Sleep information processing method and device

Abstract

The invention discloses a sleep information processing method and device, and relates to the technical field of sleep monitoring and signal processing.The sleep information processing method comprises the steps that the instantaneous phase of an electroencephalogram signal is calculated, bone conduction deviation sound stimulation is triggered when the instantaneous phase is located in a target phase window, and the electroencephalogram signal with the high signal-to-noise ratio is obtained through reinforcement learning and dynamic optimization of the stimulation interval; constructing a time-varying causal intensity matrix on the basis of the electroencephalogram signals with the high signal-to-noise ratio and the time-frequency characteristics, performing quantum state mapping on the electroencephalogram signals with the high signal-to-noise ratio, extracting quantum entanglement characteristics, identifying sleep spindle waves, and generating enhanced sleep data through an adversarial generative network; the continuous homologous characteristics of the signal-to-noise ratio electroencephalogram signals are analyzed through an algebraic topology analysis method, and the sleep state of the user is evaluated through a support vector machine in combination with the sleep spindle waves and the enhanced sleep data. According to the method, data diversity and representativeness are effectively enhanced, and more reliable data support is provided for identification and intervention of sleep disorders.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep monitoring and signal processing, and in particular to a method and device for processing sleep information. Background Art

[0002] In recent years, with the advancement of signal processing technology, methods such as time-frequency analysis and causal strength analysis have been introduced into the field of sleep monitoring, further improving the accuracy of sleep staging. In addition, the application of artificial intelligence technology, such as support vector machine (SVM) and deep learning models, has also provided new possibilities for the automatic identification of sleep states. However, existing technologies still have shortcomings in terms of time synchronization of multimodal signals, signal noise suppression, and accurate identification of sleep spindles.

[0003] There are two main problems with existing technologies in multimodal sleep monitoring. First, the time synchronization of multimodal signals is insufficient. Although PSG can collect a variety of physiological signals, the time error between signals may lead to deviations in the analysis results due to the inconsistency of the sampling frequencies and time bases of different sensors. Secondly, the existing technology still has room for improvement in signal noise suppression and feature extraction. Traditional filters have limited effects in dealing with high-frequency noise and low-frequency baseline drift, and it is difficult to completely eliminate environmental electromagnetic interference, which will affect the signal-to-noise ratio of EEG signals, thereby reducing the recognition accuracy of characteristic EEG activities such as sleep spindles. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for processing sleep information to solve the problems of insufficient signal time synchronization and limited noise suppression effect in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for processing sleep information, which includes collecting and preprocessing multimodal physiological signals and extracting time-frequency features at the same time, wherein the multimodal physiological signals include electroencephalogram signals, eye movement signals, electromyogram signals, electrocardiogram signals and respiratory signals; calculating the instantaneous phase of the electroencephalogram signal, triggering bone conduction deviation sound stimulation when the instantaneous phase is in the target phase window, and dynamically optimizing the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio electroencephalogram signal; constructing a time-varying causal intensity matrix based on the high signal-to-noise ratio electroencephalogram signal and time-frequency features, and performing quantum state mapping on the high signal-to-noise ratio electroencephalogram signal, extracting quantum entanglement features, identifying sleep spindles, and generating enhanced sleep data through an adversarial generative network; using algebraic topological analysis to analyze the persistent homologous features of the signal-to-noise ratio electroencephalogram signal, and combining the sleep spindles and enhanced sleep data to evaluate the user's sleep state through a support vector machine.

[0007] As a preferred solution of the sleep information processing method of the present invention, wherein: performing Hilbert transform on the preprocessed EEG signal to calculate the instantaneous phase of the EEG signal; The target phase window is set based on the correlation between the rising edge phase of the slow wave in the historical multimodal physiological signal and the maximum response of the EEG signal. As a preferred solution of the sleep information processing method of the present invention, the method of dynamically optimizing the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio EEG signal includes the following steps: The Q-learning algorithm is used to dynamically optimize the stimulus interval and initialize the state space and action space; Based on the state space and action space, the reward value in the current state is calculated by real-time monitoring of the signal-to-noise ratio of the EEG signal, and the reward function is defined as the increment of the signal-to-noise ratio. At the same time, according to the update rule of Q-learning, Value table; Based on the updated The value table selects the optimal action, dynamically adjusts the stimulation interval, and obtains high signal-to-noise ratio EEG signals.

