A sleep staging method based on deep learning

Abstract

The present invention relates to the technical field of sleep diagnosis, and in particular to a sleep staging method based on deep learning, the method comprising: collecting a number of target physiological signals, preprocessing each target physiological signal to obtain a number of standard physiological signals; synchronously aligning and integrating each standard physiological signal in turn to obtain a standard sleep data set; determining the actual correlation between each physiological signal to be integrated in the standard sleep data set, and determining an actual signal combination strategy according to the actual correlation; generating an actual sleep data set according to the actual signal combination strategy, and extracting features from the actual sleep data set to generate actual feature parameters; inputting the actual feature parameters into a deep learning model to determine the actual sleep state; analyzing the actual sleep state, and performing corresponding sleep intervention according to the analysis results. The present invention realizes low-cost non-intrusive sleep monitoring, thereby improving people's physical health.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep diagnosis, and in particular to a sleep staging method based on deep learning. Background Art

[0002] Sleep quality has become one of the important indicators of people's health and quality of life; monitoring and analyzing the sleep process is a prerequisite for improving sleep quality. Sleep staging is an important means of monitoring and analyzing sleep quality, so sleep staging is of great significance to improving people's physical health; PSG is considered to be the most accurate method for diagnosing sleep-related problems and is considered the gold standard of clinical sleep medicine; however, its disadvantages are that it is expensive, complicated, time-consuming, and uncomfortable for users; in fact, in most cases, people only want to understand the general sleep status; therefore, a convenient and accurate method is needed to judge the sleep state, so as to achieve low-cost and non-intrusive sleep monitoring.

[0003] Chinese patent application publication number CN111783534A discloses a sleep staging method based on deep learning, which includes the following steps: Step 1: Acquire a single-channel EEG signal to construct a data set, oversample the EEG signal in the data set, and divide the oversampled EEG signal data set into a training set and a validation set; Step 2: Construct a convolutional neural network model for sleep staging, use the oversampled data of the single-channel EEG signal as the input of the model, and use the label of the sleep period as the output of the model; Step 3: Use the training set to pre-train the network model, use the validation set to validate the model, and adjust the hyperparameters of the model according to Bayesian optimization to obtain a preliminary optimized network model; Step 4: Construct a penalty weight loss function to perform secondary optimization on the model to obtain a secondary optimized network model; Step 5: Use the trained model to test the input single-channel EEG signal to obtain a predicted sleep period.

[0004] It can be seen that the current sleep staging methods cannot achieve low-cost and non-intrusive sleep monitoring. Summary of the invention

[0005] To this end, the purpose of the present invention is to provide a sleep staging method based on deep learning, which is used to overcome the problem that the current sleep staging method cannot achieve low-cost and non-intrusive sleep monitoring.

[0006] To achieve the above object, the present invention provides a sleep staging method based on deep learning, comprising:

[0007] Collecting a number of target physiological signals, and preprocessing each target physiological signal to obtain a number of standard physiological signals;

[0008] Each standard physiological signal is synchronously aligned and integrated in turn to obtain a standard sleep data set;

[0009] Determining the actual correlation between the physiological signals to be integrated in the standard sleep data set, and determining the actual signal combination strategy according to the actual correlation;

[0010] generating an actual sleep data set according to the actual signal combination strategy, and performing feature extraction on the actual sleep data set to generate actual feature parameters;

[0011] Inputting the actual characteristic parameters into a deep learning model to determine the actual sleep state;

[0012] The actual sleep state is analyzed, and corresponding sleep intervention is performed according to the analysis result.

[0013] Furthermore, the process of preprocessing each target physiological signal includes:

[0014] Classifying each target physiological signal according to the data processing characteristics to obtain a number of first-class target physiological signals and second-class target physiological signals;

[0015] Perform collection point cleaning and filtering processing on each type of target physiological signal in turn;

[0016] Segmenting the target physiological signals of the two categories to form a plurality of signal segments of the two categories, and determining the frequency domain distribution according to each of the signal segments of the two categories;

[0017] Marking the collection points that need to be cleaned in the target physiological signal of the first type as a type of abnormality;

[0018] Identify the frequency domain distribution, and based on the abnormal frequency domain distribution, mark the second type of target physiological signals as the second type of abnormal type; based on the normal frequency domain distribution, mark the second type of target physiological signals as the second type of normal type;

[0019] The number of the one type of abnormality is recorded as a first quantity, and a first collection level of the one type of target physiological signals is determined according to the first quantity.

[0020] Furthermore, the process of preprocessing each target physiological signal also includes:

[0021] determining whether to re-acquire and determining a corresponding re-acquisition mode according to the first acquisition level and the label type of the second type of target physiological signals;

[0022] Based on the fact that the first acquisition level meets the acquisition requirements and the marking type is the second normal type, it is determined that re-acquisition is not required;

[0023] Based on the fact that the first acquisition level meets the acquisition requirements and the marking type is the second type of abnormality, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: partial re-acquisition;

[0024] Based on the fact that the first acquisition level does not meet the acquisition requirements, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: overall re-acquisition.

[0025] Furthermore, the process of marking the collection points to be cleaned in the target physiological signal of the type includes:

[0026] For any type of target physiological signal, the target physiological signal is formed into a plurality of collection points according to the cleaning interval;

[0027] For any collection point, obtaining actual monitoring data of the collection point, and determining whether the collection point needs to be cleaned according to the actual monitoring data;

[0028] Based on the fact that the actual monitoring data is blank, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality;

[0029] Based on the fact that the actual monitoring data is not within the normal monitoring range, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality.

