An intelligent neural regulation method and system based on attention state

Abstract

An intelligent neural regulation method and system based on attention state, belonging to the field of attention state analysis and regulation. The method includes: collecting EEG signals in real time, preprocessing them, and then extracting attention-related features; establishing baseline data by collecting EEG signals of users in different attention states according to the usage task requirements in the initial stage; training an attention state evaluation model using an intelligent learning algorithm; when the user uses the device, collecting and monitoring EEG signals in real time, and using the attention state evaluation model to evaluate the EEG signals to judge the user's current attention state; judging whether neuroregulation needs to be performed according to the evaluation result; generating a neuroregulation strategy and executing the neuroregulation strategy. Based on the attention state evaluation level of electroencephalogram signals, the present invention intelligently selects the optimal non-invasive transcranial electrical stimulation neuroregulation method according to the evaluation result to improve the attention state, and has higher timeliness, accuracy and sensitivity.

Description

Technical Field

[0001] The present invention belongs to the technical field of attention state analysis and regulation, and particularly relates to an intelligent neural regulation method and system based on attention state. Background Art

[0002] A person's attention state plays a key role in many jobs and tasks. Whether the attention is concentrated directly affects the work effect and safety. For example, in car or airplane driving, the driver needs to concentrate on the equipment and environmental conditions to quickly respond to sudden situations. Therefore, it is necessary to adjust and enhance the attention state through some feasible technologies to avoid serious accidents caused by distraction during tasks or work.

[0003] Attention state analysis based on behavioral performance is a method of inferring an individual's attention state by observing and analyzing the individual's external behaviors. This method can combine different behavioral characteristics, such as eye movement, facial expression, body posture, and activity level, to evaluate the degree of attention concentration of an individual. However, factors such as environmental dependence, subjective performance characteristics, and individual differences pose great challenges to the accuracy of the data.

[0004] Adjusting the attention state through sensory feedback cues is a commonly used method. For example, sounds, visual signals, tactile vibrations, etc. are used to remind an individual to maintain or restore attention. However, when using this method for feedback cues, if the user is concentrating on the task during the task, the sensory feedback itself may become a source of interference, causing the user's attention state to decrease; also, if other factors cause the user's attention to decline during the task, their own sensory attention will also decrease, and the sensory cues may be ignored due to the user's inattentiveness, making it difficult to improve the attention state through self-regulation.

[0005] In existing methods, generally, the behavior performance of the user is observed through video to evaluate the attention state, such as whether yawning, whether the eyes are tired (Chinese Patent with Publication No. CN112401887A), whether there are actions unrelated to the current behavior (eating / drinking / calling / entertainment actions, etc. while driving) (Chinese Patent with Publication No. CN109937152A), and based on this, only a two-level evaluation of whether the user's attention state is or not is performed. In addition, there are also a small number of methods for monitoring the attention state using electroencephalogram signals as the analysis basis (Chinese Patent with Publication No. CN112957049A). In these existing methods, on the one hand, the evaluation of the user's behavior performance is easily affected by subjective performance and the recognition rate, and the monitoring device is also affected by the environment (video occlusion), reducing the recognition rate; on the other hand, in the method of analyzing the attention state through electroencephalogram signals, when the user's attention state is low, it is often through the way of sensory cues for feedback, but when the attention state is poor, the sensitivity to the senses is also poor. Summary of the Invention

[0006] To solve the problems of low accuracy and sensitivity existing in the existing attention state regulation methods, the present invention provides an intelligent neural regulation method and system based on attention state.

[0007] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0008] An intelligent neural regulation method based on attention state of the present invention mainly includes the following steps:

[0009] (1) Real-time collect the user's EEG signals and capture the brain waves related to the attention state;

[0010] (2) Preprocess the collected original EEG signals and extract the attention-related features in the EEG signals;

[0011] (3) At the initial stage when the user uses the device, collect the EEG signals of the user in different attention states according to the usage task requirements to establish baseline data; use an intelligent learning algorithm to train an attention state evaluation model based on the baseline data;

[0012] (4) When the user uses the device, the system will real-time collect and monitor the EEG signals, and at the same time use the trained attention state evaluation model to analyze and evaluate the real-time collected EEG signals to judge the user's current attention state;

[0013] (5) Generate a neural regulation strategy;

[0014] According to the real-time attention state evaluation result, judge whether neural regulation needs to be performed; when neural regulation needs to be performed, the system will generate a neural regulation strategy to select a suitable neural regulation method and neural regulation parameters;

[0015] (6) Execute the neural regulation strategy.

[0016] Further, in step (2), the collected original EEG signals are subjected to filtering processing.

[0017] Further, in step (2), the EEG signals in each time window are subjected to spectral analysis, and the power values of each frequency band are calculated using the fast Fourier transform.

