Sleep intervention methods, systems, terminal devices, and storage media based on joint features

Abstract

This invention provides a sleep intervention method, system, terminal device, and storage medium based on joint features. The method includes: extracting features from real-time respiratory signals to obtain respiratory rhythm stability features; extracting features from real-time electroencephalogram (EEG) signals to obtain slow-wave EEG features; synchronizing the respiratory rhythm stability features and the slow-wave EEG features in time, and inputting the obtained joint features into a first artificial intelligence algorithm model so that the model outputs a stimulation state judgment result for the user; when several consecutive stimulation state judgment results are all in a state that is pending confirmation of stimulation, inputting the collected real-time EEG signals into a second artificial intelligence algorithm model for signal analysis; when the analysis result is that the stimulation is possible, executing a sound stimulation task within a preset time delay; when the respiratory rhythm stability features or the slow-wave EEG features do not conform to a preset trend, ending the sound stimulation task. This invention can improve the convenience and accuracy of sleep intervention.

Description

Technical Field

[0001] This invention relates to the field of sleep regulation technology, and in particular to sleep intervention methods, systems, terminal devices and storage media based on combined features. Background Technology

[0002] Sleep intervention technology refers to a technical system that monitors and intervenes in human sleep state and sleep structure through various methods such as physical, acoustic, and biosignal modulation to improve sleep quality, optimize sleep cycles, and prolong specific beneficial sleep stages. Currently, sleep intervention technologies for deep sleep are mainly divided into three categories: The first category is passive sound / white noise playback, which attempts to create a comfortable sleep environment by playing sounds of fixed frequency and intensity during the sleep onset stage or throughout the night; the second category is sleep stimulation systems based on multi-channel bioelectrical signals, which collect multimodal bioelectrical signals such as EEG, EMG, and EEG, and combine them with complex algorithm models to determine sleep stages and trigger corresponding stimuli; the third category is a single EEG monitoring stimulation scheme, which determines sleep state based on EEG signals and implements stimulation accordingly. However, all three types of intervention techniques have significant technical shortcomings: the first type lacks dynamic matching with the user's real-time sleep state, and is prone to triggering awakenings due to sound stimulation not matching the sleep stage, thus interfering with normal sleep structure and failing to achieve the goal of prolonging deep sleep; the second type improves the targeting of interventions, but requires wearing multi-lead electrodes to collect various bioelectrical signals, resulting in complex equipment structures and cumbersome operation, placing a heavy burden on users, poor comfort, and unsuitability for long-term home use; the third type reduces equipment complexity, but EEG signals are easily interfered with by environmental noise, individual physiological differences, etc., leading to insufficient stability in sleep state judgment, low accuracy of stimulus triggering, and difficulty in guaranteeing intervention effects. Existing sleep intervention technologies have significant shortcomings in terms of ease of wear, stability of state judgment, and long-term suitability. Summary of the Invention

[0003] The present invention aims to provide a sleep intervention method, system, terminal device and storage medium based on joint features to solve the above-mentioned technical problems and improve the convenience and accuracy of sleep intervention.

[0004] To address the aforementioned technical problems, this invention provides a sleep intervention method based on joint features, comprising: The first feature is extracted from the real-time respiratory signals acquired using contactless technology to obtain respiratory rhythm stability features; A second feature extraction is performed on the real-time EEG signal to obtain the slow wave features of the EEG; wherein, the real-time EEG signal is acquired by electrode units designed in the user's preset area; When the respiratory rhythm stability feature is lower than a preset threshold, the respiratory rhythm stability feature and the slow wave feature of the electroencephalogram are synchronized in time, and the resulting joint feature is input into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state; When the first artificial intelligence algorithm model outputs several consecutive stimulation state judgment results, all of which are unconfirmed stimulable states, the continuously collected real-time EEG signals are input to the second artificial intelligence algorithm model for signal analysis. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the sound stimulation task is executed within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG characteristics, and the second sample labels are the user's arousal states. The sound stimulation task ends when the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements; where the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

[0005] In the above scheme, respiratory signals are collected without contact to extract rhythm stability features, and lightweight EEG electrode units are used to collect EEG signals to extract slow-wave features. This avoids the cumbersome wearing of multi-lead devices and obtains core features for both sleep safety and stimulation feasibility. Time synchronization eliminates feature time misalignment, and the combined features are input into an AI algorithm model trained on massive labeled samples. The model's classification ability enables accurate judgment of stimulable states, and continuous result verification further reduces the probability of misjudgment, significantly improving the stability of state recognition compared to single EEG judgment. Once a stimulable state is determined, the activation state is determined by analyzing real-time EEG signals. If an activation state is confirmed, sound stimulation is triggered within a preset low latency, achieving precise time matching between stimulation and slow-wave rhythm, maximizing slow-wave activity and prolonging deep sleep time. Simultaneously, the degree of respiratory cycle fluctuation and EEG slow-wave amplitude are continuously monitored during stimulation. Stimulation is immediately terminated when features are abnormal, forming a real-time closed-loop control that avoids the risk of awakening from both parameter design and dynamic monitoring dimensions. No complicated operation is required, and it is suitable for long-term home use. While improving the effectiveness of sleep intervention, it ensures the comfort and safety of use.

[0006] In one implementation, a first feature extraction is performed on the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features, specifically including: The user's chest and abdominal displacement data is detected using contactless technology, and a raw respiratory signal is generated based on the chest and abdominal displacement data to characterize the respiratory cycle. The raw respiratory signal is preprocessed to obtain the real-time respiratory signal; Identify the respiratory rhythm stability characteristics of real-time respiratory signals; among which, respiratory rhythm stability characteristics include the duration of a single respiratory cycle, the degree of respiratory cycle fluctuation, and the stability of respiratory rate.

[0007] The above solution uses contactless technology to collect chest and abdominal displacement data of users to generate raw respiratory signals, achieving contactless monitoring from the source of collection, reducing the burden on users, and is suitable for long-term home sleep monitoring scenarios; it extracts respiratory rhythm stability features such as the duration of a single respiratory cycle, the degree of respiratory cycle fluctuation, and the stability of respiratory frequency, without the need for contact sensors or restrictive wearing throughout the process, which not only improves the convenience and comfort of respiratory signal collection, but also objectively and stably reflects the user's true respiratory rhythm state.

[0008] In one implementation, a second feature extraction is performed on the real-time EEG signal to obtain slow-wave EEG features, specifically including: Raw electroencephalogram (EEG) signals from the user at preset brain regions are acquired by at least one pair of electrode units placed on the surface of the user's scalp. The raw EEG signal is processed to obtain real-time EEG signal and low-frequency EEG signal within a preset frequency range is extracted; the low-frequency EEG signal is used to characterize slow wave activity. The slow wave characteristics of EEG are generated based on the power and amplitude of low-frequency EEG signals.

[0009] In the above scheme, raw EEG signals are collected through at least one pair of electrode units on the surface of the scalp, abandoning the complex layout of traditional multi-lead EEG, reducing the user's wearing complexity and physical burden, and the preset brain region is the core reflection area of ​​slow wave activity in deep sleep, ensuring the targeted nature of signal acquisition; after processing the raw EEG signals, low-frequency EEG signals are extracted, accurately locking the core signals representing slow wave activity in deep sleep and filtering out irrelevant EEG components, and the slow wave features generated based on the power and amplitude of the low-frequency EEG signals can intuitively and quantitatively reflect the strength of slow wave activity, improving the targeting and accuracy of feature extraction.

