A brain regulation induction training method, system, device and medium based on a virtual reality environment

Abstract

The present application relates to a kind of brain control induction training method, system, equipment and medium based on virtual reality environment, method includes: acquisition user in virtual reality cognitive scene under multichannel electroencephalogram, through time-frequency energy spectrum analysis and inter-channel mutual information fusion generation global electroencephalogram complexity feature vector;Characteristic vector is input into the discriminant function based on real cognitive participation state and strategic deception state training, calculate out the probability of deception and generate confidence factor;Based on confidence factor and task performance data, compare the deviation of current cognitive load and target load interval, calculate difficulty adjustment gradient and generate the scene difficulty parameter of next time;Baseline feedback intensity is modulated according to confidence factor, and task interaction complexity and multi-sensory feedback intensity are updated synchronously;Virtual environment configuration parameter is applied to next training cycle, and electroencephalogram signal is recycled back to first step, forms the closed loop induction training loop with anti-deception discrimination as core.

Description

Technical Field

[0001] This invention relates to the field of cross-disciplinary technology of virtual reality and brain-computer interface, specifically to a brain modulation and induction training method, system, device and medium based on a virtual reality environment. Background Technology

[0002] In recent years, the integration of virtual reality (VR) and brain-computer interface (BCI) technologies has brought new development opportunities to the field of cognitive training. VR cognitive training methods based on BCIs construct immersive virtual environments and combine real-time acquisition and analysis of electroencephalogram (EEG) signals. This allows for the dynamic adjustment of training parameters while inducing users to perform specific cognitive tasks, achieving personalized neurofeedback training. These methods stimulate user cognitive engagement through multi-sensory stimulation, assess the user's cognitive state using neurofeedback information from EEG signals, and then adjust the task difficulty and feedback intensity in the virtual environment, forming a closed-loop training mechanism of "perception-assessment-regulation." This closed-loop induced training can enhance neuroplasticity and has demonstrated significant application value in areas such as cognitive rehabilitation, attention training, and working memory improvement.

[0003] However, existing brain-computer interface-based virtual reality cognitive training methods generally suffer from a long-neglected safety issue: users may consciously and actively adjust their own neural signal characteristics to "deceive" the system by learning the system's feedback patterns. Specifically, in traditional neurofeedback training, the system typically assesses the user's cognitive engagement state based on macroscopic characteristics such as frequency band power and event-related potential amplitude in EEG signals. Once the user masters the system's response patterns to these characteristics, they can induce the system to misjudge a "high cognitive engagement" state by deliberately relaxing to generate specific EEG rhythms, suppressing certain EEG components, or mimicking ideal EEG patterns, thereby obtaining positive feedback and reduced difficulty with minimal cognitive effort. This strategic neural signal masquerading significantly reduces the effectiveness of the training process; the false "training results" cannot be converted into real changes in neural plasticity, and may even weaken the user's genuine training motivation. Existing technologies lack effective modeling of the intrinsic generation mechanism of neural signals, making it difficult to distinguish the essential differences between the nonlinear neural dynamics generated by genuine cognitive engagement and the predictable neural patterns generated by strategic relaxation at the level of signal complexity. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a brain modulation induction training method, system, device and medium based on a virtual reality environment that can effectively identify and suppress users' strategic neural signal masquerading behavior.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a brain modulation and induction training method based on a virtual reality environment, comprising the following steps:

[0007] S1: Collect multi-channel EEG signals from users in virtual reality cognitive scenarios, decompose the multi-channel EEG signals into time series to extract the time-frequency domain energy spectrum of each channel, calculate the spectral centroid offset and power spectral density normalized entropy value of each channel based on the time-frequency domain energy spectrum, and perform inter-channel mutual information fusion to generate a global EEG complexity feature vector.

[0008] S2: Based on the comparison of the global EEG complexity feature vector input into the preset discriminant function, the deception probability of the user's current time window is calculated, and a confidence factor reflecting the authenticity of the user's neural signals is generated based on the deception probability.

[0009] The preset discriminant function is a discriminant model obtained by collecting the first EEG signal under the real cognitive participation state and the second EEG signal under the strategic deception state, and using the complexity feature vectors generated by the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training.

[0010] S3: Based on the confidence factor and the task performance data fed back in real time in the virtual scene, compare the deviation between the user's current cognitive load and the target cognitive load range, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting to generate the virtual scene difficulty parameters for the next moment.

[0011] S4: Modulate the preset baseline feedback intensity based on the confidence factor, and write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine in sync. This enables the virtual scene to dynamically change the task interaction complexity according to the virtual scene difficulty parameter and adjust the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, thereby generating updated virtual environment configuration parameters.

[0012] S5: Apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, and at the same time loop the newly collected EEG signals back to S1 to form a cognitive state closed-loop induced training circuit with confidence factor as the basis for anti-deception judgment and difficulty parameter and feedback intensity parameter as the execution object.

[0013] Secondly, the present invention provides a brain modulation and induction training system based on a virtual reality environment, the system being configured with the following modules:

[0014] The EEG feature extraction module is used to collect multi-channel EEG signals from users in virtual reality cognitive scenarios. It performs time-series decomposition on the multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel. Based on the time-frequency domain energy spectrum, it calculates the spectral centroid offset and power spectral density normalized entropy value for each channel and performs inter-channel mutual information fusion to generate a global EEG complexity feature vector.

[0015] The neural confidence discrimination module is used to calculate the deception probability of the user in the current time window by comparing the global EEG complexity feature vector input into a preset discrimination function, and to generate a confidence factor reflecting the authenticity of the user's neural signals based on the deception probability. The preset discrimination function is a discrimination model obtained by collecting the first EEG signal in the real cognitive participation state and the second EEG signal in the strategic deception state, and using the complexity feature vectors generated corresponding to the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training.

[0016] The cognitive difficulty adjustment module is used to compare the deviation between the user's current cognitive load and the target cognitive load range based on the confidence factor and the task performance data fed back in real time from the virtual scene, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting processing to generate the virtual scene difficulty parameters for the next moment.

[0017] The scene parameter modulation module is used to modulate the preset baseline feedback intensity based on the confidence factor, and synchronously write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine, so that the virtual scene can dynamically change the task interaction complexity according to the virtual scene difficulty parameter and adjust the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, and generate updated virtual environment configuration parameters.

[0018] The closed-loop induction training module is used to apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, while the newly acquired EEG signals are looped back to S1, forming a cognitive state closed-loop induction training loop with confidence factor as the anti-deception judgment basis and difficulty parameter and feedback intensity parameter as the execution object.

[0019] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned brain modulation and induction training methods based on a virtual reality environment.

[0020] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned brain modulation and induction training methods based on a virtual reality environment.

[0021] In summary, the brain modulation induction training method based on a virtual reality environment provided in this application constructs a global EEG complexity feature vector reflecting the essential difference between genuine cognitive participation and strategic deception from the perspective of the signal's intrinsic generation mechanism by performing time-series decomposition and complexity feature extraction on the user's multi-channel EEG signals. This vector accurately identifies whether the user is in a deceptive state and generates a confidence factor to quantitatively assess the authenticity of neural signals. Furthermore, by weighting and correcting the difficulty adjustment gradient and feedback intensity based on the confidence factor, the method effectively inhibits the user's behavior of actively adjusting their own neural signals to obtain false positive feedback by learning system patterns, preventing deceptive behavior from distorting the training process. Simultaneously, the modulated difficulty parameters and feedback intensity parameters are synchronously written into the virtual scene engine to dynamically change the task interaction complexity and multi-sensory feedback intensity, forming a closed-loop induction training circuit with the confidence factor as the anti-deception discrimination criterion. This ensures that the training process is always adaptively adjusted based on the user's genuine cognitive participation state, significantly improving the authenticity and effectiveness of cognitive training and the reliability of neural plasticity induction.

[0022] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0023] Figure 1 A flowchart illustrating a brain modulation and induction training method based on a virtual reality environment, provided as an embodiment of this application;

[0024] Figure 2 This is a schematic diagram of the structure of a brain modulation and induction training system based on a virtual reality environment, provided as another embodiment of this application. Detailed Implementation

[0025] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0026] Unless otherwise defined, 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. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] In one embodiment, such as Figure 1As shown, a brain modulation and induction training method based on a virtual reality environment is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0028] S1: Collect multi-channel EEG signals from users in virtual reality cognitive scenarios, perform time-series decomposition on the multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel, calculate the spectral centroid offset and power spectral density normalized entropy value of each channel based on the time-frequency domain energy spectrum, and perform inter-channel mutual information fusion to generate a global EEG complexity feature vector.

