CN119498776B

CN119498776B - Feature information extraction method and system of sleep state monitoring model

Info

Publication number: CN119498776B
Application number: CN202411482257.5A
Authority: CN
Inventors: 王子莹; 王川; 许硕贵; 徐浩丹; 王贝
Original assignee: First Affiliated Hospital of Naval Military Medical University of PLA
Current assignee: First Affiliated Hospital of Naval Military Medical University of PLA
Priority date: 2024-10-23
Filing date: 2024-10-23
Publication date: 2025-07-01
Anticipated expiration: 2044-10-23
Also published as: CN119498776A

Abstract

The invention relates to the technical field of sleep assistance, in particular to a feature information extraction method and system of a sleep state monitoring model. The method comprises the following steps of collecting multi-mode physiological signals of a tested object to obtain original sleep monitoring data, carrying out noise removal and signal enhancement processing on the original sleep monitoring data, carrying out signal segmentation to obtain multi-channel sleep signal segment data, carrying out characteristic extraction on the multi-channel sleep signal segment data based on electroencephalogram, oculogram, electromyogram, electrocardiograph and accelerometer signals to obtain multi-dimensional characteristic data, carrying out ultrasonic characteristic and acoustic characteristic extraction on the multi-channel sleep signal segment data, and generating acoustic characteristic data. According to the invention, the transition process of the sleep state is dynamically modeled and the trend is predicted, so that potential sleep disorder events can be accurately identified, and the potential sleep disorder events are subjected to influence evaluation through deep learning, thereby providing support for personalized sleep diagnosis.

Description

Feature information extraction method and system of sleep state monitoring model

Technical Field

The invention relates to the technical field of sleep assistance, in particular to a feature information extraction method and system of a sleep state monitoring model.

Background

Feature information extraction in the sleep state monitoring model is a key step for realizing accurate monitoring. By extracting rich time domain, frequency domain, time-frequency domain and nonlinear characteristics from various sensor data, the model can comprehensively describe and distinguish different sleep stages and evaluate sleep quality. With the continuous progress of sensor technology and data processing methods, the accuracy and diversity of feature extraction are further improved, and more powerful support is provided for personalized health management and accurate medical treatment.

However, in the conventional method for extracting feature information of the sleep state monitoring model, in the sleep state monitoring process, physiological signals are easily affected by noise (such as electromagnetic interference, respiration, body movement and the like), so that the quality of monitoring data is reduced, and the accuracy of feature extraction is affected. How to remove noise and extract efficient features from multi-modal signals has been a technical challenge. Sleep disorders (e.g., insomnia, sleep apnea, etc.) often occur during sleep, with different disorders appearing differently on different physiological signals. Thus, conventional models have difficulty in real-time identification and monitoring of these events. In addition, the accuracy and efficiency of sleep monitoring also cannot meet the requirements of real-time health management.

Disclosure of Invention

Accordingly, the present invention is directed to a method and a system for extracting feature information of a sleep state monitoring model, so as to solve at least one of the above-mentioned problems.

In order to achieve the above object, a feature information extraction method of a sleep state monitoring model includes the following steps:

step S1, carrying out multi-mode physiological signal acquisition on a detected object to obtain original sleep monitoring data, carrying out noise removal and signal enhancement processing on the original sleep monitoring data, and carrying out signal segmentation to obtain multi-channel sleep signal segment data;

Step S2, carrying out characteristic extraction based on electroencephalogram, oculogram, electromyogram, electrocardiogram and accelerometer signals on the multichannel sleep signal segment data to obtain multi-dimensional characteristic data;

Step S3, nonlinear dynamic characteristic extraction is carried out according to the multidimensional characteristic data to generate nonlinear characteristic data, and the acoustic characteristic data are subjected to complementary fusion processing through the nonlinear characteristic data to generate mixed sleep characteristic data;

Step S4, carrying out space-time evolution processing and event influence evaluation on the mixed sleep characteristic data to generate sleep event evaluation data, carrying out potential sleep disorder event screening according to the sleep event evaluation data, and carrying out real-time physiological index acquisition to generate sleep real-time monitoring index data;

and S5, carrying out sleep environment influence area identification on the sleep stage fluctuation data to generate sleep environment influence area data, carrying out physiological-environment influence mode coupling on the sleep environment influence area data to obtain physiological-environment coupling mode data, and combining the mixed sleep characteristic data, the physiological sleep disturbance associated data and the physiological-environment coupling mode data into characteristic information extraction data.

According to the invention, through the acquisition of the multi-mode physiological signals (EEG, EOG, EMG, ECG and the accelerometer), the sleeping state of a tested object is covered comprehensively, and the perception capability of the sleeping state change is enhanced from multiple dimensions. Preprocessing the original sleep monitoring data eliminates noise and enhances key signals, and ensures the accuracy and robustness of subsequent processing. Segmentation of the multichannel signal enables feature extraction and analysis to be performed over a finer granularity period of time, accurately capturing short-time dynamics. The multidimensional feature extraction of brain electricity, eye movement, myoelectricity, electrocardio and body movement can comprehensively reveal the changes of the brain, muscle, heart and body gestures in the sleeping process, and the acoustic features and the ultrasonic features further enhance the detection of respiratory and body movement abnormality. The nonlinear dynamics feature extraction can capture complex sleep physiological changes, and through fusion with acoustic wave features, abnormal events such as sleep apnea and the like can be more accurately identified. The system reveals the dynamic change of the sleep state through space-time evolution analysis, evaluates and identifies potential sleep disorder events in real time, and provides dynamic monitoring for important events in sleep. The fluctuation analysis of the multi-channel sleep signals is combined with physiological disturbance correlation, and the correlation between the physiological signal change and the sleep stage is deeply analyzed. Through physiological-environment coupling analysis, the influence of external environment factors such as light, noise and temperature on sleep quality is revealed, so that a more comprehensive sleep state assessment framework is constructed. Finally, based on the fusion of the multi-source signals and the multi-dimensional characteristics, the system can realize the omnibearing dynamic evaluation of the sleep quality of the individual, provide personalized intervention suggestions, and be helpful for more accurately identifying and improving the potential sleep problem.

The invention also provides a feature information extraction system of the sleep state monitoring model, which is used for executing the feature information extraction method of the sleep state monitoring model, and the feature information extraction system of the sleep state monitoring model comprises the following steps:

The system comprises a signal preprocessing module, a multi-channel sleep signal segment data acquisition module, a signal processing module and a signal processing module, wherein the signal preprocessing module is used for carrying out multi-mode physiological signal acquisition on a detected object to obtain original sleep monitoring data;

The multi-dimensional feature extraction module is used for carrying out feature extraction based on electroencephalogram, oculogram, electromyogram, electrocardiograph and accelerometer signals on the multi-channel sleep signal segment data to obtain multi-dimensional feature data;

The feature fusion module is used for carrying out nonlinear dynamic feature extraction according to the multidimensional feature data to generate nonlinear feature data, and carrying out complementary fusion processing on the acoustic feature data through the nonlinear feature data to generate mixed sleep feature data;

The sleep dynamic analysis module is used for carrying out space-time evolution processing and event influence evaluation on the mixed sleep characteristic data to generate sleep event evaluation data, carrying out potential sleep disorder event screening according to the sleep event evaluation data, carrying out real-time physiological index acquisition to generate sleep real-time monitoring index data, carrying out sleep stage fluctuation analysis and physiological disturbance association processing on the multi-channel sleep signal segment data through the sleep real-time monitoring index data, and generating physiological sleep disturbance association data;

The environment-physiological coupling module is used for carrying out sleep environment influence area identification on the sleep stage fluctuation data to generate sleep environment influence area data, carrying out physiological-environment influence mode coupling on the sleep environment influence area data to obtain physiological-environment coupling mode data, and combining the mixed sleep characteristic data, the physiological sleep disturbance related data and the physiological-environment coupling mode data into characteristic information extraction data.

The invention improves the accuracy of the original signal through noise removal and signal enhancement, and ensures the reliability of feature extraction. The signal segmentation ensures that the processing of long-time monitoring data is more efficient, and the multichannel synchronous processing ensures the time consistency of physiological signals, thereby being convenient for the fusion of multidimensional features. The extracted multidimensional physiological characteristics comprehensively reflect sleeping states, including information such as brain electricity, heart rate, eye movement, myoelectricity and body movement, and the like, and external influences such as respiration, snoring and the like can be detected by combining ultrasonic waves and acoustic characteristics, so that sleeping evaluation dimensions are enriched. By extracting nonlinear dynamics features, the system improves sensitivity to complex physiological changes and abnormal events. The combination of the multidimensional features and the acoustic wave features generates comprehensive mixed sleep feature data, and the comprehensiveness of sleep evaluation is improved. The space-time evolution analysis dynamically monitors sleep state changes and identifies potential sleep disorders or abnormal events. The real-time physiological monitoring is combined with physiological disturbance analysis, so that the system can timely detect physiological abnormality and adjust the monitoring strategy. In addition, the system identifies the influence of environmental factors on sleep, optimizes the sleep environment and personalized intervention strategies through physiological-environment coupling analysis, and finally improves the accuracy and reliability of sleep quality assessment.

Drawings

Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of a non-limiting implementation, made with reference to the accompanying drawings in which:

FIG. 1 is a flow chart illustrating steps of a feature information extraction method of a sleep state monitoring model according to the present invention;

FIG. 2 is a detailed step flow chart of step S1 in FIG. 1;

fig. 3 is a detailed step flow chart of step S2 in fig. 1.

Detailed Description

The following is a clear and complete description of the technical method of the present invention, taken in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

In order to achieve the above objective, referring to fig. 1 to 3, the present invention provides a feature information extraction method of a sleep state monitoring model, the method includes the following steps:

In the embodiment of the present invention, as described with reference to fig. 1, the step flow diagram of a feature information extraction method of a sleep state monitoring model of the present invention is provided, and in this example, the feature information extraction method of the sleep state monitoring model includes the following steps:

The embodiment of the invention utilizes multi-mode physiological signal acquisition equipment, such as a polysomnography monitor, to continuously acquire physiological signals of a tested object. These signals include electroencephalogram (EEG), oculogram (EOG), electromyogram (EMG), electrocardiogram (ECG), and accelerometer signals to comprehensively record the physiological state of the subject during sleep. During acquisition, the sampling rate of the device was set to 500Hz for EEG and EOG, 1000Hz for emg and ECG, and 100Hz for accelerometer to ensure high accuracy and high resolution of the signal. The raw sleep monitor data collected is subjected to noise removal by a digital filter and bandpass filtering (e.g., in the range of 0.5-50Hz of the EEG signal) is used to remove ambient noise and power supply interference. Then, an Empirical Mode Decomposition (EMD) method is applied to perform signal enhancement processing to improve the signal-to-noise ratio of the signal. The processed signals are segmented according to a preset time window (such as 30 seconds), and multichannel sleep signal segment data are generated, so that subsequent feature extraction and analysis are facilitated.