[0010] In one embodiment, the respiratory rhythm stability features and EEG slow wave features are time-synchronized, and the resulting joint features are input into a first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result, specifically including: Based on the acquisition timestamp, the respiratory rhythm stability features and EEG slow wave features are synchronized in time, and the synchronized respiratory rhythm stability features and EEG slow wave features are normalized to obtain joint features. The joint features are input into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model can perform classification prediction based on the received joint features and output the user's stimulus state judgment result; wherein, the stimulus state judgment result is either a state that is to be confirmed as stimulable or a state that is not stimulable.

[0011] In the above scheme, the timing of respiratory and EEG features is synchronized based on the collection timestamp, ensuring that the two types of features match the user's state at the same sleep time, avoiding feature fusion distortion caused by time misalignment, and laying the foundation for the effectiveness of joint features. The synchronized features are normalized to eliminate the differences in the dimensions and numerical ranges of different features, preventing a single feature from dominating the model's judgment, and ensuring the fairness and effectiveness of feature fusion. The joint features are input into a pre-trained artificial intelligence algorithm model for classification and prediction. By utilizing the generalization ability of the model trained on massive labeled samples, accurate and automated judgment of stimulus states is achieved. Compared with single feature judgment, the accuracy and stability of stimulus state recognition are greatly improved, and the classification output format is simple and clear, adapting to the real-time requirements of subsequent stimulus triggering.

[0012] In one implementation, continuously collected real-time EEG signals are input to a second artificial intelligence algorithm model for signal analysis. When the user's arousal state analysis result output by the second artificial intelligence algorithm model is arousible, a sound stimulation task is executed within a preset time delay, specifically including: Extract low-frequency EEG signals from real-time EEG signals within a preset frequency range and obtain the peak temporal characteristics of the low-frequency EEG signals; The average period length of the slow wave is calculated based on several consecutive peak time series characteristics, and the average period length of the slow wave is used as the time window for slow wave phase prediction. When the amplitude of the low-frequency EEG signal shows a trend of increasing from low to high, the user's arousal state analysis result is determined to be arousable, and the sound stimulation task is executed within the slow-wave phase prediction time window; wherein, the time interval between generating the phase analysis result and executing the sound stimulation task does not exceed the preset time delay, and the user's arousal state analysis result is either arousable or unexciteable.

[0013] In the above scheme, real-time EEG signals are input into a second artificial intelligence algorithm model for intelligent analysis, accurately extracting low-frequency EEG signals and their peak temporal characteristics within a preset frequency band, and calculating the slow-wave phase prediction time window accordingly. This allows for accurate determination of the user's excitable state based on the trend of EEG signal amplitude changes. Simultaneously, it requires that the time interval between generating the user's excitable state analysis result and the stimulus execution does not exceed a preset delay, achieving precise synchronization between sound stimulation and the rising phase of the slow wave. Furthermore, the low-latency design avoids stimulation failure caused by phase shift, ensuring the effectiveness of stimulation intervention.

[0014] In one implementation, the sound stimulation task is terminated when the stability characteristic of the first respiratory rhythm or the slow wave characteristic of the first EEG does not meet the preset requirements. Specifically, this includes: Extract the degree of fluctuation of the first respiratory cycle from the stability features of the first respiratory rhythm, and the first amplitude from the slow wave features of the first EEG. The sound stimulation task ends when the fluctuation of the first breathing cycle exceeds the preset threshold, or when the first amplitude decreases within a preset number of consecutive breathing cycles.

[0015] In the above scheme, the two core indicators that best reflect the stability of sleep and the effectiveness of stimulation are selected as monitoring objects: the degree of respiratory cycle fluctuation and the amplitude of slow wave EEG. The scheme adopts the judgment principle of OR logic. As long as the degree of respiratory cycle fluctuation exceeds the preset threshold or the amplitude of slow wave decreases in a continuous cycle, the stimulation is terminated immediately. This timely termination of stimulation avoids the risk of awakening caused by stimulation when the respiratory rhythm is disordered, and also prevents the interference of ineffective stimulation on normal sleep structure when the amplitude of slow wave continues to decrease. This forms a closed-loop safe regulation based on physiological characteristics, which greatly improves the safety and scientific nature of sleep intervention.

[0016] In one implementation, the sleep intervention method also includes controlling a sound stimulation task based on the cumulative number of stimulations within a single sleep cycle, specifically: The total number of sound stimulation tasks performed within a single sleep cycle is calculated cumulatively. When the total number of sound stimulation tasks exceeds a preset upper limit threshold, the sound stimulation task is terminated. The preset upper limit threshold is obtained by fitting a large number of sleep intervention data samples from users. The sleep intervention data samples include the user's sleep intervention effect, the maximum cumulative number of effective stimulations within a single sleep cycle, and the sleep duration.

[0017] The above solution quantitatively limits the number of stimulation tasks by accumulating the total number of stimulation tasks within a single sleep cycle and setting a preset upper limit threshold. This avoids high-frequency overstimulation within the same sleep cycle, prevents frequent stimulation from interfering with the user's normal sleep structure, and further reduces the risk of awakening. Secondly, this application also provides a sleep intervention device based on joint features, including: a first extraction module, a second extraction module, a state judgment module, a task execution module, and a task control module; The first extraction module is used to extract the first feature from the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features; The second extraction module is used to extract the second feature from the real-time EEG signal to obtain the slow wave features of the EEG; wherein, the real-time EEG signal is acquired by an electrode unit designed in a user-preset area; The state judgment module is used to synchronize the respiratory rhythm stability feature and the slow wave feature of EEG in time when the respiratory rhythm stability feature is lower than a preset threshold, and input the obtained joint feature into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state; The task execution module is used to input continuously collected real-time EEG signals into the second artificial intelligence algorithm model for signal analysis when the first artificial intelligence algorithm model outputs several consecutive stimulus state judgment results as unconfirmed stimulable states. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the sound stimulation task is executed within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG characteristics, and the second sample labels are the user's arousal states. The task control module is used to terminate the sound stimulation task when the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements; wherein, the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

[0018] Thirdly, this application also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the above-described sleep intervention method based on joint features.

[0019] Fourthly, this application also provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to perform the above-described sleep intervention method based on joint features. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of a sleep intervention method based on joint features provided in one embodiment of the present invention; Figure 2 This is a schematic diagram of a sleep intervention device based on joint features provided in one embodiment of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] The terms "first" and "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0024] First, some of the terms used in this application will be explained to facilitate understanding by those skilled in the art.

[0025] (1) RR interval: Specifically refers to the respiratory RR interval, which is the time interval between the peaks of two consecutive R respiratory waves in the respiratory waveform collected in the electrocardiogram. It is a core indicator reflecting the heart rhythm and heart rate respiratory rhythm. Its essence corresponds to a complete atrial → ventricular electrical activity cycle (i.e., cardiac respiratory cycle) of the heart from one inspiratory phase to the next inspiratory phase in continuous breathing.

[0026] Example 1 See Figure 1 , Figure 1 This is a flowchart illustrating a sleep intervention method based on joint features according to an embodiment of the present invention. The embodiment of the present invention provides a sleep intervention method based on joint features, including steps 101 to 105, each step being as follows: Step 101: Extract the first feature from the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features.

