Sleeping posture recognition and evaluation tuning method based on optical fiber sensing data modeling

Abstract

The invention relates to the field of sleeping posture analysis and evaluation, and discloses a sleeping posture recognition and evaluation tuning method based on optical fiber sensing data modeling, which comprises the following steps of: acquiring pressure data when a user sleeps on a target intelligent bed through optical fiber sensing equipment, constructing a dynamic pressure distribution topological graph, analyzing the dynamic pressure distribution topological graph, and determining the sleeping posture of the user according to the dynamic pressure distribution topological graph. The purpose of evaluating the sleeping posture quality of the user is achieved, the sleeping posture quality score of the user is obtained, and personalized sleeping posture adjusting and optimizing suggestions are made according to the score value. According to the invention, no equipment needs to be worn, no camera is needed, real non-inductive monitoring is realized, and the cost is low.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of sleep posture analysis and evaluation, in particular to a sleep posture recognition and evaluation optimization method based on optical fiber sensing data modeling. BACKGROUND

[0002] The traditional sleep posture recognition methods include, but are not limited to, visual image analysis, wearable sensor analysis, piezoelectric film method analysis, etc. Among them, the visual image analysis is to record sleep video by using infrared or ordinary camera, and to analyze human posture by image processing algorithm, but it has the problems of needing light or infrared fill light, difficulty in recognition after being covered, and only being able to obtain surface posture, and being unable to sense physiological information such as pressure distribution and micro-motion; the wearable sensor analysis is to paste inertial measurement unit sensors on the main parts of the body, and to calculate the posture by accelerometer and gyroscope data. However, the sensors and cables will bring discomfort, interfere with normal sleep, are not suitable for long-term home use, and have the risk of falling off; the piezoelectric film method is to lay a grid-shaped pressure sensor array under the mattress, and to recognize the posture by pressure point distribution image, but the number of sensor nodes is limited, the pressure image formed is rough, and it is difficult to accurately recognize complex postures. The above traditional methods have obvious shortcomings in terms of user experience, privacy protection, data quality, etc.

[0003] The sleep posture recognition and evaluation method based on optical fiber sensing data modeling can long-term monitor sleep posture preference and quality, and evaluate its influence on spine health and blood circulation. For sleep apnea patients, it can recognize and guide them to adopt a posture such as lateral recumbency that can alleviate symptoms, and play a non-drug intervention role. Moreover, the optical fiber sensor is completely embedded in the mattress, without the need to wear any equipment or camera, realizing true "unconscious monitoring", protecting user privacy to the greatest extent, being suitable for long-term home use, and being a low-cost and high-availability technology that helps to collect massive and real sleep data and promote sleep science and ergonomics research. Therefore, the sleep posture recognition and evaluation optimization method based on optical fiber sensing data modeling is proposed. SUMMARY

[0004] The present application overcomes the shortcomings of the prior art and provides a sleep posture recognition and evaluation optimization method based on optical fiber sensing data modeling.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: The present application provides a sleep posture recognition and evaluation optimization method based on optical fiber sensing data modeling in the first aspect, comprising the following steps: Laying optical fiber sensing equipment, performing dynamic pressure sensing data measurement on the intelligent bed, and constructing a dynamic pressure distribution topology graph; Introducing a spatio-temporal graph convolution method to perform convolution analysis on the dynamic pressure distribution topology graph, and extracting sleep posture features of the user on the target intelligent bed. The sleep posture features of the user on the target smart bed are recognized by a hidden variable modeling technology, and sleep posture quality evaluation is performed; The sleep posture quality of the user is evaluated again, and the initial evaluation report of the sleep posture quality of the user is updated according to the evaluation result to obtain a sleep posture quality score of the user; The sleep posture quality score of the user, the sleep stage portrait of the user is constructed, and the personalized optimization suggestion is formulated in combination with the sleep stage portrait.