[0029] Specifically, this embodiment collects multi-channel EEG signals from the user in a virtual reality cognitive scenario. Specifically, a 32-channel EEG acquisition device (sampling frequency set to 250Hz, bandwidth 0.5-30Hz, input impedance ≥100MΩ) is used. Electrodes are fixed to the corresponding positions on the user's scalp according to the international 10-20 system layout, with the reference electrode placed on the left mastoid process and the ground electrode placed on the right mastoid process, ensuring that the contact impedance between the electrodes and the scalp is ≤5kΩ to avoid signal interference caused by poor contact. The virtual reality cognitive scenario can be set according to training objectives, such as attention training scenarios, working memory training scenarios, etc. After the user wears a VR headset and enters the immersive virtual scenario, the EEG acquisition device collects multi-channel EEG signals from the user in real time during the scenario interaction process. The signal acquisition timestamp is recorded synchronously during the acquisition process to ensure the time synchronization of EEG signals and virtual scenario interaction behavior.

[0030] Furthermore, the system performs time-series decomposition on the acquired multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel. Considering the non-stationarity and nonlinear characteristics of EEG signals, this embodiment uses wavelet packet decomposition for time-series decomposition, selecting the db4 wavelet as the base wavelet, and setting the decomposition level to 5 levels. This decomposes the EEG signal of each channel into 32 sub-band signals with different frequencies, covering a frequency range of 0.5-30Hz, corresponding to the characteristic frequency bands of delta waves, theta waves, alpha waves, and beta waves in the EEG signal. A short-time Fourier transform is performed on each sub-band signal, with a window length of 256 sampling points and an overlap rate of 50%. The time-frequency domain energy spectrum of each sub-band signal within different time windows is calculated. The horizontal axis of the time-frequency domain energy spectrum represents time (in seconds), and the vertical axis represents frequency (in Hz). The pixel value is the signal energy at the corresponding time-frequency point, thus completely preserving the time-domain and frequency-domain characteristics of the EEG signal.

[0031] The system calculates the spectral centroid shift and the normalized entropy of the power spectral density for each channel based on the time-frequency domain energy spectrum. The calculation principle for the spectral centroid shift is as follows: the spectral centroid reflects the position of the centroid of the time-frequency domain energy spectrum, and its calculation formula is:

[0032]

[0033] In the formula, For frequency, is the time-frequency domain energy corresponding to frequency f at time t; the spectral centroid offset is the difference between the spectral centroid of the current time window and the preset baseline spectral centroid. The baseline spectral centroid is calculated by collecting EEG signals from the user's resting state (eyes closed, relaxed, for 5 minutes) before training. Its function is to reflect the degree of deviation of the current EEG signal spectrum distribution relative to the resting state. The spectral centroid will show a significant shift when real cognition is involved, while the shift is smaller and shows regularity when strategic deception is used.

[0034] Preferably, the calculation process for the normalized entropy value of the power spectral density is as follows: First, the power spectral density (PSD) of the EEG signal in each channel is estimated, and the power spectral density is calculated using the periodogram method, with a frequency resolution set to 0.9766 Hz; then, the power spectral density is normalized so that the sum of the power spectral densities at all frequency points is 1; finally, the Shannon entropy is calculated based on the normalized power spectral density, using the following formula:

[0035]

[0036] In the formula, Let be the normalized power spectral density at the i-th frequency point, and n be the number of frequency points. This entropy value reflects the complexity of the power spectrum distribution of the EEG signal. When there is real cognitive involvement, the EEG signal has high complexity and a large entropy value, while when there is strategic deception, the EEG signal is predictable and the entropy value is small.

[0037] Finally, the spectral centroid offset and normalized entropy of power spectral density of all channels are fused using inter-channel mutual information to generate a global EEG complexity feature vector. The core principle of mutual information fusion is to measure the correlation between features of different channels by calculating the mutual information between corresponding features of any two channels, eliminating redundant information and retaining effective features. Specifically, first, a feature vector for each channel is constructed (containing two dimensions: the spectral centroid offset and the normalized entropy of power spectral density of that channel). Then, the mutual information value between the feature vectors of any two channels is calculated. Based on the mutual information value, a feature fusion matrix is ​​constructed. A weighted average method is used to fuse the features of each channel. The weight coefficients are determined by the mutual information value. Finally, a global EEG complexity feature vector with a dimension of 2 is obtained, which can comprehensively reflect the inherent complexity characteristics of the user's current EEG signal.

[0038] S2: Based on the comparison of the global EEG complexity feature vector input into the preset discriminant function, the deception probability of the user's current time window is calculated, and a confidence factor reflecting the authenticity of the user's neural signals is generated based on the deception probability.

[0039] The preset discriminant function is a discriminant model obtained by collecting the first EEG signal under the real cognitive participation state and the second EEG signal under the strategic deception state, and using the complexity feature vectors generated corresponding to the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training.

[0040] Specifically, the system's preset discriminant function is a discriminant model obtained through supervised binary classification training. The training samples are derived from the first EEG signal under real cognitive engagement and the second EEG signal under strategic deception. The sample collection process is consistent with the EEG signal collection conditions in S1. The first EEG signal collection scenario involves the user actively participating in a preset cognitive task in a virtual cognitive scenario, maintaining genuine cognitive engagement throughout. The second EEG signal collection scenario involves guiding the user to understand the system's feedback patterns and then deliberately deceiving the system by relaxing, suppressing specific EEG components, or mimicking ideal EEG patterns. The system preprocesses, decomposes, extracts features, and fuses the EEG signals from the real and deception sample sets according to the method in S1, generating global EEG complexity feature vectors for the real and deception samples respectively. The real and deception samples are then labeled into different categories, constructing a binary classification training dataset.

[0041] Preferably, the system employs a support vector machine (SVM) as the discriminant model, selects a radial basis function (RBF) as the kernel function, optimizes the model parameters using a grid search method, and performs supervised learning training on the training dataset. Cross-validation is used during training to ensure the model's generalization ability, resulting in a pre-defined discriminant function after training. The system inputs the global EEG complexity feature vector generated by S1 into this pre-defined discriminant function for comparison, calculating the probability of deception within the user's current time window. Based on the deception probability, the system generates a confidence factor reflecting the authenticity of the user's neural signals. The confidence factor is generated using a linear mapping method, with a value ranging from 0 to 1, and is used to characterize the degree of authenticity of the user's current neural signals. A higher confidence factor value indicates a higher authenticity of the user's current neural signals, while a lower confidence factor value indicates that the user's current neural signals are more likely to be strategically spoofed.

[0042] S3: Based on the confidence factor and the task performance data fed back in real time in the virtual scene, compare the deviation between the user's current cognitive load and the target cognitive load range, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting to generate the virtual scene difficulty parameters for the next moment.

[0043] Specifically, the system acquires real-time task performance data from the virtual scene. This data is recorded in real-time by the virtual reality scene engine and includes three core indicators: task completion accuracy, task reaction time, and number of errors within the current time window. The system calculates the user's current cognitive load value using these three indicators combined with preset scoring criteria, employing a weighted summation method where weight coefficients are determined based on indicator importance. The system sets a target cognitive load range, determined based on the user's training objectives and baseline cognitive level, to ensure the user is under appropriate cognitive load during training. The system calculates the deviation between the user's current cognitive load and the target cognitive load range. The deviation is the difference between the current cognitive load value and the median value of the target cognitive load range. A positive deviation indicates the current cognitive load is higher than the target range, a negative deviation indicates the current cognitive load is lower than the target range, and a zero deviation indicates the current cognitive load is within the target range.

[0044] Furthermore, the system calculates a confidence-weighted difficulty adjustment gradient. The base value of the difficulty adjustment gradient is determined based on the cognitive load bias. The weighting formula is: difficulty adjustment gradient equals confidence factor multiplied by cognitive load bias multiplied by the base adjustment gradient. This weighting mechanism is used to adjust the difficulty adjustment amplitude based on the realism of neural signals. The system then superimposes the difficulty adjustment gradient with the current difficulty level and performs amplitude limiting to generate the virtual scene difficulty parameters for the next time step. Amplitude limiting is used to prevent the difficulty parameters from exceeding a reasonable range, ensuring that the difficulty parameters are between the preset minimum and maximum difficulty. The direction of difficulty parameter adjustment corresponds to the direction of cognitive load bias: a positive bias decreases the difficulty, a negative bias increases the difficulty, and a zero bias maintains the same difficulty.