[0008] As a preferred solution of the sleep information processing method of the present invention, wherein: the extraction of time-frequency features includes the following steps: Through Fourier transform, the power characteristics of the δ frequency band and the frequency ratio characteristics of the θ frequency band and the β frequency band are extracted; By analyzing the number of saccades in eye movement signals, the saccade density characteristics per unit time are extracted; The RMS amplitude feature is extracted by calculating the root mean square value of the electromyographic signal.

[0009] As a preferred solution of the sleep information processing method of the present invention, the method comprises the following steps: constructing a time-varying causal strength matrix based on high signal-to-noise ratio EEG signals and time-frequency features, performing quantum state mapping on high signal-to-noise ratio EEG signals, extracting quantum entanglement features, identifying sleep spindles, and generating enhanced sleep data through a generative adversarial network. Calculate the Granger causal strength of eye movement signals on occipital alpha waves, generate a time-varying causal strength matrix based on the Granger causal strength, and identify the dynamic causal relationship between multimodal physiological signals based on the time-varying causal strength matrix; Calculate the fidelity of quantum states in the left and right prefrontal brain regions, and identify the presence of sleep spindles by combining the dynamic causal relationship between multimodal physiological signals; According to the medical record generator and the normal generator, combined with the Lyapunov index, sleep data is generated.

[0010] As a preferred solution of the sleep information processing method of the present invention, the analysis of the continuous homologous characteristics of the signal-to-noise ratio EEG signal includes the following steps: Topological analysis was performed on multi-channel data of high signal-to-noise ratio EEG signals to construct the Vietoris-Rips complex; Set the maximum distance parameter of the Vietoris-Rips complex ; Maximum distance parameter based on Vietoris-Rips complex , compute the one-dimensional persistent homology features of the Vietoris-Rips complex.

[0011] As a preferred solution of the sleep information processing method of the present invention, wherein: the user's sleep state is evaluated by a support vector machine, comprising the following steps: Define the light sleep threshold Z1 and the deep sleep threshold Z2; The one-dimensional continuous homologous features of sleep spindles, enhanced sleep data and Vietoris-Rips complex are integrated into a comprehensive physiological data set and input into the support vector machine to predict the sleep state index and evaluate the user's sleep state.

[0012] In a second aspect, the present invention provides a sleep information processing device, including a feature extraction module, an EEG signal optimization module, a sleep data generation module and a sleep state evaluation module; the feature extraction module is used to collect multimodal physiological signals and perform preprocessing, and extract time-frequency features at the same time; The EEG signal optimization module is used to calculate the instantaneous phase of the EEG signal, trigger the bone conduction deviation acoustic stimulation when the instantaneous phase is in the target phase window, and dynamically optimize the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio EEG signal; the sleep data generation module is used to construct a time-varying causal intensity matrix based on the high signal-to-noise ratio EEG signal and time-frequency characteristics, and perform quantum state mapping on the high signal-to-noise ratio EEG signal, extract quantum entanglement features, identify sleep spindles, and generate enhanced sleep data through an adversarial generative network; the sleep state evaluation module is used to analyze the continuous homologous characteristics of the signal-to-noise ratio EEG signal using algebraic topological analysis, and combine sleep spindles and enhanced sleep data to evaluate the user's sleep state through a support vector machine.

[0013] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for processing sleep information as described in the first aspect of the present invention is implemented.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for processing sleep information as described in the first aspect of the present invention is implemented.

[0015] The beneficial effects of the present invention are as follows: the instantaneous phase of the EEG signal is calculated by Hilbert transform, and bone conduction deviation acoustic stimulation is triggered within the target phase window, and the stimulation interval is dynamically optimized in combination with the Q-learning algorithm, thereby maximizing the signal-to-noise ratio of the EEG signal. This step significantly improves the accuracy and response speed of sleep monitoring. The high signal-to-noise ratio EEG signal is mapped to quantum entanglement features through the quantum state linear combination method, and enhanced sleep data containing pathological and healthy features is generated in combination with the adversarial generative network, which effectively enhances the data diversity and representativeness, and provides more reliable data support for the identification and intervention of sleep disorders. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0017] Figure 1 Flow chart of the sleep information processing method in Example 1.

[0018] Figure 2 Schematic diagram of the sleep information processing device in Example 1.