[0006] Further, in a preferred embodiment of the present application, the optical fiber sensing device is arranged on the smart bed to measure dynamic pressure sensing data and construct a dynamic pressure distribution topology, specifically: A target smart bed is obtained, and a distributed optical fiber sensor is arranged on the target smart bed to measure pressure data and deformation data on the mattress surface of the target smart bed when the user sleeps, which are calibrated as pressure sensing data and deformation sensing data; Wherein, the pressure sensing data and deformation sensing data on the target smart bed are distributed data, a pressure-strain curve is constructed, and the absolute pressure values at different positions on the target smart bed are obtained according to the pressure-strain curve, which are calibrated as target absolute pressure values; The Otsu method is introduced to calculate the global threshold value for each position of the target absolute pressure value on the target smart bed, and the pressure distribution binarization processing is performed on the target smart bed, wherein the pressure distribution binarization processing is to calibrate the position where the target absolute pressure value is within the corresponding global threshold value as a pressure effective position, and to construct a pressure effective region in combination with all the pressure effective positions; A pressure distribution topology is constructed according to the pressure effective region, wherein the pressure distribution topology is annotated with different pressure effective positions and corresponding target absolute pressure values, a pressure effective position is defined as a vertex, and the spatial coordinates of different vertices are calculated; When the user sleeps on the target smart bed, the pressure distribution topology is updated in real time according to the change of different pressure effective positions of the target smart bed, and the real-time time point when different vertices receive pressure is introduced in the pressure distribution topology, and a dynamic pressure distribution topology is output.

[0007] Further, in a preferred embodiment of the present application, the space-time graph convolution method is introduced to perform convolution analysis on the dynamic pressure distribution topology, and the sleep posture features of the user on the target smart bed are extracted, specifically: The vertex feature of the dynamic pressure distribution topology is normalized, i.e. the target absolute pressure value and the spatial coordinates at the vertex in the dynamic pressure distribution topology are normalized to the range of (0, 1) to obtain a normalized dynamic pressure distribution topology; Spatial graph convolution processing is performed on the normalized dynamic pressure distribution topology graph. Specifically, vertex feature filtering is performed using the Laplacian matrix of the normalized dynamic pressure distribution topology graph to aggregate adjacent vertices in the normalized dynamic pressure distribution topology graph, thereby creating a relationship between adjacent vertices and achieving the purpose of spatial graph convolution. Among them, the vertex features on the normalized dynamic pressure distribution topology map after spatial graph convolution describe the pressure distribution area and the corresponding target absolute pressure value on the target smart bed under different combinations of user body parts. The normalized dynamic pressure distribution topology graph after spatial graph convolution is processed by temporal graph convolution. The temporal graph convolution process involves pre-setting T consecutive time steps to construct a vertex feature sequence of the normalized dynamic pressure distribution topology graph after spatial graph convolution for T consecutive time steps. Convolution is then performed on the vertex feature sequence to capture micro-motion features and macro-motion features in the vertex features. Finally, the convolved vertex features are fused at multiple scales to output the normalized dynamic pressure distribution topology graph after temporal graph convolution and spatial graph convolution, which is labeled as the target dynamic pressure distribution topology graph. On the target dynamic pressure distribution topology map, different vertex features are pooled and the pooled vertex features are extracted in a hierarchical manner to obtain the global features of the user on the target smart bed, that is, the sleeping posture features of the user on the target smart bed. Among them, the user's sleeping posture characteristics on the target smart bed are divided into sleeping posture types, including supine, left lateral, right lateral and prone.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of identifying the user's sleeping posture characteristics on the target smart bed and assessing the quality of the sleeping posture through latent variable modeling technology specifically involves: Collect basic user information, including height, weight, and BMI, and also collect environmental information about the target smart bed's location. A conditional variational autoencoder is constructed, and conditional information and feature information are input into the conditional variational autoencoder. The conditional information includes the user's basic information and the environmental information of the target smart bed. The feature information is the user's sleeping posture characteristics on the target smart bed. The conditional information and feature information are mapped onto the latent variable information of the conditional variational autoencoder to predict and identify sleeping posture. The method for predicting and identifying sleeping posture is to calculate the mean and variance of the feature information and train the conditional variational autoencoder through the reparameter method. That is, the mean and variance of the feature information are iteratively calculated. Each iteration generates a latent variable, and each latent variable corresponds to a sleeping posture. The sleeping posture state refers to the predicted probability of the user changing to different sleeping posture types based on the current sleeping posture type, as well as the duration of maintaining different sleeping posture types. When the number of iterations equals the target number, the generation of latent variable information stops, and all latent variable information is aggregated to construct a latent space. Different sleeping postures of users are clustered in the latent space. Combined with the target dynamic pressure distribution topology map, graph topology change analysis is performed on the different sleeping postures of users in the latent space to evaluate the quality of users' sleeping postures.

