation coefficient (CC), this composite denoising algorithm maximizes the signal-to-noise ratio while preserving nonlinear neurophysiological characteristics, thus verifying the effectiveness of the denoising algorithm.

[0201] (3) Constructing a multi-domain feature complementarity fusion system covering the time domain, frequency domain, and entropy domain significantly improves recognition performance: This application abandons the limitations of single-dimensional analysis and constructs a system covering the time domain (AR model parameters), frequency domain (... A multi-domain feature index system including bandwidth power ratio and entropy domain (cumulative multi-scale sample entropy across scales). The AR model uses the Burg algorithm to accurately capture the dynamic dependencies of EEG signals, and the parameters of a 6th-order AR model can represent the core features of the original signal; Bandwidth power ratio focuses on aspects closely related to motion imagery and Waves quantify neural activation intensity and cognitive resource consumption from a frequency domain perspective. Cross-scale cumulative multi-scale sample entropy enhances the information continuity between the current scale and all preceding scales through a cross-scale statistical accumulation mechanism. While maintaining low computational complexity, it achieves statistical information enhancement and entropy estimation stability. Under finite accumulation time scales, it optimizes the optimal accumulation time scale factor based on the overall recognition average of five machine learning classifiers: DT, LDA, LR, KNN, and SVM. Experimental results, using 5-fold cross-validation combined with paired-samples t-tests, demonstrate that the fused feature group significantly outperforms any single-feature algorithm in terms of accuracy and robustness in motor literacy recognition within the SVM classifier (p<0.001), and the extremely short error bars verify the high stability of accuracy at each fold.

[0202] (4) Establishing an objective electroencephalographic (EEG) indicator system specifically for basketball players: This application effectively addresses the problem of existing EEG research subjects being overly concentrated in the general population, precisely targeting professional competitive athletes. It uses 10 national-level basketball players (average training years 9.3±2.37 years) and 10 non-athletes with no training experience as a comparative sample with strictly controlled variables. This system breaks away from the reliance of traditional sports psychology assessments on subjective questionnaires and coaching experience, providing an objective and quantitative EEG physiological indicator system for research on athletes' neural mechanisms, assessment of athletic literacy, and quantification of training effects in competitive sports. Attached Figure Description

[0203] Figure 1 This is a framework diagram of the method for constructing electroencephalographic indicators for athletes proposed in this application.

[0204] Figure 2 This is a schematic diagram of the experimental paradigm for two-dimensional motion imagination.

[0205] Figure 3 A breakdown diagram of the basketball throwing motion.

[0206] Figure 4 This is a significance analysis chart for the KVIQ scale.

[0207] Figure 5 This includes basic information about the subjects.

[0208] Figure 6 This is a schematic diagram of the 19-channel leads of the EEG acquisition device.

[0209] Figure 7 This is a schematic diagram of the overall preprocessing process.

[0210] Figure 8 A comparative chart of quantitative analysis of evaluation indicators for different noise reduction methods.

[0211] Figure 9 The order determination curves for AR models based on FPE and AIC criteria.

[0212] Figure 10 This is a schematic diagram of wavelet packet decomposition during frequency domain bandwidth extraction.

[0213] Figure 11 A comparative chart showing the significance of frequency domain features between groups of kinesthetic and visual motion imagination.

[0214] Figure 12 A comparative statistical analysis of the recognition accuracy of SVM classifiers using different feature extraction algorithms. Detailed Implementation

[0215] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0216] Before describing the specific steps of the embodiments of this application, the general experimental environment and statistical standards used in this embodiment are first defined: All algorithm implementations, feature extractions, machine learning analyses, and related statistical analyses of EEG signals in this embodiment are completed in the MATLAB environment. For various statistical tests involved in the embodiments, the significance level is uniformly set as follows: p < 0.05 indicates a statistically significant difference; p < 0.01 indicates that the difference is statistically significant; p < 0.001 indicates that the difference is statistically significant.

