Multi-level team cooperation performance evaluation method and system based on EEG (electroencephalogram) superscanning

Abstract

The invention provides a multi-level team cooperation performance evaluation method and system based on EEG superscanning, and belongs to the field of neural engineering and human-computer interaction. The method comprises the steps that electroencephalogram signals and behavior performance signals are synchronously collected by designing a multi-person team cooperation task; calculating an average time difference based on the behavior signal to quantify a collaboration level and grade; the electroencephalogram signals are preprocessed; extracting multi-task-stage, multi-frequency-band and multi-index interbrain connectivity features to construct a high-dimensional feature set; single-stage and cross-stage fusion models are trained by adopting a machine learning method to perform collaborative level classification, and key neural features are identified by utilizing an interpretability method; and finally, constructing and analyzing a team-level brain function network based on a graph theory, and revealing neural network characteristic differences of teams with different cooperation levels. According to the method, multi-dimensional and dynamic analysis of a team collaboration neural mechanism is realized, and a new objective and quantitative approach is provided for team performance evaluation and collaboration capability improvement.

Description

Technical Field

[0001] This invention belongs to the field of neural engineering and human-computer interaction technology, and in particular relates to a method and system for evaluating the performance of multi-level team collaboration based on EEG super-scanning. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Teamwork is a crucial form of task completion in human society, particularly vital in military operations, emergency command, surgery, sports, and large-scale orchestral performances. Effective teamwork relies not only on the individual abilities of its members but also on continuous coordination, communication, and role-playing among them. Traditionally, assessments of teamwork capabilities have depended on subjective questionnaires, behavioral observations, or final task outcome metrics (such as completion time and accuracy). These methods struggle to objectively and quantitatively capture the dynamic, intrinsic neurophysiological mechanisms involved in collaboration, especially failing to reveal the real-time neural coordination characteristics among multiple individuals.

[0004] The development of cognitive neuroscience, especially the emergence of hyperscanning technology, has provided new avenues for studying the neural mechanisms of interpersonal interaction. Hyperscanning technology can simultaneously record the brain activity of two or more individuals performing social interaction tasks. Among these, electroencephalography (EEG) has become an important tool for neural synchronization in research teams due to its high temporal resolution, low cost, and good portability. Based on EEG signals, interbrain connectivity (IBC) can be calculated to quantify the degree of coupling between neural activities in different brain regions. This interbrain synchronization has been shown to be a potential neural marker for interpersonal coordination and cooperative behavior.

[0005] Currently, there are the following obvious limitations in collaborative research using EEG ultrascan technology: (1) Current research focuses on two-person collaboration scenarios, with a serious lack of exploration into the neural mechanisms of teams with four or more people. Real-world teams are mostly three or more people, and involve more complex leader-follower and multi-role interaction patterns, making it difficult to generalize the conclusions of two-person studies to real team environments. At the same time, limiting the analysis of neural connections between two people lacks an analysis of the network characteristics of the entire team (such as global efficiency, modularity, centrality, etc.) at the system level. The overall efficiency of team collaboration depends not only on the pairwise pairing relationship, but also on the overall information integration and flow pattern of the team.

[0006] (2) Most existing studies use a single IBC index, such as coherence or phase lock value (PLV), for analysis, and lack comparison and verification of multi-index fusion. Different IBC indices reflect neural coupling in different dimensions such as power, phase or directionality. It is difficult to fully capture the complex patterns of team neural synergy by relying on a single index.

[0007] (3) Most studies analyze the collaborative process as a whole, ignoring the obvious stage characteristics of the task itself, such as the task stage generally including task preparation, execution and recovery stages. Different stages may involve different cognitive processes (such as anticipation, execution control and result monitoring), and their neural synchronization patterns may also show dynamic changes. Existing holistic analysis methods are difficult to reveal the neural mechanisms of this stage specificity. Summary of the Invention