[0027] In this embodiment of the invention, a contactless method is used to collect the user's real-time respiratory signal, and feature extraction is performed to obtain respiratory rhythm stability features. The contactless collection method eliminates the need for skin contact, reducing the burden of use from the source and making it suitable for long-term home sleep monitoring scenarios. Furthermore, the respiratory signal collected without contact is a non-electrophysiological signal, exhibiting strong anti-interference capabilities, resulting in more stable extracted feature data and reducing the impact of signal noise on subsequent judgments.

[0028] In one embodiment, a first feature extraction is performed on the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features. Specifically, this includes: detecting the user's chest and abdominal displacement data using contactless technology, generating a raw respiratory signal to characterize the respiratory cycle based on the chest and abdominal displacement data; preprocessing the raw respiratory signal to obtain a real-time respiratory signal; and identifying the respiratory rhythm stability features of the real-time respiratory signal. The respiratory rhythm stability features include the duration of a single respiratory cycle, the degree of respiratory cycle fluctuation, and the stability of respiratory frequency.

[0029] In this embodiment of the invention, contactless respiratory signal acquisition is achieved using millimeter-wave radar. Its working principle is as follows: a radar transmitting module radiates millimeter-wave detection waves of a specific frequency band. When these waves illuminate the user's chest and abdomen, they are reflected due to the rhythmic rise and fall of the chest and abdomen caused by breathing. The phase and frequency of the reflected waves change synchronously with the chest and abdomen displacement. After the radar receiving module captures the reflected waves, it calculates the signal difference between the transmitted and reflected waves to accurately detect continuous chest and abdominal displacement data. Alternatively, this millimeter-wave radar can be replaced by ultra-wideband radar, pressure sensors, or optical sensors. It should be noted that the radar is generally installed directly facing the user's chest and abdomen area, and can be placed on a bedside table, bed frame railing, or under a pillow to avoid detection angle deviation. During installation, the radar probe should be kept horizontal, with the vertical angle between it and the user's chest and abdomen not exceeding 30° to reduce environmental obstruction and signal attenuation. Provided the installation requirements are met, the chest and abdominal displacement data detected by the radar is a continuous waveform that changes over time, perfectly synchronized with the human inhalation-exhalation cycle. A signal conversion algorithm maps the physical displacement into electrical / digital signals, generating a raw respiratory signal that directly characterizes the changing patterns of the respiratory cycle. During the acquisition process, the raw respiratory signal inevitably encounters three types of interference: first, circuit noise from the radar itself; second, environmental interference such as indoor airflow and slight furniture movement; and third, human interference caused by minor body movements during sleep, such as turning over and blinking. These interferences can mask the true pattern of the respiratory cycle, leading to distorted results if directly used for feature extraction. Therefore, targeted preprocessing of the raw respiratory signal is necessary to filter out invalid interference, restore the true respiratory signal characteristics, and ultimately obtain a clean, continuous, and regular real-time respiratory signal. The specific preprocessing operations include filtering and denoising: (1) using low-pass filtering to effectively filter high-frequency environmental noise and body micro-movement interference, while retaining the main components of the respiratory signal; (2) using wavelet denoising / moving average denoising to eliminate random spike noise in the signal to smooth the signal waveform and make the peaks and troughs of the respiratory cycle clearer. The RR interval is defined as the duration of a single respiratory cycle, specifically the time difference between two adjacent feature points in the real-time respiratory signal. The preprocessed real-time respiratory signal is scanned frame by frame. First, a reasonable peak threshold is set based on the statistical mean and standard deviation of the signal amplitude. This peak threshold is usually 1.2-1.5 times the peak mean. Small fluctuations in the signal are filtered out, and then local extreme points exceeding the threshold are identified and determined as respiratory peaks. Subsequently, each identified peak is marked with a millisecond-level global timestamp. Finally, the timestamp difference between two adjacent peaks is calculated. This difference is the single RR interval, in seconds. Then, a sliding window of RR intervals of preset length needs to be set. This window is a first-in-first-out dynamic window. Every time a new RR interval is collected, the oldest one in the window is immediately removed, so that the number of RR intervals in the window is always kept fixed, so as to realize the real-time dynamic update of features and adapt to the dynamic changes of breathing rhythm during sleep.The dispersion statistics of the RR interval series within the window are directly performed. The coefficient of variation is used as the criterion for evaluating the degree of respiratory cycle fluctuation. The coefficient of variation is the ratio of the standard deviation of a set of data to the arithmetic mean of the set of data, and is used to reflect the relative dispersion of the data. Then, the formula is used. Each RR interval within the window is converted into a respiratory rate to obtain a continuous respiratory rate sequence, which is then subjected to statistical analysis; among which, Respiratory rate, The RR interval is used as the criterion for assessing respiratory rate stability. The maximum range of frequency fluctuation and the relative rate of change of frequency in the respiratory rate sequence are used as the criteria for evaluating respiratory rate stability.

[0030] For example, a threshold method combined with an extreme point detection method is used to detect peaks in real-time respiratory signals: the peak threshold is set to 1.3 times the average signal amplitude, the peaks of each respiratory cycle are identified and marked with global millisecond-level timestamps, in the following order: 1719800000000ms, 1719800004900ms, 1719800009800ms, 1719800014700ms, 1719800019600ms... The time difference between adjacent peaks is calculated to obtain a continuous RR interval sequence: 4.9s, 4.9s, 4.9s, 4.9s, 5.0s, 4.8s... A sliding window with a preset length of 20 RR intervals is set, and the RR interval sequence within the window is: [4.9,4.9,4.9,4.9,5.0,4.8,4.9,5.0,4.9,4.9,5.0,4.8,4.9,4.9,5.0,4.8,4.9,4.9,5.0,4.8,4.9,5.0,4.9,4.9]. Mean. =4.91s, standard deviation ≈0.07s, coefficient of variation =(0.07 / 4.91)×100%≈1.43%, meaning the degree of fluctuation in the respiratory cycle is 1.43%. Converting the RR interval to a respiratory rate sequence, the mean... ≈12.22 times / minute, maximum frequency 12.5 times / minute, minimum frequency 12.0 times / minute, maximum fluctuation range ΔRESPmax = 0.5 times / minute, relative rate of change R RESP =(0.5 / 12.22)×100%≈4.09%, that is, the respiratory rate stability is "fluctuation range of 0.5 breaths / minute, relative change rate of 4.09%". Finally, the respiratory rhythm stability feature vector is generated [1.43%, 0.5 breaths / minute, 4.09%].

[0031] Step 102: Perform second feature extraction on the real-time EEG signal to obtain slow wave features of EEG; wherein, the real-time EEG signal is acquired by electrode units designed in the user's preset area.

[0032] In this embodiment of the invention, real-time EEG signals are acquired and feature extraction is performed using an electrode unit to obtain slow-wave EEG features. Preferably, the frontal region can be selected for EEG acquisition, abandoning the complex layout of traditional multi-lead systems, resulting in extremely low wearing burden and suitability for home use. The extraction of frontal EEG slow-wave features focuses on frequency bands related to deep sleep, accurately extracting the core slow-wave features reflecting deep sleep, directly locking the target point of sound stimulation, providing a core basis for judging the feasibility of stimulation, and ensuring that stimulation can effectively enhance slow-wave activity.

[0033] In one embodiment, raw EEG signals from the user at a preset brain region are acquired based on at least one pair of electrode units disposed on the surface of the user's scalp; the raw EEG signals are processed to obtain real-time EEG signals and low-frequency EEG signals within a preset frequency range are extracted; wherein, the low-frequency EEG signals are used to characterize slow-wave activity; and slow-wave EEG features are generated based on the power and amplitude of the low-frequency EEG signals.