[0009] Furthermore, in a preferred embodiment of the present invention, the step of combining the target dynamic pressure distribution topology map to perform graph topology change analysis on different sleeping postures of the user in the hidden space and evaluating the quality of the user's sleeping posture specifically includes: In the hidden space, the quality indicators of different sleeping postures of users are classified. The quality indicators are divided into benign indicators and unfavorable indicators. The benign indicator is that the duration of continuous maintenance of a certain sleeping posture is within the standard range, while the unfavorable indicator is that the duration of continuous maintenance of a certain sleeping posture is not within the standard range. When the user's sleeping posture is an unfavorable indicator, vertex mutation analysis is performed on the target dynamic pressure distribution topology map to identify the user's turning events. Vertex mutation analysis is to analyze the rate of change of the target absolute pressure value corresponding to the pressure distribution area on the target smart bed. If the rate of change of the target absolute pressure value remains within the preset range, it is determined that the user is able to turn over smoothly on the target smart bed; if the rate of change of the target absolute pressure value does not remain within the preset range, it is determined that the user is struggling to turn over on the target smart bed. The system performs a set analysis on the quality indicators of the user's sleeping posture and the user's turning over events, and outputs an initial assessment report of the user's sleeping posture quality. The initial assessment report of the sleeping posture quality records the user's sleeping posture at different time points during sleep, as well as the corresponding quality indicators and turning over events at different time points.

[0010] Furthermore, in a preferred embodiment of the present invention, the step of performing a secondary assessment of the user's sleeping posture quality and updating the user's initial sleeping posture quality assessment report based on the secondary assessment results to obtain the user's sleeping posture quality score specifically involves: Based on the user's basic information and the pressure distribution area on the target smart bed, the placement area of ​​the user's spine on the target smart bed is located and marked as the spine distribution area. At the same time, the distribution areas of the user's shoulders and hips are also located. Based on the distribution areas of the spine, shoulders, and hips, the center point of each area is determined, and the center points of each area are connected to obtain the shoulder-spine-hip line, while the central axis of the target smart bed is obtained. Measure the angle between the line connecting the shoulder, spine, and hip and the central axis of the target smart bed, and calibrate it as the spinal analysis angle. Based on the spinal analysis angle, calculate the alignment between the spine and the target smart bed. The system presets a standard alignment range. During the user's sleep, it calculates the time steps in which the alignment between the spine and the target smart bed is not maintained within the standard alignment range, and analyzes the length of the time steps. If the length of the time steps is greater than the standard value, the user is deemed to have unqualified spinal stability during sleep; otherwise, it is qualified. The spinal stability status of users during sleep is imported into the initial sleep posture quality assessment report of users for report updates. At the same time, through big data network, scoring standards are introduced into the initial sleep posture quality assessment report of users to generate an updated sleep posture quality assessment report. The scoring criteria include quality indicators for different sleeping postures, scores for the user's turning over events, and scores for the user's spinal stability during sleep. By combining different scores, a user's sleeping posture quality score is output.

[0011] Furthermore, in a preferred embodiment of the present invention, a sleep stage profile is constructed based on the user's sleep posture quality score, and personalized optimization suggestions are formulated in conjunction with the sleep stage profile, specifically as follows: Based on the user's sleep posture quality score, a sleep stage profile is constructed for the user. The sleep stage profile is different for different sleep stages. The sleep stage profile describes the user's turning over events, the duration of different sleep types, and the alignment of the spine with the target smart bed at different sleep times. Based on the sleep stage profile, users are classified into users who do not need optimization and users who need optimization. For users who need optimization, the corresponding sleep stage profile is imported into the big data network to retrieve optimization suggestions that can make the sleeping posture quality score of the users who need optimization pass the test, and these suggestions are stored in the sleeping posture quality update assessment report.

[0012] This invention addresses the technical deficiencies in the prior art and offers the following advantages: It collects pressure data from users sleeping on a target smart bed using fiber optic sensing devices, constructs a dynamic pressure distribution topology map, analyzes this map to assess the quality of the user's sleeping posture, obtains a sleeping posture quality score, and provides personalized sleeping posture optimization suggestions based on the score. This invention achieves true "unobtrusive monitoring" without requiring any worn devices or cameras, while also being cost-effective. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0014] Figure 1 A flowchart is shown for a method for sleep posture recognition, evaluation, and optimization based on fiber optic sensor data modeling. Figure 2 A flowchart illustrating the method for assessing the quality of a user's sleeping posture is shown. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flowchart illustrating a sleep posture recognition, evaluation, and optimization method based on fiber optic sensor data modeling is shown, including the following steps: S102: Deploy fiber optic sensing equipment to measure dynamic pressure sensing data on the smart bed and construct a dynamic pressure distribution topology map. S104: Introducing the spatiotemporal graph convolution method to perform convolution analysis on the dynamic pressure distribution topology map and extract the sleeping posture features of the user on the target smart bed; S106: Using latent variable modeling technology, identify the user's sleeping posture characteristics on the target smart bed and assess the quality of the sleeping posture. S108: Conduct a secondary assessment of the user's sleeping posture quality, and update the user's initial sleeping posture quality assessment report based on the secondary assessment results to obtain the user's sleeping posture quality score. S110: The user's sleeping posture quality score is used to construct a sleep stage profile of the user, and personalized optimization suggestions are formulated based on the sleep stage profile.