The embodiment of the invention respectively performs characteristic extraction on various physiological signals on the basis of the multi-channel sleep signal segment data after the signal segmentation. For electroencephalogram signals, energy features of the delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30-50 Hz) bands are extracted, and waveform features of the K complex and sleep spindle waves are detected. The eye diagram signal extracts the frequency and amplitude characteristics of the eye diagram signal through the identification of rapid and slow eye movements. Electromyographic signals assess muscle tone by analyzing the intensity and frequency characteristics of myoelectrical activity. The electrocardiogram signal adopts heart rate variability analysis to extract time domain and frequency domain indexes. The accelerometer signal is used for body movement analysis, and the characteristics of body movement frequency, intensity, duration and the like are extracted. After the multidimensional feature data are integrated, the integrated acoustic feature data are generated by combining the micro body movement features acquired by the ultrasonic sensor and the environmental sound features acquired by the high-sensitivity microphone. In particular applications, feature extraction algorithms such as wavelet transforms and Fast Fourier Transforms (FFTs) are applied to signal processing to ensure the accuracy and effectiveness of feature extraction.

After the multidimensional feature data are obtained, the embodiment of the invention adopts a nonlinear dynamics method to further extract the features. For example, multi-dimensional feature data is processed using phase space reconstruction techniques, a maximum lyapunov exponent is calculated to evaluate the stability of the system, recursive Quantitative Analysis (RQA) is used to extract the recursive rate and deterministic features, and sample entropy and approximate entropy are used to measure the complexity of the signal. Further, fluctuation features are extracted by detrending fluctuation analysis and long-range correlation analysis. And carrying out complementary fusion processing on the nonlinear characteristic data and the acoustic characteristic data, and generally adopting a characteristic fusion method based on a deep neural network to enhance the expression capability and discrimination performance of the characteristics. Finally, mixed sleep characteristic data is generated, which plays a key role in subsequent sleep event assessment and disturbance correlation analysis. In specific application, the deep neural network structure can select a multi-layer perceptron (MLP) or a Convolutional Neural Network (CNN), and network parameters are optimized through cross verification, so that the optimal characteristic fusion effect is realized.

The mixed sleep characteristic data is subjected to space-time evolution processing, and the time sequence analysis, a Hidden Markov Model (HMM) and other methods are utilized to generate sleep state space-time evolution event data. These data are used to identify potential sleep disruptions or abnormal events and to evaluate the impact of the event by a feature importance assessment method (e.g., gradient-based importance scoring) to generate sleep event assessment data. And then, based on the evaluation data, screening out potential sleep disorder events, configuring acquisition parameters of the multi-mode sensor, and acquiring key physiological indexes such as heart rate, respiratory rate and the like in real time to generate sleep real-time monitoring index data. By carrying out sleep stage fluctuation analysis on the real-time monitoring index data and the multichannel sleep signal segment data, adopting methods such as a Dynamic Time Warping (DTW) algorithm and the like, analyzing the fluctuation of the sleep stage, correlating physiological disturbance and generating physiological sleep disturbance correlation data. In a specific application scene, the acquisition frequency of the real-time monitoring index is set to be the heart rate and the respiratory rate once per second, so that the real-time property and the accuracy of data are ensured.

The physiological sleep disturbance associated data are used for identifying the influence area of the sleep environment, and an algorithm (such as DBSCAN) based on spatial clustering is adopted for area identification to generate the sleep environment influence area data. Then, the physiological data and the environmental data (such as temperature and noise level) are coupled by using a multivariate coupling analysis method, and a physiological-environmental influence mode is established to generate physiological-environmental coupling mode data. In practical application, the acquisition of environmental parameters is realized through intelligent household equipment, the temperature is set between 18 ℃ and 22 ℃, and the noise level is controlled between 30 dB and 40 dB so as to optimize the sleep quality. And finally, integrating the mixed sleep characteristic data, the physiological sleep disturbance related data and the physiological-environment coupling mode data to form comprehensive characteristic information extraction data. The comprehensive data can be used for further sleep quality assessment, sleep disorder diagnosis and personalized sleep intervention scheme formulation, and the application scene covers the fields of home sleep monitoring, medical diagnosis, sleep research and the like.

Preferably, step S1 comprises the steps of:

Step S11, continuously collecting a tested object by using a polysomnography to obtain sleep physiological data, wherein the sleep physiological data comprise electroencephalogram, oculogram, electromyogram, electrocardiogram and accelerometer signals;

Step S12, acquiring a body movement signal through an ultrasonic sensor, and acquiring environmental sound through a high-sensitivity microphone so as to obtain body environment acoustic data;

Step S13, merging the sleep physiological data and the physical environment acoustic data into original sleep monitoring data;

Step S14, performing signal quality evaluation on the original sleep monitoring data to obtain signal quality evaluation data, wherein the signal quality evaluation comprises signal-to-noise ratio calculation and electrode falling detection;

step S15, carrying out signal enhancement based on empirical mode decomposition on the sleep monitoring data according to the signal quality evaluation data so as to obtain enhanced sleep monitoring data;

And S16, carrying out signal segmentation on the enhanced sleep monitoring data to obtain multichannel sleep signal segment data.

As an embodiment of the present invention, referring to fig. 2, a detailed step flow diagram of step S1 in fig. 1 is shown, and in the embodiment of the present invention, step S1 includes the following steps:

The embodiment of the invention uses the polysomnography to continuously collect the measured object, and the collected data comprises an electroencephalogram (EEG), an oculogram (EOG), an Electromyogram (EMG), an Electrocardiogram (ECG) and an accelerometer signal. In the specific operation, the electrode for electroencephalogram signal acquisition is placed on the scalp according to an international 10-20 electrode placement system, and the electrode contact resistance is controlled within 5k omega. The sampling rate of EEG, EOG is set to 500Hz, the sampling rate of EMG, ECG is 1000Hz, and accelerometer signals are recorded at a frequency of 100 Hz. In the acquisition process, good contact of the electrodes is ensured, and the stability of the signals is monitored in real time so as to avoid artifact interference. The acquisition process lasts the whole sleep period, records the physiological data of each sleep stage, and provides a complete sleep physiological basis for subsequent analysis.

In order to more comprehensively evaluate the sleep state, the embodiment of the invention utilizes the ultrasonic sensor to acquire the body movement signal and combines the high-sensitivity microphone to acquire the environmental sound. In a specific application scenario, the ultrasonic sensor is installed below a bed head or a mattress, and the working frequency is set to 40kHz to detect micro body movements of a detected object, such as respiration and turning over. The high-sensitivity microphone is placed in a sleeping environment, is 1.5 meters away from the noise source, has a frequency response range of 20Hz to 20kHz, and collects external noise, snoring sounds of a tested person and the like. By synchronously recording body movement signals and environmental acoustic data, interference factors in the sleeping process can be better understood, and data support is provided for subsequent body movement and environmental impact analysis.

The embodiment of the invention combines the sleep physiological data acquired in the step S11 and the physical environment acoustic data acquired in the step S12 to generate the original sleep monitoring data. When the data are combined, the time stamps of various signals are required to be ensured to be synchronous, so that the physiological and environmental data in the same time period can be accurately corresponding to the subsequent analysis. The data storage format is a multi-channel time sequence, and all signals are uniformly recorded by adopting a time resolution of 1 ms. By integrating body movement and environmental sound information with physiological data, a global sleep monitoring data set is constructed, so that subsequent feature extraction and analysis are facilitated.

The embodiment of the invention evaluates the signal quality of the original sleep monitoring data, and firstly calculates the signal-to-noise ratio of each channel signal. In particular operation, the signal-to-noise ratio is calculated from the ratio of the noise variance to the signal power using a background noise level estimation formula in the signal, and the ideal signal-to-noise ratio for EEG and EOG signals is typically maintained above 20 dB. In addition, through real-time monitoring of EEG electrode contact impedance, whether the electrode falls off is detected, and if the impedance suddenly rises to exceed 10kΩ, the electrode falls off is marked. And then, an adaptive filter method is adopted to remove EOG and EMG artifacts, so that the purity of sleep physiological signals is ensured. Aiming at the problem of baseline drift, a wavelet transformation method is applied to remove, and particularly, a db4 wavelet basis is used for processing low-frequency drift aiming at an electroencephalogram signal, so that sleep monitoring data subjected to artifact removal and baseline correction is obtained.

according to the signal quality evaluation data, the embodiment of the invention applies an Empirical Mode Decomposition (EMD) algorithm to enhance the signal aiming at the sleep monitoring data with low signal-to-noise ratio or artifact. First, the original signal is decomposed into a series of Intrinsic Mode Functions (IMFs), the principal components are retained and noise components are removed by screening and reconstruction. In particular operation, for EEG signals, only the first 3-4 IMFs are typically retained to capture the major physiological fluctuations during sleep. The EMD processing can enhance the detail part of the signal, especially the weak signal part, and improve the usability of the signal, so that enhanced sleep monitoring data are generated, and the data are ensured to have higher credibility in subsequent analysis.

After the signal enhancement, the embodiment of the invention carries out signal segmentation on the enhanced sleep monitoring data. The segmentation is based on a standard sleep time window, typically set to 30 seconds for a period of time to ensure that each segment is able to fully contain a complete physiological change cycle. For accelerometer signals and body movement signals, the time window may be shortened to 10 seconds to capture more subtle motion changes. The segmented data will be marked as multi-channel sleep signal segments and saved by channel type and time sequence. These segment data facilitate subsequent feature extraction and sleep state classification analysis, and in actual scenarios, these segments may be used as inputs in machine learning models or deep learning models for predicting and classifying different sleep stages or states.

The multi-dimensional sleep monitor comprehensively collects electroencephalogram (EEG), oculogram (EOG), electromyogram (EMG), electrocardiogram (ECG) and accelerometer signals, covers multi-dimensional monitoring of brain, heart, muscle activity and body posture, and ensures omnibearing sensing of sleep state. The ultrasonic wave and the environmental sound are combined, so that the detection capability of the influence of respiration, body movement and external environment on sleep is further enhanced. By performing artifact removal, baseline drift correction and signal enhancement processing on the original data, the purity and the signal to noise ratio of the signals are effectively improved, and the accuracy and the robustness of subsequent data analysis are ensured. The signal segmentation divides long-term data into short-term segments, so that the system can analyze sleep characteristics in each time period in a fine granularity manner, and capture fine dynamic changes in the sleep process. This process not only improves the accuracy of recognition of sleep stages, but also helps to recognize abnormal events such as sleep apnea and the like. The fusion of the multidimensional features and the acoustic wave features increases the diversity of data, and the system can more accurately capture complex sleep state changes by combining nonlinear dynamics feature analysis. In the whole, the accuracy and the comprehensiveness of sleep state assessment are improved through the abundant physiological data and the diversified feature extraction and fusion methods, and a solid foundation is provided for personalized sleep monitoring and intervention.

Preferably, step S16 comprises the steps of:

Step 161, dividing the enhanced sleep monitoring data according to the preset time window data to obtain initial sleep signal segment data;

The embodiment of the invention divides the enhanced sleep monitoring data according to a preset time window, wherein the common time window is 30 seconds, which is a standard time period for classifying different sleep stages in sleep research. In a specific implementation, the signals of electroencephalogram, oculogram, electromyogram, etc. are divided into successive time segment segments, ensuring that each segment contains synchronized data for all channels. In order to reduce the boundary effect, an overlapping window processing method is adopted, the window overlapping rate is set to be 50%, namely, half of data of each time window segment is shared with the front segment and the rear segment, and the continuity of signal characteristics is ensured. The initial sleep signal segment data obtained by this processing will be used for further multi-channel synchronization processing.