[0018] Further, in step (3), the attention state evaluation model includes: a data preprocessing module and a shallow convolutional neural network module;

[0019] The data preprocessing module is used to organize and perform time-frequency conversion on EEG signal data: First, the data is cleaned to remove redundant and missing data, reducing the computational burden of the model on the data; then, spectral analysis is performed on the EEG signals within each time window, and the power values of each frequency band are calculated using the fast Fourier transform; the principal component analysis method is then used to reduce the number of features; finally, the data is normalized to accelerate model convergence;

[0020] The shallow convolutional neural network module is used to classify and evaluate the preprocessed data; the shallow convolutional neural network module includes: an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer; the input layer obtains the preprocessed signals according to the perception of the user's individual use and the corresponding brain regions or multi-perceptions; the convolutional layer adopts a double-layer structure, and local features are extracted through two convolutional layers, and a max pooling layer is set after each convolutional layer to reduce the feature size; the result is further optimized in the fully connected layer to reduce overfitting; the output layer uses the Softmax activation function to achieve the classification and evaluation of the attention state.

[0021] Furthermore, in step (3), the specific training process of the attention state evaluation model is as follows:

[0022] The training process of the attention state evaluation model depends on the EEG signal data of different attention states of the user's individual for training; the tasks completed by the user for the first time are respectively the complex math problem task of high cognitive load, the reading task of medium cognitive load, the image recognition task of low cognitive load, and the relaxation state of closing eyes and resting. These tasks are sequentially divided into high attention state, medium attention state, low attention state, and distracted attention state, and the EEG signal data collected under different attention states is used as the classification basis. Among them, the proportions of the training set and the test set are 80% and 20% respectively, and the shallow convolutional neural network module is trained using the cross-validation method.

[0023] Furthermore, in step (4), a threshold and the softmax classification algorithm are used to evaluate the user's current attention state, and the specific operation process is as follows:

[0024] After the data is processed by the fully connected layer, the optimized classification results are sent to the softmax output layer and the sigmoid output layer respectively, and the categories of the data are judged by probability and threshold respectively. The softmax output layer outputs a multi-class probability distribution, and then obtains the category through Argmax, where the category with the maximum probability is taken, and then the category is predicted through softmax, and the A classification is output; the sigmoid output layer outputs the probability of each category, and then judges the category through the threshold, greater than the threshold is 1, and finally the category is judged through the threshold and the B classification is output; finally, the A classification and the B classification are integrated to obtain the C classification, and the final classification result is output.

[0025] Further, in step (5), through EEG real-time monitoring and the attention state evaluation results, an adaptive optimal neuromodulation strategy is screened with the transcranial electrical stimulation strategy. The specific operation process is as follows:

[0026] After EEG real-time monitoring and attention state evaluation, the user state level is obtained. For the evaluated high attention state, medium attention state, low attention state or attention dispersion state, transcranial electrical stimulation strategies with different parameters are set: when in the high attention state, a Gaussian random current stimulation scheme in the range of 0.5-1 mA is used to maintain the attention state and optimize the allocation of attention resources; when in the medium attention state, a two-type mixed transcranial electrical stimulation scheme in the range of 0.5-1 mA is adopted, and a Gaussian random current stimulation scheme and a γ-wave (30-45 Hz) cosine / sine alternating current stimulation scheme are respectively adopted; when in the low attention state, a two-type mixed transcranial electrical stimulation scheme in the range of 1.0-1.5 mA is adopted, and a Gaussian random current stimulation scheme and a β-wave (13-30 Hz) cosine / sine alternating current stimulation scheme are respectively adopted; when in the attention dispersion state, a three-type mixed transcranial electrical stimulation scheme in the range of 1.5-2 mA is adopted, and a Gaussian random current stimulation scheme, a β-wave cosine / sine alternating current stimulation scheme and a transcranial direct current stimulation scheme are respectively adopted.

[0027] Further, after each stimulation cycle, the attention state will be evaluated again, and the transcranial electrical stimulation strategy will be optimized with the evaluation threshold result; when the first evaluation result is the high attention state, real-time periodic monitoring and evaluation will be carried out, and Gaussian random current stimulation will be performed after 1, 3, 5... cycles in turn; after the attention state level is evaluated again as the medium attention state, low attention state or attention dispersion state, the following corresponding transcranial electrical stimulation strategies will be executed: when the threshold after stimulation tends to the high attention state, the previous transcranial electrical stimulation scheme will be maintained, and this transcranial electrical stimulation scheme is also the optimal neuromodulation scheme; when the threshold after stimulation tends to remain unchanged or decrease, the parameters of the previous transcranial electrical stimulation scheme will be adjusted, and at the same time, 0.1 mA current and 1 Hz frequency will be increased, and it will be adjusted once every execution cycle until the optimal neuromodulation scheme is adjusted.