[0018] Furthermore, in a preferred embodiment of the present invention, the deployment of fiber optic sensing devices to measure dynamic pressure sensing data on the smart bed and construct a dynamic pressure distribution topology map specifically involves: The target smart bed is acquired, and distributed fiber optic sensors are deployed on the target smart bed to measure the pressure data and deformation data on the mattress surface when the user sleeps on the target smart bed, and these data are calibrated as pressure sensing data and deformation sensing data. Among them, the pressure sensing data and deformation sensing data on the target smart bed are distributed data, and a pressure-strain curve is constructed. The absolute pressure value at different positions on the target smart bed is obtained based on the pressure-strain curve and calibrated as the target absolute pressure value. The Otsu method is introduced to calculate a global threshold for the target absolute pressure value at each position on the target smart bed, and to perform pressure distribution binarization processing on the target smart bed. The pressure distribution binarization processing involves labeling the positions where the target absolute pressure value is within the corresponding global threshold as effective pressure positions, and constructing an effective pressure region by combining all effective pressure positions. A pressure distribution topology map is constructed based on the effective pressure area. Different effective pressure locations and corresponding target absolute pressure values ​​are marked on the pressure distribution topology map. An effective pressure location is defined as a vertex, and the spatial coordinates of different vertices are calculated. When a user sleeps on the target smart bed, the pressure distribution topology map is updated in real time based on the changes in the effective pressure positions of different points on the target smart bed. At the same time, the real-time time points when different vertices receive pressure are introduced into the pressure distribution topology map, and a dynamic pressure distribution topology map is output.

[0019] It should be noted that the target smart bed is equipped with distributed fiber optic sensors to collect pressure and strain data on the mattress surface. These two data points are then fused and calibrated to convert them into absolute pressure values, analyzing the actual pressure points and corresponding values ​​exerted by the user while lying on the bed. The Otsu method can calculate a global threshold for each pressure point, adapting to changes in user weight and lying position by calculating inter-class variance, thus adaptively determining the pressure threshold. The pressure distribution on the target smart bed is binarized, dividing the surface into pressure-affected areas and background areas to construct effective pressure regions. A pressure distribution topology map is then constructed based on these effective pressure regions. This topology map describes the pressure distribution points and values, with each distribution point labeled as a vertex. The set of vertices represents all effective pressure points, and each vertex has spatial coordinates and topological connections. This topology map elevates the raw fiber optic data into a structured representation rich in spatial connectivity and temporal evolution information.

[0020] Furthermore, in a preferred embodiment of the present invention, the introduction of the spatiotemporal graph convolution method to perform convolution analysis on the dynamic pressure distribution topology map and extract the sleeping posture features of the user on the target smart bed specifically includes: Vertex feature normalization is performed on the dynamic pressure distribution topology map, that is, the absolute pressure value and spatial coordinates of the target at the vertex are normalized to the range of (0,1) in the dynamic pressure distribution topology map to obtain the normalized dynamic pressure distribution topology map. Spatial graph convolution processing is performed on the normalized dynamic pressure distribution topology graph. Specifically, vertex feature filtering is performed using the Laplacian matrix of the normalized dynamic pressure distribution topology graph to aggregate adjacent vertices in the normalized dynamic pressure distribution topology graph, thereby creating a relationship between adjacent vertices and achieving the purpose of spatial graph convolution. Among them, the vertex features on the normalized dynamic pressure distribution topology map after spatial graph convolution describe the pressure distribution area and the corresponding target absolute pressure value on the target smart bed under different combinations of user body parts. The normalized dynamic pressure distribution topology graph after spatial graph convolution is processed by temporal graph convolution. The temporal graph convolution process involves pre-setting T consecutive time steps to construct a vertex feature sequence of the normalized dynamic pressure distribution topology graph after spatial graph convolution for T consecutive time steps. Convolution is then performed on the vertex feature sequence to capture micro-motion features and macro-motion features in the vertex features. Finally, the convolved vertex features are fused at multiple scales to output the normalized dynamic pressure distribution topology graph after temporal graph convolution and spatial graph convolution, which is labeled as the target dynamic pressure distribution topology graph. On the target dynamic pressure distribution topology map, different vertex features are pooled and the pooled vertex features are extracted in a hierarchical manner to obtain the global features of the user on the target smart bed, that is, the sleeping posture features of the user on the target smart bed. Among them, the user's sleeping posture characteristics on the target smart bed are divided into sleeping posture types, including supine, left lateral, right lateral and prone.