Step S162, performing multi-channel-based time synchronization and alignment processing on the initial sleep signal segment data to obtain aligned sleep signal segment data;

The embodiment of the invention performs multi-channel time synchronization and alignment processing on the initial sleep signal segment data. Since the sampling rates of different devices may be different, the sampling frequencies of the channels must first be unified. For example, the electroencephalogram may have a sampling rate of 500Hz, while the accelerometer signal has a sampling rate of 100Hz, and the data of each channel is synchronized to a uniform 500Hz by interpolation and resampling. The channel signals are then aligned using a cross-correlation algorithm to ensure that the data points at the same time on the time axis correspond to the same physiological state and environmental factors. The aligned sleep signal segment data will more accurately reflect the coordinated changes in the various physiological signals during sleep.

Step S163, aligning the sleep signal segment data for data integrity check, identifying and marking missing or distorted data segments to obtain integrity check data;

The embodiment of the invention performs data integrity check on the aligned sleep signal segment data, and identifies and marks the missing or distorted data segment. In a specific operation, the presence or absence of data loss or signal distortion is detected by checking the continuity and voltage amplitude range of the signal. For example, if the voltage amplitude of a certain segment of the electroencephalogram signal is abnormal beyond a normal physiological range (e.g., ±100 μv), then a distorted data segment may be marked. For missing data segments, identification can be made by data blanks on the time axis. These marked missing or distorted data segments will be used for subsequent data repair or processing and corresponding integrity check data generated to ensure the accuracy and reliability of subsequent data analysis.

Step S164, performing amplitude standardization processing on the aligned sleep signal segment data according to the integrity check data, and performing fast Fourier transformation to obtain sleep signal spectrum data;

The embodiment of the invention performs amplitude normalization processing on the aligned sleep signal segment data based on the integrity check data. In the specific operation, the data of each channel is normalized, and the signal amplitude range is mapped to between 0 and 1, so that the amplitude difference between different channels is eliminated, and the subsequent frequency domain analysis is facilitated. The normalized data is then subjected to a spectral analysis using a Fast Fourier Transform (FFT) to convert the time domain signal to a frequency domain signal. The specific operation parameters of the FFT are that the frequency spectrum resolution is set to be 0.1Hz, the frequency range is 0.5Hz to 50Hz, and the spectrum information of slow waves (such as delta waves, 0.5-4 Hz) and fast waves (such as alpha waves, 8-13 Hz) which are common in sleeping is extracted. The generated sleep signal spectrum data will be used to analyze the physiological activity characteristics of different frequency bands during sleep.

And step S165, carrying out multi-channel integration synchronous processing on the sleep signal spectrum data to obtain multi-channel sleep signal fragment data, wherein the multi-channel comprises an electroencephalogram, an oculogram, an electromyogram, an electrocardiogram, an accelerometer signal, an ultrasonic signal and an acoustic signal.

The embodiment of the invention carries out multi-channel integration synchronous processing on the sleep signal spectrum data so as to generate final multi-channel sleep signal fragment data. The step first aligns and integrates spectral data of an electroencephalogram (EEG), an Eyegram (EOG), an Electromyogram (EMG), an Electrocardiogram (ECG), an accelerometer signal, an ultrasonic signal, and an acoustic signal in time and frequency domain by a multi-channel synchronization algorithm. In the specific operation, the linear weighted average method is adopted to fuse the spectrum information of each channel at the same time point, so as to ensure that the correlation among the signals of each channel is maintained. In practical application, the multi-channel data integration can better capture the overall physiological state and sleep behavior of the testee. Finally, the obtained multichannel sleep signal segment data can be used for subsequent feature extraction and sleep stage classification analysis, and can reflect the sleep quality and potential problems of the tested person more comprehensively and accurately.

According to the invention, the enhanced sleep monitoring data is segmented through the preset time window, so that the accurate capture of the sleep activity in each time segment is realized, the fine granularity analysis of the signals is ensured, and the real-time monitoring and parallel processing efficiency is improved. The time synchronization and alignment process of the multi-channel data ensures the time consistency of the different sensor signals, reducing the time deviation in the analysis. Through data integrity check, the system can identify and process abnormal data, and reliability of analysis results is guaranteed. Amplitude normalization eliminates the difference of different sensor dimensions and enhances the comparability of cross-channel analysis. The Fast Fourier Transform (FFT) provides frequency domain features revealing sleep physiology activity at different frequency bands. The integrated analysis of the multichannel signals enhances the capturing capability of complex physiological changes and improves the comprehensive understanding of sleep states and the detection precision of abnormal events. In the whole, the system realizes the accurate assessment of the sleep state and the effective recognition of the sleep disorder through high-quality data processing, multidimensional feature extraction and spectrum analysis.

Preferably, step S2 comprises the steps of:

s21, extracting energy characteristics of delta, theta, alpha, beta and gamma wave bands from electroencephalogram signals in multichannel sleep signal segment data, and detecting waveforms based on a K complex and sleep spindle waves to obtain electroencephalogram characteristic data;

S22, performing rapid and slow recognition of eye movement on an eye diagram signal in the multichannel sleep signal segment data, and extracting eye movement frequency and amplitude characteristics in rapid eye movement and non-rapid eye movement states to obtain eye movement characteristic data;

step S23, performing muscle tension analysis based on the myoelectric activity intensity and frequency characteristics on electromyographic signals in the multichannel sleep signal segment data to obtain myoelectric characteristic data;

Step S24, carrying out heart rate variability analysis on the electrocardiogram signals in the multichannel sleep signal segment data, and extracting heart rate variability indexes of a time domain and a frequency domain to obtain electrocardio characteristic data;

S25, performing body movement analysis on accelerometer signals in the multichannel sleep signal segment data, and extracting body movement frequency, intensity and duration characteristics to obtain body movement characteristic data;

Step S26, combining the electroencephalogram characteristic data, the eye movement characteristic data, the myoelectricity characteristic data, the electrocardio characteristic data and the body movement characteristic data into multi-dimensional characteristic data;

Step S27, doppler frequency shift analysis is carried out on the ultrasonic signals in the multichannel sleep signal segment data, and micro body movement characteristics are extracted to obtain ultrasonic characteristic data;

And S28, detecting and classifying sound events of the environmental sound signals in the multichannel sleep signal segment data, extracting sleep sound characteristics based on breathing sound and snoring sound to obtain acoustic characteristic data, and combining the ultrasonic characteristic data and the acoustic characteristic data into acoustic characteristic data.

As an embodiment of the present invention, referring to fig. 3, a detailed step flow chart of step S2 in fig. 1 is shown, and in the embodiment of the present invention, step S2 includes the following steps:

In the multi-channel sleep signal fragment data, the embodiment of the invention analyzes an electroencephalogram (EEG) signal in detail to extract energy characteristics of delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30-50 Hz) wave bands. In a specific operation, a band-pass filter is first applied to filter out signals of each frequency band, and then a Power Spectral Density (PSD) of each frequency band is calculated to quantify energy characteristics. Subsequently, the waveform characteristics of the K complex and sleep spindle wave are identified using an automatic detection algorithm. The detection of the K complex is identified by a template matching-based method through correlation comparison with a predefined K complex template, and the detection of the sleep spindle wave is carried out by a wavelet transformation combined with a threshold judgment method so as to capture the characteristic frequency and duration characteristics of the sleep spindle wave. In a specific application scenario, the sampling rate of EEG signals is set to be 500Hz, and the bandwidths of the filters are respectively delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30-50 Hz), so that accurate extraction of energy characteristics of each frequency band is ensured. Finally, generating brain electrical characteristic data containing energy values of each frequency band, K complex and sleep spindle wave detection results for subsequent sleep stage classification and analysis.

The embodiment of the invention carries out rapid and slow eye movement identification on an eye diagram (EOG) signal in multichannel sleep signal fragment data, and extracts eye movement frequency and amplitude characteristics in Rapid Eye Movement (REM) and non-rapid eye movement (NREM) states. In particular operation, a low pass filter (e.g., 30 Hz) is first applied to remove high frequency noise, and then a time and frequency domain based approach is used to identify fast eye movement and slow eye movement. Rapid eye movement is identified using short-time energy and zero-crossing rate analysis, and slow eye movement is detected by long-time span signal fluctuations. After the recognition is completed, the frequency (times/min) and amplitude (μv) of eye movement within each time window are calculated to quantify the eye movement characteristics. In practical application, the sampling rate of the EOG signal is set to 500Hz, the time window for rapid eye movement detection is 2 seconds, and the time window for slow eye movement detection is 10 seconds. Finally, eye movement characteristic data including eye movement frequency and amplitude in REM and NREM states are generated for assessing sleep stages and monitoring eye movement related sleep abnormalities.

According to the embodiment of the invention, the Electromyography (EMG) signals in the multichannel sleep signal fragment data are subjected to muscle tension analysis based on the electromyographic activity intensity and frequency characteristics, so that electromyographic characteristic data are obtained. In particular operation, the EMG signal is first high pass filtered (e.g., 20 Hz) to remove dc components and low frequency noise, then Root Mean Square (RMS) is applied to calculate the myoelectric activity intensity, and frequency domain features such as dominant frequency and band energy distribution are extracted by Fast Fourier Transform (FFT). Then, a change in muscle tone, such as sudden muscle activity or persistent muscle relaxation, is identified using a threshold decision. In a specific application scenario, the sample rate of the EMG signal is set to 1000Hz, the rms calculation window is1 second, and the spectral resolution of the FFT is set to 1Hz. Finally, myoelectrical signature data containing myoelectrical activity intensity, dominant frequency and frequency domain distribution is generated for analyzing muscle activity in sleep and assessing muscle-related sleep disorders.

According to the embodiment of the invention, heart Rate Variability (HRV) analysis is carried out on an Electrocardiogram (ECG) signal in multichannel sleep signal segment data, and heart rate variability indexes of a time domain and a frequency domain are extracted to obtain electrocardio characteristic data. In a specific operation, QRS wave detection is first performed on an ECG signal, and a Pan-Tompkins algorithm is used to accurately identify a heartbeat interval (RR interval). In the frequency domain analysis, a Fast Fourier Transform (FFT) or a power spectral density estimation method is applied to extract low-frequency (LF, 0.04-0.15 Hz) and high-frequency (HF, 0.15-0.40 Hz) powers and LF/HF ratios. In a specific application scenario, the sampling rate of the ECG signal is set to 1000Hz, and the threshold value of QRS detection is dynamically adjusted according to the heart rate range of the measured object, so that accurate heartbeat identification is ensured. Finally, electrocardiographic characterization data including time-domain and frequency-domain HRV indicators is generated for assessing autonomic nervous system activity and cardiac health.

According to the embodiment of the invention, body movement analysis is carried out on accelerometer signals in multichannel sleep signal segment data, and body movement frequency, intensity and duration characteristics are extracted to obtain body movement characteristic data. In particular operation, the accelerometer signal is first low pass filtered (e.g., 5 Hz) to remove high frequency noise, and then the amplitude and rate of change of the signal are calculated to quantify the body movement intensity. The frequency and duration of the physical movement events, such as the number of physical movements per minute and the duration of a single physical movement, are identified using a time domain analysis method. Further, a detection algorithm based on a statistical threshold is applied to distinguish normal body movements from abnormal body movements (such as frequent turning or strenuous movements). In a specific application scene, the sampling rate of the accelerometer signal is set to be 100Hz, and the threshold value of body movement detection is adjusted in a personalized way according to the activity habit of the measured object. Finally, body movement characteristic data comprising body movement frequency, intensity and duration are generated for monitoring body movement patterns in sleep and identifying abnormal body movement behaviors.