[0030] Further, the process of identifying the frequency domain distribution includes:

[0031] For any second-category signal segment, obtaining an actual amplitude of the second-category signal segment, and judging whether the second-category signal segment meets the normal monitoring range according to the actual amplitude;

[0032] Perform amplitude judgment on each of the second-category signal segments in turn, and record the number of each of the second-category signal segments that do not meet the normal monitoring range, which is recorded as a second number;

[0033] Whether the frequency domain distribution is abnormal is determined according to the second quantity.

[0034] Furthermore, the process of synchronously aligning and integrating the standard physiological signals in sequence includes:

[0035] Identify the acquisition timestamp of each standard physiological signal;

[0036] Convert each acquisition time stamp into a unified format to obtain each standard time stamp;

[0037] Synchronously aligning each standard time stamp with the target reference time window to obtain each physiological signal to be integrated;

[0038] The physiological signals to be integrated are combined to obtain the standard sleep data set.

[0039] Furthermore, the process of determining the actual signal combination strategy according to the actual correlation degree includes:

[0040] For any two physiological signals to be integrated, determining the actual correlation between the two physiological signals to be integrated;

[0041] The correlation level between the two physiological signals to be integrated is determined according to the actual correlation degree, and the representative physiological signal is determined according to the correlation level combined with the mutual information inclusion relationship between the two physiological signals to be integrated.

[0042] Furthermore, the process of determining the actual signal combination strategy according to the actual correlation degree also includes:

[0043] For any two physiological signals to be integrated;

[0044] Based on the presence of the representative physiological signal in the two physiological signals to be integrated, the two physiological signals to be integrated are connected in parallel;

[0045] Based on the absence of the representative physiological signal in the two physiological signals to be integrated, the two physiological signals to be integrated are connected in series;

[0046] The physiological signals to be integrated in the standard sleep data set are paired and connected in sequence to form the actual signal combination strategy.

[0047] Furthermore, the process of determining the actual sleep state includes:

[0048] Comparing the actual feature parameters with the standard feature ranges in the deep learning model;

[0049] Determine the actual sleep stage and stage sleep quality based on the comparison results;

[0050] The actual sleep state includes: the actual sleep stage and the sleep quality of the stage.

[0051] Furthermore, the process of performing corresponding sleep intervention according to the analysis results includes:

[0052] Recording the actual duration of the actual sleep stage;

[0053] Sleep staging intervention or sleep stabilization intervention is performed according to the actual duration and the sleep quality of the stage.

[0054] Compared with the prior art, the beneficial effects of the present invention are that, by collecting multiple target physiological signals and performing preprocessing and integration, a more comprehensive and accurate standard sleep data set can be obtained, thereby improving the accuracy of the evaluation and analysis of the sleep state; by preprocessing each physiological signal to obtain a standardized physiological signal, it is helpful to reduce the noise and inconsistency in the original data, thereby improving the accuracy and reliability of data analysis; by synchronously aligning and integrating multiple physiological signals, different types of sleep-related data can be integrated to provide a comprehensive sleep view, and multiple physiological signals can be collected in the same time period to ensure the correspondence of data in the time dimension; time consistency between different signals is ensured by time marking so that they can be merged or processed in parallel; by determining the actual correlation between the physiological signals to be integrated, the signal combination strategy can be dynamically determined to make the data integration more in line with individual characteristics and enhance the personalization and adaptability of the data set; the correlation between different signals is analyzed to select signals that are representative and information-rich Features are extracted; representative and information-rich signal features are selected for extraction; the effectiveness and reliability of the sleep staging method in practical applications are ensured, and the physiological signal data collected during sleep are effectively prepared, providing a reliable data basis for subsequent sleep analysis and health assessment; advanced deep learning technology is used to extract features and classify states of the integrated data, which can more accurately identify and analyze sleep patterns and problems; it can better capture complex sleep state patterns and trends, and improve the accuracy and automation level of sleep state analysis; based on in-depth analysis of sleep states, personalized sleep intervention suggestions can be provided; it helps to improve sleep quality; it embodies the characteristics of data-driven medical diagnosis and health management, and helps to promote the development of intelligent health monitoring technology; it helps to realize the intelligent management and decision support of medical and health data, which is in line with the current development trend of the medical and health field; it can provide more accurate and personalized sleep assessment and intervention, which helps to improve sleep quality and health management effects.

[0055] In particular, different target physiological signals are classified into Class I and Class II target physiological signals through data processing characteristics, so that subsequent processing can be carried out in a targeted manner; abnormal collection points and Class II signal segments can be accurately marked through analysis of frequency domain distribution and comparison with actual monitoring data, which helps to accurately identify and handle abnormal situations; according to different target physiological signals, the corresponding normal monitoring range is set, taking into account the influence of factors such as age and gender, making it more versatile and adaptable; it can effectively identify collection points that need to be cleaned, ensure the accuracy and reliability of real-time monitoring data, and effectively improve the efficiency of data processing and the accuracy of results.

[0056] In particular, by calculating the absolute value of the difference between the first number and the second number and the preset standard value, a quantitative evaluation of the physiological signal processing results is achieved, making the data evaluation more objective and accurate; the acquisition level of a type of target physiological signals can be divided into three levels according to the different situations of the first number and the second number, and it can be judged whether the frequency domain distribution is abnormal; multi-level classification can more finely reflect the results and abnormal situations of data processing, which is helpful for further analysis and processing; it can evaluate in real time based on actual monitoring data, promptly discover and handle abnormal situations, and ensure the timeliness and effectiveness of monitoring; it can effectively improve the efficiency of data processing and the reliability of results.

