Non-contact electromagnetic brain wave physiotherapy method and system

Abstract

The application relates to the technical field of physiotherapy, in particular to a non-contact electromagnetic physiotherapy method and system for brain waves. The non-contact electromagnetic physiotherapy system for brain waves comprises an electroencephalogram signal acquisition module, an electroencephalogram signal feature data construction module and a pulse electromagnetic field physiotherapy module. The electroencephalogram signal data is mapped to a pulse phase time axis, and the electroencephalogram signal data sequence is corrected in combination with the repetition component constraint of the pulse phase label, so that the periodic interference component introduced in the process of the action of the pulse electromagnetic field can be effectively inhibited, the influence on the electroencephalogram acquisition result and the frequency band energy calculation is reduced, and the accuracy and stability of the electroencephalogram signal feature data extraction are improved. Based on the corrected electroencephalogram signal feature data, a reinforcement learning method is used to construct a driving parameter set of an electromagnetic field construction unit, and a reward value is calculated in combination with a standard curve of brain wave sleep to update a driving parameter construction network.

Description

Technical Field

[0001] This invention relates to the field of physiotherapy technology, specifically to a non-contact electroencephalogram (EEG) electromagnetic physiotherapy method and system. Background Technology

[0002] In relevant physical therapy programs, some technologies act on the user's head area through electrical stimulation, magnetic stimulation, or other neuromodulation methods to influence brain electrical activity and improve sleep. However, many existing technologies require contact structures that directly contact the user's scalp, such as adhesive electrodes, conductive dielectric electrodes, or stimulation components that are tightly attached to the scalp. These methods typically present the following problems in practical use: First, direct contact structures can easily cause pressure, traction, or irritation to the user's scalp, resulting in poor comfort during prolonged wear, especially unsuitable for continuous use during sleep. Second, contact structures often require a relatively stable attachment or conductive conditions, making them susceptible to factors such as hair obstruction, skin condition, changes in body position, and nighttime movement, thus affecting the actual therapeutic effect and user experience. Summary of the Invention

[0003] This invention maps EEG signal data to a pulse phase time axis and corrects the EEG signal data sequence by combining the repetition component constraint of the pulse phase label. This effectively suppresses the periodic interference components introduced during the pulse electromagnetic field action, reduces the impact on EEG acquisition results and frequency band energy calculation, and improves the accuracy and stability of EEG signal feature data extraction. Based on the corrected EEG signal feature data, a set of driving parameters for electromagnetic field construction units is constructed using reinforcement learning. The reward value is calculated by combining the EEG sleep standard curve to update the driving parameter construction network. This allows for adaptive adjustment of driving parameters such as pulse repetition frequency, pulse width, pulse amplitude, phase delay, and duty cycle according to the user's current EEG state and its changing trends. This improves the individual targeting, dynamic adaptability, and sleep induction effect of pulse electromagnetic field modulation, achieving adaptive closed-loop regulation and safe physiotherapy of the user's EEG activity.

[0004] This invention provides a non-contact electroencephalogram (EEG) electromagnetic therapy method, comprising: The user's brain is monitored by an EEG monitoring unit, and the EEG signal data sequence corresponding to the sampling window is obtained. The EEG signal data sequence includes several EEG signal data arranged according to timestamps. The brainwave signal data is mapped to the pulse phase time axis by the timestamps corresponding to the brainwave signal data. The pulse phase label corresponding to the timestamp of each brainwave signal data is determined. The brainwave signal data sequence is corrected according to the repetition component constraint of the pulse phase label, and brainwave signal feature data is constructed. The pulse phase time axis is the time axis determined according to the driving parameter set of the electromagnetic field construction unit. Several timestamps corresponding to the starting points of reference pulses are marked on the pulse phase time axis. Each reference pulse starting point corresponds to one pulse cycle. The starting point of the reference pulse is the time point when the pulse excitation begins after the driving parameter set of the electromagnetic field construction unit is applied to the electromagnetic field construction unit. Using reinforcement learning, based on EEG signal feature data, a corresponding electromagnetic field construction unit driving parameter set is constructed, and the electromagnetic field construction unit driving parameter set is applied to the electromagnetic field construction unit, so that the electromagnetic field construction unit constructs a pulsed electromagnetic field, which then acts on the user's brain region.

[0005] Preferably, the brainwave signal data is mapped to the pulse phase time axis using the timestamps corresponding to the brainwave signal data, and the pulse phase label corresponding to the timestamp of each brainwave signal data is determined. This specifically includes the following steps: Iterate through all the timestamps corresponding to the EEG signal data, and record the timestamp corresponding to the selected EEG signal data as the target timestamp. On the pulse phase time axis, find the reference pulse start point that is before the target timestamp and closest to the target timestamp and record it as the target reference point. Divide the pulse period corresponding to the target reference point into several pulse phase segments according to a preset time interval. Each pulse phase segment is numbered along the pulse phase time axis starting from the target reference point, and the number corresponding to the pulse phase segment where the target timestamp is located is used as the pulse phase label corresponding to the timestamp of the EEG signal data.