[0217] This application provides a method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task. The framework of this method is shown in the figure below. Figure 1As shown; this application designs a two-dimensional motor imagery evoked experiment to explore the specific differences in basketball players when kinesthetic and visual-motor imagery activate different brain regions; the collected effective motor imagery EEG signals are preprocessed, and then based on time-domain AR model parameters and frequency domain... The bandwidth power ratio and the entropy domain cross-scale accumulated multi-scale sample entropy are calculated and fused to obtain a multi-domain feature fusion vector; finally, the multi-domain feature fusion matrix is ​​imported into SVM for classification and recognition. The specific steps are as follows:

[0218] Step 1: Design a two-dimensional motion imagination experiment paradigm;

[0219] The flowchart of the two-dimensional motion imagination experiment paradigm designed in this embodiment is shown below. Figure 2 As shown. The experimental paradigm is based on the following three steps:

[0220] Step 1.1, Baseline Assessment of Subjects' Imagination Ability: To fully assess subjects' general motor imagination ability, a modified MIQ-R was used to measure the difficulty of four basic gross motor movements (knee raise, jump, arm movement, and bending over) in relation to kinesthetic and visual imagination before the kinesthetic and visual imagination experiments. A 7-point Likert scale was used to rate the difficulty of the imagination, ranging from 1 (very difficult to perceive / see) to 7 (very easy to perceive / see). Subjects ticked the corresponding scores on the scale based on their actual subjective feelings. Figure 2 As shown in A;

[0221] Step 1.2, Implementation of the Kinesthetic and Visual-Motor Imagination Task: The kinesiological and visual-motor imagination experiment was conducted. Each kinesiological imagination experiment lasted 11 seconds, with 9 experiments constituting one group, for a total of 5 groups. Each experiment included a 2-second focused guidance process (a "+" sign appeared on the screen), a 3-second kinesthetic motion cue process, and a 4-second closed-eye kinesiological imagination process (feeling oneself performing a basketball shooting motion). Figure 3 The diagram breaks down the basketball throwing motion, focusing on the tension changes in the arm muscles and joints from flexion to extension during the shot, and the proprioceptive sensation of the fingertips brushing the ball. A 2-second relaxation period is included. The visual-motor imagery experiment lasts 152 seconds per session, conducted three times. Each session includes a 2-second guided focus period (a "+" sign appears on the screen), a 60-second guided visual scene period, a 60-second closed-eye visual-motor imagery period (the brain visualizes the image and scene, focusing on details in time and space to form a visual image stream), and a 30-second relaxation period. Figure 2 As shown in Figure B; the final recorded time for both kinesthetic and visual-motor imagery was 180 seconds.

[0222] Step 1.3, Subjective Review and Assessment of the Vividness of Imagination: After the kinesthetic and visual-motor imagery experiments, participants completed the KVIQ scale based on the vividness of their motor imagery during the experiments, such as... Figure 2 As shown in C. Figure 4 The graph shows the significance of the KVIQ scale. After a nonparametric Mann-Whitney U test, it can be seen from the graph that there are significant differences in vividness between the two groups of motor imagery experimental groups (p<0.05), and the difference between the kinesthetic motor imagery groups is highly significant (p<0.01).

[0223] Step 2: EEG signal acquisition;

[0224] (1) Experimental Subjects: Given that the subjects' past sports experience and motor imagery ability can significantly influence the experiment, a baseline assessment of the subjects' imagery ability was conducted according to the MIQ-R in step 1 before the formal experiment. Subjects with a Likert scale score below 4 out of 7 were screened out. Ultimately, 10 national-level basketball players from the Ministry of Sports were selected as the experimental group (5 men and 5 women, average age 20.4). 0.92, age range 19-22 years old, boys train 6 times a week, girls train 5 times a week, 2 hours each time, average training years 9.3 (2.37) These 10 national-level basketball players share a unified and strong training background, resulting in high sample homogeneity and effectively controlling for the consistency of empirical variables, ensuring statistical significance of the experimental results. Simultaneously, 10 non-athletes with no competition or training experience were selected as the control group (8 men and 2 women, average age 21.1). 1.04 (Age range 20-23 years). All 20 participants were university students aged 19-23 years who underwent rigorous screening, were right-handed, had normal or corrected vision, and had no neurological or musculoskeletal disorders. Basic information of the participants is as follows: Figure 5 As shown. All subjects were fully informed of the experimental procedure for two-dimensional motor imagery before their EEG signals were collected;

[0225] (2) Experimental environment: a comfortable, quiet, and electromagnetically free shielded indoor environment;

[0226] (3) Experimental equipment: 19-channel wet electrode EEG acquisition equipment, equipped with conductive gel, syringe and international standard 10-20 system EEG cap.