[0057] In particular, based on the marking of the first collection level and the abnormal type, it is intelligently determined whether re-collection is needed and the specific mode of collection; intelligent decision-making can effectively reduce unnecessary data re-collection and improve collection efficiency and resource utilization; based on the different classifications of the first collection level, it is targeted to determine whether the collection requirements are met, and then determine the collection mode; targeting can more accurately reflect the actual situation of data quality and ensure the accuracy and reliability of the data; judgments and decisions can be made instantly during real-time monitoring, and the collection strategy can be adjusted according to the actual situation to ensure the timeliness and effectiveness of monitoring data; by reasonably determining whether re-collection is needed and the collection mode, the efficiency of data collection and the overall performance of the system can be effectively improved.

[0058] In particular, unified timestamp format conversion and time alignment ensure the temporal consistency of different physiological signals, lay the foundation for subsequent data analysis and processing, and improve the reliability and accuracy of the data; by integrating multiple standard physiological signals, a comprehensive standard sleep data set is formed; the comprehensiveness helps to gain a deeper understanding of the individual's physiological state and sleep quality, and provide richer information support; the unified format of timestamps and synchronous alignment enables data from different physiological signals to be compared and analyzed on the same timeline, improving the comparability and research value of the data; by merging the physiological signals to be integrated, the final standard sleep data set contains comprehensive and synchronized physiological data, providing a complete data foundation for subsequent sleep analysis and evaluation.

[0059] In particular, by calculating the Pearson correlation coefficient between the physiological signals to be integrated, the actual correlation between them is quantified; the quantitative evaluation can objectively reflect the degree of linear correlation between the signals, providing a reliable basis for the subsequent signal combination; the correlation of the physiological signals to be integrated is divided into three levels according to the actual correlation and the preset standard correlation; the multi-level classification can more finely describe the relationship between different signals, which is helpful to select appropriate representative physiological signals; the mutual information inclusion relationship is introduced to evaluate the nonlinear relationship between the physiological signals to be integrated; the amount of information and the degree of correlation between the signals can be more comprehensively evaluated, making the understanding of the complex relationship between the signals more in-depth and comprehensive; it can effectively simplify the data set and retain the main information, improving the efficiency and credibility of subsequent data analysis.

[0060] In particular, according to whether there are representative physiological signals in the physiological signals to be integrated, the parallel or series strategy can be flexibly selected; the ability to make adjustments based on the actual data characteristics makes the combination strategy more in line with the actual situation and data characteristics; by pairing and connecting the physiological signals in the standard sleep data set, it is ensured that the final actual signal combination strategy can cover multiple physiological signals, ensuring the integrity and comprehensiveness of the data; through the selection of parallel and series strategies, important information can be better retained and the data set structure can be optimized, which helps to improve the efficiency and accuracy of subsequent data processing and analysis.

[0061] In particular, by combining the attention mechanism and the hybrid convolutional network for feature extraction, the key local patterns and features in the sleep dataset can be captured more accurately, and it has a certain degree of translation invariance. Even if the position of the signal on the time axis changes, the model can still recognize its features; the network model adopted reduces the complexity of the model through parameter sharing and local connection, making the model lighter, easier to deploy and process in real time; the model has good generalization performance for large-scale datasets, which means that when faced with unseen data, the model can still maintain a high recognition accuracy rate; the introduction of the gated recurrent unit (GRU) to process and predict time series data can effectively handle long-term dependencies and solve the problem of gradient vanishing or exploding that may occur in traditional RNN when processing long sequences. Explosion problem; The strong memory capacity of the GRU unit enables the model to retain and process the information in the input sequence for a long time, which is particularly important for the continuous identification of states during sleep; The gating mechanism of GRU can control the flow and update process of information, enhance the expression ability of the network, and thus improve the accuracy of the sleep staging results; By comparing the actual feature parameters with the standard feature range, not only can the sleep stage be determined, but also the sleep quality of the stage can be evaluated, providing users with a more comprehensive sleep analysis; Based on the analysis results, the user's sleep pattern can be intervened, such as adjusting the duration of sleep or improving sleep quality, so as to put forward personalized sleep improvement suggestions; The accuracy of sleep staging and quality assessment is improved, and ultimately a more accurate and personalized sleep health management plan is provided to users. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of a sleep staging method based on deep learning according to an embodiment of the present invention;

[0063] Figure 2 This is a first flow chart of preprocessing in a sleep staging method based on deep learning according to an embodiment of the present invention;

[0064] Figure 3 A second flow chart of preprocessing in the sleep staging method based on deep learning according to an embodiment of the present invention;

[0065] Figure 4 A logic diagram for determining dependency levels for an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0067] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0068] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0069] See also Figure 1 As shown, this embodiment provides a sleep staging method based on deep learning, including:

[0070] Step S100, collecting a plurality of target physiological signals, and preprocessing each target physiological signal to obtain a plurality of standard physiological signals;

[0071] Step S200, synchronously aligning and integrating each standard physiological signal in sequence to obtain a standard sleep data set;

[0072] Step S300, determining the actual correlation between the physiological signals to be integrated in the standard sleep data set, and determining the actual signal combination strategy according to the actual correlation;

[0073] Step S400, generating an actual sleep data set according to the actual signal combination strategy, and performing feature extraction on the actual sleep data set to generate actual feature parameters;

[0074] Step S500, inputting the actual characteristic parameters into a deep learning model to determine the actual sleep state;

[0075] Step S600: analyzing the actual sleep state and performing corresponding sleep intervention according to the analysis result.