[0006] Preferably, the EEG signal data sequence is corrected based on the repetition component constraint of the pulse phase label, and EEG signal feature data is constructed, specifically including the following steps: Fourier transform is performed on the EEG signal data sequence to obtain the corresponding EEG signal time-frequency graph. In the EEG signal time-frequency graph, the horizontal axis is the timestamp and the vertical axis is the frequency. The horizontal axis is divided according to the time window and the vertical axis is divided according to the frequency window. The time window and the frequency window form a time-frequency unit. The value corresponding to the time-frequency unit is the power spectral density corresponding to the time window and the frequency window. Traverse all pulse phase labels. For the selected pulse phase label, form a comparative analysis set by combining all timestamps corresponding to the pulse phase label. Then traverse the frequency window. For the selected frequency window, extract the values ​​corresponding to the time-frequency units of all timestamps in the comparative analysis set to form a pulse phase frequency analysis set. Calculate the standard deviation for all elements in the pulse phase frequency analysis set, and record it as the component dispersion coefficient. If the component dispersion coefficient is higher than the dispersion threshold, form a low-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. If the component dispersion coefficient is not higher than the dispersion threshold, calculate the average value of the values ​​corresponding to all time-frequency units in the EEG signal time-frequency graph under the selected frequency window, and record it as the overall average value. If the absolute value of the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is lower than the overall distribution threshold, form a low-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. If the absolute value of the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is not lower than the overall distribution threshold, form a high-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. The algorithm iterates through the timestamps corresponding to all EEG signal data. Based on the pulse phase label and frequency window corresponding to the target timestamp, it labels the time-frequency units with low-interference and high-interference constraints. If a time-frequency unit is labeled with low-interference constraints, no operation is performed. If a time-frequency unit is labeled with high-interference constraints, it is recorded as the target time-frequency unit. Centered on the target time-frequency unit, the average value of the time-frequency units labeled with low-interference constraints in the same frequency neighborhood window of the target time-frequency unit is used to replace the value of the target time-frequency unit, thus completing the correction of the target time-frequency unit. The same frequency neighborhood window of the target time-frequency unit is the set of time-frequency units that belong to the same frequency window and are adjacent to the target time-frequency unit. This process continues until all time-frequency units labeled with high-interference constraints are corrected, thus completing the correction of the EEG signal data sequence based on the repetition component constraints of the pulse phase label. The brainwave signal feature data is constructed from the corrected brainwave signal time-frequency map. Specifically, the time-frequency unit corresponding to each frequency band is selected from the brainwave signal time-frequency map, and then the values ​​corresponding to all time-frequency units corresponding to each frequency band are summed to obtain the energy corresponding to each frequency band.

[0007] Preferably, a reinforcement learning approach is used to construct a corresponding electromagnetic field construction unit driving parameter set based on EEG signal feature data, specifically including the following steps: The system acquires the EEG sleep standard curve, calculates the reward value based on the current EEG signal feature data and the EEG sleep standard curve, updates the driving parameter construction network based on the reward value, and then sends the EEG signal feature data into the updated driving parameter construction network for processing to obtain the electromagnetic field construction unit driving parameter set corresponding to the next sampling window. Finally, it performs safety constraint mapping on the obtained electromagnetic field construction unit driving parameter set. The standard EEG sleep curve is obtained as follows: for each frequency band in the EEG signal feature data, EEG signal feature data are collected from the sleep process of several healthy subjects. Fitting points for the corresponding frequency band are constructed based on the center time of the sampling window and the energy proportion of the corresponding frequency band. Curve fitting is performed on all fitting points corresponding to the frequency band to obtain the standard EEG sleep curve for each frequency band.

[0008] Preferably, the reward value is calculated based on the current EEG signal characteristic data and the EEG sleep standard curve, specifically including the following steps: Based on the EEG signal feature data, the proportion of each frequency band in the EEG signal feature data is calculated and formed into frequency band proportion feature data. The center time of the sampling window corresponding to the EEG signal feature data is substituted into the EEG sleep standard curve. The proportion of each frequency band obtained is combined into standard frequency band proportion feature data. The similarity between the frequency band proportion feature data and the standard frequency band proportion feature data is calculated to obtain the reward value.