[0008] To overcome the shortcomings of the existing technologies, this invention provides a multi-level team collaboration performance evaluation method and system based on EEG hyperscanning. Combining EEG hyperscanning technology with multi-level analysis methods, the neural dynamic characteristics of team collaboration were systematically examined in a four-person collaborative task. By integrating brain connectivity analysis, interpretable machine learning models, and graph theory network metrics, a comprehensive, dynamic, and interpretable analysis of the neural mechanisms of team collaboration is achieved.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for evaluating the performance of multi-level team collaboration based on EEG superscanning; A multi-level team collaboration performance evaluation method based on EEG superscanning includes: Construct a team collaboration task involving at least four participants and assign leader and follower roles to the team; use an EEG ultrasound scanning system to simultaneously collect the electroencephalogram (EEG) signals and behavioral performance signals of all participants; Based on the behavioral performance signals, time indicators representing the synchronicity of behavior among individuals are extracted, the average behavioral time difference between all followers and leaders is calculated, and the level of collaboration of team collaboration tasks is divided according to the average behavioral time difference. The collected raw EEG signals were preprocessed; the continuous EEG signals were divided into multiple experimental phases with the key events of the task as the time origin, and each experimental phase was divided into three task phases: preparation phase, execution phase, and recovery phase; for each task phase, the brain connectivity index of the participants' EEG signals in different frequency bands was calculated, and a high-dimensional feature set was constructed. Based on the high-dimensional feature set, a classification model based on features of a single task stage and a classification model that integrates features of multiple task stages are trained respectively to classify and predict the level of collaboration. The model interpretability method is used to analyze the fused classification model to identify the key interbrain connectivity features that contribute the most to distinguishing different levels of collaboration and their corresponding frequency bands and brain regions. A team-level brain functional network was constructed based on the aforementioned brain connectivity index; the graph theory index of the brain functional network was calculated, and the differences in graph theory index among teams with different levels of collaboration were analyzed. Output the team collaboration quality level, key leader-follower IBC characteristics, and team brain network structure characteristics.

[0010] As a further technical solution, the acquired raw EEG signals are preprocessed, including: Bandpass filtering is applied to the EEG channel to eliminate environmental noise; An automatic and adjustable artifact removal algorithm based on wavelet decomposition is used to achieve adaptive artifact identification and local frequency band suppression.

[0011] As a further technical solution, the different frequency bands include , , , , Five frequency bands; the brain connectivity indicators include power correlation index, circulation correlation index, coherence, phase lag index, weighted phase lag index, and phase lock value.

[0012] As a further technical solution, the fusion of features from multiple task stages is achieved by calculating the class discriminative power coefficients of features from different stages and performing weighted fusion; the classification model adopts a support vector machine, and its kernel function is a hybrid kernel function, which is composed of a combination of a linear kernel and a radial basis function.

[0013] As a further technical solution, the fusion classification model is analyzed using a model interpretability method, including: using the SHAP method to perform interpretability analysis on the best model.

[0014] As a further technical solution, the training process of the support vector machine model adopts a soft margin maximization strategy, introduces a penalty coefficient to balance the margin width and misclassification rate, and combines the RBF kernel to achieve nonlinear mapping.

[0015] As a further technical solution, the graph theory metrics include global efficiency, small-world properties, leader degree centrality, and clustering coefficient.

[0016] The second aspect of this invention provides a multi-level team collaboration performance evaluation system based on EEG superscanning.

[0017] A multi-level team collaboration performance evaluation system based on EEG superscanning includes: The data acquisition module is configured to: construct a team collaboration task involving at least four participants and designate the leader and follower roles in the team; and simultaneously acquire the electroencephalogram (EEG) signals and behavioral performance signals of all participants using an EEG ultrasound scanning system. The collaboration level determination module is configured to: extract time indicators representing the synchronicity of behavior among individuals based on the behavioral performance signals, calculate the average behavioral time difference between all followers and leaders, and classify the collaboration level of team collaboration tasks according to the average behavioral time difference. The data preprocessing module is configured to: preprocess the acquired raw EEG signals; divide the continuous EEG signals into multiple experimental phases with the key events of the task as the time origin, and divide each experimental phase into three task phases: preparation phase, execution phase, and recovery phase; and for each task phase, calculate the brain connectivity index of the participants' EEG signals in different frequency bands and construct a high-dimensional feature set. The model building module is configured to: train a classification model based on a single task stage feature and a classification model that integrates features from multiple task stages, based on the high-dimensional feature set, respectively, for classifying and predicting the level of collaboration; and analyze the fused classification model using a model interpretability method to identify the key interbrain connectivity features that contribute most to distinguishing different levels of collaboration and their corresponding frequency bands and brain regions. The evaluation result output module is configured to: construct a team-level brain functional network based on the interbrain connectivity index; calculate the graph theory index of the brain functional network and analyze the differences in graph theory indexes among teams with different levels of collaboration; and output the team collaboration quality level, key leader-follower IBC features, and team interbrain network structure features.

[0018] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the multi-level team collaboration performance evaluation method based on EEG superscanning as described in the first aspect of the present invention.

[0019] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the multi-level team collaboration performance evaluation method based on EEG superscanning as described in the first aspect of the present invention.

