Hidden multi-mode brain state real-time monitoring and closed-loop intervention system

Abstract

The invention discloses a hidden multi-mode brain state real-time monitoring and closed-loop intervention system, and relates to the technical field of intelligent wearable equipment. The system comprises a cap body, and further comprises a multi-point electrode array used for collecting EEG (electroencephalogram) signals, EOG (eye movement) signals and EMG (electromyography) signals of a human body; the brain blood oxygen monitoring module is used for monitoring brain blood oxygen saturation; the signal receiving and processing module is used for preprocessing and analyzing data acquired by the multi-point electrode array and the brain blood oxygen monitoring module; the multi-mode stimulation module is used for implementing sensory and nerve stimulation on the user according to a judgment result of the state judgment module so as to intervene in a drowsiness or attention decline state; and the power supply module and the communication unit are used for providing a working power supply for the modules needing power supply in the system and carrying out data communication with external terminal equipment. The system can monitor the waking degree of the user with high precision and carry out closed-loop feedback intervention by various stimulation means.

Description

Technical Field

[0001] The present invention relates to the technical field of smart wearable devices, and in particular to a concealed multimodal real-time brain state monitoring and closed-loop intervention system. Background Art

[0002] In many modern scenarios, such as long-distance driving, shift work, classroom learning, and high-risk work environments, driver alertness and attention are crucial to safety and efficiency. For example, long-distance passenger and freight drivers often experience drowsiness, which leads to decreased alertness and cognitive impairment, increasing the risk of accidents. Therefore, real-time monitoring of driver drowsiness and timely intervention are of great practical significance.

[0003] There are many technologies that have been used to detect sleepiness, but each has its limitations. They include:

[0004] (1) Single-signal detection methods are susceptible to interference: Traditional methods include detection based on external behavior, such as using a camera to capture eye closure rates or changes in facial expressions; and detection based on physiological signals, such as pure electroencephalogram (EEG) signal analysis, eye movement / electro-oculogram (EOG) signal analysis, electromyography (EMG) signal analysis, or heart rate signal analysis. Among these methods, single-modality methods are susceptible to interference or have limited applicability. For example, relying solely on video surveillance is not effective in dark conditions or when the driver is wearing glasses; relying solely on EEG signals is effective in laboratory environments, but in daily environments, the signals are easily affected by noise, making it difficult to obtain a stable indicator of wakefulness.

[0005] (2) Multimodal methods require complicated equipment and are inconvenient to wear: Although some multimodal combined detection methods improve the accuracy of discrimination, they usually require complex equipment (such as EEG caps with multiple wires, electrooculogram electrode patches, cameras, etc.), which are not comfortable to wear and have poor continuous use experience. In addition, they lack concealment, making them difficult for drivers, especially long-distance drivers, to accept and difficult to be truly applied.

[0006] (3) Lack of active and effective intervention methods: In addition, existing technologies mainly focus on state detection, but lack active intervention in the drowsy state. For example, existing driving drowsiness warning devices often use sound alarms or vibration reminders to remind the driver when drowsiness is detected. However, these reminders are often short-lived and require people to take action on their own to correct drowsiness. Some devices try to refresh the mind through single stimulation such as playing music or vibrating the earlobes, but the effect is limited for those with severe drowsiness. In addition, brain stimulation technologies such as transcranial electrical stimulation or low-frequency electromagnetic stimulation have been shown in clinical studies to have a certain effect on improving brain wakefulness, but traditional equipment is bulky and complex to operate, and cannot be used for operations during driving.

[0007] In summary, existing technologies have significant shortcomings in terms of convenience, accuracy, real-time performance, and proactive intervention. A system that integrates multiple physiological signal monitoring and multimodal stimulation intervention, creating a discreet system that allows people to wear it for extended periods of time without resistance while driving, while providing a timely alarm when drowsiness is detected and automatically applying appropriate combined stimulation for closed-loop control, is an important issue that needs to be addressed urgently. Summary of the Invention

[0008] The technical problem to be solved by the present invention is how to provide a concealed multimodal brain state real-time monitoring and closed-loop intervention system that can monitor the user's alertness with high precision and perform closed-loop feedback intervention using multiple stimulation means.