[0009] Preferably, updating the network based on the reward value for driving parameters specifically includes the following steps: The EEG signal feature data corresponding to the previous sampling window is fed into the driving parameter construction network for processing to obtain the driving parameter set of the electromagnetic field construction unit corresponding to the current sampling window. The current EEG signal feature data is then fed into the driving parameter construction network for processing to obtain the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window. The current EEG signal feature data and the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window are then concatenated and fed into the policy evaluation network for processing to obtain the target evaluation value. The product of the target evaluation value and the discount factor is then summed with the reward value to obtain the target reward value. The concatenated EEG signal feature data corresponding to the previous sampling window and the electromagnetic field construction unit driving parameter set corresponding to the current sampling window are then fed into the policy evaluation network for processing to obtain the evaluation value. The absolute value of the difference between the target reward value and the evaluation value is calculated and recorded as the evaluation loss value. The gradient value of the evaluation loss value with respect to the EEG signal feature data corresponding to the previous sampling window is calculated to minimize the direction of the gradient value. The driving parameter construction network is then updated using the gradient descent method.

[0010] This invention also provides a non-contact electroencephalogram (EEG) electromagnetic therapy system, comprising: The EEG signal acquisition module is used to monitor the user's brain through the EEG monitoring unit and acquire the EEG signal data sequence corresponding to the sampling window. The EEG signal data sequence includes several EEG signal data arranged according to timestamps. The EEG signal feature data construction module is used to map EEG signal data to a pulse phase time axis using the timestamps corresponding to the EEG signal data, determine the pulse phase label corresponding to the timestamp of each EEG signal data, correct the EEG signal data sequence according to the repetition component constraint of the pulse phase label, and construct EEG signal feature data. The pulse phase time axis is a time axis determined according to the driving parameter set of the electromagnetic field construction unit. Several timestamps corresponding to the start points of reference pulses are marked on the pulse phase time axis. Each reference pulse start point corresponds to one pulse cycle. The reference pulse start point is the time point when the pulse excitation begins after the driving parameter set of the electromagnetic field construction unit is applied to the electromagnetic field construction unit. The pulsed electromagnetic field therapy module uses reinforcement learning to construct a corresponding electromagnetic field construction unit driving parameter set based on EEG signal feature data. The electromagnetic field construction unit driving parameter set is then applied to the electromagnetic field construction unit, causing the electromagnetic field construction unit to construct a pulsed electromagnetic field, which is then applied to the user's brain region.