[0227] The lead distribution diagram of the 19-channel EEG acquisition device in this embodiment is as follows: Figure 6 As shown, the 19-lead electrode distribution conforms to the international 10-20 system standard electrode placement method, specifically including frontal electrodes Fp1 and Fp2, frontal lobe electrodes F3, F4, F7, F8, and Fz, central lobe electrodes C3, C4, and Cz, parietal lobe electrodes P3, P4, and Pz, occipital lobe electrodes O1 and O2, temporal lobe electrodes T3, T4, T5, and T6, and reference electrodes A1 and A2.

[0228] Step 3: Preprocessing of EEG signals;

[0229] The schematic diagram of the EEG signal preprocessing process in this embodiment is shown below. Figure 7 As shown, it mainly includes two parts: EEGLAB toolbox preprocessing and VMD-PE-WTD algorithm noise reduction.

[0230] Step 3.1, EEGLAB toolbox preprocessing: First, electrode positioning is performed, and 50Hz power frequency interference is removed. At the same time, a 0.5-40Hz bandpass filter is applied to remove high-frequency signal interference. Next, the sampling rate is reduced to 256Hz, and bad conductors and bad segments are interpolated and removed. Then, A1 / A2 bilateral mastoid electrodes are used for rereference. Finally, independent component analysis (ICA) is performed to remove artifacts and bad segment components related to electrooculography (EOG) and electromyography (EMG).

[0231] Step 3.2, VMD-PE-WTD algorithm for noise reduction: First, the signal processed in Step 3.1 is subjected to Variational Mode Decomposition (VMD) to obtain IMF components, where the number of modes K is set to 6 and the penalty factor alpha is set to 3000; then, according to the normalized permutation entropy formula... The normalized permutation entropy (PE) of each IMF component is calculated. A higher PE indicates greater noise, and its threshold is set to 0.8. Based on the set PE threshold, the IMF components are divided into noise components and effective signal components. Finally, wavelet threshold denoising (WTD) is performed on the noise components. Wavelet threshold denoising includes wavelet decomposition, thresholding, and wavelet reconstruction. In wavelet decomposition, the decomposition level parameter is set to 5, and the db4 wavelet is used for denoising. The threshold calculation formula is as follows: ,in It is the noise standard deviation. It is the signal length, which is ultimately reconstructed with the effective signal components to achieve further noise reduction and improve the signal-to-noise ratio. Figure 8 A comparative chart of quantitative analysis of evaluation metrics for different denoising methods is presented. Although different denoising algorithms were used for comparative analysis, since these data are essentially samples of the same nature, they meet the strict homology condition. Furthermore, considering that the Wilcoxon signed-rank test does not rely on the normal distribution assumption of the data and has high robustness, the Wilcoxon signed-rank test was used for algorithm difference comparison analysis. Statistical results show that the VMD-PE-WTD composite denoising algorithm exhibits extremely significant performance advantages in the three core evaluation metrics: signal-to-noise ratio (SNR), root mean square error (RMSE), and correlation coefficient (CC). The Wilcoxon signed-rank test shows that the composite algorithm has extremely significant statistical differences compared to the individual WTD and VMD-PE algorithms (p<0.001).

[0232] Figure 8The noise content assessment metrics are signal-to-noise ratio (SNR), root mean square error (RMSE), and correlation coefficient (CC).

[0233] The formula for calculating the signal-to-noise ratio (SNR) is:

[0234]

[0235] The root mean square error (RMSE) is calculated as follows:

[0236]

[0237] The formula for calculating the correlation coefficient CC is:

[0238]

[0239] In the above formulas For the first The original noisy signal values ​​of each sampling point For the first The signal value after noise reduction at each sampling point The arithmetic mean of the original signal sequence. The arithmetic mean of the denoised signal sequence. This represents the signal length, i.e., the number of sampling points.