[0045] S4: Modulate the preset baseline feedback intensity based on the confidence factor, and write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine in sync. This allows the virtual scene to dynamically change the task interaction complexity according to the virtual scene difficulty parameter and adjust the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, thereby generating updated virtual environment configuration parameters.

[0046] Specifically, the system sets a preset baseline feedback intensity, which is set separately for each type of multi-sensory feedback. Multi-sensory feedback in the virtual scene includes visual, auditory, and tactile feedback. The baseline feedback intensity is set based on the user's comfort feedback threshold in a real-world cognitive engagement state. The system modulates the preset baseline feedback intensity based on a confidence factor. The modulation formula is that the modulated feedback intensity parameter equals the preset baseline feedback intensity multiplied by the confidence factor. This modulation mechanism is used to adjust the feedback intensity based on the realism of neural signals. The system synchronously writes the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine. The writing process is implemented through the engine's API interface to ensure real-time parameter writing. The virtual scene engine dynamically changes the task interaction complexity based on the virtual scene difficulty parameter and adjusts the intensity of the multi-sensory feedback signals based on the feedback intensity parameter, generating updated virtual environment configuration parameters.

[0047] Changes in task interaction complexity manifest as increasing the number of tasks, shortening task completion time, and adding distractions when the difficulty parameter increases; conversely, decreasing the number of tasks, extending task completion time, and reducing distractions when the difficulty parameter decreases. Adjustments to feedback intensity parameters are reflected in changes in the brightness and contrast of virtual cues for visual feedback intensity, changes in the volume and frequency of cues for auditory feedback intensity, and changes in the vibration frequency and amplitude of VR controllers for haptic feedback intensity. The updated virtual environment configuration parameters include four core dimensions: difficulty parameter, visual feedback intensity, auditory feedback intensity, and haptic feedback intensity, which drive the virtual scene update for the next training cycle.

[0048] S5: Apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, and at the same time loop the newly collected EEG signals back to S1 to form a cognitive state closed-loop induced training circuit with confidence factor as the basis for anti-deception judgment and difficulty parameter and feedback intensity parameter as the execution object.

[0049] Specifically, the system applies the updated virtual environment configuration parameters to the virtual reality cognitive scene in the next training cycle. The virtual scene engine updates the scene based on these parameters, including adjusting the complexity of task interaction and the intensity of multi-sensory feedback. After the update, the next training cycle begins, and the user continues cognitive training in the updated virtual scene. Simultaneously, the EEG acquisition device collects new multi-channel EEG signals from the user, maintaining the same acquisition conditions as S1 to ensure the continuity and consistency of the EEG signals. After acquisition, the system loops the newly acquired EEG signals back to S1, re-decomposes the time series, extracts features, and fuses them to generate a new global EEG complexity feature vector, which then enters the next round of calculation and adjustment of deception probability, confidence factor, and training parameters. The system ultimately forms a closed-loop cognitive state induction training circuit, using the confidence factor as the anti-deception criterion and the difficulty parameter and feedback intensity parameter as the execution objects. This closed-loop system identifies users' strategic neural signal masquerading behaviors by real-time acquisition and analysis of EEG signals' complexity characteristics. It avoids false training results by weighting training parameters using confidence factors, and ensures that the training difficulty matches the user's actual cognitive level by dynamically adjusting the virtual scene difficulty and feedback intensity. This continuously stimulates the user's genuine cognitive participation, enhances neuroplasticity, and improves the effectiveness and reliability of training. During training, the system synchronously records all training parameters, EEG signals, and task performance data for subsequent training effect evaluation and parameter optimization.

[0050] In summary, the brain modulation induction training method based on a virtual reality environment provided in this application constructs a global EEG complexity feature vector reflecting the essential difference between genuine cognitive participation and strategic deception from the perspective of the signal's intrinsic generation mechanism by performing time-series decomposition and complexity feature extraction on the user's multi-channel EEG signals. This vector accurately identifies whether the user is in a deceptive state and generates a confidence factor to quantitatively assess the authenticity of neural signals. Furthermore, by weighting and correcting the difficulty adjustment gradient and feedback intensity based on the confidence factor, the method effectively inhibits the user's behavior of actively adjusting their own neural signals to obtain false positive feedback by learning system patterns, preventing deceptive behavior from distorting the training process. Simultaneously, the modulated difficulty parameters and feedback intensity parameters are synchronously written into the virtual scene engine to dynamically change the task interaction complexity and multi-sensory feedback intensity, forming a closed-loop induction training circuit with the confidence factor as the anti-deception discrimination criterion. This ensures that the training process is always adaptively adjusted based on the user's genuine cognitive participation state, significantly improving the authenticity and effectiveness of cognitive training and the reliability of neural plasticity induction.

[0051] In one embodiment, S1 of the brain modulation and induction training method based on a virtual reality environment provided by the present invention specifically includes the following steps:

[0052] S11: Collect multi-channel EEG signals from the user in a virtual reality cognitive scenario, perform short-time Fourier transform processing on each channel signal in the multi-channel EEG signal, decompose the time series of each channel signal into a spectral slice sequence arranged in time order, extract the power spectral density distribution on the frequency axis from each spectral slice, and generate the time-frequency domain energy spectrum matrix of each channel.

[0053] Specifically, the system acquires multi-channel EEG signals from users within a virtual reality cognitive scenario. EEG signal acquisition is performed using a multi-channel acquisition device, and the acquisition process is time-synchronized with the interaction within the virtual reality cognitive scenario to ensure that the EEG signals accurately correspond to the user's cognitive state in the virtual environment. The system performs a short-time Fourier transform (SFT) on each channel of the acquired multi-channel EEG signals. The SFT segments the continuous EEG time series of each channel using a sliding time window, with each time window corresponding to a signal segment. The Fourier transform converts this signal segment from the time domain to the frequency domain, obtaining the spectral information corresponding to each time window. The system decomposes the time series of each channel signal into a sequence of spectral slices arranged in chronological order. Each spectral slice corresponds to the spectral information of a time window, and the order of the spectral slices matches the sliding order of the time windows. The system extracts the power spectral density distribution on the frequency axis from each spectral slice. The power spectral density distribution reflects the power intensity corresponding to different frequency points in each spectral slice. The system arranges the power spectral density distribution of all spectral slices in each channel in chronological order to construct the time-frequency domain energy spectrum matrix for each channel. The rows of the time-frequency domain energy spectrum matrix correspond to the spectral slices in the time dimension, and the columns correspond to the power spectral density values ​​in the frequency dimension, thus fully preserving the time-domain and frequency-domain characteristics of the EEG signal in each channel.

[0054] S12: Calculate the centroid offset of the spectrum for each spectrum slice in the time-frequency domain energy spectrum matrix, calculate the Euclidean distance between the power-weighted average frequency value at each frequency point and the preset resting-state centroid of the spectrum, and perform standardized mapping processing on the entire frequency band to generate the centroid offset feature sequence of each channel.

[0055] Specifically, the system calculates the centroid offset for each spectral slice in the time-frequency domain energy spectrum matrix. The calculation process begins by determining the power value corresponding to each frequency point based on the power spectral density distribution of each spectral slice. A weighted average frequency value is then calculated for that spectral slice, with the weighting coefficients being the power spectral density values ​​at each frequency point. The system calls a preset resting-state centroid, which is obtained and stored by processing the same procedure using EEG signals collected from the user in a resting state before training. The system calculates the Euclidean distance between the power-weighted average frequency value of each spectral slice and the preset resting-state centroid. This Euclidean distance quantifies the degree of offset of the current spectral centroid relative to the resting-state centroid. The system then performs a standardized mapping on the calculated Euclidean distance across the entire frequency band. This standardization transforms the Euclidean distance to a fixed range, eliminating scale differences between different frequency intervals and ensuring the comparability of the offsets of each spectral slice. The system arranges the Euclidean distances of all spectral slices of each channel in chronological order after normalization, generating a spectral centroid shift feature sequence for each channel. This sequence can reflect the change of the spectral centroid of the EEG signal of each channel over time.