[0011] The present invention has the following advantages: This invention maps EEG signal data to a pulse phase time axis and corrects the EEG signal data sequence by combining the repetition component constraint of the pulse phase label. This effectively suppresses the periodic interference components introduced during the pulse electromagnetic field action, reduces the impact on EEG acquisition results and frequency band energy calculation, and improves the accuracy and stability of EEG signal feature data extraction. Based on the corrected EEG signal feature data, a set of driving parameters for electromagnetic field construction units is constructed using reinforcement learning. The reward value is calculated by combining the EEG sleep standard curve to update the driving parameter construction network. This allows for adaptive adjustment of driving parameters such as pulse repetition frequency, pulse width, pulse amplitude, phase delay, and duty cycle according to the user's current EEG state and its changing trends. This improves the individual targeting, dynamic adaptability, and sleep induction effect of pulse electromagnetic field modulation, achieving adaptive closed-loop regulation and safe physiotherapy of the user's EEG activity. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the non-contact electroencephalography (EEG) electromagnetic therapy system used in an embodiment of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0014] Example 1: A non-contact electroencephalogram (EEG) electromagnetic therapy method, comprising: The brain is monitored by an EEG monitoring unit, and the EEG signal data sequence corresponding to the sampling window is obtained. The sampling window is set by the operator. The EEG signal data sequence includes several EEG signal data arranged according to timestamps. It should be noted that the EEG monitoring unit here is generally set as an inductive electrode, specifically a capacitive non-contact electrode, which can collect the EEG signals of the user's brain through electromagnetic coupling. The brainwave signal data is mapped to the pulse phase time axis by the timestamps corresponding to the brainwave signal data. The pulse phase label corresponding to the timestamp of each brainwave signal data is determined, and the brainwave signal data sequence is corrected according to the repetition component constraint of the pulse phase label. Brainwave signal feature data is then constructed. From the acquired brainwave signals, alpha, gamma, delta, theta, and beta band energy data can be extracted to form brainwave feature data, which characterizes the user's brainwave state. The pulse phase time axis is determined based on the driving parameter set of the electromagnetic field construction unit. Several timestamps corresponding to the start points of reference pulses are marked on the pulse phase time axis. Each reference pulse start point corresponds to one pulse cycle. The reference pulse start point is the time point at which pulse excitation begins after the electromagnetic field construction unit's driving parameter set is applied to the electromagnetic field construction unit. After the electromagnetic field construction unit's driving parameter set is applied to the electromagnetic field construction unit, the electromagnetic field construction unit will periodically perform current excitation. Each current excitation will form a pulse excitation. The range corresponding to one pulse excitation on the pulse phase time axis is the pulse cycle. By mapping the brainwave signal data to the pulse phase time axis using the timestamp corresponding to the brainwave signal data, the position of the timestamp corresponding to the brainwave signal data can be determined on the pulse phase time axis. The reference pulse start point is located based on the timestamp corresponding to the brainwave signal data closest to it. The relative offset of the points allows analysis to determine the position of the timestamp corresponding to the EEG signal data within the pulse excitation, i.e., the pulse phase label. To enable real-time analysis of the impact of the electromagnetic field construction unit's driving parameter set on the user's EEG, the EEG signal data acquisition in this application is performed simultaneously with the pulsed electromagnetic field. However, the pulsed electromagnetic field can affect the acquisition of EEG signals through pathways such as spatial radiation coupling, human conduction coupling, electrode parasitic capacitance coupling, and common-mode injection by the front-end amplifier. For example, while the user is receiving pulsed electromagnetic field modulation, capacitive non-contact electrodes are used to acquire EEG signals. Due to the periodic pulse output characteristic of the pulsed electromagnetic field, the rapidly changing field strength at the start or end of each pulse can... It can be coupled to the EEG sampling link through the human body-electrode parasitic capacitance path and the common mode channel of the acquisition front end, so that the original sampling waveform will have spikes, baseline drift or decay oscillations at the corresponding time. In the frequency domain, the stimulation-related components may also form additional spectral peaks near the pulse repetition frequency and its harmonics, or spread to a wider frequency band due to the excessively steep pulse edge, thereby affecting the energy estimation of the θ band, α band and β band. Therefore, the acquired EEG data signal also includes the interference components corresponding to these pulse electromagnetic fields. However, these interference components will have similar performance under the same pulse phase. Therefore, the stable components that recur under the same pulse phase can be suppressed to correct the EEG signal data sequence. Employing reinforcement learning, this method uses EEG signal feature data and a corresponding electromagnetic field construction unit driving parameter set. This parameter set includes pulse repetition frequency, pulse width, pulse amplitude, phase delay, and duty cycle. Applying this parameter set to the electromagnetic field construction unit causes it to generate a pulsed electromagnetic field. This pulsed electromagnetic field acts on the user's brain region, adjusting their EEG activity, alleviating insomnia, and achieving a therapeutic effect. It's worth noting that the electromagnetic field construction unit is a thin, flexible coil that can be sewn into items such as hats and headbands. When energized, it generates a pulsed electromagnetic field that acts on the user's brain without direct contact. Mapping brainwave signal data to a pulse phase time axis using timestamps corresponding to each brainwave signal data point, and determining the pulse phase label corresponding to each timestamp of brainwave signal data, specifically includes the following steps: The process iterates through the timestamps corresponding to all EEG signal data, designating the timestamp corresponding to the selected EEG signal data as the target timestamp. On the pulse phase time axis, the reference pulse start point closest to the target timestamp is located and designated as the target reference point. The pulse period corresponding to the target reference point is divided into several pulse phase segments according to a preset time interval. This preset time interval is set by the operator, typically consistent with the time window in the EEG signal time-frequency diagram, facilitating subsequent queries of time-frequency units. Each pulse phase segment is numbered from the target reference point along the pulse phase time axis, numbered 0, 1, 2, 3…20, and the number corresponding to the pulse phase segment containing the target timestamp is used as the pulse phase label corresponding to the timestamp of the EEG signal data. On the original sampling time axis, mixed EEG signals within different pulse periods are scattered at different absolute time positions. Introducing the pulse phase time axis maps each sampling point from absolute time to a relative position relative to the reference pulse, allowing data in the same or similar driving phases within different pulse periods to be grouped into the same phase category. This facilitates subsequent analysis of stable components that repeatedly occur under the same pulse phase. The EEG signal data sequence is corrected based on the repetition component constraint of the pulse phase label, and EEG signal feature data is constructed. The specific steps include the following: Fourier transform is performed on the EEG signal data sequence to obtain the corresponding EEG signal time-frequency graph. In the EEG signal time-frequency graph, the horizontal axis is the timestamp and the vertical axis is the frequency. The horizontal axis is divided according to the time window, which is generally set to 0.5s. The vertical axis is divided according to the frequency window, which is generally set to 0.5Hz. The time window and the frequency window are constructed into a time-frequency unit. The value corresponding to the time-frequency unit is the power spectral density corresponding to the time window and the frequency window. Iterate through all pulse phase labels. For each selected pulse phase label, compile all timestamps corresponding to that label into a comparative analysis set. Then, iterate through the frequency window. For each selected frequency window, extract the values ​​corresponding to the time-frequency units of all timestamps in the comparative