[0076] The embodiments of the present invention can obtain a more comprehensive and accurate standard sleep data set by collecting multiple target physiological signals and performing preprocessing and integration, thereby improving the accuracy of the evaluation and analysis of the sleep state; by preprocessing each physiological signal to obtain a standardized physiological signal, it is helpful to reduce the noise and inconsistency in the original data, thereby improving the accuracy and reliability of data analysis; by synchronously aligning and integrating multiple physiological signals, different types of sleep-related data can be integrated to provide a comprehensive sleep view, and multiple physiological signals can be collected in the same time period to ensure the correspondence of data in the time dimension; by time marking, the time consistency between different signals can be ensured for merging or parallel processing; by determining the actual correlation between the physiological signals to be integrated, the signal combination strategy can be dynamically determined, so that the data integration is more in line with individual characteristics and the personalization and adaptability of the data set are enhanced; the correlation between different signals is analyzed, and representative and information-rich signal features are selected for extraction; Determine and select representative and information-rich signal features for extraction; ensure the effectiveness and reliability of the sleep staging method in practical applications, effectively prepare the physiological signal data collected during sleep, and provide a reliable data basis for subsequent sleep analysis and health assessment; use advanced deep learning technology to extract features and classify states of the integrated data, which can more accurately identify and analyze sleep patterns and problems; can better capture complex sleep state patterns and trends, and improve the accuracy and automation level of sleep state analysis; based on in-depth analysis of sleep states, can provide personalized sleep intervention suggestions; help improve sleep quality; embody the characteristics of data-driven medical diagnosis and health management, and help promote the development of intelligent health monitoring technology; help realize the intelligent management and decision support of medical and health data, in line with the current development trend of the medical and health field; can provide more accurate and personalized sleep assessment and intervention, which helps to improve sleep quality and health management effects.

[0077] See also Figure 2 As shown, it is a first preprocessing flow chart; wherein, the process of preprocessing each target physiological signal includes:

[0078] Step S111, classifying each target physiological signal according to the data processing characteristics to obtain a plurality of first-class target physiological signals and second-class target physiological signals;

[0079] Step S112, performing collection point cleaning and filtering processing on each type of target physiological signal in turn;

[0080] Step S113, segmenting the two-category target physiological signal to form a plurality of two-category signal segments, and determining the frequency domain distribution according to each of the two-category signal segments;

[0081] Step S114, marking the collection points that need to be cleaned in the target physiological signal of the first type as a type of abnormality;

[0082] Step S115, identifying the frequency domain distribution, and marking the second type of target physiological signals as the second type of abnormal type based on the abnormal frequency domain distribution; and marking the second type of target physiological signals as the second type of normal type based on the normal frequency domain distribution;

[0083] Step S116, recording the number of the one type of abnormality as a first number, and determining a first collection level of the one type of target physiological signals according to the first number.

[0084] In the specific implementation process, when the second type of target physiological signals are segmented, the segmentation is performed according to the cutting standard of cutting once every 3 minutes;

[0085] In this embodiment, each target physiological signal includes: heart rate, breathing, body movement and snoring; the data processing features include: point processing and segment processing; each target physiological signal is classified according to the data processing features to obtain the first category of target physiological signals and the second category of target physiological signals; wherein the first category of target physiological signals includes: heart rate, breathing and body movement; the second category of target physiological signals includes: snoring.

[0086] Specifically, the process of marking the collection points to be cleaned in the target physiological signal of the type described in this embodiment includes:

[0087] For any type of target physiological signal, the target physiological signal is formed into a plurality of collection points according to the cleaning interval;

[0088] In the specific implementation process, the cleaning interval is set to 1min;

[0089] For any collection point, obtaining actual monitoring data of the collection point, and determining whether the collection point needs to be cleaned according to the actual monitoring data;

[0090] Based on the fact that the actual monitoring data is blank, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality;

[0091] Based on the fact that the actual monitoring data is not within the normal monitoring range, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality;

[0092] Wherein, when the minimum value of the normal monitoring range ≤ the actual monitoring data ≤ the maximum value of the normal monitoring range, it is determined that the collection point does not need to be cleaned;

[0093] When the actual monitoring data is less than the minimum value of the normal monitoring range or the actual monitoring data is greater than the maximum value of the normal monitoring range, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality;

[0094] In this embodiment, when the target physiological signal is heart rate, the corresponding normal monitoring range should be set to 50-100 times per minute; when the target physiological signal is respiration, the corresponding normal monitoring range should be set to 10 to 25 times per minute; when the target physiological signal is body movement, the corresponding normal monitoring range should be set to 1-7 times per minute; the setting of the normal monitoring range is also affected by the influence of age and gender; in this embodiment, the influence of age and gender is to calculate the mean value based on historical experimental data to obtain the normal monitoring range in which the influence of age and gender does not exceed 3%;

[0095] The process of identifying the frequency domain distribution includes:

[0096] For any second-category signal segment, obtaining an actual amplitude of the second-category signal segment, and judging whether the second-category signal segment meets the normal monitoring range according to the actual amplitude;

[0097] Perform amplitude judgment on each of the second-category signal segments in turn, and record the number of each of the second-category signal segments that do not meet the normal monitoring range, which is recorded as a second number;

[0098] determining whether the frequency domain distribution is abnormal according to the second quantity;

[0099] Among them, in this embodiment, when the second type of target physiological signal is snoring, the corresponding normal monitoring range should be set to 30-60 decibels; the specific setting is also affected by the influence of age and gender;

[0100] When the minimum value of the normal monitoring range ≤ the actual amplitude ≤ the maximum value of the normal monitoring range, it is determined that the second type of signal segment is normal and meets the normal monitoring range;

[0101] When the actual amplitude is less than the minimum value of the normal monitoring range or the actual amplitude is greater than the maximum value of the normal monitoring range, it is determined that the second type of signal segment is abnormal and does not conform to the normal monitoring range.