The embodiment of the invention integrates the electroencephalogram characteristic data, the eye movement characteristic data, the myoelectricity characteristic data, the electrocardio characteristic data and the body movement characteristic data extracted in the steps S21-S25 to form multi-dimensional characteristic data. In a specific operation, various feature data are standardized (for example, Z-score standardized) firstly, so as to eliminate dimension differences among different features. The various feature data are then sequentially arranged according to a time window (e.g., 30 seconds) to form a multi-dimensional vector containing all feature dimensions. For example, within a 30 second time window, the electroencephalogram features include energy values and waveform detection results for each frequency band, the eye movement features include frequency and amplitude, the myoelectric features include activity intensity and frequency, the electrocardiographic features include HRV indicators, and the body movement features include frequency, intensity and duration. Finally, a comprehensive multidimensional characteristic data set is generated and used as input of subsequent sleep stage classification and abnormality detection, and comprehensive analysis and evaluation of various physiological signals are ensured.

The embodiment of the invention performs Doppler frequency shift analysis on the ultrasonic signals in the multichannel sleep signal segment data so as to extract the micro body movement characteristics and obtain ultrasonic characteristic data. In a specific operation, the ultrasonic signal is first subjected to band-pass filtering (e.g., 20kHz to 40 kHz) to remove background noise, and then a doppler shift algorithm is applied to detect small movements of the object to be measured, such as breathing and slight turning. By comparing the frequency changes of the transmitted and received ultrasonic signals, the speed and direction of body movement are calculated, thereby quantifying the frequency and amplitude of minute body movements. In a specific application scenario, the working frequency of the ultrasonic sensor is set to 40kHz, and the signal sampling rate is 100kHz, so that high-precision body movement detection is ensured. Finally, ultrasonic characteristic data containing the frequency and the amplitude of the micro body movement are generated and used for supplementing and enhancing body movement information in the traditional physiological signals, so that the accuracy of overall sleep monitoring is improved.

The embodiment of the invention carries out sound event detection and classification on the environmental sound signals in the multichannel sleep signal segment data, extracts the sleep sound characteristics based on breathing sound and snoring sound to obtain acoustic characteristic data, and then combines the ultrasonic characteristic data and the acoustic characteristic data in the step S27 to generate acoustic characteristic data. In a specific operation, the environmental sound signals are preprocessed, including noise reduction and normalization, and then different sound events such as normal respiration, snoring and environmental noise are identified and classified by applying a sound event detection algorithm (such as a Support Vector Machine (SVM) or a Deep Neural Network (DNN)) based on machine learning. For breathing sounds and snoring sounds, the characteristic parameters such as frequency, amplitude and duration are further extracted. In a specific application scene, the sampling rate of the microphone is set to be 44.1kHz, and the sound event classification model is trained by a large amount of sleep environment data so as to improve the recognition accuracy. And finally, fusing the extracted acoustic characteristic data with the ultrasonic characteristic data obtained through Doppler frequency shift analysis, adopting characteristic splicing or weighted average and other methods to generate comprehensive acoustic characteristic data, and further analyzing the influence of the environment on sleep and identifying sound abnormal events related to the sleep.

The invention can accurately analyze the quality change of each sleep stage by extracting the energy characteristics of delta, theta, alpha, beta, gamma and other frequency bands, and identify key events of deep sleep and Rapid Eye Movement (REM) stages by combining K complex and sleep spindle wave detection. Analysis of eye movement characteristics helps to distinguish REM and NREM stages, assess sleep depth, dream activity and physical functional recovery. The myoelectric activity intensity and frequency characteristics effectively identify muscle states of different sleep stages, help judge the muscle relaxation degree in deep sleep, and detect sleep disorders such as periodic limb movements. Extraction of Heart Rate Variability (HRV) features evaluates the balance of the autonomic nervous system revealing changes in activity of sympathetic and parasympathetic nerves during sleep. The analysis of body movement characteristics can monitor the number of wakefulness and the activity level at night, and provides basis for judging light sleep and deep sleep. The multichannel signal fusion comprises electroencephalogram, eye movement, myoelectricity, electrocardio and body movement characteristic data, forms a comprehensive visual angle, comprehensively analyzes physiological changes in sleep and improves analysis capability of complex sleep states. Detecting minute body movements by ultrasonic signals, particularly in deep sleep, and analyzing breath sounds and snoring sounds to identify respiratory disorders such as sleep apnea syndrome. The acoustic signal can also detect the effect of ambient noise on sleep quality. Through comprehensive analysis of various physiological signals and external interference, the system improves accuracy of sleep quality assessment and enhances recognition capability of sleep disorder.

Preferably, step S3 comprises the steps of:

Step S31, carrying out phase space reconstruction on the multidimensional feature data to obtain reconstructed phase space data, and calculating the maximum Lyapunov exponent based on the reconstructed phase space data to obtain system stability feature data;

The embodiment of the invention carries out phase space reconstruction on the multidimensional characteristic data, and the reconstruction method maps the time sequence data into Gao Weixiang space by selecting proper time delay and embedding dimension based on Takens theorem. The time delay is typically determined by an autocorrelation function or a mutual information method, and the embedding dimension can be estimated by a pseudo-nearest neighbor algorithm. In practical application, for reconstruction of multidimensional feature data including electroencephalogram, electrocardiograph, body movement and the like, the time delay is set to 10 sampling points, and the embedding dimension is 3. After reconstructing the phase space, calculating the maximum Lyapunov exponent based on the reconstructed phase space data to measure the chaos and stability of the system. The calculation of the Lyapunov exponent uses the Wolf algorithm to measure the exponential divergence rate of the trajectory in phase space. Finally, system stability characteristic data is obtained, and dynamic stability characteristics of physiological signals in a sleep state are reflected.

S32, carrying out recursion quantitative analysis on the multidimensional feature data, and extracting recursion rate and deterministic features to obtain recursion feature data;

The embodiment of the invention carries out Recursion Quantitative Analysis (RQA) on multidimensional characteristic data, and firstly carries out normalization processing on the characteristic data so as to adapt to the construction of a recursion graph. The recursion map is generated by computing a distance matrix between the phase space trajectory points, using a fixed distance threshold (e.g., 0.1) to determine the recursion of the points. Based on the recurrence map, the recurrence rate (i.e., the frequency with which the trajectory returns to the previous state) and deterministic features (i.e., the proportion of consecutive recurrence lines) are extracted, which quantify the complexity and certainty of the system. In practice, the distance threshold and the resolution of the recursion map are adjusted according to the data characteristics. Finally, recursive feature data comprising a recursive rate and deterministic features is generated for further evaluating the periodicity and stability of the multi-dimensional feature data.

Step S33, calculating sample entropy and approximate entropy according to the multidimensional feature data to obtain entropy feature data, performing trend removal fluctuation analysis on the multidimensional feature data, and extracting long-range correlation features to obtain fluctuation feature data;

the embodiment of the invention calculates sample entropy (SampEn) and approximate entropy (ApEn) of the multidimensional feature data. Sample entropy is used for measuring complexity and randomness of data by comparing similarity of sub-sequences with different lengths in a time sequence, and approximate entropy is used for detecting predictability of signals. In practical application, the embedding dimension of the sample entropy and the approximate entropy is set to 2, and the tolerance is set to 20% of the data standard deviation. Then, a Detrending Fluctuation Analysis (DFA) is performed on the multidimensional feature data, and the method extracts long-range correlation features by segmenting a time series and calculating a fluctuation amplitude of each segment, reflecting self-similarity of signals. In the application scenario, the segment length is set to a multiple of 4 seconds to 16 seconds. Finally, feature data comprising entropy features and fluctuation features are obtained for identifying the complexity and long-term relevance of physiological signals in sleep states.

Step S34, fusing the system stability characteristic data, the recursion characteristic data, the entropy characteristic data and the fluctuation characteristic data to obtain nonlinear characteristic data;

The embodiment of the invention performs feature fusion on the system stability feature data, the recursion feature data, the entropy feature data and the fluctuation feature data extracted in the steps S31-S33. In a specific operation, firstly, all feature data are standardized so as to ensure the dimension consistency of different types of features. Then, different features are combined into a nonlinear feature vector by using a dimension reduction method such as a weighted average method or Principal Component Analysis (PCA). In practical applications, the principal component number of the PCA is set to a number that can account for more than 90% of the variance to maximize information retention. And finally, generating fused nonlinear characteristic data, reflecting the nonlinear dynamics characteristic of the whole system, and taking the nonlinear dynamics characteristic as the input of the subsequent step.

S35, carrying out wavelet packet transformation on the sound wave characteristic data, and extracting time-frequency domain characteristics to obtain time-frequency sound wave characteristic data;

The embodiment of the invention carries out wavelet packet transformation on the acoustic wave characteristic data extracted in the step S28, and decomposes signals into different time-frequency domains. The method comprises the steps of selecting a proper mother wavelet (such as Daubechies wavelet) first, determining a decomposition layer number (generally 3 layers), and decomposing an original signal into sub-signals with different frequency bands through multi-level wavelet packet decomposition. Each sub-signal extracts its time and frequency domain features, such as energy, frequency center, and time-frequency energy distribution. In an actual application scene, the sampling rate of an acoustic wave signal is 44.1kHz, the number of layers of wavelet packet decomposition is3, and the extracted time-frequency characteristics comprise the energy duty ratio and the frequency change trend of each frequency band. Finally, generating time-frequency sound wave characteristic data, and accurately reflecting the change rule of sound wave signals in different time and frequency dimensions.

Step S36, characteristic weighting is carried out on the time-frequency sound wave characteristic data by utilizing nonlinear characteristic data to obtain weighted sound wave characteristic data;

The embodiment of the invention utilizes the nonlinear characteristic data in the step S34 to perform characteristic weighting on the time-frequency sound wave characteristic data in the step S35. In a specific operation, the nonlinear characteristic data and the time-frequency acoustic wave characteristic data are firstly subjected to characteristic matching so as to ensure the alignment of the nonlinear characteristic data and the time-frequency acoustic wave characteristic data in the time dimension. Then, a feature weighting algorithm (e.g., a correlation-based weighted average) is applied to emphasize acoustic features that are sensitive to sleep states by applying weights of the nonlinear features to the time-frequency acoustic features. In an actual application scene, a linear regression model is adopted for weight calculation, and the weight is dynamically adjusted according to the correlation between the nonlinear characteristic and the time-frequency characteristic. Finally, weighted acoustic signature data is generated that enhances acoustic signature associated with sleep state changes.

And S37, carrying out feature fusion based on the deep neural network on the nonlinear feature data and the weighted sound wave feature data so as to obtain mixed sleep feature data.