[0020] The above one or more technical solutions have the following beneficial effects: (1) This invention is the first to combine brain-brain synchronization analysis at the leader-follower pairing level with brain network analysis at the team-wide level, constructing a two-tiered "pairing-team" evaluation framework. This method can not only reveal the neural coupling strength between specific roles (such as a leader and a follower), but also evaluate the information integration efficiency and network structure characteristics of the team from the perspective of the whole system, thus overcoming the shortcomings of existing technologies that are mostly limited to two-person interaction analysis, and providing a more comprehensive and in-depth understanding of multi-person team collaboration. By dividing the collaboration process into preparation, execution and recovery phases, and extracting the neural features of each phase, this invention can capture the dynamic evolution of neural coordination patterns under different cognitive phases (such as expected planning, real-time execution and feedback monitoring). It breaks through the limitations of existing methods that treat collaboration as a whole for general analysis, and achieves high-resolution analysis of the team collaboration process, providing precise neural basis for intervention and training for specific phases.

[0021] (2) This invention comprehensively uses six brain connectivity indicators and five EEG frequency bands for combined analysis, making full use of multi-dimensional information such as power, phase, and directionality, and avoiding the limitations of a single indicator. By combining machine learning and SHAP interpretability technology, it can automatically screen out the key neural features most relevant to the quality of collaboration and quantify their contribution. This not only improves the accuracy and generalization ability of the classification model, but also makes the evaluation results both predictive and interpretable.

[0022] (3) By introducing graph theory analysis, this invention reveals that the neural basis of efficient collaboration lies in the balance between global integration and local specialization, as well as the key role of low-frequency neural oscillations in maintaining this balance, providing new theoretical guidance and biomarkers for optimizing team composition and intervention strategies.

[0023] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 This is a flowchart of the method in the first embodiment.

[0026] Figure 2 This is a SHAP summary diagram of the top 20 most important IBC features in the first embodiment.

[0027] Figure 3 This is a representative connection topology diagram of the brainwave channels between leaders and followers corresponding to the top-ranked features of SHAP in the first embodiment.

[0028] Figure 4 The results are based on a differential graph theory analysis of the inter-team brain network structure at multiple levels in the first embodiment.

[0029] Figure 5 This is a system structure diagram of the second embodiment. Detailed Implementation

[0030] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0032] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0033] Example 1 This embodiment discloses a multi-level team collaboration performance evaluation method based on EEG super-scanning; like Figure 1 As shown, the multi-level team collaboration performance evaluation method based on EEG superscanning includes: Step S1: Construct a team collaboration task involving at least four participants and assign leader and follower roles to the team; use an EEG ultrasound scanning system to simultaneously collect the EEG signals and behavioral performance signals of all participants. Step S2: Based on the behavioral performance signals, extract time indicators that characterize the synchronicity of behavior among individuals, calculate the average behavioral time difference between all followers and leaders, and classify the level of collaboration of team collaboration tasks according to the average behavioral time difference. Step S3: Preprocess the acquired raw EEG signals; using the key events of the task as the time origin, divide the continuous EEG signals into multiple experimental phases, and divide each experimental phase into three task phases: preparation phase, execution phase, and recovery phase; for each task phase, calculate the brain connectivity index of the participants' EEG signals in different frequency bands and construct a high-dimensional feature set. Step S4: Based on the high-dimensional feature set, a classification model based on features of a single task stage and a classification model that integrates features of multiple task stages are trained respectively to classify and predict the level of collaboration. The model interpretability method is used to analyze the fused classification model to identify the key interbrain connectivity features that contribute the most to distinguishing different levels of collaboration and their corresponding frequency bands and brain regions. Step S5: Construct a team-level brain functional network based on the brain connectivity index; calculate the graph theory index of the brain functional network and analyze the differences in graph theory index among teams with different collaboration levels; output the team collaboration quality level, leader-follower key IBC features, and team brain network structure features.

[0034] Specifically, it also includes the following: Step S1: Construct a team collaboration task involving at least four participants and assign leader and follower roles to the team; use an EEG ultrasound scanning system to simultaneously collect the electroencephalogram (EEG) signals and behavioral performance signals of all participants.

[0035] In step S1, 400 participants were selected for a drumming experiment. The 400 participants were randomly divided into 100 teams of four. In each team, one member was randomly designated as the leader and the other three members were the followers. The leader received external cues to guide the drumming rhythm. The followers could neither observe the leader's drumming movements nor know the rhythm in advance. They could only respond immediately after hearing the leader's drumming and try their best to synchronize with the leader's rhythm.