[0056] S13: Calculate the normalized entropy value of the power spectral density for each spectrum slice, and calculate the Shannon entropy after probabilistic normalization of the power spectral density value at each frequency point to generate the power spectral entropy feature sequence of each channel.

[0057] Specifically, the system calculates the normalized entropy value of the power spectral density for each spectral slice in the time-frequency domain energy spectrum matrix. The calculation process first processes the power spectral density distribution of each spectral slice, performing probabilistic normalization on the power spectral density value at each frequency point so that the sum of the normalized power spectral density values ​​at all frequency points is one, obtaining the probability value corresponding to each frequency point. Based on the probabilistically normalized power spectral density values, the system calculates the Shannon entropy. The Shannon entropy is obtained by performing a logarithmic operation on the probability values ​​of all frequency points and then weighted summing, used to quantify the complexity of the power spectral density distribution. The system sequentially calculates the normalized entropy value of the power spectral density for all spectral slices of each channel, arranging the entropy values ​​corresponding to each spectral slice in chronological order to generate a power spectral entropy feature sequence for each channel. This sequence reflects the change in the complexity of the power spectral distribution of the EEG signal in each channel over time; under conditions of genuine cognitive participation and strategic deception, this sequence will exhibit different patterns of change.

[0058] S14: Perform permutation conditional mutual information calculation on the spectral centroid shift feature sequence and power spectral entropy feature sequence between each channel. That is, perform symbolic coarsening and conditional probability distribution estimation on the feature sequence of each pair of channels, calculate the amount of uncertainty reduced by the feature sequence of one channel under the condition that the sequence of another channel is known, and use the conditional mutual information value as a quantitative index of the nonlinear coupling strength between channels to generate the coupling feature matrix between channels.

[0059] Specifically, the system performs permutation conditional mutual information calculation on the spectral centroid shift feature sequences and power spectral entropy feature sequences between each channel. The calculation process is carried out for each channel pair. First, the corresponding feature sequences of each channel pair undergo symbolic coarsening, converting continuous feature sequences into discrete symbol sequences, reducing data complexity while preserving core variation patterns. The system then estimates the conditional probability distribution of each pair of channel feature sequences after symbolic coarsening, determining the probability distribution of one channel's feature symbol sequence given the characteristic symbol sequence of another channel. Based on the conditional probability distribution, the system calculates the reduction in uncertainty of one channel's feature sequence given the sequence of the other channel; this reduction in uncertainty is the permutation conditional mutual information value. The system uses the conditional mutual information value as a quantitative indicator of the nonlinear coupling strength between channels; a larger conditional mutual information value indicates a stronger nonlinear coupling strength between the two channels, and vice versa. The system calculates the conditional mutual information values ​​for the spectral centroid shift feature sequence and power spectral entropy feature sequence of all channel pairs, and constructs a matrix based on the arrangement order of these conditional mutual information values ​​to generate the inter-channel coupling feature matrix. This matrix can comprehensively reflect the degree of nonlinear correlation between each channel.

[0060] S15: The spectral centroid shift feature sequence of each channel, the power spectral entropy feature sequence of each channel, and the inter-channel coupling feature matrix are fused by matrix concatenation and principal component dimensionality reduction. The dimensionality-reduced feature vector is then used as the global EEG complexity feature vector.

[0061] Specifically, the system collects the spectral centroid shift feature sequences, power spectral entropy feature sequences, and inter-channel coupling feature matrices for each channel, and performs feature fusion processing on these three types of features. The fusion process first uses a matrix concatenation method, converting the spectral centroid shift feature sequences and power spectral entropy feature sequences of each channel into matrix form. After dimensionality matching with the inter-channel coupling feature matrix, these are concatenated to form a unified high-dimensional feature matrix. This high-dimensional feature matrix contains the time-domain features of all channels and the coupling features between channels. The system then performs principal component dimensionality reduction (PCR) on the concatenated high-dimensional feature matrix. PCR dimensionality reduction extracts principal components from the high-dimensional feature matrix, retaining feature components that reflect the core information of the data, eliminating redundant information, reducing feature dimensionality, and avoiding the curse of dimensionality from affecting subsequent discrimination models. The system uses the PCR-reduced feature vector as the global EEG complexity feature vector, which integrates the time-frequency domain features of each channel and the coupling features between channels.

[0062] In one embodiment, S2 of the brain modulation and induction training method based on a virtual reality environment provided by the present invention specifically includes the following steps:

[0063] S21: Based on time window segmentation technology, the global EEG complexity feature vector is decomposed into multi-scale sliding window decomposition. Within each time window, the feature vector is split into overlapping sub-sequence segments along the time axis. The statistical moment parameters such as mean, variance, and kurtosis of each sub-sequence segment are calculated. The statistical moment parameters of each sub-sequence segment are compared with the preset resting state baseline statistical moment interval to generate abnormal fluctuation label sequences for each time window.

[0064] Specifically, the system employs a time window segmentation technique to perform multi-scale sliding window decomposition on the global EEG complexity feature vector. This decomposition continuously divides the global EEG complexity feature vector by setting the sliding step size and window length, resulting in multiple time windows arranged chronologically. Within each time window, the system splits the feature vector into overlapping sub-sequence segments along the time axis. The degree of overlap between these sub-sequence segments is determined based on the temporal resolution of the feature vector, ensuring that the feature information within each time window is fully covered. The system calculates the statistical moment parameters for each sub-sequence segment. These parameters quantify the distribution characteristics of the sub-sequence segments, with the mean and variance calculated using the following formulas:

[0065]

[0066]

[0067] In the formula, Let be the feature value in the subsequence segment, and n be the length of the subsequence segment. The mean of the subsequence segments. This represents the variance of the subsequence segments. The system calls a preset resting-state baseline statistical moment interval. This interval is determined and stored by calculating the statistical moment parameters after the system collects the user's EEG signals in a resting state before training, following the same processing procedure. The system compares the statistical moment parameters of each subsequence segment with the preset resting-state baseline statistical moment interval to determine whether the statistical moment parameters of each subsequence segment exceed the baseline interval. If they do, they are marked as abnormal; otherwise, they are marked as normal. The system arranges the marking results of all subsequence segments within each time window in chronological order, generating an abnormal fluctuation marking sequence for each time window. This sequence reflects the abnormal fluctuations of the global EEG complexity feature vector within each time window.

[0068] S22: Feature filtering is performed on the global EEG complexity feature vector based on the abnormal fluctuation label sequence. The filtered feature vector is mapped to a high-dimensional feature space through the radial basis kernel function and input into a binary classification discriminant function constructed with support vector machine for classification decision processing. The original decision value output by the discriminant function is converted into a posterior probability estimate through the Platt scaling method to generate the deception probability value of the user's current time window.

[0069] Specifically, the system performs feature filtering on the global EEG complexity feature vector based on abnormal fluctuation labeling sequences. The filtering process retains the feature vector components corresponding to sub-sequence segments marked as abnormal within the abnormal fluctuation labeling sequence, while removing feature components marked as normal and lacking effective information, thus achieving redundant feature vector removal. The system then maps the filtered feature vectors to a high-dimensional feature space using a radial basis function (RBF) kernel. It is used to solve linearly inseparable problems in low-dimensional feature spaces, and its calculation formula is as follows:

[0070]

[0071] In the formula, The input is the filtered feature vector. For support vectors, The kernel width parameter controls the radial range of the kernel function. The squared Euclidean distance between the support vectors and the input feature vectors is used. The system inputs the feature vectors, mapped to a high-dimensional feature space, into a binary classification discriminant function constructed using a support vector machine for classification decision processing. The calculation formula is:

[0072]

[0073] In the formula, This is the binary classification discriminant function, i.e., the support vector machine decision function. These are Lagrange multipliers used to adjust the weights of each support vector. The class labels for the support vectors are used to distinguish between genuine cognitive participation and strategic deception samples. denoted by , b is the number of support vectors, and 'b' is the bias term used to adjust the decision boundary of the discriminant function. The system uses the Platt scaling method to convert the raw decision values ​​output by the discriminant function into posterior probability estimates, generating the deception probability value for the user's current time window. The formula for calculating the deception probability value is:

[0074]

[0075] in, For the probability value of deception, and To estimate the Platt scaling parameters fitted by maximum likelihood estimation, For adaptive correction coefficients, The variance of the eigenvectors, This is the preset benchmark variance parameter.