[0019] Figure 3 This is a flow chart of the reinforcement learning dynamic optimization of stimulation intervals in Example 1.

[0020] Figure 4 This is a flow chart of the support vector machine evaluating sleep state in Example 1. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and 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.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, reference Figure 1~Figure 4 , which is the first embodiment of the present invention, provides a method for processing sleep information, comprising the following steps: S1. Collect multimodal physiological signals and preprocess them, and extract time-frequency features.

[0025] Multimodal physiological signals include EEG signals, eye movement signals, EMG signals, ECG signals and respiratory signals.

[0026] Furthermore, EEG signals are collected through an electrode array deployed on the scalp, eye movement signals are recorded through electrodes placed around the eyes, EMG signals detect muscle electrical activity through electrodes attached to the muscle surface, ECG signals measure heart electrical activity through electrodes on the chest, and respiratory signals monitor breathing rate and depth through a chest strap or airflow sensor.

[0027] It should be noted that the above data are all obtained with the user's consent and are used for legitimate purposes; Preprocessing includes signal noise reduction, filtering and time alignment. The specific steps are as follows: First, the collected original multimodal physiological input is input into the quantum spin Hall effect filter to retain the effective multimodal physiological signal components by eliminating environmental electromagnetic interference (such as 50Hz industrial frequency noise); for example, the high-frequency noise in the EEG signal is significantly suppressed, thereby improving the signal-to-noise ratio of the EEG signal.

[0028] A bandpass filter was used for the eye movement signal, and the frequency band was set to 0.1-10 Hz to retain the effective eye movement characteristics; a high-pass filter was used for the electromyography signal, and the cutoff frequency was set to 20 Hz to remove the low-frequency baseline drift.

[0029] A unified time reference (such as GPS or network time protocol) is used to provide synchronized timestamps for all devices, ensuring that the time error of multimodal physiological signals is controlled within ±5ms; for example, EEG signals, eye movement signals, and EMG signals are precisely aligned on the time axis to ensure the consistency of multimodal analysis.

[0030] The time-frequency characteristics include delta wave power and theta / beta frequency ratio characteristics, scan density characteristics and RMS amplitude characteristics.

[0031] Furthermore, the EEG signals were subjected to Fourier transform to convert them into frequency domain signals to obtain the power spectral density distribution. The power spectral density of the delta wave band (0.5-4 Hz) was extracted in the frequency domain, and its integral value was calculated as the delta wave power. At the same time, the power spectral density of the theta wave band (4-8 Hz) and the beta wave band (13-30 Hz) was extracted by calculating the power ratio of the theta wave to the beta wave.

[0032] After preprocessing the eye movement signals, an event detection algorithm is used to identify rapid eye movement (saccade) events, and the start and end points of the saccade are determined by setting amplitude and speed thresholds; the number of saccades per unit time is counted, and the saccade density is calculated.

[0033] The EMG signal is processed in segments, and the length of each segment is usually set to a fixed time window (such as 1 second). The root mean square value of each EMG signal is calculated as an intensity indicator of muscle activity. For example, the root mean square value of the chin EMG signal can be used to monitor muscle tension and assist in identifying micro-arousal events during sleep.

[0034] S2. By calculating the instantaneous phase of the EEG signal, the bone conduction deviation acoustic stimulation is triggered when the instantaneous phase is in the target phase window, and the stimulation interval is dynamically optimized through reinforcement learning to obtain a high signal-to-noise ratio EEG signal.

[0035] Perform Hilbert transform on the preprocessed EEG signal and calculate its instantaneous phase. The expression is: ; in, Indicates the EEG signal at time The instantaneous phase value at time is the Cauchy principal value integral, Indicates time The EEG signal at is the integration variable, Represents the integral variable The EEG signal value at .

[0036] The target phase window is set based on the correlation between the rising edge phase (approximately 0 to π / 2) of the slow wave (0.5-2 Hz) in the historical multimodal physiological signal and the maximum response of the EEG signal. .

[0037] Correlation refers to the positive relationship between the rising edge phase (approximately 0 to π / 2) of the slow wave (0.5-2 Hz) and the response intensity of the EEG signal; for example, when the stimulus is applied at the rising edge phase of the slow wave, the amplitude of the event-related potential ERP is significantly higher than that at other phases, indicating that this phase window is the key time to induce the optimal response of neural activity.