The embodiment of the invention performs feature fusion based on a Deep Neural Network (DNN) on the nonlinear feature data in the step S34 and the weighted acoustic feature data in the step S36. First, a deep neural network model is constructed, the model comprising an input layer, a plurality of hidden layers and an output layer, the number of neurons in the hidden layers being set according to the feature data dimension and complexity (e.g., 128 neurons per layer). The input layer receives the nonlinear characteristics and the weighted acoustic wave characteristics respectively, and the characteristic fusion is carried out through the nonlinear activation function (such as ReLU) of the hidden layer. The model is trained by a back propagation algorithm, the loss function uses Mean Square Error (MSE), and the optimizer selects Adam. In practical application, training data is divided into a training set and a verification set through cross verification, the learning rate is set to be 0.001 in the training process, and the iteration times are 1000 times. Finally, fused mixed sleep characteristic data is generated for more accurately describing the sleep state and the related physiological and environmental characteristics thereof, and input is provided for subsequent sleep quality assessment and abnormality detection.

According to the invention, complex dynamic behaviors in the sleeping process are revealed through phase space reconstruction, the stability of a sleeping system is measured by using the maximum Lyapunov exponent, and the sleeping system is particularly outstanding in detecting problems such as apnea and the like. Recursive Quantitative Analysis (RQA) effectively identifies repetitive sleep stages and abnormal events such as body movement and apnea, enhancing analysis of sleep structure stability. Nonlinear characteristics such as sample entropy, approximate entropy and the like quantify the complexity of physiological signals, trend fluctuation analysis reveals long-range dependence, and potential sleep problems are identified. Wavelet packet transformation extracts the time-frequency characteristics of breathing sounds and snoring, which helps to find sleep breathing disorders early. The nonlinear characteristics, the time-frequency characteristics and the Deep Neural Network (DNN) are combined, so that the detection accuracy of the sleep event is further improved. The DNN automatically extracts key modes in the complex data, and the identification and classification of abnormal sleep events are optimized. Overall, the present invention provides an efficient, accurate sleep monitoring solution, particularly superior in sleep disorder detection and sleep quality assessment.

Preferably, step S4 comprises the steps of:

s41, performing sleep state space-time evolution processing according to the mixed sleep characteristic data to generate sleep state space-time evolution event data;

According to the embodiment of the invention, the space-time evolution processing of the sleep state is performed according to the mixed sleep characteristic data so as to generate the space-time evolution event data of the sleep state. In specific operation, firstly, mixed sleep characteristic data is input into a space-time evolution model, and the model adopts deep learning architectures such as a long-short-time memory network (LSTM) or a graph rolling network (GCN) and the like so as to capture dynamic changes of a sleep state in time and space. The model identifies transition trends and patterns of sleep states by analyzing feature changes within a continuous time window. For example, in an actual application scenario, the number of hidden layer units of the LSTM network is set to 128, the learning rate is 0.001, training is performed by using an Adam optimizer, and the number of iterations is 200, so as to ensure that the model can effectively capture the time sequence characteristics of the sleep state. During the processing, the model generates event data comprising sleep state transition, duration and spatial distribution, and the data reflects evolution dynamics of different stages in the sleep process and provides a basis for subsequent abnormal event identification.

Step S42, potential sleep interruption or abnormal event identification is carried out on the mixed sleep characteristic data through the sleep state space-time evolution event data, and event influence evaluation based on characteristic importance is carried out, so that sleep event evaluation data are generated;

The embodiment of the invention utilizes the sleep state space-time evolution event data generated in the step S41 to identify potential sleep interruption or abnormal events of the mixed sleep characteristic data, and carries out event influence evaluation based on characteristic importance to generate sleep event evaluation data. In a specific operation, firstly, an anomaly detection algorithm (such as an isolated forest or Support Vector Machine (SVM)) is applied to identify events which are significantly different from a normal sleep mode, such as frequent sleep interruption or abnormal sleep stage transition, in the time-space evolution event data. The extent of impact of these abnormal events on overall sleep quality is then analyzed using feature importance based assessment methods (e.g., SHAP values or feature importance scores). For example, in practical application, the tree number of the isolated forest algorithm is set to 100, and the maximum feature number is 5, so as to improve the accuracy and efficiency of detection. By the method, sleep event evaluation data comprising the types, the occurrence frequency and the influence degree of the abnormal event on the sleep quality are generated, and basis is provided for further sleep disorder screening and intervention.

Step S43, potential sleep disorder event screening is carried out according to the sleep event evaluation data, and real-time physiological index acquisition is carried out to generate sleep real-time monitoring index data;

According to the sleep event evaluation data generated in the step S42, potential sleep disorder events are screened out, real-time physiological indexes are collected, and sleep real-time monitoring index data are generated. In a specific operation, firstly, cluster analysis (such as K-means or spectral clustering) is carried out on sleep event evaluation data, and potential sleep disorder modes with similar characteristics are identified. Screening thresholds are then set based on these patterns, for example, with event occurrence frequencies exceeding 3 times per hour or specific abnormal event scores exceeding a certain threshold (e.g., 0.8) as a screening criterion. After screening potential obstacle events, configuring acquisition parameters of a multi-mode sensor, and mainly monitoring physiological indexes related to the obstacles, such as heart rate, respiratory rate and blood oxygen saturation. In a specific application, the real-time acquisition parameters are set to be heart rate once per second, respiratory rate twice per second and blood oxygen saturation once per second, data are transmitted in real time through the wireless sensor, and the edge computing equipment is used for real-time signal processing and analysis. Finally, sleep real-time monitoring index data containing key physiological indexes are generated and used for dynamically monitoring and timely intervening in sleep disorder events.

And S44, carrying out sleep stage fluctuation analysis on the multichannel sleep signal segment data to generate sleep stage fluctuation data, and carrying out physiological disturbance association processing on the sleep stage fluctuation data through the sleep real-time monitoring index data to generate physiological sleep disturbance association data.

The embodiment of the invention carries out sleep stage fluctuation analysis on the multichannel sleep signal segment data to generate sleep stage fluctuation data, carries out physiological disturbance association processing on the sleep stage fluctuation data through the sleep real-time monitoring index data, and generates physiological sleep disturbance association data. In specific operation, a Dynamic Time Warping (DTW) algorithm is first applied to carry out fluctuation analysis on sleep stages in multi-channel sleep signal segment data, and frequent transitions and duration changes of the sleep stages are identified. For example, setting the window length of the DTW to 30 seconds allows a range of time offsets to match the sleep stage patterns of different time periods. Then, in combination with the sleep real-time monitoring index data generated in step S43, correlation between the physiological index change and the sleep stage fluctuation is evaluated using a correlation analysis or a causal inference method (e.g., granger causal test). For example, in practical applications, a correlation threshold of 0.6 is set, and significantly associated physiological disturbance factors, such as heart rate sudden changes and frequent transitions of REM phases, are identified. Through the association processing, physiological sleep disturbance associated data comprising a sleep stage fluctuation mode and relevant physiological disturbance factors thereof are generated, and data support is provided for deep understanding of sleep quality and establishment of personalized intervention strategies.

The invention can monitor the change of the sleep state in real time by carrying out the space-time evolution processing on the mixed sleep characteristic data and capture the dynamic evolution in the whole sleep period. In particular, during transitions between different sleep stages, such as Rapid Eye Movement (REM) and non-rapid eye movement (NREM), analysis of the spatial and temporal evolution may reveal the duration, frequency of occurrence and pattern of each stage. The time-space evolution process can sharply identify sleep discontinuities or minor abnormal changes such as brief arousals or micro-movements. This is particularly useful for identifying phenomena that affect sleep quality, such as frequent arousal at night, periodic limb movements, etc. Through the time-space evolution event data, potential interruption or abnormal events which may occur in sleep, such as apnea, sleep interruption, night awakening and the like, can be automatically identified. By combining the multidimensional features in the mixed sleep feature data, the occurrence time and intensity of the sleep event can be more accurately positioned. By using the feature importance assessment method, the system can automatically analyze which features have important effects on sleep disruption or occurrence of abnormal events. Such as heart rate fluctuations, apneas, or abnormal changes in myoelectrical activity. Through such evaluation, the system can identify key physiological factors, helping to further screen for potential sleep disorders. After analysis of the sleep event assessment data, possible sleep disorder events, such as sleep apnea syndrome, insomnia or periodic limb movements, etc., may be automatically screened. Such automatic screening may reduce human intervention and help to more efficiently discover potential problems in sleep. During sleep monitoring, the system can dynamically evaluate according to physiological indexes (such as heart rate, respiratory rate, body movement and the like) acquired in real time. The real-time feedback mechanism can immediately respond to new abnormal events, and ensures the immediate adjustment and early warning of the sleep state. Through real-time physiological index collection, the system can perform personalized monitoring according to the sleep characteristics of an individual and identify potential health risks according to specific physiological conditions of the individual. Sleep stage fluctuation analysis can identify and track fluctuations in different stages of the sleep process (such as light sleep, deep sleep and REM stages), and this analysis helps to understand the duration of each stage, the regularity of interconversions, and the characteristics of the fluctuation between stages, which is critical to study the stability and quality of the different sleep stages. By correlating the sleep real-time monitoring index data with sleep stage fluctuation data, the system can identify the relevance of physiological disturbances (e.g., apneas, heart rate disturbances) to a particular sleep stage. This helps reveal the effects of certain physiological events on sleep stage changes to further evaluate whether these disturbances may lead to a decline in sleep quality or the occurrence of an obstacle. By analyzing the relationship between sleep stage fluctuation and physiological disturbance, the system can effectively integrate multichannel sleep signal data (such as EEG, EMG, ECG, accelerometer data and the like) so as to generate high-precision physiological sleep disturbance associated data. This data can be used to diagnose complex sleep disorders such as sleep apnea syndrome, REM behavioral disorders, and the like.

Preferably, step S4 comprises the steps of:

step S411, extracting sleep state transition characteristics based on time sequence from the mixed sleep characteristic data to obtain state transition sequence data;

The embodiment of the invention extracts the sleep state transition characteristics based on the time sequence for the mixed sleep characteristic data, thereby obtaining state transition sequence data. In a specific operation, the mixed sleep characteristic data is firstly sorted according to time sequence, and the sleep characteristic in each time period is marked as a corresponding sleep state (such as NREM, REM or wakefulness). Then, the data is processed segment by using a sliding window method (for example, the window size is set to 30 seconds, the step length is set to 10 seconds), the state transition characteristics in each time window are extracted, and the starting time, the ending time and the state transition frequency of the sleep state are recorded. The transition characteristics of each time window are recorded as state transition sequence data by a time stamp for subsequent modeling and analysis.

Step S412, modeling the state transition sequence data by using a hidden Markov model so as to obtain sleep state probability distribution data;

The embodiment of the invention models the state transition sequence data by using a Hidden Markov Model (HMM) so as to obtain sleep state probability distribution data. In a specific operation, parameters of the hidden Markov model are initialized first, including the number of states (set to 3, corresponding to NREM, REM and awake states), and a transition probability matrix and an initial state probability distribution between the states. Next, the state transition sequence data extracted in step S411 is input into a model, and parameter estimation and model training are performed using a Baum-Welch algorithm. After training, the model can output the sleep state probability distribution of each time point according to the input characteristic sequence. In practical application, the hidden state number of the HMM is set to be 3, and the maximum iteration number is 100, so that the model can accurately capture the state transition characteristics of different sleep stages, and the generated sleep state probability distribution data is used for further state transition analysis.