[0036] Furthermore, during the experiment, the total duration was 480 seconds, with the drum being struck once every 10 seconds. The EEG acquisition device and drum sound sensor were used to collect the EEG signals and behavioral performance signals of all participants at a sampling rate of 1000 Hz. The behavioral performance signals were represented as drum sound signals, and real-time alignment was performed through a hardware-level synchronization module to ensure that the latency between devices was controlled at the millisecond level.

[0037] Step S2: Based on the behavioral performance signals, extract time indicators that characterize the synchronicity of behavior among individuals, calculate the average behavioral time difference between all followers and leaders, and classify the level of collaboration of team collaboration tasks according to the average behavioral time difference.

[0038] The peak values ​​of the drum sound signals were extracted and converted into binary pulse sequences by setting a threshold, thus obtaining the drumming time points for each participant (up to 48 times). Due to external interference, some participants drummed less than 48 times; these data were discarded. Next, the mean time difference (MTD) between each follower and their team leader over the 48 drumming sessions was calculated as a standard to measure the level of team collaboration. Based on the MTD values, synchronization levels were divided into high synchronization, medium synchronization, and low synchronization. Accordingly, all participants were divided into three groups: "Good," "Middle," and "Bad," which served as labels for subsequent models. Statistical analysis was used to identify which indicators showed significant differences and which indicators could measure collaboration ability.

[0039] Step S3: Preprocess the acquired raw EEG signals; using the key events of the task as the time origin, divide the continuous EEG signals into multiple experimental phases, and divide each experimental phase into three task phases: preparation phase, execution phase, and recovery phase; for each task phase, calculate the brain connectivity index of the participants' EEG signals in different frequency bands and construct a high-dimensional feature set.

[0040] The acquired raw EEG signals were bandpass filtered, and the sampling rate was reduced to 200 Hz to decrease computational complexity. Although filtering eliminated some environmental noise, artifacts such as blinking, heartbeats, and motion artifacts still existed during the acquisition. Traditional independent component analysis methods can effectively remove these artifacts, but they require manual intervention, are subjective, and are labor-intensive and time-consuming. In this embodiment, to improve data quality and subsequent modeling effects during EEG signal preprocessing, an Automatic and Adjustable Artifact Removal Algorithm (ATAR) based on wavelet decomposition is introduced. This algorithm achieves adaptive identification and local frequency band suppression of artifacts through wavelet packet decomposition, and has advantages such as no manual intervention required, adjustable parameters, and single-channel applicability. Specifically: The input EEG signal is first processed by a high-pass filter to remove slow drift artifacts. The signal is then divided into overlapping windows for segmented processing.

[0041] For each signal window, multi-level wavelet packet decomposition is performed using a specified wavelet basis to obtain the local time-frequency coefficients. As shown in the following formula:

[0042] In the formula, It is the input signal. Indicates the first L Layer i The first small wave packet k One coefficient, For the first L Layer i wavelet basis functions n This is the time sampling index.

[0043] For each group of wavelet coefficients, a threshold is calculated based on the interquartile range of its amplitude distribution. Artifact coefficients are adaptively suppressed using one of three modes: hard thresholding, linear reduction, or soft thresholding. Hard thresholding directly sets coefficients exceeding the threshold to zero; linear reduction linearly attenuates the coefficients; and soft thresholding uses a hyperbolic tangent function to smoothly compress high-amplitude components, as shown in the following equation:

[0044] In the formula, The exponential function mapping for the threshold, The attenuation constant is and These are the upper and lower bounds of the threshold. Interquartile range, The final threshold is selected.

[0045] The wavelet coefficients after artifact processing are reconstructed by inverse transform into the cleaned signal, and then synthesized into a complete EEG signal by overlapping addition.

[0046] Through the above processing, typical artifacts such as eye movement, electromyography, and motor activity were effectively suppressed, significantly improving signal quality and the performance of subsequent modeling. After ATAR processing, all channels of data were uniformly rereferenced using whole-brain averaging. Subsequently, each participant's 8 minutes of EEG data was divided into up to 48 segments according to the drumming time, with each segment lasting 10 seconds. To reveal the dynamic changes in neural synchronization under collaborative tasks, each drumming was divided into a preparation phase (Pre), an execution phase (Exp), and a recovery phase (Post), with the leader's drumming time as the time origin. Each phase included 2 seconds of EEG data, and uniform baseline correction was performed.

[0047] Furthermore, to comprehensively characterize the neural coupling state among team members, six IBC indicators were selected: power correlation index, cyclic correlation index, coherence, phase lag index, weighted phase lag index, and phase lock value. These indicators complement each other and provide multi-dimensional collaborative assessment.

[0048] The power correlation index (Pow_Corr) measures the correlation between the instantaneous power changes of two signals in the same frequency band, reflecting the intensity of co-activation between brain regions. The calculation formula is as follows:

[0049] Where P represents the signal power in the target frequency band. For covariance, The standard deviation is denoted as .