[0038] The instantaneous phase refers to the angular position of the EEG signal at a certain moment, usually expressed in radians (0 to 2π).

[0039] When the instantaneous phase of the EEG signal is detected In target window When the bone conduction deviation sound stimulation is automatically triggered, the stimulation time Depend on Determine, among which Indicates the instantaneous phase.

[0040] The Q-learning algorithm is used to dynamically optimize the stimulation interval. The objective function is the maximum signal-to-noise ratio of the EEG signal. The specific steps include: Based on the instantaneous phase of EEG signals and the current stimulus interval, initialize the state space; Initialize the action space based on the increment or decrement of the stimulus interval; Increment and decrement refer to the time of increase and decrease in the stimulation interval; By monitoring the signal-to-noise ratio of the EEG signal in real time, the reward value in the current state is calculated, and the reward function is defined as the increment of the signal-to-noise ratio. At the same time, according to the update rule of Q-learning, Value table, the expression is: ; in, is in state Next action of value, represents the learning rate, Indicates the current state Next action The reward value obtained after (i.e., the increment of the EEG signal-to-noise ratio), is the discount factor (range 0 to 1), is the next state, The next state The action performed, is in the next state All possible actions The largest value.

[0041] Should be explained, update The value table refers to the adjustment of state-action pairs according to the update rules of the Q-learning algorithm. value to reflect the change of expected cumulative reward under the current strategy; specifically, after each action is executed, according to the reward value (i.e., the increment of EEG signal-to-noise ratio) and the maximum value of the next state Value, update the current state-action pair This process is directly related to the dynamic optimization of the inter-stimulus interval, because After the value table is updated, the Q-learning algorithm selects the one with the highest The action of optimizing the stimulation interval time (i.e., the optimal stimulation interval time adjustment) gradually approaches the goal of maximizing the signal-to-noise ratio of the EEG signal; for example, when a stimulation interval time adjustment leads to a significant improvement in the signal-to-noise ratio, the action The value will increase, and the Q-learning algorithm will choose this action to adjust and ultimately maximize the signal-to-noise ratio.

[0042] It should also be noted that the action refers to the specific operation of adjusting the stimulation interval, such as increasing or decreasing by 100 milliseconds, and the state refers to the combination of the phase information of the current EEG signal and the stimulation interval.

[0043] Based on the updated The value table selects the optimal action, dynamically adjusts the stimulation interval, and obtains high signal-to-noise ratio EEG signals.

[0044] S3. Based on high signal-to-noise ratio EEG signals and time-frequency characteristics, a time-varying causal strength matrix is ​​constructed, and quantum state mapping of high signal-to-noise ratio EEG signals is performed to extract quantum entanglement features and identify sleep spindles. At the same time, enhanced sleep data is generated through a generative adversarial network.

[0045] Time series analysis was performed on high signal-to-noise ratio EEG signals and time-frequency features, and the Granger causal analysis method was used to calculate the Granger causal strength of eye movement signals on the occipital alpha wave (8-12 Hz), which was expressed as: ; in, Eye movement signal Occipital alpha waves The Granger causal strength of Represents historical eye movement signals Historical occipital alpha waves The predicted variance of occipital alpha waves under different conditions, Representation based on historical occipital alpha waves Occipital alpha waves The prediction variance of Indicates the current occipital alpha wave, It is the historical occipital alpha wave, is the historical eye movement signal, is the current eye movement signal; Furthermore, when calculating the Granger causal strength of the eye movement signal E on the occipital alpha wave g, the signal is segmented using a fixed time window, and the Granger causal strength is calculated separately in each time window.

[0046] The Granger causal strength in each time window is filled into the corresponding position of the causal strength matrix. The rows and columns of the matrix represent different multimodal physiological signals respectively. As the time window slides, the causal strength matrix is ​​dynamically updated to generate a time-varying causal strength matrix.

[0047] Furthermore, first, multimodal physiological signals (such as EEG signals, eye movement signals, and electromyographic signals) are time-segmented, with each segment length being a fixed time window (such as 1 second); then, the Granger causal strength between different physiological signals is calculated within each time window, such as the causal strength of eye movement signals on occipital alpha waves. , the causal strength of electromyographic signals on electroencephalographic signals, etc.; then, the Granger causal strength values ​​in each time window are filled into the corresponding positions of the causal strength matrix. The rows and columns of the causal strength matrix represent different multimodal physiological signals, respectively. For example, the first row of the time-varying causal strength matrix represents the causal influence of eye movement signals on other multimodal physiological signals, and the second row represents the causal influence of electromyographic signals on other signals; finally, as the time window slides, the causal strength matrix is ​​dynamically updated to generate a time-varying causal strength matrix.