Step S413, constructing a sleep state transition matrix based on the sleep state probability distribution data to obtain state transition characteristic data;

The embodiment of the invention carries out sleep state transition matrix construction based on the sleep state probability distribution data to obtain state transition characteristic data. In the specific operation, firstly, the sleep state probability distribution of each time period is counted, and the transition frequency among the states is calculated. For example, the probability of transitioning from the NREM state to the REM state is determined by the transition frequency in the state probability distribution. Through the statistics, a sleep state transition matrix is constructed, and each element in the matrix represents the transition probability between two sleep states. To ensure the accuracy of the matrix, a large amount of historical data may be used for the estimation. In practical applications, the rows of the matrix represent the current state (e.g., NREM, REM), the columns represent the next state, and the data of the transition matrix is used as the basis for the subsequent timing modeling.

Step S414, extracting local time characteristics based on a sliding time window from the mixed sleep characteristic data to obtain time window characteristic data;

The embodiment of the invention performs local time feature extraction based on the sliding time window on the mixed sleep feature data so as to obtain the time window feature data. In a specific operation, the length and the step length of the sliding time window are defined first, for example, the window size is set to 60 seconds, the step length is set to 20 seconds, and the data of the whole sleep period is subjected to segmentation processing. And carrying out statistical analysis on the characteristic data (such as heart rate, eye movement, myoelectricity and the like) in each time window, and extracting local time characteristics such as the mean value, standard deviation, maximum value, minimum value and the like of the characteristics. Then, the local time features in each time window are saved as time window feature data for time series modeling in combination with state transition features. In practical applications, the strategy of feature extraction may be different for different physiological signals, for example, feature extraction of heart rate signals may focus on variability, and eye movement signals may focus on the frequency of rapid and slow eye movements.

Step S415, performing time sequence modeling based on a long-short time memory network according to the state transition characteristic data and the time window characteristic data to generate a sleep state space-time evolution model, wherein the sleep state space-time evolution model comprises transition probability and duration time distribution of different sleep stages;

According to the embodiment of the invention, according to the state transition characteristic data and the time window characteristic data, a long-short-time memory network (LSTM) is utilized for time sequence modeling, and a sleep state space-time evolution model is generated. In a specific operation, state transition feature data and time window feature data are firstly input into an LSTM model, an input layer of the model is multidimensional feature data, and a hidden layer is set to 128 neurons so as to capture time sequence dependency of a sleep state. In the training process, the LSTM adjusts the weight parameters through back propagation, so that the model can accurately predict the sleep state at the next moment. In practical application, the training data set of the LSTM model is divided into 80% for training, 20% for verification, and the optimizer uses Adam, and the learning rate is set to 0.001. After training is completed, the generated sleep state space-time evolution model comprises transition probabilities and duration distributions of different sleep stages, and provides basis for predicting sleep state evolution trend.

And S416, carrying out sleep state evolution trend prediction according to the sleep state space-time evolution model and the nonlinear characteristic data, and carrying out event classification and marking to generate different types of sleep state space-time evolution event data, wherein the sleep state space-time evolution event data comprises normal sleep events, abnormal sleep events and external environment influence events.

According to the embodiment of the invention, the sleep state evolution trend is predicted according to the sleep state time-space evolution model and the nonlinear characteristic data, and the event classification and the marking are performed to generate different types of sleep state time-space evolution event data. In the specific operation, firstly, the time sequence prediction is performed on the future sleep state by using the space-time evolution model generated in the step S415, and the prediction result includes the sleep state and the transition probability of the sleep state at each moment. Then, combining nonlinear characteristic data (such as the maximum Lyapunov index, recursive characteristic and the like), carrying out trend analysis, and judging the stability and possible abnormal change of the sleep state. Based on these analysis results, the system classifies and marks different types of events (e.g., normal sleep events, abnormal sleep events, or external environmental impact events). For example, an exception event may be a frequent sleep interrupt or an abnormal sleep state transition. Finally, the sleep state space-time evolution event data generated by the system can be used for further sleep quality assessment and intervention decision.

The invention can finely describe the change and transfer process of different sleep stages, such as the sequence from light sleep to deep sleep to Rapid Eye Movement (REM) stage and the duration thereof, by extracting the time sequence features of the mixed sleep feature data. This information is the basis for modeling of subsequent sleep state transitions. The extracted state transition sequence provides accurate time series data for subsequent modeling, so that changes in sleep state can be dynamically monitored and evaluated. The Hidden Markov Model (HMM) is a probability-based time sequence modeling method, and can effectively capture randomness and hidden rules of a sleep state. By modeling the state transition sequence data, probability distributions for sleep stages, such as transition probabilities for light sleep, deep sleep, and REM sleep, can be derived. HMM helps reveal and model sleep stage transfer processes that are difficult to observe directly, which can greatly enhance understanding of complex sleep states. By constructing the sleep state transition matrix, the transition rules between different sleep stages, such as the probability of transition from REM stage to light sleep, can be clearly demonstrated. This helps to better understand the overall structure of the sleep cycle. The transition matrix data may be used to evaluate the stability and predictability of sleep states. For example, a higher probability of sustained deep sleep indicates a more stable sleep state, which helps to assess sleep quality. The mixed sleep characteristic data is subjected to local time characteristic extraction through the sliding time window, so that sleep state changes in a short time can be captured. This has an important role in identifying transient sleep events or stage changes, especially when detecting suddenly occurring abnormal sleep events. The sliding time window approach enables finer timing analysis by the system, avoiding ignoring short but important sleep fluctuations or anomalies. Long and short term memory networks (LSTM) are powerful tools for processing time series data, especially those that are adept at modeling long time dependent sequences. The LSTM is used for carrying out time sequence modeling on state transition characteristics and time window characteristic data, the system can capture long-time dependence and transition trend of different sleep stages, and an accurate sleep state space-time evolution model is constructed. The space-time evolution model not only can describe the transition probability of each sleep stage, but also can predict the duration distribution of each stage, thereby providing a powerful basis for evaluating the integrity of the sleep cycle. By combining nonlinear characteristic data and a sleep state space-time evolution model, the system can predict the trend of the future sleep state. This feature is particularly helpful in early warning of potential sleep disruptions or abnormal events, such as apneas or sudden arousals. The system can classify and flag different types of sleep states according to the prediction result, including normal sleep events, abnormal sleep events (such as frequent arousal or sleep disorder), and external environmental influences (such as noise, light, etc.). This helps the user or physician to better understand the root cause of the sleep problem. For abnormal events (such as apnea or heart rate abnormality) possibly causing health risks, the system can detect early through space-time evolution prediction and give out a warning, so that timely intervention is facilitated, and sleep safety is improved.

Preferably, step S43 comprises the steps of:

Step S431, carrying out cluster analysis for identifying potential sleep disorder modes on the sleep event evaluation data to obtain sleep disorder mode data;

the embodiment of the invention carries out cluster analysis for identifying potential sleep disorder modes on the sleep event evaluation data to obtain sleep disorder mode data. In a specific operation, firstly, the sleep event evaluation data are subjected to standardized processing so as to ensure that the ranges of different characteristic data are consistent. And then, carrying out cluster analysis on the data by adopting a K-means clustering algorithm without supervision learning, setting the initial cluster number to be 5, and classifying similar sleep disorder events into one type by iteratively optimizing the position of a cluster center. The optimal cluster number is determined by an elbow rule. The result of the clusters generates a plurality of sleep disorder patterns, and the main characteristics (such as abnormal sleep interruption frequency, rapid eye movement abnormality and the like) in each cluster are recorded. The finally generated sleep disorder mode data are used for subsequent event screening and threshold setting, and the finally generated sleep disorder mode data can be used for analyzing personalized sleep disorder modes of different users in an application scene.

Step S432, setting a sleep disorder event screening threshold based on the sleep disorder mode data to obtain screening threshold data;

the embodiment of the invention sets the sleep disorder event screening threshold value based on the sleep disorder mode data, so as to obtain screening threshold value data. In specific operations, each sleep disorder mode is first analyzed to determine an abnormal value range of a specific physiological signal (such as heart rate and respiratory rate) in each type of disorder mode. And calculating the abnormal range of each physiological signal by carrying out statistical analysis on the historical data, and setting a screening threshold value. For example, for heart rate variability, a heart rate variability coefficient exceeding 90% of the standard range is set as the screening threshold for sleep disturbance events. Then, new sleep event evaluation data are screened according to the thresholds, and events with abnormality exceeding the set threshold are reserved as potential obstacle events. Screening threshold data provides a criterion for efficient screening of potential sleep disorders.

Step S433, determining physiological indexes to be monitored in a key way according to the potential obstacle event data, and generating a monitoring index list;

according to the embodiment of the invention, physiological indexes which need to be monitored in a key way are determined according to the potential obstacle event data, and a monitoring index list is generated. In the specific operation, the screened potential obstacle event data is firstly analyzed, and physiological characteristics possibly causing sleep obstacle are identified. For example, heart rate, respiratory rate, eye movement frequency, etc. may be indicators of high relevance to potential disorders. Next, by correlation analysis and expert system recommendation, the physiological index most relevant to the specific obstacle pattern is determined. In one application scenario, a rapid eye movement frequency anomaly may suggest some neurological problems, and thus the eye movement frequency may be included in the monitoring index list. The generated monitoring index list contains all physiological indexes which need to be monitored in an important way and is used as the basis for the acquisition of the subsequent sensor.

Step S434, carrying out acquisition parameter configuration of the multi-mode sensor according to the monitoring index list so as to obtain sensor configuration data;

According to the embodiment of the invention, the acquisition parameters of the multi-mode sensor are configured according to the monitoring index list, so that the sensor configuration data are obtained. In the specific operation, firstly, according to the monitoring index list generated in step S433, a suitable multi-mode sensor is selected for physiological data acquisition. For example, an electrocardiographic sensor is selected for heart rate monitoring, an acceleration sensor is selected for body movement, and an ocular sensor is selected for eye movement. Then, acquisition parameters such as sampling frequency, transmission interval, etc. are set according to the characteristics of different sensors. For electrocardiographic data, the sampling frequency may be set to 500Hz, and the sampling frequency of acceleration data may be set to 50Hz. The sensor configuration data contains acquisition parameters of all devices for ensuring that the devices are able to efficiently and accurately perform physiological data acquisition.

And S435, collecting the physiological index of the tested object in real time by using the sensor configuration data, and processing the signal in real time so as to obtain sleep real-time monitoring index data.

The embodiment of the invention utilizes the sensor configuration data to collect the real-time physiological index of the tested object and process the real-time signal so as to obtain the sleep real-time monitoring index data. In the specific operation, firstly, the configured sensor equipment is worn on a tested object, and the sensor starts to acquire physiological signals in real time according to set parameters. The collected data is transmitted to a data processing terminal through wireless transmission, and real-time signal processing is carried out on the terminal. The processing includes filtering, denoising, feature extraction and other operations. For example, the electrocardiosignal may be filtered to remove high frequency noise by a low pass filter, and heart rate data may be extracted by a QRS wave detection algorithm. The finally generated sleep real-time monitoring index data comprises real-time change conditions of a plurality of physiological signals, such as heart rate variability, respiratory rate, eye movement frequency and the like, and is used for further sleep quality analysis and intervention.