[0028] The time ratios corresponding to the Gaussian random current stimulation scheme in the high, medium, low attention states and the attention dispersion state are 100%, 50%, 30% and 10% in sequence; the time ratio corresponding to the transcranial direct current stimulation scheme in the attention dispersion state is 30%, and the remaining time is for its corresponding complementary / sine alternating current stimulation scheme; at the same time, the stimulation time period ratio remains unchanged as the real-time monitoring threshold biases towards the high attention state, or the complementary / sine alternating current stimulation scheme occupancy time increases when the threshold corresponds to an unchanged attention state or biases towards the low attention state to obtain an optimal neuromodulation scheme; while when the threshold remains unchanged or decreases in the attention dispersion state, the time ratio corresponding to the transcranial direct current stimulation scheme increases.

[0029] An intelligent neuromodulation system based on attention state provided by the present invention is used to implement the intelligent neuromodulation method based on attention state described above. This system mainly includes: an electroencephalogram acquisition module, a main control module, and a neuromodulation module; the main control module sends a start instruction for EEG signal acquisition to the electroencephalogram acquisition module, the electroencephalogram acquisition module starts to acquire EEG signals and sends them to the main control module, the main control module performs signal processing and feature extraction, and at the same time conducts model training and real-time attention state evaluation, and judges whether neuromodulation needs to be performed according to the real-time attention state evaluation result; when neuromodulation needs to be performed, a neuromodulation strategy will be generated in the neuromodulation module, the main control module sends a neuromodulation parameter instruction to the neuromodulation module, and the neuromodulation module will execute the neuromodulation strategy according to the neuromodulation parameter instruction and feedback a regulation end signal to the main control module.

[0030] Furthermore, when a neuromodulation strategy for the user's attention state needs to be executed, the electroencephalogram acquisition module and the neuromodulation module alternately start and stop the cyclic regulation scheme operation.

[0031] The beneficial effects of the present invention are:

[0032] The present invention is based on the personalized evaluation level of the attention state of electroencephalogram signals, and intelligently selects the optimal non-invasive transcranial electrical stimulation neuromodulation method according to the evaluation result to improve the attention state. The present invention realizes the hierarchical evaluation of the attention state, and improves the attention by using the neuromodulation method when the attention state is poor.

[0033] The present invention mainly solves the real-time monitoring of electroencephalogram signals for the attention state level evaluation algorithm, and uses the optimal mode of transcranial electrical stimulation according to its evaluation level result to improve the user's attention state, and can meet diverse needs according to the user's environmental (such as auditory, visual, tactile, etc.) perception requirements.

[0034] Compared with the method of analyzing the attention state based on behavioral performance, the present invention uses electroencephalogram (EEG) signals for analysis, which can achieve higher timeliness (the measurement time resolution of EEG can reach milliseconds), and can reflect the changes in brain activities in real time. At the same time, by directly measuring brain activities, it can accurately reflect the physiological basis of the attention state, avoid the interference and errors that may be introduced by behavioral performance, and EEG is very sensitive to minute changes in attention, and can detect subtle changes that cannot be captured by behavioral performance, making it have higher accuracy and sensitivity. The analysis of behavioral performance depends on external manifestations and is easily affected by factors such as the external environment and individual emotions, while EEG is not interfered by these factors and can provide more stable and reliable data. Most importantly, EEG can identify the hidden attention state that the user does not show in behavior, such as attention dispersion but no obvious changes in behavior. It can be seen that compared with the method of analyzing the attention state based on behavioral performance, using EEG physiological signal analysis has higher timeliness, accuracy and sensitivity, does not depend on external behavioral performance, can identify the hidden attention state, and has broad application prospects and practical advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart of an intelligent neural regulation method based on attention state provided by the present invention.

[0036] Figure 2 It is a block diagram of an intelligent neural regulation system based on attention state provided by the present invention.

[0037] Figure 3 It is a schematic diagram of the structural composition of a shallow convolutional neural network module.

[0038] Figure 4 It is a specific operation flowchart for evaluating the user's current attention state using a threshold and the softmax classification algorithm. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The present invention will be further described in detail below with reference to the accompanying drawings.

[0040] In the first aspect, the present invention provides an intelligent neural regulation method based on attention state. The present invention mainly includes two important parts:

[0041] (1) Personalized attention state level evaluation method

[0042] By collecting the electroencephalogram (EEG) signals of the user in real time and preprocessing them, performing spectral analysis on the EEG signals in each time window, and using the fast Fourier transform (FTT) to calculate the power values of each frequency band.

[0043] In the initial stage of the user using the device, EEG signals of the user in different attention states will be collected, and an attention state evaluation model will be established, including states of relaxation, mild concentration, and high concentration.