[0009] To solve the above technical problems, the technical solution adopted by the present invention is: a concealed multimodal real-time brain state monitoring and closed-loop intervention system, including a cap body, and also including:

[0010] A multi-point electrode array is embedded in the cap to collect EEG, EOG, and EMG signals from the human body.

[0011] The cerebral blood oxygen monitoring module is located in the cap and is used to monitor the cerebral blood oxygen saturation SpO2;

[0012] The signal receiving and processing module is used to pre-process and analyze the data collected by the multi-point electrode array and the cerebral blood oxygen monitoring module, and perform artifact removal, feature extraction, and data compression on the data. It is also used to classify and judge the human brain state based on the extracted feature set using an AI model. If the user is determined to be drowsy or inattentive, the intervention process will be initiated; if the user is determined to be awake and normal, the system will maintain monitoring and will not trigger stimulation;

[0013] A multimodal stimulation module is used to implement sensory and neural stimulation on the user based on the judgment results of the state discrimination module to intervene in the state of drowsiness or decreased attention;

[0014] The power supply module and the communication unit are used to provide working power to the modules in the system that need power, transmit the data processed by the system to the external terminal device and receive the control commands transmitted by the external terminal device.

[0015] The beneficial effects of adopting the above technical solution are: the system simultaneously collects multi-source signals such as EEG, EOG, EMG, and cerebral blood oxygen saturation, achieving higher detection accuracy and stability compared to single-signal detection. The complementary fusion of information from each modality enables the system to more comprehensively identify the physiological characteristics of drowsiness. For example, it combines the enhanced mid- and low-frequency EEG with signs such as frequent blinking in EOG and decreased muscle tone in EMG to improve the accuracy of drowsiness detection and reduce false positives and missed detections. At the same time, through signal separation and feature extraction algorithms, the system effectively overcomes the noise interference problem in daily environments and can extract stable indicators of wakefulness from noisy backgrounds.

[0016] Compared with the passive method or single stimulation method that only prompts the alarm, the present invention provides an active intervention mechanism with synchronous or combined output of electrical stimulation, low-frequency electromagnetic and sound stimulation. When it is detected that the user's attention is declining, the system automatically applies covert multimodal stimulation to quickly restore the user to consciousness. Multiple stimulation methods can work synergistically. Especially for those who are severely sleepy, transcranial electrical stimulation and low-frequency electromagnetic stimulation can directly mobilize brain neural activity to make up for the short-term deficiency of simple sound reminders. At the same time, closed-loop feedback control ensures timely adjustment and cessation of stimulation: the stimulation is stopped immediately after the user wakes up to avoid excessive intervention, and the stimulation intensity is increased when the sleepiness persists. Therefore, the intervention is both efficient and intelligently adaptive. Actual tests show that compared with the intervention of electrical stimulation, the combined sound stimulation can instantly improve the subjective wakefulness score of the subjects, and can more effectively prevent the risk of accidents caused by sleepiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Figure 1 is a principle block diagram of the system according to an embodiment of the present invention;

[0019] Figure 2 2 is a schematic structural diagram of a cap body in a system according to an embodiment of the present invention;

[0020] Figure 3 2 is a schematic structural diagram of a cap body in a system according to an embodiment of the present invention;

[0021] Figure 4 2 is a schematic structural diagram of a cap body in a system according to an embodiment of the present invention;

[0022] Figure 5 Schematic diagram of the autoencoder neural network structure in the system according to an embodiment of the present invention;

[0023] Figure 6 is a processing flow chart of the system according to an embodiment of the present invention;

[0024] Among them: 1. Cap body; 2. Multi-point electrode array; 3. Brain blood oxygen monitoring module; 4. Signal receiving and processing module; 5. Sound stimulator; 6. Electrical stimulation electrode; 7. Electromagnetic stimulation coil. DETAILED DESCRIPTION

[0025] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] Overall, such as Figure 1 As shown, the embodiment of the present invention discloses a concealed multimodal real-time brain state monitoring and closed-loop intervention system, including a cap body 1, the specific structure of the cap body is as follows Figure 2-Figure 4 As shown, the system further includes:

[0028] The multi-point electrode array 2 is embedded in the cap body 1 and is used to collect EEG signals, EOG signals and EMG signals of the human body;

[0029] The cerebral blood oxygen monitoring module 3 is located in the cap body 1 and is used to monitor the cerebral blood oxygen saturation SpO2;

[0030] The signal receiving and processing module 4 is used to pre-process and analyze the data collected by the multi-point electrode array 2 and the cerebral blood oxygen monitoring module 3, and perform artifact removal, feature extraction, and data compression on the data. It is also used to classify and judge the human brain state using an AI model based on the extracted feature set. When the user is determined to be drowsy or inattentive, the intervention process is initiated; if the user is determined to be awake and normal, the system maintains monitoring and does not trigger stimulation;

[0031] A multimodal stimulation module is used to implement sensory and neural stimulation on the user based on the judgment results of the state discrimination module to intervene in the state of drowsiness or decreased attention;

[0032] The power supply module and the communication unit are used to provide working power to the modules in the system that need power, transmit the data processed by the system to the external terminal device and receive the control commands transmitted by the external terminal device.

[0033] The following describes in detail the various modules, functions and implementation methods of the system described in this application in combination with the specific content:

[0034] Multi-point Electrode Array 2: The electrode array consists of non-invasive dry electrodes embedded within the cap, covering the frontal, temporal, and occipital regions. It collects EEG, EOG, and EMG signals and provides a stable differential reference. The electrode array is arranged to cover multiple regions of the user's head, such as the forehead (frontal lobe region), the temples (temporal lobe region), and the occipital region of the back of the head, to collect physiological signals from corresponding brain regions. Different electrodes within the electrode array can be used to collect EEG, EOG, and EMG signals. For example, electrodes located on the forehead not only capture frontal EEG activity but also sense potential changes caused by eye movements (i.e., EOG signals); electrodes located on the occipital region capture EEG signals from the back of the head; and electrodes distributed near the brim of the cap sense EMG signals caused by facial expression muscles or neck muscle activity. The electrode array also includes a reference electrode to provide a stable differential measurement reference. This multi-point arrangement enables simultaneous acquisition of multi-channel physiological data, including EEG, EOG, and EMG.

[0035] Cerebral blood oxygen monitoring module 3: An optical sensor for monitoring brain blood oxygen saturation is also integrated on the front inner side of the cap. The sensor includes a near-infrared light emitter and receiver pair. Through the principle of near-infrared light transmission and reflection measurement (near-infrared spectroscopy), changes in blood oxygen saturation SpO2 in the superficial area of ​​the cerebral cortex are detected in real time. This module provides brain hemodynamic information and can be used as a supplement to electrical signals such as EEG to assess brain sleepiness (for example, when local brain activity decreases, changes in blood oxygen supply can be captured).

[0036] Signal receiving and processing module: This module includes a physiological signal analog front-end circuit, an analog-to-digital converter (ADC), and an edge AI processor NMA-1. The weak analog signals collected by each electrode are first amplified and filtered by a low-noise analog front-end, converted into digital signals by the analog-to-digital converter ADC, and then submitted to the NMA-1 for processing. The analog front-end circuit can use a dedicated amplifier chip for biological signals to provide appropriate gain (such as approximately 1000 times) for each electrode signal and a 0.5Hz-45Hz bandpass filter to remove DC drift and high-frequency noise, so that signals such as EEG, EOG, and EMG remain within the effective analysis bandwidth.

[0037] The analog-to-digital converter (ADC) then synchronously samples the multi-channel analog signals at a sampling rate of, for example, 500 Hz, generating digital signals. These digital signals enter the edge AI processor NMA-1, which performs the intelligent preprocessing and analysis described in this invention. The NMA-1 runs the locally deployed Neo-St model, based on the Xet architecture, and implements various artifact removal (such as blinks, electromyography, and line noise), feature extraction, and data compression.