According to the invention, through cluster analysis, the sleep event evaluation data is subjected to pattern recognition, so that different sleep disorder patterns (such as apnea, frequent arousal, REM behavior disorder and the like) can be found. Such pattern recognition not only helps to automatically classify different types of sleep disorders, but also reveals the potential relevance of each type of disorder. The cluster analysis is helpful for generating personalized sleep disorder modes according to sleep data of different individuals, so that basis is provided for personalized sleep problem diagnosis. Based on the results of the cluster analysis, the system can set a screening threshold to perform event screening for a particular sleep disorder pattern. For example, for sleep apnea, a specific respiratory interruption frequency or heart rate fluctuation amplitude threshold may be set to ensure that the screened events have a high correlation. The setting of the screening threshold can be adaptively adjusted according to different sleep characteristics of each person, so that the limitation of a single threshold is avoided, and the screening precision is improved. And determining physiological indexes to be monitored in an important way, wherein the system can automatically generate a physiological index list to be monitored in an important way according to the screened potential obstacle event. By pertinently selecting key physiological indexes for monitoring, unnecessary monitoring items are reduced, a monitoring flow is optimized, and the monitoring efficiency and the accuracy of data processing are improved. According to the generated monitoring index list, the system can dynamically configure the acquisition parameters of the multi-mode sensor, and ensure that the sensor can accurately acquire the physiological indexes which are monitored in an important way. This not only optimizes the use of the sensor, but also improves the effectiveness of the data acquisition. Through accurate parameter configuration, the system can avoid unnecessary data redundancy and acquisition noise, ensure definition and accuracy of signals, and further promote the effect of subsequent data processing and analysis. By utilizing the optimized sensor configuration, the system can collect the physiological indexes of the detected object in real time, and ensure that relevant physiological data can be accurately captured when a potential sleep disorder event is detected, thereby improving the detection and confirmation rate of the event. The system can perform real-time signal processing on the acquired physiological data, and rapidly identify abnormal conditions. This real-time processing capability is critical to cope with sudden sleep disorders (e.g., sleep apnea, arrhythmia, etc.), and can provide effective support for timely intervention. The data collected and processed in real time can provide feedback for subsequent sleep state monitoring and sleep disorder evaluation, a closed-loop monitoring mechanism is formed, and continuous optimization of monitoring strategies and configuration of the system is ensured.

Preferably, step S44 includes the steps of:

Step S441, performing sleep cycle variation analysis based on a dynamic time warping algorithm on the multi-channel sleep signal segment data to obtain sleep cycle data;

The embodiment of the invention analyzes the sleep cycle change of the multichannel sleep signal segment data based on the dynamic time warping algorithm, thereby obtaining the sleep cycle data. In a specific operation, a multi-channel sleep signal segment is first acquired, including an electroencephalogram signal (EEG), an eye movement signal (EOG), an electromyogram signal (EMG), and the like. Then, the multidimensional signals are aligned and periodically analyzed by using a Dynamic Time Warping (DTW) algorithm, the similarity and the difference between different channels are calculated, and then the boundary and the change rule of each sleep period are identified. Accurate sleep cycle data is obtained by analyzing signal characteristics over a continuous plurality of sleep cycles. Then, based on the period data, the duration and the transition frequency of each sleep stage are statistically analyzed, and sleep structure data is generated. In an application scenario, the method is suitable for identifying multiple sleep cycles of a user at night and characteristic changes thereof, such as a transition from Rapid Eye Movement (REM) to deep sleep (NREM).

Step S442, predicting the sleep stage change trend based on time sequence for the sleep structure data so as to obtain sleep stage fluctuation data;

The embodiment of the invention predicts the sleep stage change trend based on the time sequence for the sleep structure data, thereby obtaining the sleep stage fluctuation data. In the specific operation, firstly, time series modeling is carried out on sleep structure data obtained in the earlier stage, and the change trend of each sleep stage in the future is predicted based on historical data by using an autoregressive moving average (ARIMA) model or a Long and Short Time Memory (LSTM) network and other prediction models. By analyzing the periodic features and fluctuations in the time series, sleep stage fluctuations of the user over a period of time in the future are predicted. The resulting sleep stage fluctuation data reflects the trend of the changes in the different stages (e.g., shallow sleep, deep sleep, REM) that may occur in future sleep. This step is particularly suitable for monitoring the user for long sleep stage changes in order to identify potential sleep quality problems in advance.

Step S443, carrying out physiological disturbance correlation analysis on the sleep stage fluctuation data according to the sleep real-time monitoring index data, and identifying the correlation between the sleep stage and the physiological signal change to generate physiological disturbance correlation data;

According to the embodiment of the invention, physiological disturbance correlation analysis is carried out on the sleep stage fluctuation data according to the sleep real-time monitoring index data, and the correlation between the sleep stage and the physiological signal change is identified, so that physiological disturbance correlation data is generated. In the specific operation, firstly, the sleep real-time monitoring index data (such as heart rate, respiratory rate, blood oxygen saturation and the like) are acquired, and then the data and the sleep stage fluctuation data obtained in the previous step are subjected to correlation analysis. And analyzing the relation between the change of each physiological index and a specific sleep stage by statistical methods such as pearson correlation coefficient, mutual information and the like, and identifying a mode such as whether heart rate acceleration is related to a REM stage or whether respiratory rate change is related to a shallow sleep stage. The finally generated physiological disturbance associated data reflects the specific association condition of the sleep stage and the key physiological signal change, and is convenient for subsequent risk assessment.

Step 444, performing mode classification on the physiological disturbance related data according to a preset disturbance mode library to generate disturbance mode data, and performing physiological disturbance risk assessment on physiological signal disturbance of each sleep stage according to the disturbance mode data to obtain physiological disturbance assessment data;

According to the embodiment of the invention, the physiological disturbance related data is subjected to pattern classification according to the preset disturbance pattern library, so that disturbance pattern data is generated. In a specific operation, a physiological disturbance pattern library is first established, which includes a common physiological signal disturbance pattern (such as a wake corresponding to acceleration of heart rate, a sleep apnea corresponding to irregular breathing, etc.). And then, matching the physiological disturbance related data with templates in a pattern library by using a pattern matching algorithm, identifying the type of the current physiological disturbance, and generating corresponding disturbance pattern data. And carrying out risk assessment on physiological signal disturbance occurring in each sleep stage according to the disturbance mode data. The assessment method may generate physiological disturbance assessment data based on a weighted scoring system in combination with the frequency, duration and severity of the disturbance. In an application scenario, this step may identify high risk sleep problems such as frequent apneas or excessive heart rate fluctuations.

And step S445, generating physiological sleep disturbance related data comprising the physiological index influence weight, the sleep stage transition probability and the disturbance degree according to the physiological disturbance evaluation data.

According to the embodiment of the invention, physiological sleep disturbance related data comprising physiological index influence weights, sleep stage transition probabilities and disturbance degrees is generated according to the physiological disturbance evaluation data. In a specific operation, the physiological disturbance evaluation data is first weighted and analyzed, and the influence weight of each physiological index is set (for example, the influence weight of heart rate may be higher and the body movement weight is lower). Next, based on the transition probability data of the sleep stages, disturbance events that may occur in each sleep stage are evaluated in combination with the degree of disturbance. The finally generated physiological sleep disturbance associated data contains the influence degree of different physiological indexes in different sleep stages, and provides the transition probability of disturbance in each stage. These data may be used to further optimize sleep interventions, helping users improve sleep quality and reduce potential health risks.

The Dynamic Time Warping (DTW) algorithm of the invention can process unsynchronized time series data, and is suitable for processing sleep signals of different time periods or different individuals. By analyzing the multichannel sleep signal segment data through the algorithm, the fine change of the sleep period can be accurately captured, and the reliability of the sleep period data is improved. By counting the duration and transition frequency of each sleep stage (such as shallow sleep, deep sleep, REM, etc.), the sleep structure of the tested object can be more comprehensively reflected, and basic data is provided for subsequent sleep quality assessment. Through time series analysis, the system can predict the change trend of the sleep stage. For example, the system may identify whether a shallow sleep state is about to go into deep sleep or REM phase. the prediction capability can help a user or doctor to know fluctuation conditions of the sleep state in advance, and provides a precedent opportunity for intervention of the sleep problem. The fluctuation trend of the sleep stage is predicted, so that the system can timely respond when abnormal fluctuation is recognized, and the monitoring strategy is dynamically adjusted, and data loss or monitoring delay caused by the bursty problem is avoided. By correlating the real-time monitored physiological index with sleep stage fluctuation data, potential links between specific physiological signals (e.g., heart rate, respiratory rate, etc.) and sleep stage changes can be revealed. For example, the system may identify a phenomenon in which heart rate variability increases significantly during the REM phase, thereby providing a reference for subsequent interventions. Each person's physiological disturbance may have a unique pattern of association with sleep stages, and this analysis step is able to identify individual physiological disturbance features through association analysis, providing support for personalized sleep monitoring. By pattern classification of the physiological perturbation associated data, the system is able to quickly identify common or abnormal physiological perturbation patterns in sleep stages based on a perturbation pattern library. For example, the system can identify physiological signal disturbances based on the apnea pattern. Such standardized pattern classification helps to improve the recognition efficiency and accuracy of the system. By performing risk assessment on the identified disturbance patterns, the system can quantify the potential impact of different physiological disturbances on sleep health. The assessment mechanism not only can provide clear risk levels (such as low risk, medium risk or high risk) for users, but also can provide further diagnostic basis for doctors. The step combines the influence weight of the physiological signals, the transition probability of the sleep stage and the disturbance degree together to generate a set of complete physiological sleep disturbance related data. By the comprehensive analysis, the system can more accurately quantify the influence of physiological signals on different sleep stages and provide a comprehensive and quantified evaluation result for sleep health. The physiological sleep disturbance related data provides a reliable basis for subsequent intervention measures. From this data, the physician can determine which abnormal fluctuations in physiological indicators may have a significant impact on a particular sleep stage, and thereby formulate a more targeted treatment or monitoring regimen. By means of dynamic time warping algorithm and time sequence analysis, the system can accurately track the change of sleep period and predict the fluctuation of sleep stage. This comprehensive analysis helps to fully understand the health of the sleeping structure. Through the association analysis and the pattern classification of the physiological disturbance, the system can identify the potential sleep disorder and physiological abnormality, which is important for early detection and prevention of sleep related diseases, the design of the steps S441 to S445 can generate personalized risk assessment reports according to individual data, thereby providing scientific basis for accurate intervention, and the physiological sleep disturbance association data can not only provide comprehensive diagnosis information for doctors, but also support a real-time monitoring system, so that the doctors can timely adjust monitoring schemes or take intervention measures when needed.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting feature information of a sleep state monitoring model, comprising the following steps:

Step S1: collecting multimodal physiological signals of the subject to obtain original sleep monitoring data; removing noise and performing signal enhancement processing on the original sleep monitoring data, and performing signal segmentation to obtain multi-channel sleep signal segment data;

Step S2: extracting features of the multi-channel sleep signal segment data based on electroencephalogram, oculogram, electromyogram, electrocardiogram and accelerometer signals to obtain multi-dimensional feature data; extracting ultrasonic features and acoustic features of the multi-channel sleep signal segment data to generate sound wave feature data;

Step S3: extracting nonlinear dynamic features according to the multidimensional feature data to generate nonlinear feature data; performing complementary fusion processing on the sound wave feature data through the nonlinear feature data to generate mixed sleep feature data. Step S3 includes:

Step S31: reconstructing the multidimensional characteristic data in phase space to obtain reconstructed phase space data; calculating the maximum Lyapunov exponent based on the reconstructed phase space data to obtain system stability characteristic data;