[0240] Step 4: Feature extraction and fusion;

[0241] Feature extraction is performed on the preprocessed EEG signals from step 3. The time-domain AR model order determination curve based on the FPE and AIC criteria is shown below. Figure 9 As shown. Based on the FPE and AIC criteria, an optimization of the AR model's order was performed. It was found that as the order increased, the function values ​​of both criteria exhibited a convergence characteristic of first decreasing sharply and then stabilizing. To achieve the best balance between model fitting accuracy and computational complexity, based on the principle of model simplification, the minimum order at which the curve shows a significant convergence inflection point was selected as the optimal solution. Combining the trends of the two curves, it can be seen that when the order is 6, the criterion values ​​have reached a significant convergence state, with subsequent decreases being minimal. Therefore, the optimal order of the AR model was finally determined to be 6. Finally, the AR model parameter characteristics were calculated using the Burg algorithm. Since it is necessary to base on... Wave (8~12Hz) and Frequency domain features were extracted from the wavelet (12~30Hz). A 6-level wavelet packet decomposition was performed on the preprocessed EEG signal with a sampling rate of 256Hz to refine the subband bandwidth of the EEG signal to 2Hz, thus obtaining the target rhythm. The wavelet basis was selected as db4. Figure 10The frequency domain wavelet packet decomposition yields wave and Since the frequency band of EEG signals is concentrated in the range of 0-32Hz, this embodiment demonstrates the process from the second layer, where nodes (6,4) and (6,5) are reconstructed. Wave, reconstructed from nodes (6,6) to (6,14) Wavelet, and then obtained based on wavelet packet decomposition wave and Wave extraction frequency domain Bandwidth power ratio; entropy domain extraction Band-cross-scale cumulative multi-scale sample entropy sum Multi-scale sample entropy accumulated across frequency bands. Nonparametric Mann-Whitney U tests were performed on the frequency domain characteristics of kinesthetic and visual-motor imagery to verify inter-group differences, such as... Figure 11 As shown in the figure, there are significant differences between the two groups of motion imagery experiments based on their respective channel selections, indicating that the frequency domain... Band power ratio characteristics can serve as biomarkers characterizing athletes' athletic ability; the entropy domain is related to... and The features of the two frequency bands were averaged after training with multiple machine learning models (DT, LDA, LR, KNN, SVM) in the Matlab classifier. The optimal cumulative timescale factor for cross-scale cumulative multi-scale sample entropy in kinematic and visual motion imagery experiments They are 5 and 4 respectively. The optimal cumulative timescale factor for cross-scale cumulative multi-scale sample entropy in kinematic and visual motion imagery experiments The scores are 8 and 7 respectively. Based on this, both kinesthetic and visual-motor imagery have 9 characteristic indicators.

[0242] Based on the differences in brain activation patterns between kinesthetic and visual-motor imagery tasks, nine electrode channels were selected for the kinesthetic imagery experiment and five electrode channels were selected for the visual-motor imagery experiment. Feature vectors of 81 and 45 dimensions were obtained for kinesthetic and visual-motor imagery, respectively.

[0243] Both kinesthetic and visual-motor imagery experiments used a 4-second time window as the sample segment. Both groups of motor imagery experiments, based on the same group of subjects, obtained 450 EEG signal samples.

[0244] The feature vectors of basketball players and non-players are fused to construct a multi-domain fusion feature fusion matrix, which generates... Kinesthetic imagery sample dataset and A dataset of visual motion imagery samples.

[0245] Step 5: SVM identification;

[0246] The kinematic and visual motion imagery sample dataset, which is composed of the multi-domain feature fusion matrix calculated and verified in step 4, is imported into the Matlab classifier SVM for classification to achieve the recognition of motion literacy.

[0247] In the two datasets, basketball players' motion imagery samples were labeled as 1, and non-athlete motion imagery samples were labeled as 2. Five-fold cross-validation was used for classification and identification.

[0248] Figure 12 A statistical comparison and analysis of the classification accuracy of SVM classifiers with different feature extraction algorithms is presented.