[0044] When the user uses the device, the system will collect and monitor EEG signals in real time and compare them with the attention state evaluation model, and use a threshold and softmax classification algorithm to evaluate the user's current attention state. The evaluation result will be fed back to the user in real time through the user interface. At the same time, the current attention state will be converted into appropriate instructions and an intelligent neuroregulation method will be used to adjust the user's attention state.

[0045] (2) Intelligent and personalized neuroregulation method

[0046] Transcranial electrical stimulation is selected in the neuroregulation method, that is, the cerebral cortex is stimulated by low-intensity current to affect nerve activities, the user's electroencephalogram activities are fed back in real time and their attention states are hierarchically distinguished. At the same time, based on the comparison of the differences in the attention state levels with the benchmark, the neuroregulation method is intelligently executed, especially the setting of the neuroregulation parameters of transcranial electrical stimulation, so as to help the user adjust the attention state.

[0047] According to the initialized user attention state setting, the neuroregulation parameters of transcranial electrical stimulation are intervened according to the personalized neuroregulation plan customized for the user, including setting the transcranial electrical stimulation current plan (such as tDCS, tRCS, and tACS, etc.) for perceptual requirements, current intensity, stimulation period, and stimulation duration, and the optimal stimulation is achieved through personalized customization. At the same time, the neuroregulation method takes into account the individual differences of the user and provides immediate intervention and adjustment, especially continuously optimizing and improving the attention state regulation effect according to the user feedback data. Among these transcranial electrical stimulation methods, a closed-loop regulation based on a human-intelligent neuroregulation system is formed through the evaluation of the attention state level.

[0048] The transcranial electrical stimulation (tES) neurofeedback regulation method has significant advantages compared to regulation methods such as audio-visual and other sensory cues. First, tES directly acts on the brain and regulates neuronal activities through microcurrents, thereby changing the physiological state of the brain. This direct regulation of brain activities is more effective than indirectly affecting the brain through sensory cues such as audio-visual. Second, tES can produce obvious effects in a short time, while methods such as audio-visual cues need to go through multiple steps such as sensory input and information processing. Since tES directly acts on neurons, it can change the attention state faster. The effect of tES can last for a long time. Even after the stimulation ends, the activities of neurons are still affected, while the effects of audio-visual cues are usually short-lived. Moreover, tES can precisely act on specific regions of the brain, such as the frontal lobe, parietal lobe, and other brain regions related to attention, so as to achieve more efficient regulation. Finally, the tES method does not require the user to actively participate or respond to sensory cues, reducing the occupation of the user's attention. Sensory stimuli such as audio-visual cues may be affected by external factors such as environmental noise and light. tES is carried out in a relatively quiet environment and is not affected by these interferences. Therefore, compared with sensory cues such as audio-visual, the transcranial electrical stimulation neurofeedback regulation method has many advantages such as the ability to more directly and effectively regulate brain nerve activities, faster immediate effects and longer duration, higher precision, and the need for no active participation and sensory cooperation of the user. This makes tES have significant advantages and application potential in other neuroregulations such as attention state regulation.

[0049] As Figure 1 shown, an intelligent neuroregulation method based on attention state provided by the present invention specifically includes the following steps:

[0050] (1) Physiological signal acquisition;

[0051] The EEG signals of the user are collected in real time through an EEG (electroencephalogram) sensor to capture brainwaves related to the attention state. For example, when there are visual and auditory demands, the EEG signals in the visual and auditory regions will be focused on.

[0052] (2) Signal processing and feature extraction;

[0053] (2.1) Preprocess the collected original EEG signals;

[0054] Among them, the preprocessing mainly includes filtering.

[0055] Among them, the methods of filtering mainly include: frequency filtering and principal component analysis. But it is not limited to this.

[0056] (2.2) Extract attention-related features from EEG signals; specifically: perform spectral analysis on the EEG signals within each time window, and calculate the power values of each frequency band using the Fast Fourier Transform (FFT).

[0057] Among them, the specific operation process of calculating the power values of each frequency band using the Fast Fourier Transform (FFT) is as follows:

[0058] By setting or adjusting the time period by the user, specifically, collect EEG signals with a time period of 10 - 20 minutes. After signal filtering, calculate the power values of each frequency band (including δ, θ, α, β, γ) of the EEG signals within this time window using the Fast Fourier Transform.

[0059] (3) Personalized model training;

[0060] At the initial stage when the user uses the device, collect the EEG signals of the user in different attention states for the usage task requirements (visual, auditory, tactile, etc.) to establish personalized baseline data; then use intelligent learning algorithms to train a personalized attention state assessment model based on the baseline data, including states of relaxation, mild concentration, and high concentration.

[0061] The attention state assessment model in the present invention mainly includes: a data preprocessing module and a shallow convolutional neural network module.