[0034] In this embodiment of the invention, non-invasive acquisition of EEG signals is achieved through local electrode layout. The electrode unit adopts a flexible patch design, which gently adheres to the scalp to conduct electrophysiological signals, eliminating the complex wearing method of traditional multi-lead EEG, thereby reducing the user's burden and adapting to long-term home sleep monitoring scenarios. The electrode unit needs to be placed on the surface of the user's forehead skin and must contain at least one pair of electrodes. The electrodes correspond to the preset brain regions of the international 10-20 EEG system, with the core being the frontal pole region Fp1 and Fp2, which can be paired with the temporal region T3 and T4 as reference electrodes. This brain region is the core reflection area of ​​slow-wave activity in deep sleep and can accurately capture low-frequency EEG signals related to deep sleep, meeting the slow-wave feature extraction requirements without the need to collect data from other brain regions such as the parietal and occipital lobes. During the acquisition process, raw EEG signals inevitably become contaminated with various types of interference noise, such as power frequency interference, electrooculogram (EOG) interference, electromyogram (EMG) interference, and electrode contact noise. This noise can mask the true characteristics of slow-wave activity. Furthermore, the raw signal contains a large number of high-frequency EEG components unrelated to deep sleep. Therefore, signal processing is necessary to restore a clean, real-time EEG signal. The core signal processing operation consists of two steps: The first step is basic signal processing, obtaining the real-time EEG signal. This involves sequentially performing signal amplification, analog-to-digital conversion, power frequency interference suppression, and noise filtering. First, a preamplifier amplifies the weak raw EEG signal to a identifiable range. Then, an analog-to-digital converter transforms the analog signal into a digital signal. Next, notch filtering suppresses 50 / 60Hz power frequency interference. Finally, wavelet denoising and adaptive filtering algorithms eliminate non-EEG interference such as EOG and EMG, ultimately resulting in a real-time EEG signal with no significant noise and a clear waveform. The second step is frequency band selection, which involves extracting low-frequency EEG signals and performing bandpass filtering on the real-time EEG signals. The filtering range is precisely locked within a preset frequency range of 0.3-10Hz, which is the core frequency range of slow waves in deep sleep. High-frequency EEG components and extremely low-frequency drift signals outside this range are filtered out. The resulting low-frequency EEG signal can directly characterize the user's slow-wave activity, and its waveform fluctuations, amplitude, and power changes are highly correlated with the intensity of slow-wave activity in deep sleep. Slow-wave EEG features are extracted based on a sliding time window, which is aligned with the window for extracting respiratory features, enabling real-time dynamic updates of the features to adapt to the dynamic changes in slow-wave activity during sleep.Slow-wave power is a core indicator for measuring the intensity of slow-wave activity. It represents the energy of low-frequency EEG signals in the 0.5-8Hz range per unit frequency. During calculation, power spectrum analysis is performed on the low-frequency EEG signals within the sliding window to extract the average power value of the 0.5-8Hz band. The higher the slow-wave power value, the stronger the slow-wave activity during deep sleep, and the more feasible it is to enhance slow waves through sound stimulation. Slow-wave amplitude is a direct indicator for measuring the degree of fluctuation of slow-wave waveforms. It represents the voltage amplitude change of low-frequency EEG signals within the sliding window. During calculation, the average value of the difference between the peak and trough values ​​of low-frequency EEG signals within the window, or the statistical value of the effective amplitude, is extracted. The more stable and higher the slow-wave amplitude, the more regular the slow-wave activity, and the more significant the effect of stimulation on enhancing slow waves. Finally, the slow wave power and slow wave amplitude values ​​calculated within the sliding window are combined to form a standardized EEG slow wave feature. This feature is a numerical vector and is the core basis for determining whether a user has the feasibility of sound stimulation. Only when the slow wave power and amplitude reach the preset threshold does it indicate that the user has effective slow wave activity, and only then is it possible to enhance slow waves and prolong deep sleep time through sound stimulation.

[0035] For example, a flexible forehead patch electrode is worn by the user. The electrode unit includes a pair of sampling electrodes, corresponding to Fp1 (left frontal pole) and Fp2 (right frontal pole) of the international 10–20 system, respectively. The electrodes are placed 1-2 cm above the brow bone on the user's forehead, ensuring a good, seamless fit with the skin. After the system detects that the user has entered a sleep stage, it automatically adjusts the sampling frequency to 250 Hz. The electrode unit captures the electrical activity of neurons in the frontal pole region of the forehead, converting it into a continuous analog electrical signal to obtain the raw EEG signal. The signal has no obvious contact noise and can initially identify low-frequency waveform fluctuations. The original EEG signal was processed sequentially as follows: ① The preamplifier amplified the signal by 1000 times and then converted it into a digital EEG signal by 16-bit analog-to-digital conversion; ② A 50Hz notch filter was used to suppress power frequency interference and wavelet denoising was used to eliminate the oculomotor interference caused by the user's slight blinking; ③ The denoised digital signal was subjected to a 0.3-10Hz bandpass filter to filter out irrelevant high-frequency components such as 10-13Hz alpha waves and 14-30Hz β waves, and finally obtained a low-frequency EEG signal representing slow wave activity. A 300ms sliding time window was set and aligned with the extraction window of the respiratory rhythm stability characteristics to facilitate subsequent time synchronization. The 0.5-8Hz low-frequency EEG signal within the window was quantitatively analyzed: (1) Slow wave power was calculated: Power spectrum analysis was performed on the low-frequency signal within the window, and the average power value of the 0.5-4Hz frequency band was extracted, with a result of 18μV² / Hz; (2) Slow wave amplitude was calculated: The average value of the peak-to-trough difference of the low-frequency signal within the window was extracted, with a result of 22μV. By combining the two quantization values, the final generated EEG slow wave feature vector is [18μV² / Hz, 22μV]. This feature can be directly used for time synchronization and fusion with the respiratory rhythm stability feature, serving as the core feasibility basis for judging the user's stimulable state.

[0036] Step 103: When the respiratory rhythm stability feature is lower than the preset threshold, the respiratory rhythm stability feature and the slow wave feature of EEG are synchronized in time, and the resulting joint feature is input into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state.

[0037] In this embodiment of the invention, meeting the standard for respiratory rhythm stability is a prerequisite. The respiratory rhythm stability feature and slow-wave EEG feature are time-synchronized and calibrated, then fused into a joint feature and input into a pre-trained neural network model. The model outputs the user's current sleep stimulation state judgment result. Using respiratory rhythm stability feature as a pre-screening condition eliminates cases of abnormal respiratory rhythm, reducing the probability of invalid judgments and inappropriate stimulation from the source. Time synchronization matches the two types of features to the user's state at the same sleep time, avoiding judgment errors caused by time misalignment and ensuring the effectiveness of the joint feature. Dual feature fusion compensates for the shortcomings of single EEG feature judgment, which is susceptible to noise and individual differences, improving the accuracy and stability of stimulation state recognition. Combined with a neural network model trained on massive labeled samples, it has good generalization ability and can adapt to the individual differences of different users, such as the general population and people with cognitive impairments, reducing the risk of missed or false judgments.

[0038] In one embodiment, the respiratory rhythm stability features and EEG slow wave features are time-synchronized, and the resulting joint features are input into a first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs a user's stimulus state judgment result. Specifically, this includes: synchronizing the respiratory rhythm stability features and EEG slow wave features based on the acquisition timestamp, and normalizing the synchronized respiratory rhythm stability features and EEG slow wave features to obtain joint features; inputting the joint features into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model performs classification prediction based on the received joint features and outputs a user's stimulus state judgment result; wherein, the stimulus state judgment result is either a state to be confirmed as stimulable or a state as not to be stimulated.