[0249] In the feature validity and model performance evaluation phase of this application, the model classification accuracy was generated using a 5-fold cross-validation strategy. For the classification performance of each single EEG feature and multi-domain fusion feature under the same data partition (i.e., the same fold), this application used a paired-samples t-test to analyze significant differences between groups.

[0250] To ensure the statistical rigor of the parameter tests, this study first evaluated the normality hypothesis of the differences between paired samples (i.e., the difference between the accuracy of multi-domain fusion features and the accuracy of each individual feature). Considering the sample size limitation (N=5) imposed by 5-fold cross-validation, the conventional normality test has low power; therefore, this study chose the Anderson-Darling test, which is better suited to small samples. The results show that in the kinesthetic imagery (KMI) and visual-motor imagery (VMI) tasks, the test results of the differences between the comparison groups did not significantly reject the normality hypothesis (p>0.05), which meets the basic mathematical premise of the paired-samples t-test.

[0251] Furthermore, in the cross-validation evaluation paradigm of machine learning based on EEG signals, for the limited number of sample folds, paired-samples t-tests are used to compare the performance differences of different feature extraction algorithms on the same dataset, which can effectively control the systematic errors caused by random data splitting. The test results show that the p-values ​​are all less than 0.001. This statistical evidence objectively and scientifically confirms that the time-domain-frequency-entropy domain multi-domain feature fusion system proposed in this application has extremely significant and non-accidental superiority in recognition accuracy and model robustness.

[0252] like Figure 12As shown, the black lines at the top of each bar chart represent error bars, used to characterize the standard deviation of the 5-fold cross-validation classification accuracy. The error bars visually reflect the dispersion of the model's classification results under different data partitions. The smaller the span of the error bars, the smaller the fluctuation in accuracy of each fold cross-validation, and the lower the degree to which the model is affected by random sample partitioning, thus demonstrating stronger generalization ability and robustness of the model. As can be seen from the figures, the multi-domain feature fusion method proposed in this application not only achieved the highest average classification accuracy in both kinesthetic and visual motor imagery tasks, but also exhibited a relatively small span of error bars. This fully demonstrates that the multi-domain feature fusion model constructed in this application not only possesses excellent recognition accuracy but also exhibits extremely high model stability and data reliability when recognizing athletes' EEG motor literacy.

[0253] The above statistical indicators provide a theoretical basis for assessing the athletic ability of basketball players and demonstrate their potential application value.

[0254] As described above, although this application has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the application itself. Various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A method for constructing an EEG physiological index of an athlete based on a two-dimensional motor imagery task, characterized in that, Includes the following steps: Step 1, Experimental Paradigm Design: Design experiments on kinesthetic and visual-motor imagery; Step 2, EEG signal acquisition: Raw EEG signals are acquired using an EEG device; Step 3, EEG signal preprocessing: Preprocessing based on EEGLAB toolbox and denoising verification of VMD-PE-WTD algorithm, where VMD-PE-WTD algorithm verified its excellent denoising performance. Preprocessing is mainly based on the analysis and processing of effective EEG data of motor imagery. Step 4, Feature Extraction and Fusion: Calculate and analyze the parameters of the 6th-order AR model in the time domain and the frequency domain respectively. The proposed frequency band power ratio and entropy domain and The feature values ​​of cross-scale cumulative multi-scale sample entropy of the frequency band are used to extract EEG feature values ​​of different numbers of electrode channels according to the differences in experimental tasks. Among them, the EEG features of 9 electrode channels are extracted for the kinesthetic imagination experiment, including the frontal lobe F3, F4, Fz channels, the central area C3, C4, Cz channels, and the parietal lobe P3, P4, Pz channels. The EEG features of 5 electrode channels are extracted for the visual-motor imagination experiment, including the parietal lobe P3, P4, Pz channels and the occipital lobe O1, O2 channels. Thus, a multi-domain feature fusion matrix is ​​constructed. Step 5, SVM recognition: Import the feature matrix calculated and analyzed in step 4 into the Matlab classifier SVM for classification to achieve recognition of motor literacy.

2. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 1, characterized in that, The requirements for experimental subjects, experimental environment, and experimental equipment during EEG signal acquisition in step 2 are as follows: (1) Experimental subjects: 10 first-level basketball players and 10 non-athletes with no competition or training experience were selected from the sports department. All 20 subjects were college students aged 19-23 years old, right-handed, with normal or corrected vision, and no neurological or musculoskeletal diseases. (2) Experimental environment: a comfortable, quiet, and electromagnetically free shielded indoor environment; (3) Experimental equipment: 19-channel wet electrode EEG acquisition equipment.

3. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1, EEGLAB Toolbox Processing: First, electrode positioning is performed, 50Hz power frequency interference is removed, and a 0.5-40Hz bandpass filter is applied to remove high-frequency signal interference; next, the sampling rate is reduced to 256Hz, and bad conductors and bad segments are interpolated and removed; then, A1 / A2 bilateral mastoid electrodes are used for rereference; finally, independent component analysis is performed to remove artifacts and bad segment components related to electrooculography and electromyography. Step 3.2, VMD-PE-WTD algorithm denoising: First, the signal processed in Step 3.1 is subjected to variational mode decomposition to obtain IMF components, where the number of modes K is set to 6 and the penalty factor alpha is set to 3000; then, the normalized permutation entropy of each IMF component is calculated, and the normalized permutation entropy threshold is set to 0.

8. Based on the set permutation entropy threshold, the IMF components are divided into noise components and effective signal components; finally, wavelet threshold denoising is performed on the noise components. Wavelet threshold denoising includes wavelet decomposition, thresholding, and wavelet reconstruction. The decomposition level is set to 5, and the db4 wavelet is selected for denoising. The threshold calculation formula is as follows: ,in It is the noise standard deviation. It is the signal length. The sign is logarithmic, and the signal is finally reconstructed from the effective signal components to obtain the denoised signal.

4. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 3, characterized in that, The algorithm steps for permutation entropy described in step 3.2 are as follows: Step 1: Phase space reconstruction; Let the original time series be... Given the embedding dimension and time delay The sequence is reconstructed in phase space to obtain a set of reconstruction matrices, where the embedding dimension is... Set to 3, time delay Set to 1; The reconstructed first Each component Represented as: ； in, ,and each They are all one A subsequence of dimension; Step 2: Ascending order arrangement and symbolic mapping; For each reconstructed component of The elements are sorted in ascending order of their numerical values; Assumption The original index position corresponding to the element in the middle is The sorted sequence is: ； If there are equal values, sort them according to their index numbers, that is, the smaller index is placed first; Therefore, each component All of them can be mapped to a unique permutation pattern or symbol sequence. : ； Step 3: Calculate the probability distribution of the permutation pattern; exist In 3D space, there are a total of 100 permutation patterns. Count the possible combinations and statistically analyze each permutation pattern. The number of times it appears in all reconstructed vectors is denoted as . ; Calculate the probability of each permutation pattern occurring. : ； Step 4: Calculate the permutation entropy; According to the definition of Shannon entropy, the embedding dimension of this time series is calculated. The entropy of the arrangement : ； Step 5: Normalization process; because The maximum value is When all permutation patterns occur with equal probability, they are normalized to facilitate comparison under different parameters, resulting in the normalized permutation entropy. : ； The range of values ​​is : The closer to 0, the more regular and simple the sequence; The closer to 1, the more random and complex the sequence.

5. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 1, characterized in that, Step 4 includes the following steps: Step 4.1, Feature Extraction: Starting from the kinesthetic and visual-motor imagery experimental tasks, electrode positioning is performed according to the 10-20 international standard lead system placement method. To accurately capture the athlete-specific neural activation patterns, the kinesthetic imagery experiment selects nine electrode channels: left F3, right F4, midline Fz in the frontal lobe of the anterior part of the brain; left C3, right C4, midline Cz in the central area across the vertex motor cortex; and left P3, right P4, midline Pz in the parietal lobe located in the upper posterior part of the brain. This ensures complete coverage of the entire cortical mapping from the frontal lobe for motor intention planning to the central area for motor execution and then to the parietal lobe for proprioceptive feedback. The visual-motor imagery experiment selects five electrode channels: left P3, right P4, midline Pz in the parietal lobe located in the upper posterior part of the brain; and left O1 and right O2 in the occipital lobe covering the visual center of the posterior brain. This aims to capture visual-spatial depth processing features. Subsequently, the Burg algorithm was used to extract the parameter features of the 6th-order AR model in the time domain, and wavelet packet decomposition was used to generate the key frequency bands for motion visualization. and ,extract Band power ratio characteristics, and based on and The cross-scale cumulative multi-scale sample entropy features proposed by the two frequency bands were finally calculated. 81 feature parameters were obtained from kinesthetic motion imagination and 45 feature parameters were obtained from visual motion imagination. Step 4.2, Feature Fusion: The EEG feature parameters of basketball players and non-athletes extracted in Step 4.1 are fused to obtain a multi-domain feature fusion vector, and then a multi-domain feature fusion matrix is ​​constructed to generate a kinesthetic motion imagery sample feature dataset and a visual motion imagery sample feature dataset, respectively.

6. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 5, characterized in that, The AR model parameters mentioned in step 4.1 include: (1) Construction of AR model; Let the time series be The order is Then AR( The model can be represented as: ； in: These are the parameters of the AR model, reflecting the dynamic characteristics of the signal; For the traversal order Variables within a range; The model order; With zero mean and variance The white noise sequence represents the deviation between the AR model's predicted values ​​and the actual values, characterizing the part of the signal that is not explained by the model; The response of white noise through a linear system, in this case, is an EEG signal, representing the response using past data. Known values The weighted sum predicts the current value; The transfer function of the system is then: ； in: For complex variables in the Z-transform; EEG signals Z-transform; Z-transform of a zero-mean white noise sequence; The power spectrum is: ； in, This refers to the digital angular frequency in signal power spectrum analysis. In summary, can be The unique representation is that AR model coefficients can be used as features of EEG signals; (2) Burg algorithm calculates AR model parameters; The Burg algorithm is described as follows: First, introduce the order variable. , define the first Step, The forward prediction error and the backward prediction error are respectively and The specific recursive formula is as follows: Initial prediction error, order : ； From the order of the steps recurrence order , ,right definition: ； ； in, That is, the first The reflection coefficient of the first order is a real number under real signals, equivalent to the first order. The first AR model coefficients ; The core principle of the Burg algorithm is to minimize the power of the forward prediction error. Power of backward prediction error The arithmetic mean, i.e., the average prediction error power. Minimum: ； make right The partial derivative of is zero, that is It can be derived that the power of the average prediction error is... The smallest First-order reflection coefficient formula: ； Calculate the reflection coefficient Then, the Levinson-Durbin recursive formula can be used to calculate the current nth... All autoregressive coefficients of the model and updating residual variance : ； ； By from Calculate stepwise recursively to the required model order. You can then obtain the complete AR model parameters. and excitation white noise variance ; (3) Determining the order of the AR model; To determine the optimal model order in the AR model The residual variances of each order calculated using the Burg algorithm described above for the AR model parameters are then used. Substitute into the evaluation criteria and introduce the order variable. And use the two major criteria of FPE and AIC to perform traversal optimization; The FPE criterion, used to balance prediction error and the number of parameters, measures the magnitude of the error when using the current model to predict future data. Its expression is: ； in The data length of the signal. The order of the AR model; order The residual variance of the time model fit; This is a complexity penalty term; The value at a certain optimal order The location will reach its minimum; The AIC criterion, based on the idea of ​​maximum likelihood estimation, comprehensively considers the fitting error and the number of parameters. It is used to measure the amount of loss information between the fitted model and the true mechanism that generated the data. Its expression is: ； The fitting error term representing the model, as the order increases... Increase, residual variance Decrease it, and the value of this term will decrease; The penalty term, representing the complexity of the model, increases by 2 for each additional order. The value at a certain optimal order The minimum will be reached.

7. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 5, characterized in that, Step 4.1 The steps for calculating the bandwidth power ratio are as follows: (1) Calculate the power spectral density; EEG signals With zero mean and variance The response of white noise through a linear system, i.e., the AR model, has the following transfer function: ； in, The system polynomial of the AR model, i.e. The location of its root determines the stability and frequency domain characteristics of the system. The order of the AR model; In a linear time-invariant discrete-time system, according to signal processing theory, the input signal at this time... power spectrum Output signal power spectrum and the frequency response function of the model system The following relationship must be satisfied: ； The power spectrum of the white noise input to the AR model is The parameters of the AR model are obtained using the Burg algorithm. , ;Will Substituting the transfer function into the AR model yields And the relationship between the input and output power spectrum of the aforementioned linear system. Combining these, the power spectrum of the AR model based on the Burg algorithm can be obtained as follows: ； Then by Therefore, the power spectral density formula for the AR model based on the Burg algorithm is: ； in, For actual frequency, The sampling frequency; (2) Calculate the bandwidth power ratio; The subband bandwidth of the EEG signal was refined to 2Hz using 6-layer wavelet packet decomposition, with the wavelet basis chosen as db4, thus obtaining the desired decomposition. wave and Wave; EEG signals frequency band and The power spectral density of the frequency band is respectively used and The bandwidth power of the two target frequency bands is shown below: ； ； calculate The bandwidth power ratio is expressed as follows: 。 8. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 5, characterized in that, The calculation steps for the proposed cross-scale cumulative multi-scale sample entropy in step 4.1 are as follows: Given a length of One-dimensional discrete time series Let the upper limit of the target scale factor be... Embedding dimension is , The similarity tolerance coefficient ranges from 0.1 to 0.

25. The specific calculation steps of the algorithm are as follows: Step 1: Signal coarsening processing; For a given scale factor ,in The original time series Divided into lengths of The scale is obtained by drawing non-overlapping data windows and calculating the average value of the data points within each window. coarse-grained sequence below The calculation formula is as follows: ； in, Indicates the length of the coarsened sequence. This is a floor function used to remove sequences ending in less than one scale window. Redundant data points; Step 2: Phase space reconstruction; At each scale Below, based on coarse-grained sequences Reconstructed separately peacekeeping A sequence of phase space vectors of dimension 1. The dimensional vector is constructed as follows: ； The dimensional vector is constructed as follows: ； Step 3: Calculate the Chebyshev distance; Chebyshev distance is used to measure the similarity between any two vectors in the reconstruction space. Any two vectors in the phase space and The distance between them is defined as the maximum absolute value of the difference between their corresponding elements: ； Similarly, The Chebyshev distance between the dimensional vectors is: ； in, Strictly exclude self-matching; Step 4: Count the number of similar matching pairs; Calculate coarse-grained sequences The standard deviation of is denoted as . Define similarity tolerance Iterate through all possible combinations of vector pairs, satisfying Statistical analysis peacekeeping The number of matching pairs whose subdimensional distance is less than or equal to the threshold is denoted as . For scale Down Total number of matches for a dimensional vector: ； remember For scale Down Total number of matches for a dimensional vector: ； in, This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. Step 5: Calculate the cumulative matching metric and the final feature value; Regarding the cumulative characteristics of this algorithm, for the current cumulative scale... , Calculate from scale 1 to scale 2 respectively. of peacekeeping The cumulative sum of the matching logarithms of the dimension: ； ； Based on the above cumulative states, the ratio of cumulative matching probabilities is calculated and the negative natural logarithm is taken to obtain the final result reflecting the time series. On the cumulative time scale Metrics of dynamic complexity : ； If it occurs at a certain cumulative scale In this case, at this scale Defined as 0; cumulative time scale Increasing from 1 to This will give you a complete set of vectors that reflect the characteristics of the original time series.

9. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 5, characterized in that, The optimal order for the AR model is set to 6; select parameter values. , , , Calculate the cumulative multi-scale sample entropy features across scales.

10. The method for constructing electroencephalographic indicators of athletes based on a two-dimensional motor imagery task according to claim 1, characterized in that, The specific operation of step 5 is as follows: Step 5.1, Data partitioning: In the two sample datasets, the basketball player motion imagery samples are labeled as 1, and the non-athlete motion imagery samples are labeled as 2. The two datasets are randomly divided into 5 equal parts. Step 5.2, Alternate Validation: Use 1 set as the validation set and the remaining 4 sets as the training set each time, repeating 5 times; Step 5.3, Results Integration: Calculate the average of the 5 validation results as the final inter-group performance difference evaluation index.

Images