[0021] It's important to note that the constructed topology graph first undergoes graph structure data normalization, including standardizing vertex features to eliminate the dimensional influence of different user weights and lying positions. After normalization, spatial graph convolution is used to learn the spatial structural features of sleeping postures. Since traditional CNNs can only perform convolutions on regular grid data, and topology graphs are irregular non-Euclidean structures, standard convolution cannot be directly applied. Therefore, the Laplacian matrix of the graph is used to filter the signals. The core idea is that when each node updates its features, it aggregates the feature information of its direct neighbors. After aggregating adjacent vertices, all vertices are associated, achieving the purpose of spatial graph convolution. Temporal graph convolution captures the dynamic changes in sleeping postures. The convolution kernel slides along the time axis to capture features over time. Finally, pooling vertex features achieves hierarchical feature extraction. Different levels of vertex features differ, including the changing patterns of individual pressure points and limb movement coordination.

[0022] Furthermore, in a preferred embodiment of the present invention, the step of performing a secondary assessment of the user's sleeping posture quality and updating the user's initial sleeping posture quality assessment report based on the secondary assessment results to obtain the user's sleeping posture quality score specifically involves: Based on the user's basic information and the pressure distribution area on the target smart bed, the placement area of ​​the user's spine on the target smart bed is located and marked as the spine distribution area. At the same time, the distribution areas of the user's shoulders and hips are also located. Based on the distribution areas of the spine, shoulders, and hips, the center point of each area is determined, and the center points of each area are connected to obtain the shoulder-spine-hip line, while the central axis of the target smart bed is obtained. Measure the angle between the line connecting the shoulder, spine, and hip and the central axis of the target smart bed, and calibrate it as the spinal analysis angle. Based on the spinal analysis angle, calculate the alignment between the spine and the target smart bed. The system presets a standard alignment range. During the user's sleep, it calculates the time steps in which the alignment between the spine and the target smart bed is not maintained within the standard alignment range, and analyzes the length of the time steps. If the length of the time steps is greater than the standard value, the user is deemed to have unqualified spinal stability during sleep; otherwise, it is qualified. The spinal stability status of users during sleep is imported into the initial sleep posture quality assessment report of users for report updates. At the same time, through big data network, scoring standards are introduced into the initial sleep posture quality assessment report of users to generate an updated sleep posture quality assessment report. The scoring criteria include quality indicators for different sleeping postures, scores for the user's turning over events, and scores for the user's spinal stability during sleep. By combining different scores, a user's sleeping posture quality score is output.

[0023] It's important to note that while the spine is prone to curvature during sleep, maintaining an incorrect position for extended periods can lead to spinal health problems. Therefore, sleep posture quality analysis needs to consider both spinal curvature and alignment. First, the spine's position is located. Then, the positions of the shoulders and hips are used to draw a line aligned with the bed's central axis for alignment analysis, determining if there are any instabilities in spinal stability. The assessment report is updated based on stability data. Finally, the scores from different criteria are combined to obtain a final score, representing the user's sleep posture quality score.

[0024] Furthermore, in a preferred embodiment of the present invention, a sleep stage profile is constructed based on the user's sleep posture quality score, and personalized optimization suggestions are formulated in conjunction with the sleep stage profile, specifically as follows: Based on the user's sleep posture quality score, a sleep stage profile is constructed for the user. The sleep stage profile is different for different sleep stages. The sleep stage profile describes the user's turning over events, the duration of different sleep types, and the alignment of the spine with the target smart bed at different sleep times. Based on the sleep stage profile, users are classified into users who do not need optimization and users who need optimization. For users who need optimization, the corresponding sleep stage profile is imported into the big data network to retrieve optimization suggestions that can make the sleeping posture quality score of the users who need optimization pass the test, and these suggestions are stored in the sleeping posture quality update assessment report.

[0025] It's important to note that the sleep posture quality score reflects a person's usual sleeping posture, i.e., their personal sleeping habits. Based on these habits, appropriate adjustments can be made. For example, if spinal alignment is low, meaning the spine is in a lateral curvature position while sleeping, it can be suggested that the user use a suitable cushion or other support while sleeping to prevent lateral curvature.

[0026] Figure 2 A flowchart illustrating a method for assessing the quality of a user's sleeping posture is shown, including the following steps: S202: Using latent variable modeling technology, identify the user's sleeping posture characteristics on the target smart bed and assess the quality of the sleeping posture. S204: Combining the target dynamic pressure distribution topology map, perform graph topology change analysis on different sleeping postures of users in the hidden space to evaluate the quality of users' sleeping postures.