[0050] The Circular Correlation (CCorr) index is used to measure the phase consistency between signals. The specific calculation formula is as follows:

[0051] Where θ represents the instantaneous phase at time t, and N represents the total number of signal sampling points.

[0052] Coherence (Coh) indicates whether the amplitude and phase of two signals are stably coupled over a long period of time in a certain frequency band. It can be used to evaluate continuous cooperation or information sharing mechanisms. The specific calculation formula is as follows:

[0053] in, and This represents the power spectrum of the signal at the analysis frequency f. This represents the cross-power spectrum of the two signals.

[0054] The Phase Lag Index (PLI) is used to measure the directional consistency of the phase difference between two signals. The specific calculation formula is as follows:

[0055] in, Let be the instantaneous phase difference between the two, and sign be the sign function.

[0056] The Weighted Phase Lag Index (WPLI) is suitable for directional phase synchronization analysis in real experimental environments. The specific calculation formula is as follows:

[0057] in, This represents the complex cross spectrum of the two signals. This indicates the extraction of the imaginary part.

[0058] Phase Locking Value (PLV) is used to measure whether the phase difference between two signals is stable within a specific frequency band. The calculation formula is as follows:

[0059] in, The PLV value represents the instantaneous phase of the signal, N represents the total number of sampling points, and the closer the PLV value is to 1, the more stable the phase-locked loop is.

[0060] Analysis coverage includes , , , , Five frequency bands were used. Within each team, IBC indices were extracted from 48 EEG segments of all paired combinations, and their mean was taken to represent the level of neural coordination. The resulting matrix had dimensions of (3, 6, 5, 30, 30), corresponding to task stage (3), index type (6), frequency band (5), and 30×30 channel combination of two individuals, respectively. The analysis covered both leader-follower and follower-follower relationships. The entire IBC calculation process was completed based on the HyPyP analysis framework. The data were tested for significance to identify EEG indices that could assess different levels of coordination. The Shapiro-Wilk test was used to determine normality, and the Levene test was used to assess homogeneity of variance. ANOVA was then performed, and Tukey's post-hoc test was used for inter-group comparisons. The significance level was set at p<0.05.

[0061] Step S4: Based on the high-dimensional feature set, a classification model based on features of a single task stage and a classification model that integrates features of multiple task stages are trained respectively to classify and predict the level of collaboration. The model interpretability method is used to analyze the fused classification model to identify the key interbrain connectivity features that contribute the most to distinguishing different levels of collaboration and their corresponding frequency bands and brain regions.

[0062] Collaboration quality was categorized based on the mean time difference (MTD) between the leader and followers, with participants divided into three classes: Good (n=88), Middle (n=114), and Bad (n=86).

[0063] To quantify the quality of team collaboration, a machine learning model was constructed for in-depth analysis based on the EEG synchronization characteristics between leaders and followers. The extracted IBC indicator dimensions are: stage × number of paired samples × indicator type × frequency band × channel × channel, i.e., 3×288×6×5×30×30. After flattening all features, a high-dimensional feature matrix of size 3×288×27000 was constructed. Considering that the original features contain a large amount of redundant and irrelevant information, which may affect the generalization performance of the model, a feature selection step was introduced. First, the F-score was calculated through one-way analysis of variance (ANOVA). The F-score is a statistic used in ANOVA to compare the differences in means among three or more groups. It is calculated by the ratio of between-group variance to within-group variance; a larger F-score indicates a more significant difference between groups. The top 200 features with the most significant differences in each stage were selected; then, combined with 10-fold cross-validation, forward selection was used for further selection, retaining the top 100 most informative features in each stage, resulting in a dimensionality-reduced dataset (3×288×100). Label vectors These represent low, medium, and high levels of collaboration, respectively.

[0064] Subsequently, single-stage and cross-stage modeling were performed. In the single-stage modeling, three independent models were trained for the feature sets of the Pre, Exp and Post stages respectively, and each model was built based on only 100 features of that stage.

[0065] Cross-stage fusion modeling weights and fuses features from the three stages according to the class discriminant power coefficient (CSC) to generate a 300-dimensional comprehensive feature vector (ALL). The CSC is obtained by calculating the ratio of the sum of squared Euclidean distances between samples of different classes to the within-class variance, and then normalized to weights wpre, wexp, and wpost (the sum of the weights is 1). The fused feature vector highlights information from stages with stronger discriminative power, improving the predictive ability of the cross-stage model.