analysis set to form a pulse phase frequency analysis set. Calculate the standard deviation for all elements in the pulse phase frequency analysis set, denoted as the component dispersion coefficient. If the component dispersion coefficient is higher than the dispersion threshold (set by the operator), it indicates that under the same pulse phase, the corresponding frequency components do not exhibit similar behavior, meaning the pulse electromagnetic field has not affected that frequency component. Then, combine the pulse phase labels and frequency windows corresponding to the pulse phase frequency analysis set into a low-interference constraint. If the component dispersion coefficient is not higher than the dispersion threshold, calculate the average value of all time-frequency units in the EEG signal time-frequency graph within the selected frequency window, denoted as the overall average. If the absolute value of the difference between the average of all elements in the pulse phase frequency analysis set and the overall average is lower than the overall distribution threshold, it indicates that the pulse phase... The frequency component distribution in the pulse phase frequency analysis set is similar to the overall distribution. The pulse phase label and frequency window corresponding to the pulse phase frequency analysis set are used to form a low-interference constraint. The overall distribution threshold is generally set to 1.25 of the overall average value. In the natural fluctuations of EEG, there may be frequency components with small variance and stable repetition. Since the overall distribution is stable and repetitive, these frequency components are not considered as interference from the pulse electromagnetic field. Therefore, the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is compared with the overall distribution threshold to determine whether the frequency component distribution in the pulse phase frequency analysis set is similar to the overall distribution under the current pulse phase. If the absolute value of the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is not lower than the overall distribution threshold, it indicates that the corresponding frequency components show similar behavior under the same pulse phase, that is, the pulse electromagnetic field affects the frequency component. The pulse phase label and frequency window corresponding to the pulse phase frequency analysis set are used to form a high-interference constraint. The algorithm iterates through the timestamps corresponding to all EEG signal data. Based on the pulse phase label and frequency window corresponding to the target timestamp, it marks the time-frequency units with low-interference constraints and high-interference constraints. If a time-frequency unit is marked with low-interference constraints, no operation is performed. If a time-frequency unit is marked with high-interference constraints, it is recorded as the target time-frequency unit. Centered on the target time-frequency unit, the average value of the time-frequency units marked with low-interference constraints in the same frequency neighborhood window of the target time-frequency unit is used to replace the value of the target time-frequency unit, thus completing the correction of the target time-frequency unit. The same frequency neighborhood window of the target time-frequency unit is a set of time-frequency units that belong to the same frequency window and are adjacent to the target time-frequency unit. The size of the same frequency neighborhood window can be set to 9, with four selected before and four after. This process continues until all time-frequency units marked with high-interference constraints are corrected, thus completing the correction of the EEG signal data sequence based on the repetition component constraints of the pulse phase label. The brainwave signal feature data is constructed from the corrected brainwave signal time-frequency map. Specifically, the corresponding time-frequency units are selected from the brainwave signal time-frequency map according to the preset α band, γ band, δ band, θ band, and β band. Then, the values ​​of all time-frequency units corresponding to each frequency band are summed to obtain the energy corresponding to each frequency band. Using reinforcement learning, a set of driving parameters for electromagnetic field construction units is constructed based on EEG signal feature data. The specific steps include: The system acquires the EEG sleep standard curve and calculates a reward value based on the current EEG signal feature data and the EEG sleep standard curve. The reward value characterizes the degree of consistency between the user's EEG signal feature data change trend and the change trend in the EEG sleep standard curve. The higher the consistency, the better the electromagnetic field construction unit driving parameter set is for adjusting the user's sleep. The driving parameter construction network is updated based on the reward value, and then the EEG signal feature data is sent to the updated driving parameter construction network for processing to obtain the electromagnetic field construction unit driving parameter set corresponding to the next sampling window. The obtained electromagnetic field construction unit driving parameter set is then subjected to safety constraint mapping. Safety constraint mapping means that each parameter in the electromagnetic field construction unit driving parameter set has a set safety range. If a parameter in the obtained electromagnetic field construction unit driving parameter set exceeds the corresponding safety range, it must be scaled to the safety range to ensure the safe use of pulse electromagnetic fields. The standard EEG sleep curve is obtained as follows: For each frequency band in the EEG signal feature data, EEG signal feature data is collected from the sleep process of several healthy subjects. Fitting points for each frequency band are constructed based on the center time of the sampling window and the energy proportion of the corresponding frequency band. Curve fitting is performed on all fitting points corresponding to the frequency band. The curve fitting method can be least squares fitting to obtain the standard EEG sleep curve for each frequency band. It should be noted that the center time of the sampling window is the time when the healthy subject falls asleep with their eyes closed. The reward value is calculated based on current EEG signal characteristic data and EEG sleep standard curves, specifically including the following steps: Based on the EEG signal feature data, the proportion data corresponding to each frequency band in the EEG signal feature data is calculated and formed into frequency band proportion feature data. The center time of the sampling window corresponding to the EEG signal feature data is substituted into the EEG sleep standard curve. The proportion data corresponding to each frequency band obtained are combined into standard frequency band proportion feature data. The similarity between the frequency band proportion feature data and the standard frequency band proportion feature data is calculated. The calculation method adopts the cosine similarity algorithm to obtain the reward value. The network is updated based on the reward value, specifically including the following steps: The EEG signal feature data corresponding to the previous sampling window is fed into the driving parameter construction network for processing to obtain the electromagnetic field construction unit driving parameter set corresponding to the current sampling window. The current EEG signal feature data is then fed into the driving parameter construction network for processing to obtain the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window. The current EEG signal feature data and the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window are then concatenated and fed into the policy evaluation network for processing to obtain the target evaluation value. The product of the target evaluation value and the discount factor is then summed with the reward value to obtain the target reward value. The EEG signal feature data corresponding to the previous sampling window and the electromagnetic field construction unit driving parameter set corresponding to the current sampling window are then concatenated and fed into the policy evaluation network for processing to obtain the evaluation value. The absolute value of the difference between the target reward value and the evaluation value is calculated and recorded as the evaluation loss value. The gradient value of the evaluation loss value with respect to the EEG signal feature data corresponding to the previous sampling window is calculated to minimize the direction of the gradient value. The driving parameter construction network is then updated using the gradient descent method. The driving parameter construction network and the policy evaluation network are pre-trained in the following manner: During the development process, in sleep trials on healthy subjects, EEG signal feature data corresponding to the previous sampling window, electromagnetic field construction unit driving parameter set corresponding to the current sampling window, and reward value were collected as training samples. Under several training samples, self-supervised pre-training was performed on the driving parameter construction network and the policy evaluation network by updating the driving parameter construction network based on the reward value.