[0027] Furthermore, in a preferred embodiment of the present invention, the step of identifying the user's sleeping posture characteristics on the target smart bed and assessing the quality of the sleeping posture through latent variable modeling technology specifically involves: Collect basic user information, including height, weight, and BMI, and also collect environmental information about the target smart bed's location. A conditional variational autoencoder is constructed, and conditional information and feature information are input into the conditional variational autoencoder. The conditional information includes the user's basic information and the environmental information of the target smart bed. The feature information is the user's sleeping posture characteristics on the target smart bed. The conditional information and feature information are mapped onto the latent variable information of the conditional variational autoencoder to predict and identify sleeping posture. The method for predicting and identifying sleeping posture is to calculate the mean and variance of the feature information and train the conditional variational autoencoder through the reparameter method. That is, the mean and variance of the feature information are iteratively calculated. Each iteration generates a latent variable, and each latent variable corresponds to a sleeping posture. The sleeping posture state refers to the predicted probability of the user changing to different sleeping posture types based on the current sleeping posture type, as well as the duration of maintaining different sleeping posture types. When the number of iterations equals the target number, the generation of latent variable information stops, and all latent variable information is aggregated to construct a latent space. Different sleeping postures of users are clustered in the latent space. Combined with the target dynamic pressure distribution topology map, graph topology change analysis is performed on the different sleeping postures of users in the latent space to evaluate the quality of users' sleeping postures.

[0028] It should be noted that basic user information is collected to construct a conditional variational autoencoder, aiming to model and interpret users' sleeping posture states. The encoder contains extracted sleeping posture features of the users on the target smart bed, used as feature information, along with conditional information, to map onto latent variables for constructing sleeping posture states. All latent variables exist in the cause space, clustering the sleeping posture states of different users.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of combining the target dynamic pressure distribution topology map to perform graph topology change analysis on different sleeping postures of the user in the hidden space and evaluating the quality of the user's sleeping posture specifically includes: In the hidden space, the quality indicators of different sleeping postures of users are classified. The quality indicators are divided into benign indicators and unfavorable indicators. The benign indicator is that the duration of continuous maintenance of a certain sleeping posture is within the standard range, while the unfavorable indicator is that the duration of continuous maintenance of a certain sleeping posture is not within the standard range. When the user's sleeping posture is an unfavorable indicator, vertex mutation analysis is performed on the target dynamic pressure distribution topology map to identify the user's turning events. Vertex mutation analysis is to analyze the rate of change of the target absolute pressure value corresponding to the pressure distribution area on the target smart bed. If the rate of change of the target absolute pressure value remains within the preset range, it is determined that the user is able to turn over smoothly on the target smart bed; if the rate of change of the target absolute pressure value does not remain within the preset range, it is determined that the user is struggling to turn over on the target smart bed. The system performs a set analysis on the quality indicators of the user's sleeping posture and the user's turning over events, and outputs an initial assessment report of the user's sleeping posture quality. The initial assessment report of the sleeping posture quality records the user's sleeping posture at different time points during sleep, as well as the corresponding quality indicators and turning over events at different time points.

[0030] It's important to note that user sleep quality assessment primarily involves analyzing the number of times a user turns over during sleep, the number of turning events, and the duration of maintaining a particular sleep position. First, the duration of maintaining a sleep position is categorized as either beneficial or detrimental. Beneficial indicators include maintaining a single sleep position for a relatively long period, while detrimental indicators include excessively short periods and frequent shifts, which may suggest restlessness, pain, or discomfort. By analyzing vertex mutations in the topology graph, turning events are identified, determining whether the user's turning is smooth or relatively labored. Different users have different sleep position quality indicators and different number of turning events, representing different sleep quality levels. A quality assessment report is then constructed, recording the user's sleep position at different times during sleep, along with the corresponding quality indicators and the number of turning events at those times.