[0038] Figure 5 This diagram illustrates the autoencoder architecture used in this invention, comprising the encoder, latent layer, and decoder. The encoder network transforms the raw input signal into a low-dimensional latent representation h; the decoder network reconstructs an output X' from h that approximates the input. By limiting the dimensionality of the latent layer, the network is forced to learn the key characteristic structures of the input signal and filter out noise and artifacts. Using this system, we achieved power consumption of 0.2 watts and inference time of 9 milliseconds when processing 1-second EEG footage, significantly improving the performance of artifact-resistant, real-time portable EEG monitoring.

[0039] Front-end feature extraction and state discrimination: The edge AI processor NMA-1 uses an AI model to classify and determine brain state based on the extracted feature set. Preferably, the initial model is obtained through transfer learning: a general wakefulness discrimination model is trained using a large amount of existing subject data and then transferred to the current user. The processor, combined with the user's initial calibration data, adjusts the model parameters online (e.g., using incremental learning or dynamic threshold adjustment) to form a personalized model. This model can use classifiers such as deep neural networks, support vector machines, or decision trees to integrate latent multimodal features for comprehensive judgment. For example, the autoencoder latent layer output can be input into a fully connected neural network along with the explicitly calculated features described above to output the user's current brain function state level—for example, expressed as alert, mildly sleepy, or severely sleepy—or a continuous wakefulness score. The system can also incorporate interpretability mechanisms to analyze the basis for the model's decisions, for example, identifying "increased blinking frequency and increased theta wave power" as the primary trigger for drowsiness judgment and providing this information as a note for the user's reference.

[0040] In the input vector of this state discrimination model, the autoencoder's latent features h and the explicit features of each modality together constitute a complete feature set. Among them, EEG-related features often have a greater weight in discrimination, as changes in EEG frequency bands directly reflect the degree of brain wakefulness. EOG features (such as blink frequency and duration of eye closure) are extremely sensitive to sleepiness and can provide immediate warning signals. EMG features (such as fluctuations in facial EMG amplitude during yawning) serve as auxiliary tools to further improve the reliability of judgment. SpO2 changes are generally slow, but combined with the aforementioned fast electrical signal features, they can capture long-term fatigue trends. The latent vector h output by the autoencoder plays a role in fusing information from various modalities in the model. It includes comprehensive patterns such as the EEG principal component and therefore has a high feature contribution in the classification network. During training, the model automatically adjusts the weight parameters of each input feature to optimize the performance of distinguishing between wakefulness and sleepiness.

[0041] Threshold determination and trigger conditions: When the model determines that the user is drowsy or inattentive (exceeding a preset threshold), intervention begins. If the user is deemed alert and alert, the system maintains monitoring without triggering stimulation. The threshold can be dynamically adjusted based on user needs and adaptively optimized based on individual historical data to reduce false positives and false negatives.

[0042] Hidden multimodal stimulation module and stimulation strategy: The multimodal stimulation module includes a variety of stimulation devices for implementing sensory and neural stimulation to the user according to the instructions of the processor to intervene in the drowsy state. Specifically, it includes: an acoustic stimulator 5, an electrical stimulation electrode 6, and an electromagnetic stimulation coil 7 integrated into the inside of the hat. The acoustic stimulator is, for example, a miniature bone conduction earphone, which is hidden and installed on the edge of the brim of the hat (can be close to the temples on both sides to be close to the ears), and is used to play a reminder sound or a sound pulse of a specific frequency to stimulate the auditory system. Electrical stimulation is applied through two pairs of electrodes on the bilateral frontal lobes, giving a microcurrent of no more than 2mA to stimulate the brain, which has the effect of awakening the cerebral cortex. The electromagnetic stimulation coil refers to a small planar coil embedded in the brim of the hat. When a current pulse passes through, a transient magnetic field is generated around the coil, thereby performing low-frequency electromagnetic induction stimulation on specific areas of the brain; its position and orientation are designed, preferably located in the center of the forehead brim, adjacent to the frontal brain area (such as the prefrontal cortex), to effectively stimulate the brain area related to wakefulness.