[0062] The data preprocessing module mainly realizes the arrangement and time-frequency conversion of EEG signal data. Specifically: First, the data preprocessing module cleans the data, removes redundant and missing data, and reduces the data operation burden on the model; then perform spectral analysis on the EEG signals within each time window, and calculate the power values of each frequency band using the Fast Fourier Transform; then use the Principal Component Analysis (PCA) method to reduce the number of features; finally, perform normalization processing on the data to accelerate the model convergence.

[0063] The shallow convolutional neural network module is used to classify and evaluate the preprocessed data. As Figure 3 shown, this shallow convolutional neural network module mainly includes: an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer. Among them, the input layer obtains the preprocessed signal according to the user's individual perception and the corresponding brain regions (visual corresponding to the occipital lobe, auditory corresponding to the temporal lobe, and tactile corresponding to the parietal lobe, etc.) or multi-sensory perception; the convolutional layer adopts a double-layer structure, extracts local features through two convolutional layers, and sets a max pooling layer after each convolutional layer to reduce the feature size; the fully connected layer further optimizes the result to reduce overfitting; the output layer uses the Softmax activation function to achieve the classification and evaluation of the attention state.

[0064] The environment in which the hardware of the attention state evaluation model in the present invention is used is based on a portable mobile hardware system, so a lightweight evaluation model is adopted. At the same time, the attention state evaluation model is based on a personalized evaluation scheme, which enables the attention state evaluation model not to process a large amount of data, but only individual attention state difference data, thus reducing the requirement for the computational complexity of the attention state evaluation model and maintaining the evaluation performance of the attention state evaluation model.

[0065] Among them, the specific training process of the attention state evaluation model is as follows:

[0066] The specific training process of the attention state evaluation model depends on the EEG signal data of different attention states of individual users for training. When the user first uses the system, they will be required to complete a specified number of different tasks, which can regulate the user's attention state in different situations. The EEG acquisition module obtains the EEG data set corresponding to the user's state, and the attention state evaluation model uses the data in these EEG data sets for training and verification to obtain a personalized and stable attention state evaluation model. Among them, the tasks completed by the user for the first time are the complex mathematical problem task of high cognitive load, the reading task of medium cognitive load, the image recognition task of low cognitive load, and the relaxation state of closing eyes and resting. These tasks are divided into four attention states: high attention state, medium attention state, low attention state, and attention dispersion state in sequence. The EEG signal data collected corresponding to different attention states will be used as the classification basis. The proportions of the training set and the test set in the four data sets of different attention states collected are 80% and 20% respectively, and the shallow convolutional neural network module is trained using the cross-validation method to further improve the generalization ability of the individual user data and reduce the deviation caused by data division.

[0067] (4) Real-time attention state evaluation;

[0068] When the user uses the device, the system will collect and monitor the EEG signal in real time, and at the same time use the trained attention state evaluation model to analyze and evaluate the EEG signal collected in real time to judge the user's current attention state. Specifically, the EEG signal collected in real time is compared with the trained attention state evaluation model, and the threshold and softmax classification algorithm are used to evaluate the user's current attention state.

[0069] Among them, the specific operation process of using the threshold and softmax classification algorithm to evaluate the user's current attention state is as follows:

[0070] In order to further improve the evaluation classification accuracy of the attention state and optimize its classification method, the present invention uses the threshold and softmax classification algorithm to evaluate the attention state, as Figure 4As shown. After the data of the model is processed by the fully connected layer, the optimized classification results are sent to the softmax output layer and the sigmoid output layer respectively. The category of the data is judged by probability and threshold respectively. The softmax output layer outputs the multi-class probability distribution, and then the category is obtained through Argmax, where the category with the maximum probability is taken, and then the category is predicted through softmax, and the A classification is output; the sigmoid output layer outputs the probability of each category, and then the category is judged through the threshold, greater than the threshold is 1, and finally the category is judged through the threshold and the B classification is output; finally, the above A classification and B classification are integrated to obtain the C classification, and the final classification result is output, and the classification accuracy is improved through the integration result.

[0071] (5) Neuroregulation strategy generation;

[0072] According to the real-time attention state evaluation result, it is judged whether neuroregulation needs to be performed. When neuroregulation needs to be performed, the system will generate a personalized neuroregulation strategy to select appropriate neuroregulation methods and neuroregulation parameters. Among them, the user's personalized neuroregulation strategy will be adaptively optimized for the neuroregulation strategy through EEG real-time monitoring and attention state evaluation results and transcranial electrical stimulation strategy during the previous several neuroregulation processes, and finally realize the personalized customization of the neuroregulation strategy for the user's attention state.