[0031] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for sleep posture recognition, evaluation, and optimization based on fiber optic sensor data modeling, characterized in that, Includes the following steps: Fiber optic sensing devices are deployed to measure dynamic pressure sensing data on the smart bed and to construct a dynamic pressure distribution topology map. Spatiotemporal graph convolution method is introduced to perform convolution analysis on dynamic pressure distribution topology map and extract the sleeping posture features of users on the target smart bed; Using latent variable modeling technology, the user's sleeping posture characteristics on the target smart bed are identified, and the quality of the sleeping posture is assessed, specifically: Collect basic user information, including height, weight, and BMI, and also collect environmental information about the target smart bed's location. A conditional variational autoencoder is constructed, and conditional information and feature information are input into the conditional variational autoencoder. The conditional information includes the user's basic information and the environmental information of the target smart bed. The feature information is the user's sleeping posture characteristics on the target smart bed. The conditional information and feature information are mapped onto the latent variable information of the conditional variational autoencoder to predict and identify sleeping posture. The method for predicting and identifying sleeping posture is to calculate the mean and variance of the feature information and train the conditional variational autoencoder through the reparameter method. That is, the mean and variance of the feature information are iteratively calculated. Each iteration generates a latent variable, and each latent variable corresponds to a sleeping posture. The sleeping posture state refers to the predicted probability of the user changing to different sleeping posture types based on the current sleeping posture type, as well as the duration of maintaining different sleeping posture types. When the number of iterations equals the target number, the generation of latent variable information stops, and all latent variable information is aggregated to construct a latent space. Different sleeping postures of users are clustered in the latent space. Combined with the target dynamic pressure distribution topology map, graph topology change analysis is performed on the different sleeping postures of users in the latent space to evaluate the quality of users' sleeping postures. A secondary assessment of the user's sleeping posture quality is conducted, and the initial assessment report of the user's sleeping posture quality is updated based on the results of the secondary assessment to obtain the user's sleeping posture quality score. The system scores users' sleeping posture quality to create a sleep stage profile, and then uses this profile to develop personalized optimization suggestions.

2. The sleeping posture recognition and evaluation optimization method based on fiber optic sensor data modeling as described in claim 1, characterized in that, The fiber optic sensing device is deployed to measure dynamic pressure sensing data on the smart bed and construct a dynamic pressure distribution topology map, specifically as follows: The target smart bed is acquired, and distributed fiber optic sensors are deployed on the target smart bed to measure the pressure data and deformation data on the mattress surface when the user sleeps on the target smart bed, and these data are calibrated as pressure sensing data and deformation sensing data. Among them, the pressure sensing data and deformation sensing data on the target smart bed are distributed data, and a pressure-strain curve is constructed. The absolute pressure value at different positions on the target smart bed is obtained based on the pressure-strain curve and calibrated as the target absolute pressure value. The Otsu method is introduced to calculate a global threshold for the target absolute pressure value at each position on the target smart bed, and to perform pressure distribution binarization processing on the target smart bed. The pressure distribution binarization processing involves labeling the positions where the target absolute pressure value is within the corresponding global threshold as effective pressure positions, and constructing an effective pressure region by combining all effective pressure positions. A pressure distribution topology map is constructed based on the effective pressure area. Different effective pressure locations and corresponding target absolute pressure values ​​are marked on the pressure distribution topology map. An effective pressure location is defined as a vertex, and the spatial coordinates of different vertices are calculated. When a user sleeps on the target smart bed, the pressure distribution topology map is updated in real time based on the changes in the effective pressure positions of different points on the target smart bed. At the same time, the real-time time points when different vertices receive pressure are introduced into the pressure distribution topology map, and a dynamic pressure distribution topology map is output.

3. The method for sleep posture recognition, evaluation, and optimization based on fiber optic sensor data modeling as described in claim 1, characterized in that, The introduced spatiotemporal graph convolution method performs convolution analysis on the dynamic pressure distribution topology map to extract the user's sleeping posture features on the target smart bed, specifically: Vertex feature normalization is performed on the dynamic pressure distribution topology map, that is, the absolute pressure value and spatial coordinates of the target at the vertex are normalized to the range of (0,1) in the dynamic pressure distribution topology map to obtain the normalized dynamic pressure distribution topology map. Spatial graph convolution processing is performed on the normalized dynamic pressure distribution topology graph. Specifically, vertex feature filtering is performed using the Laplacian matrix of the normalized dynamic pressure distribution topology graph to aggregate adjacent vertices in the normalized dynamic pressure distribution topology graph, thereby creating a relationship between adjacent vertices and achieving the purpose of spatial graph convolution. Among them, the vertex features on the normalized dynamic pressure distribution topology map after spatial graph convolution describe the pressure distribution area and the corresponding target absolute pressure value on the target smart bed under different combinations of user body parts. The normalized dynamic pressure distribution topology graph after spatial graph convolution is processed by temporal graph convolution. The temporal graph convolution process involves pre-setting T consecutive time steps to construct a vertex feature sequence of the normalized dynamic pressure distribution topology graph after spatial graph convolution for T consecutive time steps. Convolution is then performed on the vertex feature sequence to capture micro-motion features and macro-motion features in the vertex features. Finally, the convolved vertex features are fused at multiple scales to output the normalized dynamic pressure distribution topology graph after temporal graph convolution and spatial graph convolution, which is labeled as the target dynamic pressure distribution topology graph. On the target dynamic pressure distribution topology map, different vertex features are pooled and the pooled vertex features are extracted in a hierarchical manner to obtain the global features of the user on the target smart bed, that is, the sleeping posture features of the user on the target smart bed. Among them, the user's sleeping posture characteristics on the target smart bed are divided into sleeping posture types, including supine, left lateral, right lateral and prone.