[0015] This application maps EEG signal data to a pulse phase time axis and corrects the EEG signal data sequence by combining the repetition component constraint of the pulse phase label. This effectively suppresses the periodic interference components introduced during the pulse electromagnetic field action, reduces the impact on EEG acquisition results and frequency band energy calculation, and improves the accuracy and stability of EEG signal feature data extraction. Based on the corrected EEG signal feature data, a set of driving parameters for electromagnetic field construction units is constructed using reinforcement learning. The reward value is calculated by combining the EEG sleep standard curve to update the driving parameter construction network. This allows for adaptive adjustment of driving parameters such as pulse repetition frequency, pulse width, pulse amplitude, phase delay, and duty cycle according to the user's current EEG state and its changing trend. This improves the individual targeting, dynamic adaptability, and sleep induction effect of pulse electromagnetic field modulation, achieving adaptive closed-loop regulation and safe physiotherapy of the user's EEG activity.

[0016] Example 2: A non-contact electroencephalogram (EEG) electromagnetic therapy system, such as... Figure 1 As shown, it includes: The EEG signal acquisition module is used to monitor the user's brain through the EEG monitoring unit and acquire the EEG signal data sequence corresponding to the sampling window. The sampling window is set by the operator. The EEG signal data sequence includes several EEG signal data arranged according to timestamps. It should be noted that the EEG monitoring unit here is generally set as a sensing electrode, specifically a capacitive non-contact electrode, which can acquire the EEG signals of the user's brain through electromagnetic coupling. The EEG signal feature data construction module maps EEG signal data to a pulse phase time axis using timestamps. It determines the pulse phase label corresponding to each timestamp of the EEG signal data, corrects the EEG signal data sequence based on the repetition component constraint of the pulse phase label, and constructs EEG signal feature data. From the acquired EEG signals, it extracts alpha, gamma, delta, theta, and beta band energy to form the EEG signal feature data, which characterizes the user's brainwave state. The pulse phase time axis is determined based on the driving parameter set of the electromagnetic field construction unit, and several reference pulses are marked on the pulse phase time axis. The timestamp corresponding to the starting point is used to define a pulse cycle for each reference pulse starting point. The reference pulse starting point is the time when pulse excitation begins after the electromagnetic field construction unit's driving parameter set is applied to the electromagnetic field construction unit. After the electromagnetic field construction unit's driving parameter set is applied to the electromagnetic field construction unit, the electromagnetic field construction unit will periodically perform current excitation. Each current excitation will form a pulse excitation. The range corresponding to a pulse excitation on the pulse phase time axis is the pulse cycle. By mapping the brainwave signal data to the pulse phase time axis using the timestamps corresponding to the brainwave signal data, the position of the timestamp corresponding to the brainwave signal data can be determined on the pulse phase time axis. The position is determined based on the nearest timestamp corresponding to the brainwave signal data. The relative offset of the reference pulse start point can be used to analyze the position of the timestamp corresponding to the EEG signal data within the pulse excitation, i.e., the pulse phase label. To enable real-time analysis of the impact of the electromagnetic field construction unit driving parameter set on the user's EEG, the EEG signal data acquisition in this application is performed simultaneously with the pulsed electromagnetic field. However, the pulsed electromagnetic field can affect the acquisition of EEG signals through pathways such as spatial radiation coupling, human conduction coupling, electrode parasitic capacitance coupling, and common-mode injection by the front-end amplifier. For example, while the user is receiving pulsed electromagnetic field modulation, capacitive non-contact electrodes are used to acquire EEG signals. Due to the periodic pulse output characteristic of the pulsed electromagnetic field, the rapidly changing pulses at the start or end of each pulse... The field strength may be coupled to the EEG sampling link through the human-electrode parasitic capacitance path and the common-mode channel of the acquisition front end, causing the original sampling waveform to have spikes, baseline drift, or decaying oscillations at the corresponding time. In the frequency domain, the stimulation-related components may also form additional spectral peaks near the pulse repetition frequency and its harmonics, or spread to a wider frequency band due to the excessively steep pulse edge, thereby affecting the energy estimation of the θ band, α band, and β band. Therefore, the acquired EEG data signal also includes the interference components corresponding to these pulse electromagnetic fields. However, these interference components will have similar performance under the same pulse phase. Therefore, the stable components that recur under the same pulse phase can be suppressed to correct the EEG signal data sequence. The pulsed electromagnetic field therapy module uses reinforcement learning to construct a corresponding electromagnetic field building unit driving parameter set based on EEG signal feature data. This parameter set includes pulse repetition frequency, pulse width, pulse amplitude, phase delay, and duty cycle. Applying this parameter set to the electromagnetic field building unit causes it to generate a pulsed electromagnetic field. This pulsed electromagnetic field acts on the user's brain region, adjusting brain activity, alleviating insomnia, and achieving a therapeutic effect. It should be noted that the electromagnetic field building unit is a thin, flexible coil that can be sewn into items such as hats and headbands. When energized, it generates a pulsed electromagnetic field that acts on the user's brain without direct contact.