[0076] S23: The deception probability value is smoothed by moving average to eliminate instantaneous noise fluctuations. The smoothed deception probability value is then accumulated by point-by-point deviation integration with the preset deception probability benchmark value. The confidence reduction coefficient of the current window is calculated based on the mapping relationship between the accumulated deviation area and the confidence decay curve, generating a confidence factor that reflects the authenticity of the user's neural signal. The confidence factor decreases monotonically as the accumulated deviation area increases.

[0077] Specifically, the system performs a moving average smoothing process on the generated deception probability value. Moving average smoothing calculates a weighted average of the deception probability values ​​over multiple consecutive time windows by setting a smoothing window, thus eliminating the impact of instantaneous noise fluctuations on the deception probability value. The moving average calculation formula is as follows:

[0078]

[0079] In the formula, Let be the smoothed probability of deception at time t, and m be the length of the smoothing window. The original deception probability value is given at time t - k. The system calls a preset deception probability baseline value, which is determined and stored based on the statistical results of the training samples. The system performs point-by-point deviation integration and accumulation processing on the smoothed deception probability value and the preset deception probability baseline value, calculating the deviation between the deception probability value and the baseline value within each time window. The accumulated deviation is calculated through integration, and the formula for calculating the area of ​​accumulated deviation is:

[0080]

[0081] In the formula, S is the cumulative deviation area. and These are the start and end times of the current time window, respectively. This is a preset baseline value for the deception probability. The system calculates the confidence reduction coefficient for the current window based on the mapping relationship between the cumulative deviation area and the confidence decay curve. The confidence decay curve defines the correspondence between the cumulative deviation area and the reduction coefficient; the larger the cumulative deviation area, the smaller the confidence reduction coefficient. The system combines the confidence reduction coefficient with the baseline confidence value to generate a confidence factor reflecting the authenticity of the user's neural signal. The confidence factor monotonically decreases as the cumulative deviation area increases; a larger cumulative deviation area indicates a higher probability of deception and a lower confidence factor, and vice versa. This confidence factor is used for weighted adjustment of subsequent training parameters.

[0082] In one embodiment, S3 of the brain modulation and induction training method based on a virtual reality environment provided by the present invention specifically includes the following steps:

[0083] S31: Based on the frontal electrode channel signal in the multi-channel EEG signal, extract the power spectral density values ​​of the theta and beta bands in each window, calculate the ratio of theta power to the beta power as the EEG cognitive load index, and read the user task response accuracy and response delay time of the current window from the virtual scene engine. Perform normalized weighted fusion processing on the accuracy and delay time to generate the task behavior cognitive load index. Linearly weighted fusion of the EEG cognitive load index and the task behavior cognitive load index through preset weight coefficients to generate a comprehensive cognitive load value.

[0084] Specifically, the system filters the prefrontal cortex electrode channel signals from multi-channel EEG signals. These signals are significantly correlated with the user's cognitive load and are the core signal source for cognitive load assessment. The system performs frequency band separation processing on the prefrontal cortex electrode channel signals, extracting the power spectral density values ​​of the theta and beta bands within each time window. Changes in the power spectral density of the theta and beta bands directly reflect the level of the user's cognitive load. The system calculates the ratio of theta power to beta power as an EEG cognitive load indicator; the calculation formula is as follows:

[0085]

[0086] In the formula, EEG cognitive load indicators This represents the power spectral density value in the theta band. This represents the power spectral density value in the beta band. The system reads the user task response accuracy and response latency time of the current window from the virtual scene engine. These two parameters are core indicators reflecting the cognitive load of user task behavior. The system normalizes the user task response accuracy and response latency time separately, transforming the two parameters to the same value range and eliminating dimensional differences. The system performs weighted fusion processing on the normalized accuracy and latency time to generate a task behavior cognitive load index. The system linearly weights and fuses the EEG cognitive load index and the task behavior cognitive load index using preset weight coefficients to generate a comprehensive cognitive load value, the calculation formula of which is:

[0087]

[0088] In the formula, To comprehensively assess cognitive load, The preset weighting coefficients for the EEG cognitive load indicators, Preset weighting coefficients for task behavior cognitive load indicators. The cognitive load index for task behavior is pre-defined, and the weighting coefficients are determined and stored in the system based on the focus of the cognitive load assessment.

[0089] S32: Compare the overall cognitive load value with the upper and lower boundaries of the preset target cognitive load range. If the overall cognitive load value is higher than the upper limit of the target load range, a positive adjustment error value is generated. If the overall cognitive load value is lower than the lower limit of the target load range, a negative adjustment error value is generated. If the overall cognitive load value falls within the target load range, the adjustment error value is zero, and a cognitive load deviation signal is generated.

[0090] Specifically, the system invokes a preset target cognitive load interval. This preset interval is determined based on the user's training objectives and baseline cognitive level. The target cognitive load interval includes an upper and lower limit to define a reasonable cognitive load range. The system compares the overall cognitive load value with the upper and lower boundaries of the preset target cognitive load interval. This deviation comparison determines the range of the overall cognitive load value by calculating the difference between the overall cognitive load value and the upper and lower boundaries. The system sets rules for generating adjustment error values: if the overall cognitive load value is higher than the upper limit of the target load interval, a positive adjustment error value is generated, which is the difference between the overall cognitive load value and the upper limit of the target load interval; if the overall cognitive load value is lower than the lower limit of the target load interval, a negative adjustment error value is generated, which is the difference between the overall cognitive load value and the lower limit of the target load interval; if the overall cognitive load value falls within the target load interval, the adjustment error value is zero. The system generates a cognitive load deviation signal based on the above rules. The formula for calculating the cognitive load deviation signal is:

[0091]

[0092] In the formula, For cognitive load deviation signal, The upper limit of the target cognitive load range. This is the lower limit of the target cognitive load range. This deviation signal directly reflects the degree of deviation between the comprehensive cognitive load value and the reasonable range.

[0093] S33: Multiply the cognitive load deviation signal by a preset proportional gain coefficient to obtain the basic difficulty adjustment gradient, then multiply the basic difficulty adjustment gradient by a confidence factor for weighted attenuation processing to generate a confidence-weighted difficulty adjustment gradient; after algebraically superimposing the current difficulty level and the difficulty adjustment gradient, perform amplitude limiting clamping processing through preset upper and lower boundaries of the difficulty value to generate the virtual scene difficulty parameters for the next moment.

[0094] Specifically, the system calls a preset proportional gain coefficient, which controls the influence of the cognitive load deviation signal on the difficulty adjustment gradient. This preset proportional gain coefficient is preset and stored by the system based on training experience. The system multiplies the cognitive load deviation signal by the preset proportional gain coefficient to obtain the basic difficulty adjustment gradient, calculated using the following formula:

[0095]

[0096] In the formula, Adjust the gradient based on the basic difficulty level. The preset proportional gain coefficient, This is the cognitive load deviation signal. The system calls the confidence factor generated by S23, multiplies the basic difficulty adjustment gradient by the confidence factor, and performs weighted attenuation processing to generate a confidence-weighted difficulty adjustment gradient. The calculation formula is as follows:

[0097]

[0098] In the formula, The difficulty adjustment gradient is weighted by confidence level. As a confidence factor, weighted attenuation processing adjusts the difficulty adjustment amplitude based on the authenticity of the neural signal, avoiding erroneous adjustments caused by deception. The system obtains the current difficulty level of the virtual scene, algebraically superimposes the current difficulty level with the confidence-weighted difficulty adjustment gradient, and obtains the preliminary difficulty parameters for the next time step. The system calls the preset upper and lower boundaries of the difficulty value to perform amplitude clamping processing on the preliminary difficulty parameters to ensure that the difficulty parameters are within a reasonable range, generating the virtual scene difficulty parameters for the next time step, the calculation formula of which is:

[0099]

[0100] In the formula, This is the difficulty parameter for the virtual scene in the next moment. The current difficulty level. This is the preset lower limit for difficulty. This is the preset upper limit for difficulty values. This is a clamping function used to limit the initial difficulty parameters between the upper and lower boundaries.

[0101] In one embodiment, S4 of the brain modulation and induction training method based on a virtual reality environment provided by the present invention specifically includes the following steps:

[0102] S41: Multiply the confidence factor by the preset baseline feedback strength to perform intensity modulation processing, generate a preliminary modulation feedback strength value, and perform exponential moving average filtering processing on the preliminary modulation feedback strength value and the feedback strength value at the previous moment to generate a smooth transition feedback strength parameter.