[0043] Once the intervention condition is triggered, the processor generates a stimulation control instruction and implements wake-up stimulation to the user through the stimulation output module. The stimulation strategy can be a pre-set combination scheme, or it can be adaptively adjusted according to real-time feedback. The preferred strategies include: first applying sound stimulation (playing prompt sounds or sound pulses of a specific frequency through bone conduction headphones to stimulate hearing and increase alertness), followed by transcranial electrical stimulation (applying low-intensity DC or AC current to the forehead for several seconds through electrodes selected in the cap body to increase the excitability of the cerebral cortex), and if necessary, combined with low-frequency electromagnetic stimulation (driving the brim coil to generate short magnetic field pulses to stimulate the brain).

[0044] The above stimulations can be applied gradually in sequence, forming a graded intervention mechanism: for example, the first level is sound stimulation, the second level adds mild electrical stimulation, and the third level superimposes low-frequency electromagnetics to gradually increase the intensity of intervention. Each level of stimulation lasts for several seconds and the effect is observed. If the user's wakefulness index recovers after a certain level of intervention, further stimulation is stopped; if there is still no improvement, the next level of stimulation is entered. When there is still no obvious effect after the highest level of intervention, the system may determine that the user may be in a state of extreme sleepiness and trigger an emergency warning (such as notifying the user through the APP that they need to rest immediately or call others to pay attention). During the entire intervention process, the system keeps monitoring the user's physiological reactions every second. Once it detects that the user's condition has improved to a safe range, the stimulation can be stopped in time to avoid excessive intervention. All intervention triggering and response data will be recorded for future optimization of stimulation strategies and model parameters.

[0045] Power module and communication unit: The system is powered by a portable power supply, such as a rechargeable lithium battery embedded in the back of the cap, which provides the required DC power for the signal acquisition and stimulation units. The cap is equipped with a power management circuit and a switch to ensure power safety and battery life. Preferably, the system also includes a wireless communication unit (such as a Bluetooth module), and the processor can send the collected physiological data and judgment results to an external terminal device (such as a smartphone APP or computer) for display, and can accept user-set parameters or mode instructions, thereby realizing human-computer interaction and remote data recording and analysis.

[0046] like Figure 6 As shown in the process, this system uses a four-modal combination of EEG + EOG + EMG + brain blood oxygen SpO2, and cleverly embeds each module into the hat. The four-modal combination facilitates the subsequent algorithm to extract richer physiological information, which can significantly improve the accuracy of sleepiness prediction. The peaked cap is different from the previous helmet-style, headband-style, or head-wrapped EEG recording caps. It is a common form in daily life, which can minimize the psychological burden when wearing it, increase comfort, and facilitate promotion and use in driving scenarios.

[0047] This application has the following main advantages:

[0048] Concealment and comfort: The present invention uses a duckbill cap to cleverly and conceal a variety of sensors and stimulators into the cap body, achieving a fusion of functionality and everyday wear. Compared to traditional EEG devices that require wearing cumbersome electrode caps or ear-clip vibration devices, this system looks similar to an ordinary hat, does not cause a sense of shyness, reduces the psychological burden on the wearer, and is suitable for drivers, especially long-distance drivers, for continuous long-term use. This design improves user acceptance and makes real-time monitoring and intervention possible in driving scenarios, while overcoming the limitations of existing technologies such as bulky wear and conspicuous appearance.

[0049] High-precision monitoring through covert multimodal fusion: This system simultaneously collects multiple signals, including EEG, EOG, EMG, and cerebral blood oxygen saturation, achieving higher detection accuracy and stability than single-signal detection. The complementary integration of information from each modality enables the system to more comprehensively identify physiological characteristics of drowsiness. For example, it combines enhanced mid- and low-frequency EEG with frequent blinking in EOG and decreased muscle tone in EMG, thereby improving the accuracy of drowsiness detection and reducing false positives and negatives. Based on EEG datasets and actual sleepiness tests, the system achieved a sensitivity and specificity of over 85% for drowsiness detection. Furthermore, through signal separation and feature extraction algorithms, the system effectively overcomes the challenge of noise interference in everyday environments, capable of extracting stable indicators of alertness from noisy backgrounds. For example, the NMA-1 artifact removal significantly improves the stability and reliability of EEG, compensating for the poor accuracy of existing devices in the presence of artifacts.

