CN113261951B - Sleeping posture identification method and device based on piezoelectric ceramic sensor - Google Patents

Abstract

The present invention provides a method and device for recognizing sleeping posture based on piezoelectric ceramic sensors, including: when a user is in a stable sleep state, based on a piezoelectric ceramic sensor system, obtaining a mixed cardiac shock signal of the user's chest and abdomen area; The cardiac shock signal is used to determine the user's cardiopulmonary activity distribution characteristics; the cardiopulmonary activity distribution characteristics and the preset environment vector characteristics are input into the trained sleeping posture recognition classification network model to obtain the user's sleeping posture recognition result. In the method of the present invention, when the user is in a stable sleep state, the mixed cardiac shock signal of the user's chest and abdomen area is collected by a multi-channel piezoelectric ceramic sensor, and the signal processing is performed to obtain the distribution characteristics of the cardiopulmonary activity of the user, so as to obtain the distribution characteristics of the cardiopulmonary activity. Input the trained sleeping posture recognition and classification network model with preset environment vector features, obtain the user's sleeping posture recognition result, and realize non-invasive real-time user sleeping posture monitoring, which has strong universality and environmental anti-interference ability.

Description

A sleeping posture recognition method and device based on piezoelectric ceramic sensor

technical field

The invention relates to the technical field of sleep monitoring, in particular to a method and device for recognizing sleeping posture based on piezoelectric ceramic sensors.

Background technique

With the improvement of living standards, more and more people begin to pay attention to their sleep status.

For sleep monitoring, currently the commonly used methods to monitor the user's sleeping posture and related data states include Poly Somno Graph (PSG) and Cyclic Alternating Pattern (CAP) in sleep, which use gyroscopes to assist video monitoring. The system determines the user's sleeping position throughout the night. This method obtains the user's physiological parameters through contact measurement. The user needs to wear more than a dozen electrodes and a tension sensor all night for testing to get a complete report. The whole testing process is cumbersome, and it is easy to cause the "first night effect" for users. At the same time, such an intrusive test also brings physical and psychological burdens to the user, which directly affects the user's normal sleeping posture data state.

Therefore, how to better realize sleep posture recognition has become the focus of research in the industry.

SUMMARY OF THE INVENTION

The present invention provides a method and device for recognizing a sleeping posture based on a piezoelectric ceramic sensor, so as to better realize the sleeping posture recognition.

The present invention provides a sleeping posture recognition method based on a piezoelectric ceramic sensor, comprising:

When the user is in a stable sleep state, the hybrid cardiac shock signal of the thoracic and abdominal region of the user is obtained based on the piezoelectric ceramic sensor system, wherein the piezoelectric ceramic sensor system includes a plurality of piezoelectric ceramic sensors;

determining the distribution characteristics of the user's cardiopulmonary activity based on the mixed cardiac shock signal;

Inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleeping posture recognition classification network model to obtain the user's sleeping posture recognition result;

Wherein, the trained sleeping posture recognition and classification network model is obtained by training according to the cardiopulmonary activity distribution features and the environmental vector feature samples carrying the sleeping posture labels.

According to a method for identifying a sleeping posture based on a piezoelectric ceramic sensor provided by the present invention, when the user is in a stable sleep state, the hybrid cardiac shock signal of the user's chest and abdomen area is obtained based on the piezoelectric ceramic sensor system, specifically: :

acquiring first pressure information applied to the piezoelectric ceramic sensor system by the user's thoracic and abdominal region;

determining the first voltage information corresponding to the first pressure information according to the mapping relationship between the pressure and the output voltage;

Judging the sleep state of the user according to the variation of the first voltage information, and determining the second voltage information of the user in a stable sleep state;

According to the variation of the second voltage information, the mixed cardiac shock signal of the user in a stable sleep state is determined.

According to a sleeping posture recognition method based on a piezoelectric ceramic sensor provided by the present invention, based on the mixed cardiac shock signal, the distribution characteristics of the cardiopulmonary activity of the user are determined, specifically:

Based on the mixed cardiac shock signal, calculate the cardiopulmonary activity intensity feature of the user;

According to the characteristics of the cardiopulmonary activity intensity of the user and the parameter elimination processing parameters in the piezoelectric ceramic sensor system, the distribution characteristics of the cardiopulmonary activity of the user are determined.

According to a method for recognizing a sleeping posture based on a piezoelectric ceramic sensor provided by the present invention, based on the mixed cardiac shock signal, the characteristics of the cardiopulmonary activity intensity of the user are calculated, specifically:

extracting information from the mixed cardiac shock signal to obtain the cardiopulmonary activity signal voltage information of the user;

Calculate the cardiopulmonary activity stress amplitude information of the user according to the user's cardiorespiratory activity signal voltage information and approximate estimation based on signal differentiation and McLaughlin's formula;

The feature extraction of the preset signal length is performed on the cardiopulmonary activity stress amplitude information of the user to obtain the cardiopulmonary activity intensity feature of the user.