[0102] The embodiment of the present invention classifies different target physiological signals into Class I and Class II target physiological signals through data processing characteristics, so as to carry out subsequent processing in a targeted manner; through the analysis of frequency domain distribution and the comparison of actual monitoring data, it can accurately mark abnormal collection points and Class II signal segments, which is helpful to accurately identify and handle abnormal situations; according to different target physiological signals, the corresponding normal monitoring range is set, considering the influence of factors such as age and gender, making it more universal and adaptable; it can effectively identify the collection points that need to be cleaned, ensure the accuracy and reliability of real-time monitoring data, and effectively improve the efficiency of data processing and the accuracy of results.

[0103] Specifically, the process of determining the first collection level of the target physiological signal of the first type according to the first quantity in this embodiment includes:

[0104] Calculate a first difference absolute value according to the first quantity and a preset first standard value; the first difference absolute value=|first quantity-first standard value|;

[0105] If the absolute value of the first difference is less than or equal to a preset first evaluation value, the first acquisition level is determined to be level 2;

[0106] If the absolute value of the first difference is greater than the first evaluation value and the first quantity is less than the first standard value, then the first acquisition level is determined to be level one;

[0107] If the absolute value of the first difference is greater than the first evaluation value and the first quantity is greater than the first standard value, then the first acquisition level is determined to be level three;

[0108] In this embodiment, the first standard value is set to 8, and the first evaluation value is set to 2; the specific setting is also affected by the actual sleep detection accuracy; in this embodiment, the actual sleep detection accuracy is that the consistency between the detected sleep state and the actual sleep state must reach more than 90%;

[0109] The process of judging whether the frequency domain distribution is abnormal according to the second quantity includes:

[0110] Calculate a second difference absolute value according to the second quantity and a preset second standard value; the second difference absolute value=|second quantity-second standard value|;

[0111] If the absolute value of the second difference is less than or equal to a preset second evaluation value, it is determined that the frequency domain distribution is normal;

[0112] If the absolute value of the second difference is greater than the second evaluation value and the second quantity is less than the second standard value, then it is determined that the frequency domain distribution is normal;

[0113] If the absolute value of the second difference is greater than the second evaluation value and the second quantity is greater than the second standard value, then determining that the frequency domain distribution is abnormal;

[0114] In this embodiment, the second standard value is set to 4, and the second evaluation value is set to 1; the specific setting is also affected by the actual sleep detection accuracy; in this embodiment, the actual sleep detection accuracy is that the consistency between the detected sleep condition and the actual sleep condition must reach more than 90%.

[0115] Specifically, the embodiment of the present invention realizes a quantitative evaluation of the physiological signal processing results by calculating the absolute value of the difference between the first number and the second number and the preset standard value, so that the evaluation of the data is more objective and accurate; it can divide the collection level of a type of target physiological signals into three levels according to the different situations of the first number and the second number, and can judge whether the frequency domain distribution is abnormal; multi-level classification can more finely reflect the results and abnormal situations of data processing, which is helpful for further analysis and processing; it can evaluate in real time according to the actual monitoring data, discover and handle abnormal situations in time, and ensure the timeliness and effectiveness of monitoring; it can effectively improve the efficiency of data processing and the reliability of results.

[0116] Specifically, see Figure 3 As shown, it is a second preprocessing flow chart; wherein, the process of preprocessing each target physiological signal also includes:

[0117] Step S121, determining whether to re-acquire and determining a corresponding re-acquisition mode according to the first acquisition level and the label type of the second type of target physiological signals;

[0118] Step S122, based on the fact that the first collection level meets the collection requirements and the marking type is the second normal type, it is determined that re-collection is not required;

[0119] Step S123, based on the fact that the first acquisition level meets the acquisition requirements and the mark type is the second type of abnormality, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: partial re-acquisition;

[0120] Step S124: based on the fact that the first acquisition level does not meet the acquisition requirements, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: overall re-acquisition.

[0121] When the first acquisition level is level one or level two, it is determined that the first acquisition level meets the acquisition requirements; when the first acquisition level is level three, it is determined that the first acquisition level does not meet the acquisition requirements.

[0122] Specifically, the embodiment of the present invention intelligently determines whether re-collection is needed and the specific mode of collection based on the first collection level and the mark of the abnormal type; the intelligent decision-making can effectively reduce unnecessary data re-collection and improve the efficiency of collection and resource utilization; according to the different classifications of the first collection level, it is targeted to determine whether the collection requirements are met, and then determine the collection mode; the targeting can more accurately reflect the actual situation of data quality and ensure the accuracy and reliability of the data; it can make judgments and decisions in real time during real-time monitoring, adjust the collection strategy according to the actual situation, and ensure the timeliness and effectiveness of the monitoring data; by reasonably determining whether re-collection is needed and the collection mode, the efficiency of data collection and the overall performance of the system can be effectively improved.