[0103] Specifically, the system calls the generated confidence factor and simultaneously calls the preset baseline feedback intensity. The preset baseline feedback intensity is set according to the different types of multi-sensory feedback and is used to provide a benchmark reference for the feedback intensity, and is stored in the system. The system multiplies the confidence factor by the preset baseline feedback intensity to perform intensity modulation processing. This processing establishes a correlation between the feedback intensity and the authenticity of the neural signal, generating a preliminary modulated feedback intensity value. The calculation formula is as follows:

[0104]

[0105] In the formula, This is the initial modulation feedback strength value. As the confidence factor, The system presets the baseline feedback intensity. It obtains the feedback intensity parameters from the previous time step, which are generated and stored in the previous scene configuration step to ensure the continuity of the feedback intensity. The system performs an exponential moving average filter on the initial modulation feedback intensity value and the feedback intensity value from the previous time step. This filter eliminates instantaneous fluctuations in the feedback intensity, achieving a smooth transition and preventing sudden changes from affecting the user's training experience. The calculation formula is as follows:

[0106]

[0107] In the formula, Feedback strength parameters for smooth transition, The exponential moving average coefficient is used to control the weighting of the current initial modulation value relative to the value at the previous time step. The feedback intensity parameters are from the previous moment.

[0108] S42: Convert the virtual scene difficulty parameter into a sequence of task interaction complexity control instructions, and convert the feedback intensity parameter into a sequence of multi-sensory feedback signal intensity control instructions; wherein, the task interaction complexity control instruction sequence includes stimulus presentation rate control instructions, working memory load control instructions, and response time window control instructions, and the multi-sensory feedback signal intensity control instruction sequence includes visual feedback brightness instructions, auditory feedback volume instructions, and tactile feedback intensity instructions.

[0109] Specifically, the system acquires the virtual scene difficulty parameter S, which reflects the difficulty level of the training scene at the next moment. The system converts the virtual scene difficulty parameter into a sequence of task interaction complexity control instructions. The conversion process is achieved through a preset parameter-instruction mapping relationship, which is set and stored based on the correlation between the difficulty parameter and task interaction complexity. The task interaction complexity control instruction sequence includes stimulus presentation rate control instructions, working memory load control instructions, and response time window control instructions. The stimulus presentation rate control instructions are used to adjust the presentation frequency of cognitive stimuli in the virtual scene, the working memory load control instructions are used to adjust the cognitive task load that the user needs to process, and the response time window control instructions are used to adjust the time limit for the user to complete the cognitive task. The system also acquires the feedback intensity parameter and converts it into a sequence of multi-sensory feedback signal intensity control instructions. The conversion process is also achieved through a preset parameter-instruction mapping relationship. The multi-sensory feedback signal intensity control command sequence includes visual feedback brightness commands, auditory feedback volume commands, and haptic feedback intensity commands. Visual feedback brightness commands adjust the brightness parameters of visual cues in the virtual scene, auditory feedback volume commands adjust the volume parameters of auditory cues in the virtual scene, and haptic feedback intensity commands adjust the haptic vibration intensity parameters of the VR device. After the system completes the conversion between the two types of control command sequences, it performs format standardization processing to ensure that the command sequences conform to the receiving specifications of the virtual scene engine.

[0110] S43: The task interaction complexity control instruction sequence and the multi-sensory feedback signal strength control instruction sequence are synchronized and written into the control register of the virtual scene engine after being aligned in time, so that the virtual scene engine can dynamically reconfigure scene parameters according to the written instruction sequence and generate updated virtual environment configuration parameters.

[0111] Specifically, the system performs time alignment processing on the task interaction complexity control instruction sequence and the multi-sensory feedback signal strength control instruction sequence. This time alignment is based on the timestamps of the instruction sequences, ensuring that both types of instruction sequences execute synchronously at the same time node. This avoids scene configuration disorder caused by instruction execution delays, and their time alignment relationship satisfies… In the formula Let be the execution timestamp of the i-th instruction in the task interaction complexity control instruction sequence. This is the execution timestamp of the i-th instruction in the multi-sensory feedback signal strength control instruction sequence. The system synchronously writes the time-aligned two types of instruction sequences into the control register of the virtual scene engine. The writing process is implemented through the API interface of the virtual scene engine to ensure the real-time performance and accuracy of instruction writing. The control register is used to store the scene control instructions to be executed, which can be read by the virtual scene engine in real time.

[0112] The virtual scene engine reads the instruction sequence from the control register in real time and dynamically reconfigures scene parameters based on the written instruction sequence, synchronously adjusting the task interaction complexity and multi-sensory feedback intensity. The system obtains the reconfigured scene parameters through the virtual scene engine's parameter feedback interface, integrating these parameters into updated virtual environment configuration parameters. These updated virtual environment configuration parameters include task interaction complexity parameters and multi-sensory feedback intensity parameters, which drive the virtual reality cognitive scene operation in the next training cycle, ensuring that the scene configuration matches the user's actual cognitive state and training needs.

[0113] In one embodiment, step S5 of the brain modulation and induction training method based on a virtual reality environment provided by the present invention specifically includes the following steps:

[0114] S51: Unpack the updated virtual environment configuration parameters into a task logic parameter table and a sensory feedback parameter table, and write the task logic parameter table into the scene state buffer of the virtual reality rendering pipeline to overwrite the old configuration. At the same time, distribute the sensory feedback parameter table to the driver registers of the three feedback channels of vision, hearing and touch to generate the updated virtual reality cognitive scene.

[0115] Specifically, the system obtains the updated virtual environment configuration parameters generated by S43. These parameters are the integrated scene configuration data, containing information related to task interaction complexity and multi-sensory feedback intensity. The system unpacks the updated virtual environment configuration parameters using a preset parameter parsing algorithm, splitting the virtual environment configuration parameters into a task logic parameter table and a sensory feedback parameter table. The unpacking process satisfies... In the formula The updated virtual environment is configured with parameters: Task_Table (task logic parameter table), Feedback_Table (sensory feedback parameter table), and ⊕ (parameter unpacking operator) used to categorize and split configuration parameters. The task logic parameter table contains the core operational parameters for cognitive tasks in the virtual scene, while the sensory feedback parameter table contains the intensity control parameters for the visual, auditory, and tactile feedback channels.

[0116] Furthermore, the system writes the task logic parameter table into the scene state buffer of the virtual reality rendering pipeline. The scene state buffer stores the runtime configuration parameters of the virtual scene, and the writing process overwrites the old configuration parameters in the buffer to ensure that the new task logic parameters take effect. Simultaneously, the system distributes the sensory feedback parameter table to the driver registers of the visual, auditory, and tactile feedback channels. Each feedback channel has an independent driver register used to store the feedback intensity control parameters for that channel. The system reads the task logic parameter table from the scene state buffer through the virtual reality rendering pipeline to drive the virtual scene to render and update. At the same time, it reads the sensory feedback parameter table from the driver registers of each feedback channel to drive each feedback channel to load the new feedback intensity parameters, ultimately generating an updated virtual reality cognitive scene for the user to enter the next round of training.

[0117] S52: During the operation of the updated virtual reality cognitive scene, the multi-channel EEG signal acquisition operation in S1 is re-executed. The newly acquired EEG signals are used as input data for the next iteration. The task performance data of the current window is read from the virtual scene engine and fed back to step S3 as the basis for calculating the task behavior cognitive load index, forming a closed-loop data flow channel of perception-computation-regulation.

[0118] Specifically, the system initiates the updated virtual reality cognitive scene. During scene operation, the multi-channel EEG signal acquisition operation in S1 is re-executed. The operating parameters of the acquisition device remain consistent with those in S1 to ensure the consistency and comparability of the newly acquired EEG signals with those previously acquired. The acquisition process is synchronized with the scene operation to ensure that the EEG signals accurately correspond to the user's cognitive state in the new scene. The system temporarily stores the newly acquired multi-channel EEG signals as input data for the next iteration, used for generating the global EEG complexity feature vector and deception detection. Simultaneously, the system reads user task performance data from the virtual scene engine for the current time window. This data includes core parameters such as the user's task response accuracy and response latency in the current scene, which are crucial for calculating the task behavior cognitive load index. The system standardizes the read task performance data to eliminate format differences and then feeds it back to step S3 as the basis for calculating the task behavior cognitive load index in S31. Through these operations, the system constructs a closed-loop data flow channel of perception-computation-regulation, where the data flow relationship satisfies… In the formula, Data_new represents the newly acquired EEG signal and the task performance data of the current window. This channel enables the cyclical flow of data, ensuring that the adjustment of training parameters in each round is based on the user's current real cognitive state.