[0039] In this embodiment of the invention, using the timestamp of one type of feature as a benchmark, the respiratory rhythm stability feature is preferred. In the dataset of the other type of feature, feature values ​​with a timestamp difference ≤ 50ms are searched, and pairing is completed according to the nearest matching principle. If a very small number of features are missing, linear interpolation is used to complete the feature values ​​corresponding to the timestamp, ensuring that each time node has paired respiratory and EEG features. This method achieves accurate matching between the two types of features and the user's sleep state at the same time, avoiding feature fusion distortion caused by time misalignment. The synchronized features are normalized, mapping all feature values ​​to a standardized range of 0-1. The features after time synchronization and normalization are combined according to a fixed dimension to form a joint feature, which is a purely numerical vector and the standard input format for neural network models. Then, a pre-trained neural network model performs binary classification prediction on the standardized joint feature, automatically outputting the user's current sleep stimulation state. The first artificial intelligence algorithm model can use lightweight networks such as random forests, logistic regression, backpropagation neural networks, and shallow CNN networks, adaptable to the hardware computing capabilities of portable home devices, and trained on a massive labeled training dataset. The training dataset takes respiratory-EEG joint features from different users as input and outputs corresponding stimulus state labels, including only two categories: "to be confirmed as stimulable" and "not stimulable." Labels are assigned by professionals using polysomnography monitoring results to ensure the accuracy of the training data. After training, the model exhibits good generalization ability, adapting to individual differences among users such as the general population, those with mild cognitive impairment, and the elderly. It effectively identifies valid information within the features, reducing the risk of missed or false positives from single-feature judgments. The standardized joint features are input into a pre-trained neural network model along fixed dimensions. Based on the feature patterns learned during training, the model extracts features, calculates weights, and performs classification inference on the input joint features, ultimately outputting a single stimulus state judgment result. The result is either "to be confirmed as stimulable" or "not stimulable": when the model determines that the joint features simultaneously satisfy both the EEG slow-wave feature reaching the stimulable threshold and the respiratory rhythm stability feature reaching the safety threshold, it outputs "to be confirmed as stimulable"; if either feature fails to reach the threshold, such as no effective slow-wave activity or unstable respiratory rhythm, it outputs "not stimulable." The model's inference process is real-time and fast, with inference time much shorter than the sliding window period for feature extraction, ensuring the real-time nature of sleep intervention and meeting the needs of closed-loop regulation.

[0040] Step 104: When the first artificial intelligence algorithm model outputs several consecutive stimulation state judgment results, all of which are unconfirmed stimulable states, the continuously collected real-time EEG signals are input to the second artificial intelligence algorithm model for signal analysis. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the sound stimulation task is executed within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG characteristics, and the second sample labels are the user's arousal states.

[0041] In this embodiment of the invention, the first artificial intelligence algorithm model outputs a confirmed stimulable state several times consecutively as a trigger condition. The real-time EEG signal is analyzed, and if the user's current stimulable state is determined to be stimulable, a sound stimulation task is triggered and executed within a preset time delay. The core is to lock in the appropriate timing through multiple rounds of state verification, and then accurately match the slow-wave rising phase to implement sound stimulation. Using several consecutive stimulation state judgments as a precondition for phase analysis, rather than a single judgment result, can effectively filter out misjudgments caused by occasional feature fluctuations, ensuring that subsequent operations are only initiated when the user's sleep state continuously meets the stimulation conditions, thus improving the reliability of stimulation triggering.

[0042] In one embodiment, continuously acquired real-time EEG signals are input to a second artificial intelligence algorithm model for signal analysis. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousible, a sound stimulation task is executed within a preset time delay. Specifically, this includes: extracting low-frequency EEG signals from the real-time EEG signals within a preset frequency range and obtaining the peak temporal characteristics of the low-frequency EEG signals; calculating the average slow-wave period length based on several consecutive peak temporal characteristics, and using the average slow-wave period length as the slow-wave phase prediction time window; when the amplitude in the low-frequency EEG signals shows a trend of increasing from low to high, determining that the user arousal state analysis result is arousible, and executing the sound stimulation task within the slow-wave phase prediction time window; wherein, the time interval between generating the phase analysis result and executing the sound stimulation task does not exceed a preset time delay, and the user arousal state analysis result is either arousible or unarousible.

[0043] Peak time sequence features refer to the millisecond-level global timestamps corresponding to the peaks of each slow wave waveform in low-frequency EEG signals. Slow waves, as a core EEG feature of deep sleep, exhibit a regular rising-peaking-falling cyclical change in waveform. Peaks are key feature points of the slow wave cycle, accurately marking the end of each slow wave cycle and the beginning of the next. In this embodiment of the invention, continuous peak detection is performed on the real-time acquired low-frequency EEG signals, and the timestamp of each peak is recorded to form a continuous peak time sequence feature sequence. This sequence directly reflects the cyclical change pattern of slow waves and is the core basis for calculating the average cycle length. The average slow wave cycle length refers to the average duration of several consecutive slow wave cycles. During calculation, the most recent 5-10 consecutive slow wave peak time sequence features are selected, and the timestamp difference between two adjacent peaks is calculated sequentially, i.e., the length of a single slow wave cycle. The arithmetic mean of these cycle lengths is then calculated to obtain the average slow wave cycle length. This average value represents the average cycle length of slow waves in the user's current deep sleep stage. This window is directly used as the slow-wave phase prediction time window, with its length exactly matching the average slow-wave cycle length. Its purpose is to limit the time range for phase recognition and stimulus triggering, ensuring that the stimulus always falls within a single slow-wave cycle and preventing cross-cycle stimulation from causing ineffectiveness. For example, if the average slow-wave cycle length is 800ms, the phase prediction window is 800ms. The system only identifies the rising phase and triggers the stimulus within this window, ensuring the timing of the stimulus matches the slow-wave cycle. Slow-wave phases are divided into rising phase, peak phase, and falling phase. The rising phase is the optimal time for sound stimulation, as it maximizes the amplification of slow-wave amplitude and prolongs deep sleep. Within the set slow-wave phase prediction time window, the amplitude trend of low-frequency EEG signals is continuously monitored: when the amplitude is detected to continuously increase from a continuous low value, such as a trough in the slow-wave cycle, towards a high value, it is immediately identified as a rising phase. This determination does not require complex phase angle calculations; it relies solely on amplitude trend judgment, adapting to the lightweight computing needs of home devices and enabling rapid recognition and low-latency response. It is important to note that the judgment process filters out brief amplitude fluctuations, such as instantaneous amplitude increases caused by electromyographic interference, and only identifies amplitude increase trends lasting ≥50ms to avoid misjudgment. Once a rising phase is determined, the sound stimulation task is triggered within a preset delay, typically ≤30ms. This delay is much shorter than the duration of the slow-wave rising phase, ensuring that the stimulation falls within the effective range of the rising phase. The preset delay setting must take into account the hardware response speed, such as the audio module startup time and phase synchronization requirements, and is typically set to 10-30ms, and can be adaptively adjusted according to the actual response speed of the device. The execution parameters of the stimulation task must be adapted to the needs of deep sleep intervention, generally using short, low-intensity, narrow-band sound pulses.For example, a pure tone of 30-50Hz, lasting 50-100ms, with a volume ≤40dB, should be used to avoid high-intensity / prolonged sound triggering user arousal. Simultaneously, the stimulus task is tied to a slow-wave phase prediction time window, executing only within that window. If the rising phase is not detected outside the window, the stimulus is abandoned, and the next slow-wave cycle is used for reassessment. The core requirement is that the time interval between rising phase determination and stimulus task execution must strictly not exceed a preset delay to ensure precise synchronization between the stimulus and the rising phase, maximizing the slow-wave enhancement effect. It should be noted that the second AI algorithm model can also use lightweight networks such as random forests, logistic regression, backpropagation (BP) neural networks, and shallow CNN networks.