[0066] A team collaboration quality assessment model is constructed using Support Vector Machines (SVM), employing a hybrid kernel strategy by adding a linear component to the Radial Basis Function (RBF) kernel. The RBF captures non-linear details, while the linear kernel preserves overall trend information, making it suitable for signals like EEG that contain both global and local patterns. The core idea of ​​SVM is to find an optimal separating hyperplane in the feature space that maximizes the margin between samples of different classes, thereby improving classification generalization ability. For non-linearly separable data, SVM uses a kernel function... The data is mapped to a high-dimensional space, where linear separability is achieved. The hybrid kernel used is defined as follows:

[0067] in The parameters are automatically adjusted based on the validation set performance. The scale of the control mapping is adjusted; a larger value indicates that the model focuses more on local details, while a smaller value results in a smoother decision boundary. In EEG collaboration quality prediction, this characteristic enables the model to capture fine-grained differences between different pairings in specific frequency bands and spatial channels.

[0068] During model construction, the data is first stratified and sampled in a 7:3 ratio to ensure a consistent proportion of the three collaborative quality classes (Good, Middle, Bad). Feature selection is performed only on the training set, while the test set is completely isolated during feature engineering to ensure the objectivity of model evaluation. The SVM model training process employs a soft-margin maximization strategy, introducing a penalty coefficient C to balance the margin width and misclassification rate, and combining it with an RBF kernel to achieve nonlinear mapping. To obtain the optimal parameter combination, a stratified grid search based on weighted F1 values ​​is used on the training set: a coarse search is first performed within the range of C∈{0.1, 1, 10, 100} and γ∈{0.001, 0.01, 0.1, 1} to locate the optimal region, followed by a fine-grained search within its neighborhood. Finally, the parameter combination with the highest weighted F1 (F1-W) value is selected to optimize minority class prediction performance under class imbalance conditions. Finally, accuracy and F1-W values ​​are calculated on the test set, and the performance differences between the three single-stage models and the fusion model are compared.

[0069] By using this parallel modeling of single-stage and cross-stage processes, combined with the nonlinear mapping capability of hybrid kernels and the parameter tuning strategy of weighted F1 value optimization, high-precision and stable prediction of leader-follower collaboration quality is achieved.

[0070] To further explain the contribution of each IBC feature to the classification model performance, this invention uses the SHAP (Shapley Additive Explanations) method to perform interpretability analysis on the optimal model (All-SVM). The SHAP value calculation formula is as follows:

[0071] Where M is the set of all input features, and the function f is the expected prediction of the model when only features from a subset S are used.

[0072] in, Figure 2 This chart shows the SHAP summary of the top 20 most important IBC features. The horizontal axis represents the mean absolute SHAP value, reflecting the overall impact of each feature on the model's prediction results. To further visualize the spatial distribution of key IBCs, Figure 3 This paper presents a representative topology of the brainwave pathways between leaders and followers corresponding to the top features of SHAP.

[0073] The results showed that the most discriminative IBC features were mainly distributed in the delta, theta, and alpha frequency bands, covering the prefrontal, central, and parietal lobes, highlighting the crucial role of low-frequency cortical synchronization in distinguishing team collaboration levels. Specifically, the discriminative features of the Exp phase were mainly concentrated in the delta and alpha frequency bands, distributed in the central frontal and centroparietal regions, reflecting neural processes such as sensorimotor integration, sensory filtering, and real-time action control; the features of the Pre phase were mainly distributed in the delta and theta frequency bands, involving the occipital and parietal lobes, and closely related to attentional preparation and motor planning; while the Post phase showed a small number of beta frequency band features, mainly involving the midline and occipital regions, possibly related to performance monitoring and feedback processing. This distribution pattern validates the functional differences among the three task phases: the Exp phase mainly reflects dynamic, real-time collaboration needs, while the Pre and Post phases more reflect strategic and reflective team collaboration processes, each supplementing the prediction of collaboration quality from different dimensions.

[0074] Step S5: Construct a team-level brain functional network based on the brain connectivity index; calculate the graph theory index of the brain functional network and analyze the differences in graph theory index among teams with different collaboration levels; output the team collaboration quality level, leader-follower key IBC features, and team brain network structure features.

[0075] To explore the overall neural synchronization characteristics at the team level, we constructed the interbrain networks of each team and performed graph theory index analysis. This method can capture the overall attributes of collaboration, such as efficiency, integration, and modularity, overcoming the limitations of leader-follower pairwise analysis. To quantify the quality of collaboration at the team level, this study calculated the average of all drumming time differences among members in each group and used it as the gold standard for measuring the overall coordination level of the team. The 96 teams were still divided into three categories: high coordination (Good, n=28), medium coordination (Middle, n=39), and low coordination (Bad, n=29).