[0073] Among them, the specific screening process of the adaptive optimal neuroregulation strategy by EEG real-time monitoring and attention state evaluation results and transcranial electrical stimulation strategy is as follows:

[0074] In the present invention, the user state level is obtained after EEG real-time monitoring and attention state evaluation. For the evaluated high attention state, medium attention state, low attention state or attention dispersion state, transcranial electrical stimulation strategies with different parameters will be set. When in the high attention state, a Gaussian random current stimulation scheme in the range of 0.5-1 mA is used to maintain the attention state and optimize the allocation of attention resources; when in the medium attention state, a two-category mixed transcranial electrical stimulation scheme in the range of 0.5-1 mA is adopted, and the Gaussian random current stimulation scheme and the γ-wave (30-45 Hz) residual / sine alternating current stimulation scheme are respectively adopted; when in the low attention state, a two-category mixed transcranial electrical stimulation scheme in the range of 1.0-1.5 mA is adopted, and the Gaussian random current stimulation scheme and the β-wave (13-30 Hz) residual / sine alternating current stimulation scheme are respectively adopted; when in the attention dispersion state, a three-category mixed transcranial electrical stimulation scheme in the range of 1.5-2 mA is adopted, and the Gaussian random current stimulation scheme, the β-wave residual / sine alternating current stimulation scheme and the transcranial direct current stimulation scheme are respectively adopted.

[0075] After each stimulation cycle, the attention state will be evaluated again, and the transcranial electrical stimulation strategy will be optimized based on the evaluation threshold result. When the first evaluation result is a high attention state, real-time periodic monitoring and evaluation will be carried out, and Gaussian random current stimulation will be performed after 1, 3, 5... cycles in sequence. After the attention state level is evaluated again as a medium attention state, a low attention state, or a distracted attention state, the following corresponding transcranial electrical stimulation strategies will be executed: When the threshold after stimulation biases towards a high attention state, the previous transcranial electrical stimulation program will be maintained, and this transcranial electrical stimulation program is also the optimal neuromodulation program; when the threshold after stimulation biases towards remaining unchanged or decreasing, the parameters of the previous transcranial electrical stimulation program will be adjusted, while increasing the current by 0.1 mA and the frequency by 1 Hz, and adjusted once for each executed cycle until the optimal neuromodulation program is adjusted. During the regulation process of the transcranial electrical stimulation program, these two parameters (current and frequency) do not exceed the limit values of the corresponding transcranial electrical stimulation program. For example, the maximum stimulation current in the medium attention state is 1 mA, and the maximum stimulation frequency is 45 Hz. Only when the attention state changes (such as from a low attention state to a medium attention state) will the corresponding transcranial electrical stimulation program be adjusted.

[0076] Among them, the time proportion of the mixed transcranial electrical stimulation program also changes dynamically. The time proportions of the Gaussian random current stimulation program corresponding to the high, medium, low attention states, and the distracted attention state are 100%, 50%, 30%, and 10% in sequence. The time proportion of the transcranial direct current stimulation program in the distracted attention state is 30%, and the remaining time is for its corresponding residual / sine alternating current stimulation program. At the same time, the proportion of the stimulation time cycle will remain unchanged as the real-time monitoring threshold biases towards a high attention state, or the proportion of the time of the residual / sine alternating current stimulation program will increase when the threshold corresponds to an unchanged attention state or biases towards a low attention state to obtain the optimal neuromodulation program. When the threshold remains unchanged or decreases in the distracted attention state, the proportion of the time corresponding to the transcranial direct current stimulation program will be increased. Each time the above adjustment of the time proportion is adjusted in real-time by 5% of the time proportion.

[0077] (6) Execution of the neuromodulation strategy;

[0078] Activate the neural regulation module C to implement the neural regulation strategy. During the neural regulation process, the EEG signals of the user are monitored in real time, and the EEG acquisition module A is used to collect the (attention state analysis and rating) EEG signals in real time with a time window. When it is necessary to execute the neural regulation strategy for the user's attention state, in order to avoid the influence of transcranial electrical stimulation on the EEG signals, the EEG acquisition module A and the neural regulation module C alternate to start and stop the cyclic regulation scheme operation, that is, the EEG acquisition module A (attention state analysis and rating) - the neural regulation module C - the EEG acquisition module A (attention state analysis and rating). Only when the attention state rating of the EEG acquisition module A meets the requirements, will only the EEG acquisition module A (attention state analysis and rating) be executed. Once the attention state rating of the EEG acquisition module A is low, the system will execute the start and stop cyclic regulation scheme to adjust the user's attention state.

[0079] In a second aspect, the present invention provides an intelligent neural regulation system based on attention state. This system is mainly used to implement an intelligent neural regulation method based on attention state described in the first aspect of the present invention.

[0080] As Figure 2 shown, an intelligent neural regulation system based on attention state of the present invention mainly includes: an EEG acquisition module A, a main control module B, and a neural regulation module C. Among them, the user interacts with the EEG acquisition module A and the neural regulation module C respectively, and the main control module B interacts with the EEG acquisition module A and the neural regulation module C respectively.