[0048] According to the time-varying causal strength matrix, the dynamic causal relationship between multimodal physiological signals can be identified. For example, in the rapid eye movement (REM) sleep stage, the Granger causal strength of eye movement signals on occipital alpha waves Significantly enhanced, indicating that eye movement signals have a stronger influence on alpha waves; while in the non-rapid eye movement sleep stage, The value of is reduced, indicating that the influence of eye movement signals on alpha waves is weakened. At the same time, the Granger causal strength of myoelectric signals on EEG signals is significantly increased during micro-awakening events, indicating that the influence of muscle activity on EEG signals is enhanced.

[0049] Through the quantum state linear combination method, the high signal-to-noise ratio EEG signal is quantum-mapped and converted into quantum entanglement characteristics. The expression is: ; in, is the quantum entanglement feature of EEG signals. is the total number of frequency components, is the frequency component In time The phase change when Indicates The quantum state basis vector corresponding to the frequency component is: is the index variable of the frequency component.

[0050] Furthermore, first, the high signal-to-noise ratio EEG signal is Fourier transformed to convert it from the time domain to the frequency domain and extract all frequency components; then, each frequency component is mapped to the corresponding quantum state basis vector, which represents the ground state of the frequency component in the quantum state space; then, the phase change of each frequency component is weightedly combined with the corresponding quantum state basis vector to generate the quantum entanglement feature of the EEG signal; finally, through normalization processing, the total probability of the quantum entanglement feature is ensured to be 1, and the final quantum state representation is generated. For example, when analyzing the EEG signals of the left and right brain regions of the prefrontal lobe, the quantum entanglement features generated by the quantum state linear combination method can be used to evaluate the similarity between brain regions and then identify sleep spindles.

[0051] Based on the characteristics of quantum entanglement , calculate the fidelity of the quantum state of the left and right brain regions of the prefrontal lobe , the expression is: ; in, represents the quantum state of the left prefrontal brain area, Represents the quantum state of the right prefrontal brain area.

[0052] Based on historical multimodal physiological signal data, the fidelity benchmark F1 and the causal strength benchmark G1 are defined; Fidelity of quantum states based on the left and right brain regions of the prefrontal cortex , combining the dynamic causal relationship between multimodal physiological signals to analyze the interaction between brain regions. >F1 and When <G1, it is determined that sleep spindles exist.

[0053] Furthermore, first, the fidelity of the quantum states of the left and right prefrontal cortex is calculated. , evaluate the similarity of the quantum states of the two brain regions; then, calculate the Granger causal strength of the eye movement signal on the occipital alpha wave, and analyze the impact of the eye movement signal on the occipital alpha wave; then, Compared with the fidelity benchmark F1, the Granger causal strength is compared with the causal strength benchmark G1; finally, when the fidelity When the fidelity is greater than the reference value F1 and the Granger causal strength is less than the causal strength reference G1, it is determined that sleep spindles exist. When the value of F1 is 0.9 (greater than the reference value F1=0.85) and the Granger causal strength is 0.6 (less than the reference value G1=0.65), it is determined that sleep spindles exist in the current EEG signal.

[0054] It should be noted that sleep spindles are a characteristic EEG activity, manifested as spindle-shaped oscillation waves of 11-16 Hz, such as regular wave groups lasting 0.5-2 seconds; the value range of the fidelity benchmark F1 is usually 0.8-0.9, and the value range of the causal strength benchmark G1 is usually 0.6-0.7.