[0076] Subsequently, team-level interbrain networks were constructed for three experimental phases (Pre, Exp, Post), six IBC indices, and five frequency bands, generating 90 different network configurations for each team. Each participant's 30 EEG channels were considered nodes, resulting in 120 nodes per team. IBC indices were used as edges to connect different individual channels, reflecting neural synchronization at the team level. The network retained only interbrain connections, focusing on inter-individual interactions. To enhance the network's sparsity and interpretability, only the top 5% of connections were retained, totaling 270 interbrain edges (selected from 5400 possible edges) to reduce noise interference and improve sensitivity.

[0077] Based on this network, four graph theory metrics are calculated: Global Efficiency (GE), Small World Property (SW), Leader Degree Centrality (LDC), and Clustering Coefficient (CC), as follows: Global efficiency measures the efficiency of information transfer between any two nodes in the entire network; that is, the shorter the path, the higher the efficiency. The specific formula is as follows:

[0078] Where N represents the total number of nodes in the graph. This represents the shortest path length from node i to node j.

[0079] Small-world property is used to measure whether a network possesses both high clustering properties and short path properties. The specific formula is as follows:

[0080] Where C represents the average clustering coefficient of the actual network, and L represents the shortest path length of the actual network. and This indicates the corresponding index for a random graph of the same size as the network.

[0081] Queue length centrality: measures the number of connections a node has to other nodes, and the specific formula is as follows:

[0082] in, This represents the edge weight between node i and node j.

[0083] Clustering coefficients reflect whether the neighbors of a node are also connected, and are an important indicator of the closure of local structures. The specific formula is as follows:

[0084] in, Indicates the degree of node i. This represents the edge weight between nodes i and j.

[0085] ANOVA and Tukey's post-hoc test were used to statistically compare the differences in graph theory indicators among the three groups. Figure 4 As shown, graph theory analysis reveals differences in the brain network structure among teams at multiple levels. The colors in the graph represent different collaboration level groups (Good, Middle, Bad), while the shaded background corresponds to indicators at different task stages. Figure 4 As shown in Figure A, GE exhibited significant inter-group differences under multiple IBC indices. The high-coordination group demonstrated significantly higher GE across all three phases (Pre, Exp, Post) and multiple frequency bands, indicating that the high-performing team maintained stronger whole-brain integration capabilities throughout the task. The increased GE reflects higher efficiency in cross-brain information integration, facilitating rapid and coordinated information transfer between neural systems. This, in turn, promotes shared attention, sensorimotor synchronization, and feedback processing, constructing a highly flexible neural infrastructure. It also emphasizes the close link between low-frequency interbrain coupling and efficient collaboration.

[0086] Example 2 This embodiment discloses a multi-level team collaboration performance evaluation system based on EEG super-scanning; like Figure 5 As shown, a multi-level team collaboration performance evaluation system based on EEG superscanning includes: The data acquisition module is configured to: construct a team collaboration task involving at least four participants and designate the leader and follower roles in the team; and simultaneously acquire the electroencephalogram (EEG) signals and behavioral performance signals of all participants using an EEG ultrasound scanning system. The collaboration level determination module is configured to: extract time indicators representing the synchronicity of behavior among individuals based on the behavioral performance signals, calculate the average behavioral time difference between all followers and leaders, and classify the collaboration level of team collaboration tasks according to the average behavioral time difference. The data preprocessing module is configured to: preprocess the acquired raw EEG signals; divide the continuous EEG signals into multiple experimental phases with the key events of the task as the time origin, and divide each experimental phase into three task phases: preparation phase, execution phase, and recovery phase; and for each task phase, calculate the brain connectivity index of the participants' EEG signals in different frequency bands and construct a high-dimensional feature set. The model building module is configured to: train a classification model based on a single task stage feature and a classification model that integrates features from multiple task stages, based on the high-dimensional feature set, respectively, for classifying and predicting the level of collaboration; and analyze the fused classification model using a model interpretability method to identify the key interbrain connectivity features that contribute most to distinguishing different levels of collaboration and their corresponding frequency bands and brain regions. The evaluation result output module is configured to: construct a team-level brain functional network based on the interbrain connectivity index; calculate the graph theory index of the brain functional network and analyze the differences in graph theory indexes among teams with different levels of collaboration; and output the team collaboration quality level, key leader-follower IBC features, and team interbrain network structure features.

[0087] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0088] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the multi-level team collaboration performance evaluation method based on EEG superscanning as described in Example 1.

[0089] Example 4 The purpose of this embodiment is to provide an electronic device.