[0081] Specifically, the main function and role of the EEG acquisition module A are: it can customize the acquisition of EEG signals in specific sensory (visual / auditory / tactile / olfactory / gustatory) brain regions.

[0082] Specifically, the main function and role of the main control module B are: 1. Evaluate the attention state level based on the collected EEG signals; 2. Send start / stop commands for EEG acquisition to the EEG acquisition module A; 3. Send the best regulation command parameters according to the attention state evaluation level.

[0083] Specifically, the main function and role of the neural regulation module C are: 1. Activate the stimulation method according to the received command; 2. Feedback the end command after the stimulation ends.

[0084] The specific working principle of an intelligent neural regulation system based on attention state of the present invention is as follows:

[0085] The main control module B sends a start instruction for EEG signal acquisition to the EEG acquisition module A. The EEG acquisition module A starts to acquire EEG signals and sends them to the main control module B. The main control module B performs signal processing and feature extraction, and at the same time conducts personalized model training and real-time attention state evaluation. According to the real-time attention state evaluation result, it judges whether it is necessary to perform neuromodulation. When neuromodulation is required, a personalized neuromodulation strategy will be generated in the neuromodulation module C. The main control module B sends a neuromodulation parameter instruction to the neuromodulation module C. The neuromodulation module C will execute the neuromodulation strategy according to the neuromodulation parameter instruction and feedback a regulation end signal to the main control module B.

[0086] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art of the present technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent neural regulation system based on attention state, which is used to implement an intelligent neural regulation method based on attention state, is characterized in that The system includes: an electroencephalogram (EEG) acquisition module, a main control module, and a neuromodulation module; the main control module sends a start instruction for EEG signal acquisition to the EEG acquisition module, the EEG acquisition module starts to acquire EEG signals and sends them to the main control module, the main control module performs signal processing and feature extraction, simultaneously conducts model training and real-time attention state evaluation, and determines whether neuromodulation needs to be performed according to the real-time attention state evaluation result; when neuromodulation needs to be performed, a neuromodulation strategy is generated in the neuromodulation module, the main control module sends a neuromodulation parameter instruction to the neuromodulation module, the neuromodulation module executes the neuromodulation strategy according to the neuromodulation parameter instruction, and feeds back a regulation end signal to the main control module; The intelligent neuromodulation method based on attention state includes the following steps: (1) Real-time collect the user's EEG signals and capture the brain waves related to the attention state; (2) Preprocess the collected original EEG signals and extract the attention-related features in the EEG signals; (3) At the initial stage when the user uses the device, collect the EEG signals of the user in different attention states according to the usage task requirements to establish baseline data; use an intelligent learning algorithm to train an attention state evaluation model based on the baseline data; The attention state evaluation model includes: a data preprocessing module and a shallow convolutional neural network module; The data preprocessing module is used to organize and perform time-frequency transformation on the EEG signal data: first, clean the data, remove redundant and missing data, and reduce the data operation burden on the model; then perform spectral analysis on the EEG signals within each time window, and calculate the power values of each frequency band using the fast Fourier transform; then use the principal component analysis method to reduce the number of features; finally, perform normalization processing on the data to accelerate the convergence of the model; The shallow convolutional neural network module is used to classify and evaluate the preprocessed data; the shallow convolutional neural network module includes: an input layer, two convolutional layers, two pooling layers, a fully connected layer, and an output layer; the input layer obtains the preprocessed signals according to the user's individual perception during use and the corresponding brain regions or multi-sensations; the convolutional layer adopts a double-layer structure, extracts local features through two convolutional layers, and sets a max pooling layer after each convolutional layer to reduce the feature size; the fully connected layer further optimizes the results to reduce overfitting; the output layer uses the Softmax activation function to achieve the classification and evaluation of the attention state; (4) When the user uses the device, the system will real-time collect and monitor the EEG signals, and at the same time use the trained attention state evaluation model to analyze and evaluate the real-time collected EEG signals to determine the user's current attention state; (5) Generate a neuromodulation strategy; According to the real-time attention state evaluation result, determine whether neuromodulation needs to be performed; when neuromodulation needs to be performed, the system will generate a neuromodulation strategy to select a suitable neuromodulation method and neuromodulation parameters; Adaptive optimal neuroregulation strategy screening is performed by using the results of real-time EEG monitoring and attention state assessment and transcranial electrical stimulation strategies. The specific operation process is as follows: After real-time EEG monitoring and attention state assessment, the user's state level is obtained. For the evaluated high attention state, medium attention state, low attention state, or attention dispersion state, transcranial electrical stimulation strategies with different parameters are set: When in the high attention state, a Gaussian random current stimulation scheme in the range of 0.5 - 1 mA is used to maintain the attention state and optimize the allocation of attention resources; when in the medium attention state, a two-category hybrid transcranial electrical stimulation scheme in the range of 0.5 - 1 mA is adopted, using the Gaussian random current stimulation scheme and the gamma wave (30 - 45 Hz) cosine / sine alternating current stimulation scheme respectively; when in the low attention state, a two-category hybrid transcranial electrical stimulation scheme in the range of 1.0 - 1.5 mA is adopted, using the Gaussian random current stimulation scheme and the beta wave (13 - 30 Hz) cosine / sine alternating current stimulation scheme respectively; when in the attention dispersion state, a three-category hybrid transcranial electrical stimulation scheme in the range of 1.5 - 2 mA is adopted, using the Gaussian random current stimulation scheme, the beta wave cosine / sine alternating current stimulation scheme, and the transcranial direct current stimulation scheme respectively; After each stimulation cycle, the attention state is evaluated again, and the transcranial electrical stimulation strategy is optimized based on the evaluation threshold result; when the first evaluation result is the high attention state, real-time periodic monitoring and evaluation are performed, and Gaussian random current stimulation is executed after 1, 3, 5... cycles in sequence; after the attention state level is evaluated again as the medium attention state, low attention state, or attention dispersion state, the following corresponding transcranial electrical stimulation strategies are executed: When the threshold after stimulation tends to the high attention state, the previous transcranial electrical stimulation scheme is maintained, and this transcranial electrical stimulation scheme is also the optimal neuroregulation scheme; when the threshold after stimulation tends to remain unchanged or decrease, the parameters of the previous transcranial electrical stimulation scheme are adjusted, while increasing the current by 0.1 mA and the frequency by 1 Hz, and it is adjusted once per execution of a cycle until the optimal neuroregulation scheme is adjusted; The time ratios corresponding to the Gaussian random current stimulation scheme in the high, medium, low attention states, and attention dispersion state are 100%, 50%, 30%, and 10% in sequence; the time ratio corresponding to the transcranial direct current stimulation scheme in the attention dispersion state is 30%, and the remaining time is the corresponding cosine / sine alternating current stimulation scheme; at the same time, the stimulation time period ratio will also remain unchanged as the real-time monitoring threshold tends to the high attention state, or the ratio of the cosine / sine alternating current stimulation scheme time will increase when the threshold corresponds to an unchanged attention state or tends to the low attention state to obtain the optimal neuroregulation scheme; and when the threshold remains unchanged or decreases in the attention dispersion state, the time ratio corresponding to the transcranial direct current stimulation scheme is increased; (6) Execute the neuroregulation strategy.