[0050] High efficiency of synchronous multi-stimulation closed-loop intervention: Compared with passive methods or single stimulation methods that only prompt an alarm, the present invention provides an active intervention mechanism with synchronous or combined output of electrical stimulation, low-frequency electromagnetic and sound stimulation. When it is detected that the user's attention is declining, the system automatically applies covert multimodal stimulation to quickly restore the user to consciousness. Multiple stimulation methods can work synergistically. Especially for those with severe sleepiness, transcranial electrical stimulation and low-frequency electromagnetic stimulation can directly mobilize brain neural activity, making up for the short-term deficiency of simple sound reminders. At the same time, closed-loop feedback control ensures timely adjustment and cessation of stimulation: stimulation is stopped immediately after the user wakes up to avoid excessive intervention, and the stimulation intensity is increased when the user continues to be sleepy. Therefore, the intervention is both efficient and intelligently adaptive. Actual tests show that compared with the intervention of electrical stimulation, the combined sound stimulation can instantly improve the subjective wakefulness score of the subject (for example, from very sleepy to mild sleepiness), which can more effectively prevent the risk of accidents caused by sleepiness.

[0051] Safety, reliability and scalability: All stimulation methods used in the present invention work within the safety threshold, and the sound pressure level, electrical stimulation current intensity, and electromagnetic field energy are strictly controlled to ensure no harm to the human body and no long-term side effects. At the same time, due to the modularization of the system architecture, other physiological parameter sensors (such as skin electricity, heart rate, etc.) or other forms of stimulation (such as tactile vibration) can be expanded in the future to further enhance the system function. Therefore, the present invention has good expansion and upgrading capabilities, and can be widely used in many fields such as long-distance driver monitoring, student learning status supervision, factory operator fatigue warning, and medical rehabilitation, and has significant social and economic benefits.

[0052] In summary, the system of the present invention addresses the shortcomings of existing devices in terms of artifact removal, real-time data processing, and covert stimulation, and proposes a comprehensive improvement plan. All performance indicators are superior to existing technical means, achieving a major breakthrough in the field of driver drowsiness monitoring and intervention.

Claims

1. A concealed multimodal real-time brain state monitoring and closed-loop intervention system, comprising a cap body (1), characterized in that: Also includes: A multi-point electrode array (2) is embedded in the cap body (1) and is used to collect EEG signals, EOG signals and EMG signals of the human body; A cerebral blood oxygen monitoring module (3), located in the cap body (1), is used to monitor cerebral blood oxygen saturation SpO2; The signal receiving and processing module (4) is used to pre-process and analyze the data collected by the multi-point electrode array (2) and the brain blood oxygen monitoring module (3), and to remove artifacts, extract features and compress the data; and to classify and judge the state of the human brain using an AI model based on the extracted feature set, and to enter an intervention process when it is determined that the user is in a state of drowsiness or decreased attention; If the user is determined to be awake and normal, the system will continue monitoring and will not trigger stimulation; A multimodal stimulation module is used to implement sensory and neural stimulation on the user based on the judgment results of the state discrimination module to intervene in the state of drowsiness or decreased attention; The power supply module and the communication unit are used to provide working power to the modules in the system that need power, transmit the data processed by the system to the external terminal device and receive the control commands transmitted by the external terminal device.

2. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 1, characterized in that: The multi-point electrode array (2) is embedded in the cap (1) in the form of non-invasive dry electrodes, covering the frontal lobe, temporal lobe and occipital lobe areas.

3. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 1, characterized in that: The cerebral blood oxygen monitoring module (3) comprises a near-infrared light transmitter and a receiver, and detects changes in blood oxygen saturation in the superficial area of ​​the cerebral cortex in real time through the principle of near-infrared light transmission and reflection measurement.

4. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 1, characterized in that: The signal receiving and processing module includes a physiological signal analog front-end processing circuit, an analog-to-digital converter, and an edge AI processor NMA-1.

5. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 4, characterized in that: The physiological signal analog front-end processing circuit uses a dedicated biological signal amplifier chip to provide gain for each electrode signal and has a 0.5Hz-45Hz bandpass filter to remove DC drift and high-frequency noise, so that the EEG, EOG, and EMG signals remain within the effective analysis bandwidth; then, the analog-to-digital converter synchronously samples the multi-channel analog signals at a certain sampling rate to generate digital signals; finally, the digital signals enter the edge AI processor NMA-1 for intelligent preprocessing and analysis; the edge AI processor NMA-1 runs the locally deployed AI model and, based on the Xet architecture, realizes the removal of various artifacts, feature extraction, and data compression.

6. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 5, characterized in that: The AI ​​model adopts an autoencoder structure, including an encoder, a latent layer and a decoder. The encoder network is used to transform the original input signal into a low-dimensional latent representation h; the decoder network is used to reconstruct the output X' close to the input from h; by limiting the dimension of the latent layer, the network is forced to learn the key feature structure of the input signal and filter out noise and artifacts.

7. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 1, characterized in that: The initial AI model is obtained through transfer learning. A general AI model for sobriety discrimination is trained using a large amount of existing test data, and then migrated and applied to the current user. The edge AI processor combines the user's individual initial calibration data to adjust the AI ​​model parameters online to form an optimized AI model.

8. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 7, characterized in that: In the input vector of the AI ​​model, the latent feature h of the autoencoder and the explicit features of each modality together constitute a complete feature set; among them, EEG-related features account for a larger weight in the discrimination; EOG features are extremely sensitive to drowsiness and are used to provide immediate warning signals; EMG features serve as an auxiliary to further improve the reliability of judgment; blood oxygen saturation SpO2 changes slowly and is used to combine EEG, EOG and EMG electrical signal features to capture long-term fatigue trends; the latent vector h output by the autoencoder plays the role of fusing information from various modalities in the model. The AI ​​model automatically adjusts the weight parameters of each input feature through training to optimize the performance of distinguishing between wakefulness and drowsiness.

9. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 1, characterized in that: The multimodal stimulation module comprises an acoustic stimulator (5), an electrical stimulation electrode (6) and an electromagnetic stimulation coil (7) integrated into the interior of the cap; the acoustic stimulator (5) comprises a miniature bone conduction earphone, which is concealed and installed on the edge of the cap brim and is used to play a reminder sound or a sound pulse of a specific frequency to stimulate the auditory system; the electrical stimulation is applied through two pairs of electrodes on the bilateral prefrontal lobes, giving a microcurrent of no more than 2 mA to stimulate the brain, thereby awakening the cerebral cortex; the electromagnetic stimulation coil (7) comprises a planar coil embedded in the cap brim, which generates a transient magnetic field around the coil when a current pulse passes through, thereby performing low-frequency electromagnetic induction stimulation on a specific area of ​​the brain.

10. The covert multimodal real-time brain status monitoring and closed-loop intervention system according to claim 9, characterized in that: The stimulation strategies applied by the multimodal stimulation module include: First, sound stimulation is applied, playing prompt sounds or sound pulses of a specific frequency through bone conduction headphones to stimulate hearing and increase alertness. Then, transcranial electrical stimulation is performed, applying low-intensity direct or alternating current to the forehead for several seconds through electrodes selected in the cap body to increase the excitability of the cerebral cortex. Finally, combined with low-frequency electromagnetic stimulation, the cap brim coil is driven to generate brief magnetic field pulses to stimulate the brain. The above-mentioned stimulations are applied gradually in a sequential order to form a graded intervention mechanism. If the user's alertness index recovers after a certain level of intervention, further stimulation will be stopped; if there is still no improvement, the next level of stimulation will be entered; when there is still no obvious effect after the highest level of intervention, the system may determine that the user may be in a state of extreme sleepiness and trigger an emergency warning; during the entire intervention process, the system keeps monitoring the user's physiological reactions every second. Once it detects that the user's condition has improved to a safe range, the stimulation will be stopped to avoid excessive intervention; all intervention trigger and response data will be recorded for future optimization of stimulation strategies and model parameters.

Images