[0090] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the multi-level team collaboration performance evaluation method based on EEG superscanning as described in Example 1.

[0091] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0092] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0093] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-level team collaboration performance evaluation method based on EEG superscanning, characterized in that, include: Create a team collaboration task with at least four participants and assign leader and follower roles to the team. The EEG super-scanning system was used to simultaneously collect the electroencephalogram (EEG) signals and behavioral performance signals of all participants; Based on the behavioral performance signals, time indicators representing the synchronicity of behavior among individuals are extracted, the average behavioral time difference between all followers and leaders is calculated, and the level of collaboration of team collaboration tasks is divided according to the average behavioral time difference. The collected raw EEG signals are preprocessed; Using the key events of the task as the time origin, continuous EEG signals were divided into multiple experimental phases, and each experimental phase was further divided into three task phases: preparation, execution, and recovery. For each task phase, the interbrain connectivity indices of the participants' EEG signals in different frequency bands were calculated to construct a high-dimensional feature set. Based on the high-dimensional feature set, a classification model based on features of a single task stage and a classification model that integrates features of multiple task stages are trained respectively to classify and predict the level of collaboration. The fusion classification model was analyzed using model interpretability methods to identify the key interbrain connectivity features that contribute most to distinguishing different levels of collaboration, as well as their corresponding frequency bands and brain regions. A team-level brain functional network is constructed based on the aforementioned interbrain connectivity index; graph theory indices of the brain functional network are calculated, and the differences in graph theory indices among teams with different levels of collaboration are analyzed; the team collaboration quality level, key leader-follower IBC features, and team interbrain network structure features are output.

2. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The acquired raw EEG signals are preprocessed, including: Bandpass filtering is applied to the EEG channel to eliminate environmental noise; An automatic and adjustable artifact removal algorithm based on wavelet decomposition is used to achieve adaptive artifact identification and local frequency band suppression.

3. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The different frequency bands include , , , , Five frequency bands; the brain connectivity indicators include power correlation index, circulation correlation index, coherence, phase lag index, weighted phase lag index, and phase lock value.

4. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The fusion of features from multiple task stages is achieved by calculating the class discriminative power coefficients of features from different stages and then performing weighted fusion. The classification model uses a support vector machine, and its kernel function is a hybrid kernel function, which is composed of a linear kernel and a radial basis function.

5. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The fusion classification model is analyzed using model interpretability methods, including: using the SHAP method to perform interpretability analysis on the best model.

6. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The training process of the support vector machine model adopts a soft margin maximization strategy, introduces a penalty coefficient to balance the margin width and misclassification rate, and combines the RBF kernel to achieve nonlinear mapping.

7. The multi-level team collaboration performance evaluation method based on EEG superscanning as described in claim 1, characterized in that, The graph theory metrics include global efficiency, small-world properties, leader degree centrality, and clustering coefficient.

8. A multi-level team collaboration performance evaluation system based on EEG superscanning, characterized in that, include: The data acquisition module is configured to: construct a team collaboration task involving at least four participants and specify the leader and follower roles within the team; The EEG super-scanning system was used to simultaneously collect the electroencephalogram (EEG) signals and behavioral performance signals of all participants; The collaboration level determination module is configured to: extract time indicators representing the synchronicity of behavior among individuals based on the behavioral performance signals, calculate the average behavioral time difference between all followers and leaders, and classify the collaboration level of team collaboration tasks according to the average behavioral time difference. The data preprocessing module is configured to preprocess the acquired raw EEG signals. Using the key events of the task as the time origin, continuous EEG signals were divided into multiple experimental phases, and each experimental phase was further divided into three task phases: preparation, execution, and recovery. For each task phase, the interbrain connectivity indices of the participants' EEG signals in different frequency bands were calculated to construct a high-dimensional feature set. The model building module is configured to: train a classification model based on a single task stage feature and a classification model that integrates features from multiple task stages, based on the high-dimensional feature set, for classifying and predicting the level of collaboration. The fusion classification model was analyzed using model interpretability methods to identify the key interbrain connectivity features that contribute most to distinguishing different levels of collaboration, as well as their corresponding frequency bands and brain regions. The evaluation result output module is configured to: construct a team-level brain functional network based on the interbrain connectivity index; calculate the graph theory index of the brain functional network and analyze the differences in graph theory indexes among teams with different levels of collaboration; and output the team collaboration quality level, key leader-follower IBC features, and team interbrain network structure features.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the multi-level team collaboration performance evaluation method based on EEG superscanning as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the multi-level team collaboration performance evaluation method based on EEG superscanning as described in any one of claims 1-7.

Images