2. The intelligent neural regulation system based on attention state according to claim 1, wherein In step (2), the collected original EEG signals are filtered.

3. An intelligent nerve regulation system based on an attention state according to claim 1, characterized in that, In step (2), the EEG signals in each time window are subjected to spectral analysis, and the power values of each frequency band are calculated using the fast Fourier transform.

4. An intelligent nerve regulation system based on an attention state according to claim 1, characterized in that, In step (3), the specific training process of the attention state evaluation model is as follows: The training process of the attention state evaluation model depends on the EEG signal data of different attention states of the user for training; the tasks completed by the user for the first time are the complex math problem task of high cognitive load, the reading task of medium cognitive load, the image recognition task of low cognitive load, and the relaxation state of closing eyes and resting. These tasks are sequentially divided into high attention state, medium attention state, low attention state, and attention dispersion state. The EEG signal data collected under different attention states is used as the classification basis. Among them, the proportions of the training set and the test set are 80% and 20% respectively, and the shallow convolutional neural network module is trained by using the cross-validation method.

5. An intelligent neural regulation system based on attention state according to claim 1, characterized in that, In step (4), the threshold and the softmax classification algorithm are used to evaluate the user's current attention state, and the specific operation process is as follows: After the data is processed by the fully connected layer, the optimized classification results are respectively sent to the softmax output layer and the sigmoid output layer, and the categories of the data are judged by probability and threshold respectively. The softmax output layer outputs the multi-class probability distribution, and then the category is obtained through Argmax, where the category with the maximum probability is taken, and then the category is predicted through softmax, and the A classification is output; the sigmoid output layer outputs the probability of each category, and then the category is judged through the threshold, greater than the threshold is 1, and finally the category is judged through the threshold and the B classification is output; finally, the A classification and the B classification are integrated to obtain the C classification and the final classification result is output.

6. The intelligent neural regulation system based on the attention state according to claim 1, wherein When it is necessary to execute the neuromodulation strategy of the user's attention state, the EEG acquisition module and the neuromodulation module alternately start and stop the operation of the loop modulation scheme.

Images