[0044] Step 105: When the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements, the sound stimulation task is terminated; wherein, the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

[0045] In this embodiment of the invention, during the execution of sound stimulation, the stability characteristics of respiratory rhythm and slow-wave characteristics of electroencephalography (EEG) are continuously monitored. If the trend of either characteristic does not meet the preset requirements, the sound stimulation task is immediately terminated. Simultaneous monitoring of both characteristics ensures the safety and effectiveness of the stimulation from two dimensions. Immediate termination of stimulation when characteristics are abnormal can promptly avoid over-stimulation or ineffective stimulation, preventing interference with the user's normal sleep structure and further reducing the risk of awakening. This approach abandons a fixed stimulation execution mode, dynamically adjusting stimulation behavior based on real-time physiological changes during stimulation, rather than mechanically executing stimulation tasks. This allows sleep intervention to better match the user's real-time sleep state, balancing the effectiveness and flexibility of the intervention.

[0046] In one embodiment, the sound stimulation task is terminated when the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements. Specifically, this includes: extracting the first respiratory cycle fluctuation degree in the first respiratory rhythm stability feature and the first amplitude in the first EEG slow wave feature; and terminating the sound stimulation task when the first respiratory cycle fluctuation degree exceeds a preset threshold or the first amplitude decreases within a preset number of consecutive respiratory cycles.

[0047] In this embodiment of the invention, the core indicators that best reflect the stability of sleep state and the effectiveness of stimulation are selected from the extracted respiratory rhythm stability features and EEG slow wave features as the monitoring objects. The degree of respiratory cycle fluctuation is a core indicator reflecting the stability of the user's sleep state. This indicator directly quantifies the dispersion of continuous respiratory cycles, and its numerical changes directly reflect changes in the user's sleep state, such as impending awakening or slight body movements. Compared to other respiratory indicators, it is more sensitive and timely in predicting the risk of awakening. Therefore, in real-time monitoring of the stimulation process, only the degree of respiratory cycle fluctuation is extracted as the core monitoring indicator on the respiratory side. Slow wave amplitude is a key indicator reflecting the effectiveness of sound stimulation. The core goal of sound stimulation is to enhance slow wave activity and prolong deep sleep time, and changes in slow wave amplitude directly reflect the strength of slow wave activity: if the stimulation is effective, the slow wave amplitude will remain stable or slightly increase; if the slow wave amplitude continues to decrease, it indicates that the stimulation has not had an enhancing effect and may even interfere with slow wave activity. Real-time changes in slow wave amplitude are easier to capture and calculate, adapting to the high-frequency real-time monitoring needs during stimulation; therefore, it is selected as the core monitoring indicator on the EEG side. A fixed safety preset threshold is set for the degree of respiratory cycle fluctuation. This threshold is based on a large amount of sleep monitoring data and can be adapted to different population groups: the threshold for the general population is ≤5%, and the threshold for the population with mild cognitive impairment / elderly is ≤8%. This threshold represents the critical value for a stable sleep state for the user. During stimulation, if the value of the respiratory cycle fluctuation detected in real time exceeds the preset threshold in a single instance, or if the values ​​of two consecutive sliding windows exceed the threshold, it is immediately determined that the respiratory rhythm characteristics do not conform to the preset trend. This means that the user's respiratory rhythm has suddenly become disordered, and the risk of awakening has increased significantly. Continuing stimulation at this time is very likely to trigger the user's awakening, and the stimulation termination command is immediately triggered. A trend preset requirement is set for the slow wave amplitude: it should remain stable or increase slightly. At the same time, a preset number of continuous monitoring cycles is set, generally 3-5 slow wave cycles. During stimulation, the amplitude of slow waves is continuously tracked in units of slow wave cycles. If the amplitude shows a continuous decreasing trend within a preset number of consecutive slow wave cycles (e.g., from 22μV to 20μV, then to 18μV, without any sign of recovery), it is immediately determined that the EEG slow wave characteristics do not conform to the preset trend. This means that the current sound stimulation has not enhanced the slow waves and is therefore ineffective. Continuing to perform this stimulation will not only fail to achieve the intervention goal but may also disrupt normal slow wave activity patterns and interfere with deep sleep. Therefore, the stimulation must be terminated immediately. Once any indicator is determined to be abnormal, a termination command is immediately sent to the sound stimulation module. Upon receiving the command, the sound stimulation module will immediately stop outputting sound pulses. Simultaneously, the system will record the time and reason for the stimulation termination and return to the initial stage of continuously monitoring characteristics and determining the stimulable state. If the subsequent characteristics return to the preset trend, the sound stimulation task can be retried.

[0048] In one embodiment, the total number of sound stimulation tasks performed within a single sleep cycle is calculated cumulatively. When the total number of sound stimulation tasks exceeds a preset upper limit threshold, the sound stimulation task is terminated. The preset upper limit threshold is obtained by fitting a large number of sleep intervention data samples from users. The sleep intervention data samples include the user's sleep intervention effect, the maximum cumulative number of effective stimulations within a single sleep cycle, and the sleep duration.