[0055] A dual generator architecture consisting of a pathological generator and a normal generator is used as the adversarial generative network, and the Lyapunov exponent is calculated to constrain the dynamic stability of the sleep data. The dual generator architecture includes a pathological generator and a normal generator. Generate sleep data containing pathological characteristics through the pathological generator, generate sleep data containing healthy characteristics through the normal generator, and combine the sleep data with pathological characteristics and the sleep data with healthy characteristics to generate enhanced sleep data; Furthermore, the dual generator architecture includes a pathological generator and a normal generator, wherein the pathological generator is used to generate sleep data containing pathological features (such as apnea), and the normal generator is used to generate sleep data containing healthy sleep features (such as normal breathing patterns); then, the Lyapunov exponent of the generated data is calculated by a nonlinear dynamic analysis method to ensure that the behavior of the generated data on a long-term time scale conforms to physiological laws; then, the pathological generator and the normal generator are jointly trained so that the generated sleep data contain both diverse pathological and healthy features and meet the dynamic stability conditions; finally, pathological data containing wake events or apnea events are generated to enhance the diversity and representativeness of the sleep dataset. For example, the pathological generator generates sleep data containing apnea features, and the normal generator generates sleep data that conforms to healthy sleep features, and the two together constitute an enhanced sleep dataset.

[0056] The Lyapunov exponent is calculated by nonlinear dynamics analysis method, and the expression is: ; in, is the Lyapunov exponent, is the small perturbation vector at the initial moment, It is the time that has passed The disturbance vector after is the norm of the disturbance vector, indicating the magnitude of the disturbance.

[0057] It should be noted that sleep data containing pathological characteristics is obtained by collecting physiological signals from patients with sleep disorders (such as sleep apnea, insomnia or periodic limb movement disorder), including EEG signals, eye movement signals, electromyographic signals, electrocardiographic signals and respiratory signals, recorded through polysomnography (PSG), and annotated with pathological events (such as the number of apnea times, limb movement frequency, etc.). For example, in the data of patients with sleep apnea, EEG signals show frequent awakening waves and respiratory signals show periodic interruptions.

[0058] Sleep data containing health characteristics is obtained by collecting physiological signals of healthy individuals, including EEG signals, eye movement signals, electromyographic signals, electrocardiographic signals and respiratory signals, and recorded through polysomnography (PSG). For example, in the data of healthy individuals, EEG signals show typical sleep spindles and K complex waves, and respiratory signals show a stable rhythm.

[0059] S4. The algebraic topological analysis method is used to analyze the continuous homologous characteristics of the signal-to-noise ratio EEG signals, and combined with the sleep spindles and enhanced sleep data, the user's sleep state is evaluated through the support vector machine.

[0060] Using algebraic topological analysis, we conduct topological analysis on multi-channel data of high signal-to-noise ratio EEG signals, construct the Vietoris-Rips complex, and set the maximum distance parameter of the complex based on the data distribution of historical EEG signals. , ensuring that the complex can effectively capture the topological structure of the signal.

[0061] Maximum distance parameter based on Vietoris-Rips complex , calculate the one-dimensional persistent homology features of the Vietoris-Rips complex , the expression is: ; in, represents the number of one-dimensional persistent homology features in the Vietoris-Rips complex, is the length of the duration window, is the brain area.

[0062] Furthermore, based on the data distribution of historical EEG signals, the maximum distance between signal points is determined, and the maximum distance parameter of the Vietoris-Rips complex is set accordingly. , ensuring that the Vietoris-Rips complex can effectively capture the topological structure of the signal. Then, according to the maximum distance parameter of the Vietoris-Rips complex , connect the multi-channel data points of the high signal-to-noise ratio EEG signal into a Vietoris-Rips complex to generate a topological structure. Then, by calculating the one-dimensional persistent homology group of the complex, the one-dimensional persistent homology features of the Vietoris-Rips complex are extracted.

[0063] A brain region is an area of ​​the brain with a specific function or anatomical location, such as the frontal lobe, occipital lobe, or temporal lobe.

[0064] The one-dimensional continuous homology features of sleep spindles, enhanced sleep data and Vietoris-Rips complex are integrated into a comprehensive physiological data set and input into the support vector machine to predict the sleep state index. The expression is: ; in, is the sleep state index, is the number of support vectors, is the index variable of the support vector, is the bias term of the support vector machine, It is The weight coefficients of the support vectors, Indicates The labels of the support vectors, Indicates elements of a comprehensive physiological dataset, Elements of the comprehensive physiological dataset, represents the kernel function, express and The similarity between .