[0123] Specifically, the process of synchronously aligning and integrating each standard physiological signal in sequence in this embodiment includes:

[0124] Step S201, identifying the acquisition timestamp of each standard physiological signal;

[0125] Step S202, converting each acquisition time stamp into a unified format to obtain each standard time stamp;

[0126] Step S203, synchronizing each standard time stamp with the target reference time window to obtain each physiological signal to be integrated;

[0127] Step S204: merging the physiological signals to be integrated to obtain the standard sleep data set.

[0128] During the specific implementation process, the acquisition timestamp attached to each standard physiological signal is identified; the acquisition timestamp records the specific time of data acquisition of each physiological signal, which is the basis for subsequent data alignment and integration; because different devices or sensors may use different formats and time zones when recording timestamps; therefore, all collected timestamps are converted to a unified format to ensure that they use the same time base and format for subsequent processing; all standard timestamps are aligned with a pre-set reference time window, which is a specific time period, such as a one-hour sleep period; through alignment, the time synchronization of different physiological signals can be ensured, so that they can be compared and integrated on the same time axis; all synchronized and aligned physiological signals are merged to form a complete standard sleep data set. This data set contains the integrated information of all standard physiological signals on a unified time axis, which can fully reflect the individual's sleep status;

[0129] In this embodiment, a piezoelectric film sensor is embedded in a pillow to collect target physiological signals; the material that produces the piezoelectric effect in the piezoelectric film sensor is a foamed polypropylene foam material (PPPo lypropylene), and the main manufacturing process is to mix and press the main raw materials such as polypropylene resin, polyethylene resin and natural stone powder, and generally use a film with a thickness of 30 to 70 microns; the color is pearly white, the film has good physical properties and flexibility, and is waterproof, durable, and tear-resistant; after reasonable electrical polarization, the film has obvious piezoelectric activity and ferroelectric polymer hysteresis loops, and their piezoelectric properties are derived from the special cavity structure in the material and the space charges of opposite polarities deposited on the two opposite walls of the cavity; that is, the cavity structure in the film can be polarized under a high voltage electric field to produce a piezoelectric effect, thereby obtaining a flexible piezoelectric sensor;

[0130] The embodiment of the present invention ensures the temporal consistency of different physiological signals through unified timestamp format conversion and time alignment, lays the foundation for subsequent data analysis and processing, and can improve the reliability and accuracy of the data; forms a comprehensive standard sleep data set by integrating multiple standard physiological signals; the comprehensiveness helps to have a deeper understanding of the individual's physiological state and sleep quality, and provides richer information support; the unified format of timestamps and synchronous alignment enables data of different physiological signals to be compared and analyzed on the same timeline, improving the comparability and research value of the data; by merging the physiological signals to be integrated, the final standard sleep data set contains comprehensive and synchronized physiological data, providing a complete data basis for subsequent sleep analysis and evaluation.

[0131] Specifically, the process of determining the actual signal combination strategy according to the actual correlation degree in this embodiment includes:

[0132] Step S311, for any two physiological signals to be integrated, determining the actual correlation between the two physiological signals to be integrated;

[0133] Step S312, determining a correlation level between the two physiological signals to be integrated according to the actual correlation degree, and determining a representative physiological signal according to the correlation level combined with a mutual information inclusion relationship between the two physiological signals to be integrated;

[0134] In this embodiment, the actual correlation between the two physiological signals to be integrated is determined by calculating the Pearson correlation coefficient between the two physiological signals to be integrated to determine the actual correlation; the mutual information inclusion relationship is used to evaluate the nonlinear relationship between the two physiological signals to be integrated, and can determine the amount of information contained in one signal about another signal;

[0135] See also Figure 4As shown, it is a correlation level determination logic diagram, wherein the process of determining the correlation level between two physiological signals to be integrated according to the actual correlation degree includes:

[0136] Calculate a third difference absolute value according to the actual correlation degree and the preset standard correlation degree; wherein the third difference absolute value = |actual correlation degree - standard correlation degree|;

[0137] If the absolute value of the third difference is less than or equal to a preset third evaluation value, the correlation level is determined to be level 2;

[0138] If the absolute value of the third difference is greater than the third evaluation value and the actual correlation degree is less than the standard correlation degree, determining that the correlation level is level one;

[0139] If the absolute value of the third difference is greater than the third evaluation value and the actual correlation degree is greater than the standard correlation degree, then the correlation level is determined to be level three;

[0140] In this embodiment, the standard correlation is set to 0.7; the third evaluation value is set to 0.2; the actual correlation value is within [0, 1], and the specific setting is also affected by the actual sleep detection accuracy;

[0141] For two physiological signals to be integrated: the i-th physiological signal to be integrated and the (i+1)-th physiological signal to be integrated, when the correlation level is level three and the mutual information inclusion relationship is that the i-th physiological signal to be integrated includes the (i+1)-th physiological signal to be integrated, the i-th physiological signal to be integrated is determined to be the representative physiological signal.