[0119] S53: After each iteration cycle, the confidence factor, difficulty parameter, and feedback intensity parameter of the current window are recorded in the training log, and the change curves of the confidence factor, difficulty parameter, and feedback intensity parameter on the time axis are visualized and output. The closed-loop induced training loop ends when a training termination signal triggered by the user is detected. The confidence factor is used as the basis for anti-spoofing judgment to dynamically adjust the difficulty adjustment weight in subsequent iterations, and the difficulty parameter and feedback intensity parameter are continuously written back to the virtual scene engine as execution objects.

[0120] Specifically, after each iteration cycle, the system triggers a parameter recording operation, synchronously recording the confidence factor, difficulty parameter, and feedback intensity parameter of the current time window into the training log. The training log stores the core parameters throughout the training process, facilitating subsequent training effect evaluation and parameter optimization. The system then calls the visualization module to visualize the change curves of the confidence factor, difficulty parameter, and feedback intensity parameter on the time axis. The visualization output is achieved using a preset curve plotting algorithm, and the expression for the change curves is as follows:

[0121]

[0122] In the formula, This is a curve showing the parameter variation over time. Let be the confidence factor at time t. Let S(t) be the difficulty parameter at time t, and S(t) be the feedback intensity parameter at time t. The visualized output is provided for training monitoring personnel to view, allowing them to monitor parameter changes in real time during the training process. The system continuously monitors user-triggered training termination signals, which are triggered by the user through the interactive button on the VR device. The system detects the device's interactive signals in real time to determine if a training termination command exists. Before detecting a user-triggered training termination signal, the system continuously maintains the operation of the closed-loop induced training loop. The confidence factor serves as the anti-spoofing criterion, dynamically adjusting the difficulty adjustment weight in subsequent iterations. The difficulty parameter and feedback intensity parameter are the execution objects, continuously written back to the virtual scene engine to ensure dynamic updates of training parameters. When the system detects a user-triggered training termination signal, it immediately stops the operation of the closed-loop induced training loop, while saving the training log and visualized output, completing the entire training process.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides a virtual reality-based brain modulation and induction training system for implementing the aforementioned virtual reality-based brain modulation and induction training method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the virtual reality-based brain modulation and induction training system provided below can be found in the limitations of the virtual reality-based brain modulation and induction training method described above, and will not be repeated here.

[0125] Preferably, such as Figure 2 As shown, the present invention provides a brain modulation and induction training system 600 based on a virtual reality environment, which is configured with the following modules:

[0126] The EEG feature extraction module 610 is used to collect multi-channel EEG signals of users in virtual reality cognitive scenarios, perform time series decomposition on the multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel, calculate the spectral centroid offset and power spectral density normalized entropy value of each channel based on the time-frequency domain energy spectrum, and perform inter-channel mutual information fusion to generate a global EEG complexity feature vector.

[0127] The neural confidence discrimination module 620 is used to calculate the deception probability of the user in the current time window by inputting the global EEG complexity feature vector into a preset discrimination function for comparison, and to generate a confidence factor reflecting the authenticity of the user's neural signals based on the deception probability. The preset discrimination function is a discrimination model obtained by collecting the first EEG signal in the real cognitive participation state and the second EEG signal in the strategic deception state, and using the complexity feature vectors generated corresponding to the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training.

[0128] The cognitive difficulty adjustment module 630 is used to compare the deviation between the user's current cognitive load and the target cognitive load range based on the confidence factor and the task performance data fed back in real time from the virtual scene, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting processing to generate the virtual scene difficulty parameters for the next moment.

[0129] The scene parameter modulation module 640 is used to modulate the preset baseline feedback intensity based on the confidence factor, and synchronously write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine, so that the virtual scene can dynamically change the task interaction complexity according to the virtual scene difficulty parameter and adjust the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, and generate updated virtual environment configuration parameters.

[0130] The closed-loop induction training module 650 is used to apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, and at the same time, to cycle the newly acquired EEG signals back to the EEG feature extraction module 610, forming a cognitive state closed-loop induction training loop with confidence factor as the anti-deception judgment basis and difficulty parameter and feedback intensity parameter as the execution object.

[0131] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described brain modulation and induction training method based on a virtual reality environment.

[0132] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described brain modulation and induction training method based on a virtual reality environment.

[0133] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0134] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0135] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A brain modulation and induction training method based on a virtual reality environment, characterized in that, Includes the following steps: S1: Collect multi-channel EEG signals of the user in a virtual reality cognitive scene, perform time series decomposition on the multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel, calculate the spectral centroid offset and power spectral density normalized entropy value of each channel based on the time-frequency domain energy spectrum, and perform inter-channel mutual information fusion to generate a global EEG complexity feature vector. S2: Based on the comparison of the global EEG complexity feature vector input into a preset discriminant function, the deception probability of the user's current time window is calculated, and a confidence factor reflecting the authenticity of the user's neural signals is generated based on the deception probability. The preset discrimination function is a discrimination model obtained by collecting the first EEG signal in the real cognitive participation state and the second EEG signal in the strategic deception state, and using the complexity feature vectors generated corresponding to the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training. S3: Based on the confidence factor and the task performance data fed back in real time in the virtual scene, compare the deviation between the user's current cognitive load and the target cognitive load range, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting processing to generate the virtual scene difficulty parameters for the next moment. S4: Modulate the preset baseline feedback intensity based on the confidence factor, and write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine in sync, so that the virtual scene dynamically changes the task interaction complexity according to the virtual scene difficulty parameter and adjusts the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, and generates updated virtual environment configuration parameters. S5: Apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, and at the same time loop the newly collected EEG signals back to S1 to form a cognitive state closed-loop induced training loop with the confidence factor as the anti-deception judgment basis and the difficulty parameter and the feedback intensity parameter as the execution objects. The confidence factor is obtained through the following steps: S21: Based on the time window segmentation technology, the global EEG complexity feature vector is decomposed into a multi-scale sliding window. In each time window, the feature vector is split into overlapping sub-sequence segments along the time axis. The statistical moment parameters of each sub-sequence segment are calculated. The statistical moment parameters of each sub-sequence segment are compared with the preset resting state benchmark statistical moment interval to generate abnormal fluctuation marker sequences for each time window. S22: Based on the abnormal fluctuation label sequence, the global EEG complexity feature vector is filtered, and the filtered feature vector is mapped to a high-dimensional feature space through the radial basis kernel function and input into a binary classification discriminant function constructed with support vector machine for classification decision processing. The original decision value output by the discriminant function is converted into a posterior probability estimate through the Platt scaling method to generate the deception probability value of the user's current time window. S23: The deception probability value is smoothed by moving average to eliminate instantaneous noise fluctuations. The smoothed deception probability value is then accumulated by point-by-point deviation integration with the preset deception probability benchmark value. The confidence reduction coefficient of the current window is calculated based on the mapping relationship between the accumulated deviation area and the confidence decay curve, thereby generating a confidence factor that reflects the authenticity of the user's neural signal. The confidence factor decreases monotonically as the accumulated deviation area increases.

2. The method according to claim 1, characterized in that, S1 includes: S11: Collect multi-channel EEG signals of the user in the virtual reality cognitive scene, perform short-time Fourier transform processing on each channel signal of the multi-channel EEG signal, decompose the time series of each channel signal into a spectral slice sequence arranged in time order, extract the power spectral density distribution on the frequency axis from each spectral slice, and generate the time-frequency domain energy spectrum matrix of each channel. S12: Calculate the centroid offset of the spectrum for each spectrum slice in the time-frequency domain energy spectrum matrix, calculate the Euclidean distance between the power-weighted average frequency value at each frequency point and the preset resting state centroid of the spectrum, and perform standardized mapping processing on the entire frequency band to generate the centroid offset feature sequence of each channel. S13: Calculate the normalized entropy value of the power spectral density for each spectrum slice, and calculate the Shannon entropy after probabilistic normalization of the power spectral density value at each frequency point to generate the power spectral entropy feature sequence of each channel. S14: Perform arrangement conditional mutual information calculation on the spectral centroid shift feature sequence and power spectral entropy feature sequence between each channel, that is, perform symbolic coarsening and conditional probability distribution estimation on the feature sequence of each pair of channels, calculate the amount of uncertainty reduced by the feature sequence of one channel under the condition that the sequence of another channel is known, use the conditional mutual information value as a quantitative index of the nonlinear coupling strength between channels, and generate the coupling feature matrix between channels. S15: The spectral centroid shift feature sequence of each channel, the power spectral entropy feature sequence of each channel, and the inter-channel coupling feature matrix are subjected to feature fusion processing through matrix concatenation and principal component dimensionality reduction, and the dimensionality-reduced feature vector is used as the global EEG complexity feature vector.