[0065] Furthermore, based on the historical multimodal physiological data, the support vector machine was trained using the back propagation algorithm and the gradient descent method. The specific process is as follows: first, the historical multimodal physiological data are input into the support vector machine. Then, the similarity between the samples in the training set is calculated by the kernel function to construct a similarity matrix. Next, based on the similarity matrix and sample labels, the gradient of the loss function with respect to the weight coefficient and the bias term is calculated using the back propagation algorithm, and the weight coefficient and the bias term are updated using the gradient descent method so that the support vector machine can maximize the classification interval. Finally, the test set is used to verify the classification performance of the support vector machine, and the hyperparameters (such as kernel function parameters and penalty coefficients) are adjusted to optimize the support vector machine.

[0066] express and The similarity between them is expressed as follows: ; in, Represents the similarity decay factor.

[0067] Based on the historical sleep data distribution, define the light sleep threshold Z1 and deep sleep threshold Z2; when When <Z1, the user is considered to be in the awake period; When Z1≤ When <Z2, the user is considered to be in a light sleep period; when When ≥Z2, the user is considered to be in deep sleep.

[0068] It should be noted that the similarity decay factor It is determined based on the performance optimization of the kernel function, with a value range of 0.01 to 1.0, and the specific value is selected through cross-validation; the light sleep threshold Z1 is determined based on the dividing point between the awake period and the light sleep period in the historical sleep data distribution, with a value range of 0.3 to 0.5; the deep sleep threshold Z2 is determined based on the dividing point between the light sleep period and the deep sleep period in the historical sleep data distribution, with a value range of 0.6 to 0.8.

[0069] This embodiment also provides a sleep information processing device, including: a feature extraction module, an EEG signal optimization module, a sleep data generation module and a sleep state evaluation module; Feature extraction module, used to collect multimodal physiological signals and perform preprocessing, while extracting time-frequency features; The EEG signal optimization module is used to calculate the instantaneous phase of the EEG signal, trigger the bone conduction deviation acoustic stimulation when the instantaneous phase is in the target phase window, and dynamically optimize the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio EEG signal; The sleep data generation module is used to construct a time-varying causal strength matrix based on high signal-to-noise ratio EEG signals and time-frequency features, and to perform quantum state mapping on high signal-to-noise ratio EEG signals, extract quantum entanglement features, identify sleep spindles, and generate enhanced sleep data through a generative adversarial network. The sleep state assessment module is used to analyze the continuous homologous characteristics of the signal-to-noise ratio EEG signals using algebraic topological analysis, and combine the sleep spindles and enhanced sleep data to assess the user's sleep state through a support vector machine.

[0070] This embodiment also provides a computer device suitable for the sleep information processing method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the sleep information processing method proposed in the above embodiment.

[0071] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0072] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the sleep information processing method proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0073] In summary, the present invention achieves the maximization of the signal-to-noise ratio of the EEG signal by: calculating the instantaneous phase of the EEG signal through Hilbert transform, triggering bone conduction deviation acoustic stimulation within the target phase window, and dynamically optimizing the stimulation interval time in combination with the Q-learning algorithm. This step significantly improves the accuracy and response speed of sleep monitoring. The high signal-to-noise ratio EEG signal is mapped to quantum entanglement features through the quantum state linear combination method, and combined with the adversarial generative network to generate enhanced sleep data containing pathological and healthy features, which effectively enhances the data diversity and representativeness, and provides more reliable data support for the identification and intervention of sleep disorders.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing sleep information, characterized in that: include, Collecting multimodal physiological signals and preprocessing them, while extracting time-frequency features, wherein the multimodal physiological signals include electroencephalogram signals, eye movement signals, electromyography signals, electrocardiogram signals and respiratory signals; Calculate the instantaneous phase of the EEG signal, trigger bone conduction deviation acoustic stimulation when the instantaneous phase is in the target phase window, and dynamically optimize the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio EEG signal; Based on high signal-to-noise ratio EEG signals and time-frequency characteristics, a time-varying causal strength matrix is ​​constructed, and quantum state mapping of high signal-to-noise ratio EEG signals is performed to extract quantum entanglement features and identify sleep spindles. At the same time, enhanced sleep data is generated through a generative adversarial network. The algebraic topological analysis method is used to analyze the persistent homologous features of the signal-to-noise ratio EEG signals, and combined with the sleep spindles and enhanced sleep data, the user's sleep state is evaluated through a support vector machine.