[0142] Specifically, the embodiment of the present invention quantifies the actual correlation between the physiological signals to be integrated by calculating the Pearson correlation coefficient between them; the quantitative evaluation can objectively reflect the degree of linear correlation between the signals, providing a reliable basis for subsequent signal combination; the correlation of the physiological signals to be integrated is divided into three levels according to the actual correlation and the preset standard correlation; the multi-level classification can more finely describe the relationship between different signals, which is helpful to select appropriate representative physiological signals; the mutual information inclusion relationship is introduced to evaluate the nonlinear relationship between the physiological signals to be integrated; the amount of information and the degree of correlation between the signals can be more comprehensively evaluated, so that the understanding of the complex relationship between the signals is more in-depth and comprehensive; the data set can be effectively simplified and the main information can be retained, thereby improving the efficiency and credibility of subsequent data analysis.

[0143] Specifically, the process of determining the actual signal combination strategy according to the actual correlation degree in this embodiment further includes:

[0144] Step S321, for any two physiological signals to be integrated, based on the presence of the representative physiological signal in the two physiological signals to be integrated, the two physiological signals to be integrated are connected in parallel;

[0145] Step S322, based on the absence of the representative physiological signal in the two physiological signals to be integrated, connecting the two physiological signals to be integrated in series;

[0146] Step S323 , pairing and connecting the physiological signals to be integrated in the standard sleep data set in sequence to form the actual signal combination strategy.

[0147] Specifically, the embodiments of the present invention flexibly select parallel or series strategies according to whether there are representative physiological signals in the physiological signals to be integrated; the implementation can be adjusted according to the actual data characteristics so that the combination strategy is more in line with the actual situation and data characteristics; by pairing and connecting the physiological signals in the standard sleep data set, it is ensured that the actual signal combination strategy finally formed can cover multiple physiological signals, ensuring the integrity and comprehensiveness of the data; through the selection of parallel and series strategies, important information can be better retained and the data set structure can be optimized, which helps to improve the efficiency and accuracy of subsequent data processing and analysis.

[0148] Specifically, in the present embodiment, the process of extracting features from the actual sleep dataset uses an attention mechanism and a hybrid convolutional network to extract features from each physiological signal in the actual sleep dataset; it can effectively capture local patterns and features and has a certain degree of translation invariance; and the characteristics of parameter sharing and local connection make the network more lightweight and have better generalization performance for large-scale datasets.

[0149] Specifically, in this embodiment, inputting the actual feature parameters into the deep learning model includes:

[0150] After using the attention mechanism and the hybrid convolutional network for feature extraction, a gated recurrent unit is introduced. The gated recurrent unit is used to process and predict time series data, can effectively process long-term dependencies, has strong memory capabilities, and can retain and process information in the input sequence for a long time; it has a flexible gating mechanism that can control the flow and update process of information and enhance the expression ability of the network to obtain more accurate sleep staging results.

[0151] Specifically, the process of determining the actual sleep state in this embodiment includes:

[0152] Step S501, comparing the actual feature parameters with the standard feature range in the deep learning model;

[0153] Step S502, determining the actual sleep stage and the sleep quality of the stage according to the comparison result;

[0154] The actual sleep state includes: the actual sleep stage and the sleep quality of the stage.

[0155] In the present embodiment, the standard feature range in the deep learning model is calibrated synchronously with an electrocardiograph and a brain-surge device to determine the standard feature range corresponding to different sleep stages; wherein the different sleep stages include: (I) awake state (II) rapid eye movement (REM) (III) light sleep (IV) deep sleep; in the specific implementation process, other sleep monitoring devices may also be used to obtain each standard feature range, such as: PSG equipment, etc.; this embodiment is not specifically limited;

[0156] In this embodiment, the standard feature range will be periodically trained and optimized. The specific training optimization process is: regression training is performed on the standard feature range based on a device with high accuracy in diagnosing sleep stages (for example, the brain shock device or the PSG device, etc.), so that the judgment accuracy of the deep learning model is higher.

[0157] Specifically, the process of performing corresponding sleep intervention according to the analysis results in this embodiment includes:

[0158] Step S601, recording the actual duration of the actual sleep stage;

[0159] Step S602: Perform sleep staging intervention or sleep stabilization intervention according to the actual duration and the sleep quality of the stage.

[0160] Specifically, the embodiment of the present invention combines the attention mechanism and the hybrid convolutional network for feature extraction, which can more accurately capture the key local patterns and features in the sleep data set, and has a certain degree of translation invariance. Even if the position of the signal on the time axis changes, the model can still recognize its features; the network model adopted reduces the complexity of the model through parameter sharing and local connection, making the model lighter, easier to deploy and process in real time; the model has good generalization performance for large-scale data sets, which means that when faced with unseen data, the model can still maintain a high recognition accuracy rate; the introduction of the gated recurrent unit (GRU) to process and predict time series data can effectively handle long-term dependencies and solve the gradient problem that may occur in traditional RNN when processing long sequences. The disappearance or explosion problem; the strong memory capacity of the GRU unit enables the model to retain and process the information in the input sequence for a long time, which is particularly important for the continuous identification of the state during sleep; the gating mechanism of GRU can control the flow and update process of information, enhance the expression ability of the network, and thus improve the accuracy of the sleep staging results; by comparing the actual feature parameters with the standard feature range, not only the sleep stage can be determined, but also the sleep quality of the stage can be evaluated, providing users with a more comprehensive sleep analysis; based on the analysis results, the user's sleep pattern can be intervened, such as adjusting the duration of sleep or improving sleep quality, so as to put forward personalized sleep improvement suggestions; the accuracy of sleep staging and quality assessment is improved, and ultimately a more accurate and personalized sleep health management plan is provided to users.