Step S32: performing recursive quantitative analysis on the multi-dimensional feature data, and extracting the recursive rate and deterministic features to obtain recursive feature data;

Step S33: performing sample entropy and approximate entropy calculations on the multidimensional feature data to obtain entropy feature data; performing detrending fluctuation analysis on the multidimensional feature data and extracting long-range correlation features to obtain fluctuation feature data;

Step S34: fusing the system stability characteristic data, the recursive characteristic data, the entropy characteristic data and the fluctuation characteristic data to obtain nonlinear characteristic data;

Step S35: performing wavelet packet transformation on the sound wave characteristic data, and extracting time-frequency domain features to obtain time-frequency sound wave characteristic data;

Step S36: using the nonlinear characteristic data to perform characteristic weighting on the time-frequency sound wave characteristic data to obtain weighted sound wave characteristic data;

Step S37: performing feature fusion based on a deep neural network on the nonlinear feature data and the weighted sound wave feature data, thereby obtaining mixed sleep feature data;

Step S4: performing spatiotemporal evolution processing and event impact assessment on the mixed sleep feature data to generate sleep event assessment data; screening potential sleep disorder events based on the sleep event assessment data, and collecting real-time physiological indicators to generate sleep real-time monitoring indicator data; performing sleep stage fluctuation analysis and physiological disturbance association processing on the multi-channel sleep signal segment data through the sleep real-time monitoring indicator data to generate physiological sleep disturbance association data. Step S4 includes:

Step S41: performing a sleep state spatiotemporal evolution process according to the mixed sleep feature data to generate sleep state spatiotemporal evolution event data. Step S41 includes:

Step S411: extracting sleep state transition features based on time series from the mixed sleep feature data, thereby obtaining state transition sequence data;

Step S412: Modeling the state transition sequence data using a hidden Markov model to obtain sleep state probability distribution data;

Step S413: constructing a sleep state transition matrix based on the sleep state probability distribution data to obtain state transition feature data;

Step S414: extracting local time features of the mixed sleep feature data based on a sliding time window, thereby obtaining time window feature data;

Step S415: performing time series modeling based on a long short-term memory network according to the state transition feature data and the time window feature data to generate a sleep state spatiotemporal evolution model, wherein the sleep state spatiotemporal evolution model includes transition probabilities and duration distributions of different sleep stages;

Step S416: predicting the sleep state evolution trend according to the sleep state spatiotemporal evolution model and the nonlinear characteristic data, and classifying and marking events to generate different types of sleep state spatiotemporal evolution event data, wherein the sleep state spatiotemporal evolution event data includes normal sleep events, abnormal sleep events, and external environment impact events;

Step S42: identifying potential sleep interruptions or abnormal events in the mixed sleep feature data through the sleep state spatiotemporal evolution event data, and performing event impact assessment based on feature importance to generate sleep event assessment data;

Step S43: Screening potential sleep disorder events according to the sleep event evaluation data, and collecting real-time physiological indicators to generate real-time sleep monitoring indicator data;

Step S44: performing sleep stage fluctuation analysis on the multi-channel sleep signal segment data to generate sleep stage fluctuation data; and performing physiological disturbance correlation processing on the sleep stage fluctuation data through the real-time sleep monitoring index data to generate physiological sleep disturbance correlation data;

Step S5: Identify the sleep environment impact area of the sleep stage fluctuation data to generate sleep environment impact area data; perform physiological-environmental impact pattern coupling on the sleep environment impact area data to obtain physiological-environmental coupling pattern data; merge the mixed sleep feature data, physiological sleep disturbance association data and physiological-environmental coupling pattern data into feature information extraction data.

2. The method for extracting characteristic information of a sleep state monitoring model according to claim 1, wherein step S1 comprises the following steps:

Step S11: continuously collecting sleep data of the subject using a polysomnography monitor to obtain sleep physiological data, wherein the sleep physiological data includes electroencephalogram, oculogram, electromyogram, electrocardiogram and accelerometer signals;

Step S12: collecting body motion signals through an ultrasonic sensor and collecting environmental sounds through a high-sensitivity microphone, thereby obtaining body motion environmental acoustic data;

Step S13: merging the sleep physiological data and the body motion environment acoustic data into original sleep monitoring data;

Step S14: performing signal quality assessment on the original sleep monitoring data to obtain signal quality assessment data, wherein the signal quality assessment includes signal-to-noise ratio calculation and electrode detachment detection; performing electrooculography and electromyography artifact removal on the original sleep monitoring data, and removing baseline drift using wavelet transform, thereby obtaining sleep monitoring data;

Step S15: performing signal enhancement based on empirical mode decomposition on the sleep monitoring data according to the signal quality evaluation data, thereby obtaining enhanced sleep monitoring data;

Step S16: performing signal segmentation on the enhanced sleep monitoring data to obtain multi-channel sleep signal segment data.

3. The method for extracting characteristic information of a sleep state monitoring model according to claim 2, wherein step S16 comprises the following steps:

Step S161: segmenting the enhanced sleep monitoring data according to the preset time window data to obtain initial sleep signal segment data;

Step S162: performing multi-channel time synchronization and alignment processing on the initial sleep signal segment data to obtain aligned sleep signal segment data;

Step S163: aligning the sleep signal segment data to perform data integrity check, identifying and marking missing or distorted data segments, and obtaining integrity check data;

Step S164: performing amplitude normalization processing on the aligned sleep signal segment data according to the integrity check data, and performing fast Fourier transform to obtain sleep signal spectrum data;

Step S165: performing multi-channel integration and synchronous processing on the sleep signal spectrum data to obtain multi-channel sleep signal segment data, wherein the multi-channel includes electroencephalogram, oculogram, electromyogram, electrocardiogram, accelerometer signal, ultrasonic signal and acoustic signal.

4. The method for extracting characteristic information of a sleep state monitoring model according to claim 3, wherein step S2 comprises the following steps:

Step S21: Extracting EEG signals from multi-channel sleep signal segment data , , , as well as The energy characteristics of the band are measured, and waveform detection based on K complex and sleep spindle is performed to obtain EEG characteristic data;

Step S22: performing fast and slow eye movement identification on the eye movement signal in the multi-channel sleep signal segment data, and extracting the eye movement frequency and amplitude characteristics in the rapid eye movement and non-rapid eye movement states to obtain eye movement feature data;

Step S23: performing muscle tension analysis based on the intensity and frequency characteristics of electromyographic activity on the electromyographic signals in the multi-channel sleep signal segment data to obtain electromyographic characteristic data;

Step S24: performing heart rate variability analysis on the electrocardiogram signal in the multi-channel sleep signal segment data, extracting time domain and frequency domain heart rate variability indicators, and obtaining electrocardiogram feature data;

Step S25: performing body motion analysis on the accelerometer signal in the multi-channel sleep signal segment data, extracting body motion frequency, intensity and duration characteristics, and obtaining body motion feature data;

Step S26: merging the EEG feature data, the eye movement feature data, the myoelectric feature data, the electrocardiographic feature data, and the body movement feature data into multi-dimensional feature data;

Step S27: performing Doppler frequency shift analysis on the ultrasonic signal in the multi-channel sleep signal segment data, extracting micro-body motion features, and obtaining ultrasonic feature data;

Step S28: Detect and classify sound events on the ambient sound signals in the multi-channel sleep signal segment data, extract sleep sound features based on breathing sounds and snoring sounds, and obtain acoustic feature data; merge the ultrasonic feature data and the acoustic feature data into sound wave feature data.

5. The method for extracting characteristic information of a sleep state monitoring model according to claim 4, wherein step S43 comprises the following steps:

Step S431: performing cluster analysis on the sleep event assessment data to identify potential sleep disorder patterns, and obtaining sleep disorder pattern data;

Step S432: setting a sleep disorder event screening threshold based on the sleep disorder pattern data, thereby obtaining screening threshold data; screening the sleep event assessment data for sleep disorder events according to the screening threshold data, thereby obtaining potential disorder event data;

Step S433: determining the physiological indicators that need to be monitored based on the potential obstacle event data, and generating a monitoring indicator list;

Step S434: configuring the acquisition parameters of the multimodal sensor according to the monitoring indicator list, thereby obtaining sensor configuration data;

Step S435: using the sensor configuration data to collect real-time physiological indicators of the subject, and performing real-time signal processing, so as to obtain real-time sleep monitoring indicator data.

6. The method for extracting characteristic information of a sleep state monitoring model according to claim 5, wherein step S44 comprises the following steps:

Step S441: performing sleep cycle change analysis based on a dynamic time warping algorithm on the multi-channel sleep signal segment data to obtain sleep cycle data; and performing statistics on the duration and conversion frequency of each sleep stage based on the sleep cycle data to obtain sleep structure data;

Step S442: predicting the sleep stage change trend based on the time series of the sleep structure data, thereby obtaining sleep stage fluctuation data;

Step S443: performing physiological disturbance correlation analysis on the sleep stage fluctuation data according to the real-time sleep monitoring index data, identifying the correlation between the sleep stage and the physiological signal change, and generating physiological disturbance correlation data;

Step S444: performing pattern classification on the physiological disturbance associated data according to a preset disturbance pattern library to generate disturbance pattern data; performing physiological disturbance risk assessment on the physiological signal disturbance of each sleep stage according to the disturbance pattern data, thereby obtaining physiological disturbance assessment data;

Step S445: generating physiological sleep disturbance associated data including physiological index impact weights, sleep stage transition probabilities, and disturbance degrees according to the physiological disturbance assessment data.

7. A feature information extraction system for a sleep state monitoring model, characterized in that it is used to execute the feature information extraction method for a sleep state monitoring model according to claim 1, and the feature information extraction system for a sleep state monitoring model comprises:

The signal preprocessing module is used to collect multimodal physiological signals of the subject to obtain original sleep monitoring data; remove noise and enhance the signal of the original sleep monitoring data, and perform signal segmentation to obtain multi-channel sleep signal segment data;

A multi-dimensional feature extraction module is used to extract features of multi-channel sleep signal fragment data based on electroencephalogram, oculogram, electromyogram, electrocardiogram and accelerometer signals to obtain multi-dimensional feature data; and to extract ultrasonic features and acoustic features of multi-channel sleep signal fragment data to generate sound wave feature data;

A feature fusion module is used to extract nonlinear dynamic features based on multidimensional feature data to generate nonlinear feature data; and to perform complementary fusion processing on the sound wave feature data through the nonlinear feature data to generate mixed sleep feature data;

The sleep dynamic analysis module is used to perform spatiotemporal evolution processing and event impact assessment on mixed sleep feature data to generate sleep event assessment data; screen potential sleep disorder events based on the sleep event assessment data, collect real-time physiological indicators, and generate real-time sleep monitoring indicator data; perform sleep stage fluctuation analysis and physiological disturbance correlation processing on multi-channel sleep signal fragment data through real-time sleep monitoring indicator data to generate physiological sleep disturbance correlation data;

The environmental-physiological coupling module is used to identify the sleep environment impact area of the sleep stage fluctuation data and generate sleep environment impact area data; to couple the sleep environment impact area data with the physiological-environment impact pattern to obtain the physiological-environment coupling pattern data; and to merge the mixed sleep feature data, the physiological sleep disturbance association data and the physiological-environment coupling pattern data into feature information extraction data.