2. The sleep information processing method according to claim 1, characterized in that: Perform Hilbert transform on the preprocessed EEG signal and calculate the instantaneous phase of the EEG signal; The target phase window is set based on the correlation between the rising edge phase of the slow wave in the historical multimodal physiological signals and the maximum response of the EEG signal.

3. The sleep information processing method according to claim 1, characterized in that: The method of dynamically optimizing the stimulation interval through reinforcement learning and obtaining a high signal-to-noise ratio EEG signal includes the following steps: The Q-learning algorithm is used to dynamically optimize the stimulus interval and initialize the state space and action space; Based on the state space and action space, the reward value in the current state is calculated by real-time monitoring of the signal-to-noise ratio of the EEG signal, and the reward function is defined as the increment of the signal-to-noise ratio. At the same time, according to the update rule of Q-learning, Value table; Based on the updated The value table selects the optimal action, dynamically adjusts the stimulation interval, and obtains high signal-to-noise ratio EEG signals.

4. The sleep information processing method according to claim 1, characterized in that: The time-frequency feature extraction, The following steps are included: Through Fourier transform, the power characteristics of the δ frequency band and the frequency ratio characteristics of the θ frequency band and the β frequency band are extracted; By analyzing the number of saccades in eye movement signals, the saccade density characteristics per unit time are extracted; The RMS amplitude feature is extracted by calculating the root mean square value of the electromyographic signal.

5. The sleep information processing method according to claim 3, characterized in that: The method constructs a time-varying causal strength matrix based on high signal-to-noise ratio EEG signals and time-frequency features, performs quantum state mapping on high signal-to-noise ratio EEG signals, extracts quantum entanglement features, identifies sleep spindles, and generates enhanced sleep data through a generative adversarial network, including the following steps: Calculate the Granger causal strength of eye movement signals on occipital alpha waves, generate a time-varying causal strength matrix based on the Granger causal strength, and identify the dynamic causal relationship between multimodal physiological signals based on the time-varying causal strength matrix; Calculate the fidelity of quantum states in the left and right prefrontal brain regions, and identify the presence of sleep spindles by combining the dynamic causal relationship between multimodal physiological signals; According to the medical record generator and the normal generator, combined with the Lyapunov index, sleep data is generated.

6. The sleep information processing method according to claim 5, characterized in that: The analysis of the signal-to-noise ratio of the EEG signal is based on the continuous homologous characteristics of the signal-to-noise ratio. The following steps are included: Topological analysis was performed on multi-channel data of high signal-to-noise ratio EEG signals to construct the Vietoris-Rips complex; Set the maximum distance parameter of the Vietoris-Rips complex ; Maximum distance parameter based on Vietoris-Rips complex , compute the one-dimensional persistent homology features of the Vietoris-Rips complex.

7. The sleep information processing method according to claim 6, characterized in that: The method of evaluating the user's sleep state by using a support vector machine includes the following steps: Define the light sleep threshold Z1 and the deep sleep threshold Z2; The one-dimensional continuous homologous features of sleep spindles, enhanced sleep data and Vietoris-Rips complex are integrated into a comprehensive physiological data set and input into the support vector machine to predict the sleep state index and evaluate the user's sleep state.

8. A sleep information processing device, based on the sleep information processing method according to any one of claims 1 to 7, characterized in that: Including feature extraction module, EEG signal optimization module, sleep data generation module and sleep state assessment module; Feature extraction module, used to collect multimodal physiological signals and perform preprocessing, while extracting time-frequency features; The EEG signal optimization module is used to calculate the instantaneous phase of the EEG signal, trigger the bone conduction deviation acoustic stimulation when the instantaneous phase is in the target phase window, and dynamically optimize the stimulation interval through reinforcement learning to obtain a high signal-to-noise ratio EEG signal; The sleep data generation module is used to construct a time-varying causal strength matrix based on high signal-to-noise ratio EEG signals and time-frequency features, and to perform quantum state mapping on high signal-to-noise ratio EEG signals, extract quantum entanglement features, identify sleep spindles, and generate enhanced sleep data through a generative adversarial network. The sleep state assessment module is used to analyze the continuous homologous characteristics of the signal-to-noise ratio EEG signals using algebraic topological analysis, and combine the sleep spindles and enhanced sleep data to assess the user's sleep state through a support vector machine.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the sleep information processing method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the sleep information processing method according to any one of claims 1 to 7 are implemented.

Images