[0161] The calculation compensation parameters and the calculation adjustment parameters described in the present invention have two functions: one is to balance the left and right dimensions of the formula, and the other is to adjust the numerical results. No specific assignment is performed in this embodiment. In addition, the calculation formulas in this embodiment are used to intuitively reflect the adjustment relationship between the numerical values, such as positive correlation and negative correlation. Unless otherwise specified, the parameter values ​​that are not specifically limited to numerical values ​​are all positive.

[0162] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0163] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A sleep staging method based on deep learning, characterized in that: include: Collecting a number of target physiological signals, and preprocessing each target physiological signal to obtain a number of standard physiological signals; Each standard physiological signal is synchronously aligned and integrated in turn to obtain a standard sleep data set; Determining the actual correlation between the physiological signals to be integrated in the standard sleep data set, and determining the actual signal combination strategy according to the actual correlation; generating an actual sleep data set according to the actual signal combination strategy, and performing feature extraction on the actual sleep data set to generate actual feature parameters; Inputting the actual characteristic parameters into a deep learning model to determine the actual sleep state; Analyzing the actual sleep state, and performing corresponding sleep intervention according to the analysis result; The process of determining the actual signal combination strategy according to the actual correlation degree includes: For any two physiological signals to be integrated, determining the actual correlation between the two physiological signals to be integrated; Determine a correlation level between the two physiological signals to be integrated according to the actual correlation degree, and determine a representative physiological signal according to the correlation level combined with a mutual information inclusion relationship between the two physiological signals to be integrated; Based on the presence of the representative physiological signal in the two physiological signals to be integrated, the two physiological signals to be integrated are connected in parallel; Based on the absence of the representative physiological signal in the two physiological signals to be integrated, the two physiological signals to be integrated are connected in series; The physiological signals to be integrated in the standard sleep data set are paired and connected in sequence to form the actual signal combination strategy.

2. The sleep staging method based on deep learning according to claim 1, characterized in that: The process of preprocessing each target physiological signal includes: Classifying each target physiological signal according to the data processing characteristics to obtain a number of first-class target physiological signals and second-class target physiological signals; Perform collection point cleaning and filtering processing on each type of target physiological signal in turn; Segmenting the target physiological signals of the two categories to form a plurality of signal segments of the two categories, and determining the frequency domain distribution according to each of the signal segments of the two categories; Marking the collection points that need to be cleaned in the target physiological signal of the first type as a type of abnormality; Identify the frequency domain distribution, and based on the abnormal frequency domain distribution, mark the second type of target physiological signals as the second type of abnormal type; based on the normal frequency domain distribution, mark the second type of target physiological signals as the second type of normal type; The number of the one type of abnormality is recorded as a first quantity, and a first collection level of the one type of target physiological signals is determined according to the first quantity.

3. The sleep staging method based on deep learning according to claim 2, characterized in that: The process of preprocessing each target physiological signal also includes: determining whether to re-acquire and determining a corresponding re-acquisition mode according to the first acquisition level and the label type of the second type of target physiological signals; Based on the fact that the first acquisition level meets the acquisition requirements and the marking type is the second normal type, it is determined that re-acquisition is not required; Based on the fact that the first acquisition level meets the acquisition requirements and the marking type is the second type of abnormality, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: partial re-acquisition; Based on the fact that the first acquisition level does not meet the acquisition requirements, it is determined that re-acquisition is required, and the re-acquisition mode is determined to be: overall re-acquisition.

4. The sleep staging method based on deep learning according to claim 3, characterized in that: The process of marking the collection points to be cleaned in the target physiological signal of the type includes: For any type of target physiological signal, the target physiological signal is formed into a plurality of collection points according to the cleaning interval; For any collection point, obtaining actual monitoring data of the collection point, and determining whether the collection point needs to be cleaned according to the actual monitoring data; Based on the fact that the actual monitoring data is blank, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality; Based on the fact that the actual monitoring data is not within the normal monitoring range, it is determined that the collection point needs to be cleaned, and the collection point is marked as the first type of abnormality.

5. The sleep staging method based on deep learning according to claim 4, characterized in that: The process of identifying the frequency domain distribution includes: For any second-category signal segment, obtaining an actual amplitude of the second-category signal segment, and judging whether the second-category signal segment meets the normal monitoring range according to the actual amplitude; Perform amplitude judgment on each of the second-category signal segments in turn, and record the number of each of the second-category signal segments that do not meet the normal monitoring range, which is recorded as a second number; Whether the frequency domain distribution is abnormal is determined according to the second number.

6. The sleep staging method based on deep learning according to claim 5, characterized in that: The process of synchronously aligning and integrating each standard physiological signal in sequence includes: Identify the acquisition timestamp of each standard physiological signal; Convert each acquisition time stamp into a unified format to obtain each standard time stamp; Synchronously aligning each standard time stamp with the target reference time window to obtain each physiological signal to be integrated; The physiological signals to be integrated are combined to obtain the standard sleep data set.

7. The sleep staging method based on deep learning according to claim 6, characterized in that: The process of determining the actual sleep state includes: Comparing the actual feature parameters with the standard feature ranges in the deep learning model; Determine the actual sleep stage and stage sleep quality based on the comparison results; The actual sleep state includes: the actual sleep stage and the sleep quality of the stage.

8. The sleep staging method based on deep learning according to claim 7, characterized in that: The process of performing corresponding sleep intervention according to the analysis results includes: recording the actual duration of the actual sleep stage; Sleep staging intervention or sleep stabilization intervention is performed according to the actual duration and the sleep quality of the stage.

Images