[0017] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A non-contact electroencephalogram (EEG) electromagnetic therapy system, characterized in that, include: The EEG signal acquisition module is used to monitor the user's brain through the EEG monitoring unit and acquire the EEG signal data sequence corresponding to the sampling window. The EEG signal data sequence includes several EEG signal data arranged according to timestamps. The EEG signal feature data construction module is used to map EEG signal data to a pulse phase time axis using the timestamps corresponding to the EEG signal data, determine the pulse phase label corresponding to the timestamp of each EEG signal data, correct the EEG signal data sequence according to the repetition component constraint of the pulse phase label, and construct EEG signal feature data. The pulse phase time axis is a time axis determined according to the driving parameter set of the electromagnetic field construction unit. Several timestamps corresponding to the starting points of reference pulses are marked on the pulse phase time axis. Each reference pulse starting point corresponds to one pulse cycle. The starting point of the reference pulse is the time point when the pulse excitation begins after the driving parameter set of the electromagnetic field construction unit is applied to the electromagnetic field construction unit. The pulsed electromagnetic field therapy module uses reinforcement learning to construct a corresponding electromagnetic field construction unit driving parameter set based on EEG signal feature data. The electromagnetic field construction unit driving parameter set is then applied to the electromagnetic field construction unit, causing the electromagnetic field construction unit to construct a pulsed electromagnetic field, which then acts on the user's brain region. Mapping brainwave signal data to a pulse phase time axis using timestamps corresponding to each brainwave signal data point, and determining the pulse phase label corresponding to each timestamp of brainwave signal data, specifically includes the following steps: Iterate through all the timestamps corresponding to the EEG signal data, and record the timestamp corresponding to the selected EEG signal data as the target timestamp. On the pulse phase time axis, find the reference pulse start point that is before the target timestamp and closest to the target timestamp and record it as the target reference point. Divide the pulse period corresponding to the target reference point into several pulse phase segments according to a preset time interval. Each pulse phase segment is numbered along the pulse phase time axis starting from the target reference point, and the number corresponding to the pulse phase segment where the target timestamp is located is used as the pulse phase label corresponding to the timestamp of the EEG signal data.