[0049] In this embodiment of the invention, a single sleep cycle does not refer to the sleep stage cycle at the EEG level, but rather a statistical cycle of fixed length set by combining the practicality of home sleep monitoring and the intervention patterns of sound stimulation. This cycle length is based on the physiological patterns of human sleep and a large amount of intervention experimental data, while also supporting personalized adaptation: the basic single sleep cycle length for the general population and those with mild cognitive impairment / the elderly is uniformly set at 30 minutes. This length conforms to the basic cyclical patterns of human sleep stages and avoids insufficient stimulation due to a cycle that is too short, or excessive stimulation due to a cycle that is too long. Furthermore, the single sleep cycle length can be fine-tuned to 20 minutes or 40 minutes based on the user's sleep characteristics, such as the duration of slow-wave activity and the stability of respiratory rhythm, to adapt to the sleep differences of different users. The statistics for a single sleep cycle are continuously rolling, meaning that the statistics for the next 30-minute cycle begin immediately after the previous one ends. For example, 00:00-00:30 is the first sleep cycle, 00:30-01:00 is the second sleep cycle, and so on, ensuring uninterrupted statistics of stimulation counts throughout the entire process and avoiding statistical gaps. When a complete process of "rising phase determination - triggering within the preset time delay - normal sound stimulation output" is completed, and the stimulation does not terminate prematurely due to abnormal breathing / EEG characteristics, it is counted as a valid stimulation and included in the cumulative count. If the stimulation terminates immediately after triggering due to abnormal characteristics, or is not triggered during the rising phase, it is not counted in the cumulative count to avoid invalid operations interfering with the statistical results. An independent stimulation count cumulative counter is set for each sleep cycle. The counter automatically increments by 1 for each valid stimulation completed within the cycle, achieving real-time accumulation. When a single sleep cycle ends, the counter is immediately reset to zero, and a new count begins in the new sleep cycle, ensuring that the count statistics for each cycle are independent and do not cross-cycle. Preferably, while accumulating the number of stimulations, the trigger time, stimulation parameters, and corresponding slow-wave cycle of each effective stimulus can also be recorded to form a stimulus count traceability table. This facilitates subsequent analysis of sleep intervention effects and provides data support for personalized threshold adjustments. A preset upper limit threshold is set for the cumulative number of stimulations within a single sleep cycle. This threshold is determined based on population differences and sleep stage characteristics, taking into account both intervention effectiveness and sleep safety. The preset upper limit threshold for the cumulative number of stimulations within a single sleep cycle for the general population is 10 times; for individuals with mild cognitive impairment / the elderly, whose sleep states are more sensitive and easily disturbed by stimuli, the threshold is appropriately reduced to 8 times to minimize the impact of stimulation on their sleep structure. When the cumulative number of stimulations within a single sleep cycle reaches or exceeds the preset upper limit threshold, a termination command is immediately sent to the sound stimulation module. Regardless of whether the respiratory rhythm characteristics and EEG slow-wave characteristics at this time conform to the preset trend, subsequent sound stimulation triggering and execution are stopped. After the end of a single sleep cycle, the counter is reset to zero, and stimulation can be retried when a new cycle starts, resuming the normal intervention process.It's important to note that this threshold is not arbitrarily set, but rather derived through big data fitting: using a sleep intervention dataset containing a large number of users, the sample includes three key pieces of information: the user's actual sleep intervention effect (e.g., improvement in deep sleep duration and sleep structure), the maximum safe number of stimulations that the user can tolerate within a single sleep cycle, and the user's sleep duration / total sleep duration. By performing regression fitting and statistical analysis on this data, the optimal upper limit for the number of stimulations is obtained, ensuring the intervention effect without triggering awakenings or sleep disturbances. This approach ensures that the threshold is not a fixed value, but a scientifically derived value based on real-world population data, making it suitable for different groups such as ordinary adults, the elderly, and people with cognitive impairments. By cumulatively statistically analyzing and limiting the number of sound stimulations within a single sleep cycle, high-frequency sound stimulation within the same sleep cycle can be avoided, preventing excessive stimulation from interfering with the user's normal sleep structure and triggering awakenings, while ensuring the appropriateness and scientific validity of the sleep intervention. In this embodiment of the invention, a sleep intervention device based on joint features is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described sleep intervention method based on joint features.

[0050] In this embodiment of the invention, a computer-readable storage medium is also provided, which includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the above-described sleep intervention method based on joint features when it is running.

[0051] For example, the sleep intervention for an elderly user is performed as follows: (1) Single sleep cycle: 30 minutes is a statistical cycle, and this cycle starts from 00:00 and ends at 00:30; (2) Preset upper limit threshold: Based on the fitting of sleep intervention data of 1,000 elderly users, the intervention effect, the maximum safe number of stimulations, and the total sleep duration are considered, and the upper limit threshold for a single cycle for the elderly is finally determined to be 8 times. When the cumulative number of sound stimulation tasks reaches the preset upper limit threshold of 8 times, all analysis and stimulation triggering processes are immediately stopped. Even if the subsequent breathing and EEG characteristics are still normal, sound stimulation will not be performed again until the end of this cycle at 00:30 and the counter is cleared. The stimulation function will be reopened for the next cycle.

[0052] For example, a computer program may be divided into one or more modules, one or more of which are stored in memory and executed by a processor to carry out the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a sleep intervention device based on joint features.

[0053] Sleep intervention devices based on joint features can be computing devices such as desktop computers, laptops, PDAs, and cloud servers. These devices may include, but are not limited to, processors, memory, and displays. Those skilled in the art will understand that the above components are merely examples of sleep intervention devices based on joint features and do not constitute a limitation on such devices. The device may include more or fewer components, combinations of certain components, or different components. For example, sleep intervention devices based on joint features may also include input / output devices, network access devices, buses, etc.

[0054] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the sleep intervention device based on joint features, connecting all parts of the device through various interfaces and lines.

[0055] The memory can be used to store computer programs and / or modules. The processor implements various functions of the sleep intervention device based on joint characteristics by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0056] In this invention, modules for sleep intervention based on joint features, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. Those skilled in the art can understand and implement this invention without any inventive effort.

[0057] This invention provides a sleep intervention method based on joint features. It uses contactless respiratory signal acquisition to extract rhythm stability features, and lightweight EEG electrode units to acquire EEG signals and extract slow-wave features. This avoids the cumbersome wearing of multi-lead devices while obtaining core features for both sleep safety and stimulation feasibility. Time synchronization eliminates feature time misalignment, and the joint features are input into an artificial intelligence algorithm model trained on massive labeled samples. The model's classification ability enables accurate judgment of stimulable states, and continuous result verification further reduces the probability of misjudgment. Compared to single EEG judgment, this significantly improves the stability of state recognition. Once a stimulable state is determined, the activation state is determined by analyzing real-time EEG signals. If an stimulable state is detected, sound stimulation is triggered within a preset low latency, achieving precise time matching between stimulation and slow-wave rhythm, maximizing slow-wave activity and prolonging deep sleep time. Simultaneously, the degree of respiratory cycle fluctuation and EEG slow-wave amplitude are continuously monitored during stimulation. Stimulation is immediately terminated if any features are abnormal, forming a real-time closed-loop control. This approach avoids the risk of awakening from both parameter design and dynamic monitoring perspectives. No complicated operation is required, and it is suitable for long-term home use. While improving the effectiveness of sleep intervention, it ensures the comfort and safety of use.

[0058] Example 2 See Figure 2 , Figure 2This is a schematic diagram of a sleep intervention device based on joint features according to an embodiment of the present invention. The sleep intervention device based on joint features provided in this embodiment includes: a first extraction module 201, a second extraction module 202, a state judgment module 203, a task execution module 204, and a task control module 205; The first extraction module 201 is used to extract the first feature from the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features; The second extraction module 202 is used to perform second feature extraction on the real-time EEG signal to obtain slow wave features of the EEG; wherein, the real-time EEG signal is acquired by an electrode unit designed in a user-preset area; The state judgment module 203 is used to synchronize the respiratory rhythm stability feature and the slow wave feature of the electroencephalogram in time when the respiratory rhythm stability feature is lower than a preset threshold, and input the obtained joint feature into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state; The task execution module 204 is used to input continuously collected real-time EEG signals into the second artificial intelligence algorithm model for signal analysis when the first artificial intelligence algorithm model outputs several consecutive stimulus state judgment results as unconfirmed stimulable states. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the sound stimulation task is executed within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG features, and the second sample labels are the user's arousal states. The task control module 205 is used to terminate the sound stimulation task when the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements; wherein, the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

[0059] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0060] This invention provides a sleep intervention device based on combined features. It uses contactless respiratory signal acquisition to extract rhythm stability features, and lightweight EEG electrode units to acquire EEG signals and extract slow-wave features. This avoids the cumbersome wearing of multi-lead devices while obtaining core features for both sleep safety and stimulation feasibility. Time synchronization eliminates feature time misalignment, and the combined features are input into an artificial intelligence algorithm model trained on massive labeled samples. The model's classification ability enables accurate judgment of stimulable states, and continuous result verification further reduces the probability of misjudgment. Compared to single EEG judgment, this significantly improves the stability of state recognition. Once a stimulable state is determined, the activation state is determined by analyzing real-time EEG signals. If an stimulable state is detected, sound stimulation is triggered within a preset low latency, achieving precise time matching between stimulation and slow-wave rhythm, maximizing slow-wave activity and prolonging deep sleep time. Simultaneously, the degree of respiratory cycle fluctuation and EEG slow-wave amplitude are continuously monitored during stimulation. Stimulation is immediately terminated if any features are abnormal, forming a real-time closed-loop control. This approach avoids the risk of awakening from both parameter design and dynamic monitoring perspectives. No complicated operation is required, and it is suitable for long-term home use. While improving the effectiveness of sleep intervention, it ensures the comfort and safety of use.