4. The sleeping posture recognition and evaluation optimization method based on fiber optic sensor data modeling as described in claim 1, characterized in that, in, The method involves combining the target dynamic pressure distribution topology map to perform graph topology change analysis on different sleeping postures of the user in the hidden space, and evaluating the quality of the user's sleeping posture. Specifically: In the hidden space, the quality indicators of different sleeping postures of users are classified. The quality indicators are divided into benign indicators and unfavorable indicators. The benign indicator is that the duration of continuous maintenance of a certain sleeping posture is within the standard range, while the unfavorable indicator is that the duration of continuous maintenance of a certain sleeping posture is not within the standard range. When the user's sleeping posture is an unfavorable indicator, vertex mutation analysis is performed on the target dynamic pressure distribution topology map to identify the user's turning events. Vertex mutation analysis is to analyze the rate of change of the target absolute pressure value corresponding to the pressure distribution area on the target smart bed. If the rate of change of the target absolute pressure value remains within the preset range, it is determined that the user is able to turn over smoothly on the target smart bed; if the rate of change of the target absolute pressure value does not remain within the preset range, it is determined that the user is struggling to turn over on the target smart bed. The system performs a set analysis on the quality indicators of the user's sleeping posture and the user's turning over events, and outputs an initial assessment report of the user's sleeping posture quality. The initial assessment report of the sleeping posture quality records the user's sleeping posture at different time points during sleep, as well as the corresponding quality indicators and turning over events at different time points.

5. The method for sleep posture recognition, evaluation, and optimization based on fiber optic sensor data modeling as described in claim 1, characterized in that, The process involves conducting a secondary assessment of the user's sleeping posture quality and updating the initial assessment report based on the results of this secondary assessment to obtain the user's sleeping posture quality score. Specifically: Based on the user's basic information and the pressure distribution area on the target smart bed, the placement area of ​​the user's spine on the target smart bed is located and marked as the spine distribution area. At the same time, the distribution areas of the user's shoulders and hips are also located. Based on the distribution areas of the spine, shoulders, and hips, the center point of each area is determined, and the center points of each area are connected to obtain the shoulder-spine-hip line, while the central axis of the target smart bed is obtained. Measure the angle between the line connecting the shoulder, spine, and hip and the central axis of the target smart bed, and calibrate it as the spinal analysis angle. Based on the spinal analysis angle, calculate the alignment between the spine and the target smart bed. The system presets a standard alignment range. During the user's sleep, it calculates the time steps in which the alignment between the spine and the target smart bed is not maintained within the standard alignment range, and analyzes the length of the time steps. If the length of the time steps is greater than the standard value, the user is deemed to have unqualified spinal stability during sleep; otherwise, it is qualified. The spinal stability status of users during sleep is imported into the initial sleep posture quality assessment report of users for report updates. At the same time, through big data network, scoring standards are introduced into the initial sleep posture quality assessment report of users to generate an updated sleep posture quality assessment report. The scoring criteria include quality indicators for different sleeping postures, scores for the user's turning over events, and scores for the user's spinal stability during sleep. By combining different scores, a user's sleeping posture quality score is output.

6. The method for sleep posture recognition, evaluation, and optimization based on fiber optic sensor data modeling as described in claim 1, characterized in that, Based on users' sleeping posture quality scores, a sleep stage profile is constructed, and personalized optimization suggestions are formulated based on the sleep stage profile, specifically: Based on the user's sleep posture quality score, a sleep stage profile is constructed for the user. The sleep stage profile is different for different sleep stages. The sleep stage profile describes the user's turning over events, the duration of different sleep types, and the alignment of the spine with the target smart bed at different sleep times. Based on the sleep stage profile, users are classified into users who do not need optimization and users who need optimization. For users who need optimization, the corresponding sleep stage profile is imported into the big data network to retrieve optimization suggestions that can make the sleeping posture quality score of the users who need optimization pass the test, and these suggestions are stored in the sleeping posture quality update assessment report.

Images