3. The method according to claim 1, characterized in that, The deception probability value is calculated using the following formula: ； ； ； in, For the probability value of deception, Here, V is the decision function for the support vector machine, and V is the input feature vector. For support vectors, For Lagrange multipliers, For the class labels of support vectors, Here, b is the radial basis kernel function, and b is the bias term. The number of support vectors, For kernel width parameter, and To estimate the Platt scaling parameters fitted by maximum likelihood estimation, For adaptive correction coefficients, The variance of the eigenvectors, This is the preset benchmark variance parameter.

4. The method according to claim 1, characterized in that, S3 includes: S31: Based on the frontal electrode channel signal in the multi-channel EEG signal, extract the power spectral density values ​​of the theta band and the beta band in each window, calculate the ratio of theta power to the beta power as the EEG cognitive load index, and read the user task response accuracy and response delay time of the current window from the virtual scene engine. Perform normalized weighted fusion processing on the accuracy and delay time to generate the task behavior cognitive load index. Perform linear weighted fusion of the EEG cognitive load index and the task behavior cognitive load index through preset weight coefficients to generate a comprehensive cognitive load value. S32: The comprehensive cognitive load value is compared with the upper and lower boundaries of the preset target cognitive load range. If the comprehensive cognitive load value is higher than the upper limit of the target load range, a positive adjustment error value is generated. If the comprehensive cognitive load value is lower than the lower limit of the target load range, a negative adjustment error value is generated. If the comprehensive cognitive load value falls within the target load range, the adjustment error value is zero, and a cognitive load deviation signal is generated. S33: Multiply the cognitive load deviation signal by a preset proportional gain coefficient to obtain the basic difficulty adjustment gradient, then multiply the basic difficulty adjustment gradient by the confidence factor and perform weighted attenuation processing to generate a confidence-weighted difficulty adjustment gradient; after algebraically superimposing the current difficulty level and the difficulty adjustment gradient, perform amplitude limiting clamping processing through preset upper and lower boundaries of difficulty values ​​to generate the virtual scene difficulty parameters for the next moment.

5. The method according to claim 1, characterized in that, S4 includes: S41: Multiply the confidence factor by the preset baseline feedback strength and perform intensity modulation processing to generate a preliminary modulation feedback strength value. Then, perform exponential moving average filtering processing on the preliminary modulation feedback strength value and the feedback strength value at the previous moment to generate a smooth transition feedback strength parameter. S42: Convert the virtual scene difficulty parameter into a task interaction complexity control instruction sequence, and convert the feedback intensity parameter into a multi-sensory feedback signal intensity control instruction sequence; wherein, the task interaction complexity control instruction sequence includes stimulus presentation rate control instruction, working memory load control instruction and response time window control instruction, and the multi-sensory feedback signal intensity control instruction sequence includes visual feedback brightness instruction, auditory feedback volume instruction and tactile feedback intensity instruction; S43: The task interaction complexity control instruction sequence and the multi-sensory feedback signal strength control instruction sequence are aligned in time and synchronously written into the control register of the virtual scene engine, so that the virtual scene engine can dynamically reconfigure the scene parameters according to the written instruction sequence and generate updated virtual environment configuration parameters.

6. The method according to any one of claims 1-5, characterized in that, S5 includes: S51: Unpack the updated virtual environment configuration parameters into a task logic parameter table and a sensory feedback parameter table, and write the task logic parameter table into the scene state buffer of the virtual reality rendering pipeline to overwrite the old configuration. At the same time, distribute the sensory feedback parameter table to the driver registers of the three feedback channels of vision, hearing and touch to generate an updated virtual reality cognitive scene. S52: During the operation of the updated virtual reality cognitive scene, the multi-channel EEG signal acquisition operation in S1 is re-executed, the newly acquired EEG signal is used as the input data for the next iteration, and the task performance data of the current window is read from the virtual scene engine and fed back to step S3 as the basis for calculating the task behavior cognitive load index, forming a closed-loop data flow channel of perception-computation-regulation. S53: After each iteration cycle, the confidence factor, difficulty parameter, and feedback intensity parameter of the current window are recorded in the training log, and the change curves of the confidence factor, difficulty parameter, and feedback intensity parameter on the time axis are visualized and output. The closed-loop induced training loop ends when a training termination signal triggered by the user is detected. The confidence factor is used as the anti-deception discrimination basis to dynamically adjust the difficulty adjustment weight in subsequent iterations, and the difficulty parameter and feedback intensity parameter are continuously written back to the virtual scene engine as execution objects.

7. A brain modulation and induction training system based on a virtual reality environment, characterized in that, The system includes: The EEG feature extraction module is used to collect multi-channel EEG signals from users in virtual reality cognitive scenarios, perform time-series decomposition on the multi-channel EEG signals to extract the time-frequency domain energy spectrum of each channel, calculate the spectral centroid offset and power spectral density normalized entropy value of each channel based on the time-frequency domain energy spectrum, and perform inter-channel mutual information fusion to generate a global EEG complexity feature vector. The neural confidence discrimination module is used to calculate the deception probability of the user's current time window by comparing the global EEG complexity feature vector with a preset discrimination function, and to generate a confidence factor reflecting the authenticity of the user's neural signals based on the deception probability. The preset discrimination function is a discrimination model obtained by collecting a first EEG signal under the real cognitive participation state and a second EEG signal under the strategic deception state, and using the complexity feature vectors generated corresponding to the first EEG signal and the second EEG signal respectively as samples for binary classification supervised learning training. The cognitive difficulty adjustment module is used to compare the deviation between the user's current cognitive load and the target cognitive load range based on the confidence factor and the task performance data fed back in real time from the virtual scene, calculate the confidence-weighted difficulty adjustment gradient, and then superimpose the difficulty adjustment gradient with the current difficulty level and perform amplitude limiting processing to generate the virtual scene difficulty parameters for the next moment. The scene parameter modulation module is used to modulate the preset baseline feedback intensity based on the confidence factor, and synchronously write the modulated feedback intensity parameter and the virtual scene difficulty parameter into the virtual scene engine, so that the virtual scene dynamically changes the task interaction complexity according to the virtual scene difficulty parameter and adjusts the intensity of the multi-sensory feedback signal according to the feedback intensity parameter, and generates updated virtual environment configuration parameters. The closed-loop induced training module is used to apply the updated virtual environment configuration parameters to the virtual reality cognitive scene of the next training cycle, and at the same time, to cycle the newly acquired EEG signals back to the EEG feature extraction module, forming a cognitive state closed-loop induced training loop with the confidence factor as the anti-deception judgment basis and the difficulty parameter and the feedback intensity parameter as the execution objects. The confidence factor is obtained through the following steps: The global EEG complexity feature vector is decomposed into a multi-scale sliding window based on time window segmentation technology. In each time window, the feature vector is split into overlapping sub-sequence segments along the time axis. The statistical moment parameters of each sub-sequence segment are calculated, and the statistical moment parameters of each sub-sequence segment are compared with the preset resting state benchmark statistical moment interval to generate abnormal fluctuation label sequences for each time window. Based on the abnormal fluctuation label sequence, the global EEG complexity feature vector is filtered. The filtered feature vector is mapped to a high-dimensional feature space through a radial basis kernel function and input into a binary classification discriminant function constructed with a support vector machine for classification decision processing. The original decision value output by the discriminant function is converted into a posterior probability estimate through the Platt scaling method to generate the deception probability value of the user's current time window. The deception probability value is smoothed by moving average to eliminate instantaneous noise fluctuations. The smoothed deception probability value is then accumulated by point-by-point deviation integration with a preset deception probability benchmark value. The confidence reduction coefficient of the current window is calculated based on the mapping relationship between the accumulated deviation area and the confidence decay curve, generating a confidence factor that reflects the authenticity of the user's neural signal. The confidence factor decreases monotonically as the accumulated deviation area increases.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images