2. The non-contact electroencephalogram (EEG) electromagnetic therapy system according to claim 1, characterized in that, The EEG signal data sequence is corrected based on the repetition component constraint of the pulse phase label, and EEG signal feature data is constructed. The specific steps include the following: Fourier transform is performed on the EEG signal data sequence to obtain the corresponding EEG signal time-frequency graph. In the EEG signal time-frequency graph, the horizontal axis is the timestamp and the vertical axis is the frequency. The horizontal axis is divided according to the time window and the vertical axis is divided according to the frequency window. The time window and the frequency window form a time-frequency unit. The value corresponding to the time-frequency unit is the power spectral density corresponding to the time window and the frequency window. Traverse all pulse phase labels. For the selected pulse phase label, form a comparative analysis set by combining all timestamps corresponding to the pulse phase label. Then traverse the frequency window. For the selected frequency window, extract the values ​​corresponding to the time-frequency units of all timestamps in the comparative analysis set to form a pulse phase frequency analysis set. Calculate the standard deviation for all elements in the pulse phase frequency analysis set, and record it as the component dispersion coefficient. If the component dispersion coefficient is higher than the dispersion threshold, form a low-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. If the component dispersion coefficient is not higher than the dispersion threshold, calculate the average value of the values ​​corresponding to all time-frequency units in the EEG signal time-frequency graph under the selected frequency window, and record it as the overall average value. If the absolute value of the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is lower than the overall distribution threshold, form a low-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. If the absolute value of the difference between the average value of all elements in the pulse phase frequency analysis set and the overall average value is not lower than the overall distribution threshold, form a high-interference constraint between the pulse phase label and the frequency window corresponding to the pulse phase frequency analysis set. The algorithm iterates through the timestamps corresponding to all EEG signal data. Based on the pulse phase label and frequency window corresponding to the target timestamp, it labels the time-frequency units with low-interference and high-interference constraints. If a time-frequency unit is labeled with low-interference constraints, no operation is performed. If a time-frequency unit is labeled with high-interference constraints, it is recorded as the target time-frequency unit. Centered on the target time-frequency unit, the average value of the time-frequency units labeled with low-interference constraints in the same frequency neighborhood window of the target time-frequency unit is used to replace the value of the target time-frequency unit, thus completing the correction of the target time-frequency unit. The same frequency neighborhood window of the target time-frequency unit is the set of time-frequency units that belong to the same frequency window and are adjacent to the target time-frequency unit. This process continues until all time-frequency units labeled with high-interference constraints are corrected, thus completing the correction of the EEG signal data sequence based on the repetition component constraints of the pulse phase label. The brainwave signal feature data is constructed from the corrected brainwave signal time-frequency map. Specifically, the time-frequency unit corresponding to each frequency band is selected from the brainwave signal time-frequency map, and then the values ​​corresponding to all time-frequency units corresponding to each frequency band are summed to obtain the energy corresponding to each frequency band.

3. The non-contact electroencephalogram (EEG) electromagnetic therapy system according to claim 2, characterized in that, Using reinforcement learning, a set of driving parameters for electromagnetic field construction units is constructed based on EEG signal feature data. The specific steps include: The system acquires the EEG sleep standard curve, calculates the reward value based on the current EEG signal feature data and the EEG sleep standard curve, updates the driving parameter construction network based on the reward value, and then sends the EEG signal feature data into the updated driving parameter construction network for processing to obtain the electromagnetic field construction unit driving parameter set corresponding to the next sampling window. Finally, it performs safety constraint mapping on the obtained electromagnetic field construction unit driving parameter set. The standard EEG sleep curve is obtained as follows: for each frequency band in the EEG signal feature data, EEG signal feature data are collected from the sleep process of several healthy subjects. Fitting points for the corresponding frequency band are constructed based on the center time of the sampling window and the energy proportion of the corresponding frequency band. Curve fitting is performed on all fitting points corresponding to the frequency band to obtain the standard EEG sleep curve for each frequency band.

4. The non-contact electroencephalogram (EEG) electromagnetic therapy system according to claim 3, characterized in that, The reward value is calculated based on current EEG signal characteristic data and EEG sleep standard curves, specifically including the following steps: Based on the EEG signal feature data, the proportion of each frequency band in the EEG signal feature data is calculated and formed into frequency band proportion feature data. The center time of the sampling window corresponding to the EEG signal feature data is substituted into the EEG sleep standard curve. The proportion of each frequency band obtained is combined into standard frequency band proportion feature data. The similarity between the frequency band proportion feature data and the standard frequency band proportion feature data is calculated to obtain the reward value.

5. A non-contact electroencephalogram (EEG) electromagnetic therapy system according to claim 4, characterized in that, The network is updated based on the reward value, specifically including the following steps: The EEG signal feature data corresponding to the previous sampling window is fed into the driving parameter construction network for processing to obtain the driving parameter set of the electromagnetic field construction unit corresponding to the current sampling window. The current EEG signal feature data is then fed into the driving parameter construction network for processing to obtain the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window. The current EEG signal feature data and the target electromagnetic field construction unit driving parameter set corresponding to the next sampling window are then concatenated and fed into the policy evaluation network for processing to obtain the target evaluation value. The product of the target evaluation value and the discount factor is then summed with the reward value to obtain the target reward value. The concatenated EEG signal feature data corresponding to the previous sampling window and the electromagnetic field construction unit driving parameter set corresponding to the current sampling window are then fed into the policy evaluation network for processing to obtain the evaluation value. The absolute value of the difference between the target reward value and the evaluation value is calculated and recorded as the evaluation loss value. The gradient value of the evaluation loss value with respect to the EEG signal feature data corresponding to the previous sampling window is calculated to minimize the direction of the gradient value. The driving parameter construction network is then updated using the gradient descent method.

Images