[0061] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A sleep intervention method based on joint features, characterized in that, include: The first feature is extracted from the real-time respiratory signals acquired using contactless technology to obtain respiratory rhythm stability features; A second feature extraction is performed on the real-time EEG signal to obtain slow-wave EEG features; wherein, the real-time EEG signal is acquired through electrode units designed in a user-preset area; When the respiratory rhythm stability feature is lower than a preset threshold, the respiratory rhythm stability feature and the slow wave feature of the electroencephalogram are synchronized in time, and the resulting joint feature is input into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state; When the first artificial intelligence algorithm model outputs several consecutive stimulation state judgment results, all of which are unconfirmed stimulable states, the continuously collected real-time EEG signals are input to the second artificial intelligence algorithm model for signal analysis. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the sound stimulation task is executed within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG features, and the second sample labels are the user's arousal states. When the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements, the sound stimulation task is terminated; wherein, the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

2. The joint feature-based sleep intervention method of claim 1, wherein, The first feature extraction of the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features specifically includes: The user's chest and abdominal displacement data is detected using contactless technology, and a raw respiratory signal is generated based on the chest and abdominal displacement data to characterize the respiratory cycle. The raw respiratory signal is preprocessed to obtain the real-time respiratory signal; Identify the respiratory rhythm stability characteristics of the real-time respiratory signal; wherein, the respiratory rhythm stability characteristics include the duration of a single respiratory cycle, the degree of respiratory cycle fluctuation, and the stability of respiratory rate.

3. The joint feature-based sleep intervention method of claim 1, wherein, The second feature extraction of the real-time EEG signal to obtain slow-wave EEG features specifically includes: Raw electroencephalogram (EEG) signals from the user at preset brain regions are acquired by at least one pair of electrode units placed on the surface of the user's scalp. The original EEG signal is processed to obtain the real-time EEG signal and low-frequency EEG signals within a preset frequency range are extracted; wherein, the low-frequency EEG signals are used to characterize slow-wave activity. The slow-wave characteristics of the brain are generated based on the power and amplitude of the low-frequency EEG signal.

4. The joint feature-based sleep intervention method of claim 1, wherein, The step of synchronizing the respiratory rhythm stability features and the slow-wave EEG features in time, and inputting the resulting joint features into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result, specifically includes: Based on the acquisition timestamp, the respiratory rhythm stability feature and the EEG slow wave feature are synchronized in time, and the synchronized respiratory rhythm stability feature and EEG slow wave feature are normalized to obtain the joint feature. The joint features are input into the first artificial intelligence algorithm model, so that the first artificial intelligence algorithm model performs classification prediction based on the received joint features and outputs the user's stimulus state judgment result; wherein, the stimulus state judgment result is a state to be confirmed as stimulable or a state as not stimulable.

5. The joint feature-based sleep intervention method of claim 1, wherein, The step of continuously collecting real-time EEG signals and inputting them into a second artificial intelligence algorithm model for signal analysis, and then executing a sound stimulation task within a preset time delay when the user's arousal state analysis result output by the second artificial intelligence algorithm model is arousible, specifically includes: Extract the low-frequency EEG signal from the real-time EEG signal within a preset frequency range and obtain the peak temporal characteristics of the low-frequency EEG signal; The average period length of the slow wave is calculated based on several consecutive peak time-series characteristics, and the average period length of the slow wave is used as the slow wave phase prediction time window. When the amplitude of the low-frequency EEG signal shows a trend of increasing from low to high, the user's arousal state analysis result is determined to be arousable, and the sound stimulation task is executed within the slow-wave phase prediction time window; wherein, the time interval between generating the phase analysis result and executing the sound stimulation task does not exceed the preset time delay, and the user's arousal state analysis result is either arousable or unarousable.

6. The joint feature-based sleep intervention method of claim 1, wherein, The sound stimulation task is terminated when the stability characteristic of the first respiratory rhythm or the slow wave characteristic of the first EEG does not meet the preset requirements, specifically including: Extract the degree of fluctuation of the first respiratory cycle from the first respiratory rhythm stability feature, and the first amplitude from the first slow brain wave feature; The sound stimulation task ends when the fluctuation of the first respiratory cycle exceeds a preset threshold, or when the first amplitude decreases within a preset number of consecutive respiratory cycles.

7. The joint feature-based sleep intervention method of claim 1, wherein, The sleep intervention method also includes controlling the sound stimulation task based on the cumulative number of stimulations within a single sleep cycle, specifically: The total number of sound stimulation tasks performed within a single sleep cycle is calculated cumulatively. When the total number of sound stimulation tasks exceeds a preset upper limit threshold, the sound stimulation task is terminated. The preset upper limit threshold is obtained by fitting a large number of sleep intervention data samples from users. The sleep intervention data samples include the user's sleep intervention effect, the maximum cumulative number of effective stimulations within a single sleep cycle, and the sleep duration.

8. A sleep intervention device based on combined features, characterized by include: The module comprises a first extraction module, a second extraction module, a status judgment module, a task execution module, and a task control module. The first extraction module is used to extract the first feature from the real-time respiratory signal acquired using contactless technology to obtain respiratory rhythm stability features; The second extraction module is used to extract a second feature from the real-time EEG signal to obtain slow-wave EEG features; wherein, the real-time EEG signal is acquired by an electrode unit designed in a user-preset area; The state judgment module is used to synchronize the respiratory rhythm stability feature and the slow wave feature of the electroencephalogram (EEG) in time when the respiratory rhythm stability feature is lower than a preset threshold, and input the resulting joint feature into the first artificial intelligence algorithm model so that the first artificial intelligence algorithm model outputs the user's stimulus state judgment result; wherein, the first artificial intelligence algorithm model is trained on a dataset containing a large number of first user data samples and first sample labels, the first user data samples are the user's joint features, and the first sample labels are the user's stimulus state; The task execution module is used to input continuously collected real-time EEG signals into the second artificial intelligence algorithm model for signal analysis when the first artificial intelligence algorithm model outputs several consecutive stimulation state judgment results as unconfirmed stimulable states. When the user arousal state analysis result output by the second artificial intelligence algorithm model is arousable, the module executes the sound stimulation task within a preset time delay. The second artificial intelligence algorithm model is trained on a dataset containing a large number of second user data samples and second label samples. The second user data samples are the user's slow-wave EEG features, and the second sample labels are the user's arousal states. The task control module is used to terminate the sound stimulation task when the first respiratory rhythm stability feature or the first EEG slow wave feature does not meet the preset requirements; wherein, the first respiratory rhythm stability feature is the respiratory rhythm stability feature during the execution of the sound stimulation task, and the first EEG slow wave feature is the EEG slow wave feature during the execution of the sound stimulation task.

9. A terminal device, comprising: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the sleep intervention method based on joint features as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the sleep intervention method based on joint features as described in any one of claims 1 to 7.

Images