According to a sleeping posture recognition method based on a piezoelectric ceramic sensor provided by the present invention, the cardiopulmonary activity distribution of the user is determined according to the cardiopulmonary activity intensity feature of the user and the parameter elimination processing parameters in the piezoelectric ceramic sensor system. Features, specifically:

Calculate the cardiopulmonary activity amplitude information collected by each sensor according to the user's cardiorespiratory activity intensity feature and the parameter elimination processing parameters in the piezoelectric ceramic sensor system;

Wherein, the parameter elimination processing parameters are obtained by calculating the ratio of the parameters of the remaining sensors to the reference parameters;

Wherein, the reference parameter is set by selecting the sensor parameter with the smallest parameter and working normally in the piezoelectric ceramic sensor system;

According to the cardiopulmonary activity amplitude information collected by the various sensors, the cardiorespiratory activity distribution characteristics of the user are determined.

According to a sleeping posture recognition method based on a piezoelectric ceramic sensor provided by the present invention, before inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleeping posture recognition classification network model, the method further includes:

Obtain multiple samples of cardiorespiratory activity distribution features and environmental vector features carrying sleeping posture labels; take each cardiopulmonary activity distribution feature and environmental vector feature samples carrying sleeping posture labels as a set of training samples to obtain multiple sets of training samples, and use Multiple sets of training samples are used to train the network model of sleeping posture recognition and classification.

According to a sleeping posture recognition method based on a piezoelectric ceramic sensor provided by the present invention, the sleep posture recognition classification network model is trained by using multiple sets of training samples, specifically:

For any set of training samples, input the training samples into the sleeping posture recognition and classification network model, and output the prediction probability corresponding to the training samples;

Using a preset loss function, calculate the loss value according to the predicted probability corresponding to the training sample and the sleeping position label in the training sample;

If the loss value is less than the preset threshold, the training of the sleeping posture recognition and classification network model is completed.

The present invention also provides a sleeping posture recognition device based on the piezoelectric ceramic sensor, comprising:

The hybrid cardiac shock signal acquisition module is used to obtain the hybrid cardiac shock signal of the user's chest and abdomen area based on the piezoelectric ceramic sensor system when the user is in a stable sleep state, wherein the piezoelectric ceramic sensor system includes a plurality of Piezoelectric ceramic sensor;

a cardiopulmonary activity distribution feature generation module, configured to determine the user's cardiopulmonary activity distribution feature based on the mixed cardiac shock signal;

A sleeping posture recognition result generation module, used for inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleeping posture recognition classification network model to obtain the user's sleeping posture recognition result;

Wherein, the trained sleeping posture recognition and classification network model is obtained by training according to the cardiopulmonary activity distribution features and the environmental vector feature samples carrying the sleeping posture labels.

The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, when the processor executes the program, the piezoelectric-based system as described in any of the above-mentioned methods is implemented when the processor executes the program. The steps of the method of sleeping posture recognition of ceramic sensor.

The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of any of the above-mentioned methods for recognizing sleeping posture based on piezoelectric ceramic sensors .

The method and device for recognizing the sleeping posture based on the piezoelectric ceramic sensor provided by the present invention, when the user is in a stable sleep state, collects the mixed cardiac shock signal of the user's chest and abdomen area through the multi-channel piezoelectric ceramic sensor, and after information processing, obtains The user's cardiorespiratory activity distribution characteristics, in order to input the user's cardiorespiratory activity distribution characteristics and preset environment vector characteristics into the trained sleeping posture recognition classification network model, obtain the user's sleeping posture recognition results, and realize non-invasive real-time user sleeping posture monitoring, It has strong universality and environmental anti-interference ability.

Description of drawings

In order to explain the present invention or the technical solutions in the prior art more clearly, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are the For some embodiments of the invention, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without any creative effort.

1 is a schematic flowchart of a method for recognizing a sleeping posture based on a piezoelectric ceramic sensor provided by the present invention;

2 is a schematic flowchart of steps of a method for recognizing a sleeping posture based on a piezoelectric ceramic sensor provided by an embodiment of the present invention;

3 is a schematic diagram showing the comparison of output voltage curves of a user and an equal weight object on a piezoelectric ceramic sensor system in an embodiment of the present invention;

4 is a schematic diagram of a respiratory frequency band signal and an original mixed signal in a mixed cardiac shock signal provided by an embodiment of the present invention;

FIG. 5 is a schematic diagram of the envelope of the shock signal and its heartbeat cycle in the mixed shock signal provided by an embodiment of the present invention;

6 is a schematic diagram of a piezoelectric ceramic sensor voltage amplifying circuit provided by an embodiment of the present invention;

7 is a schematic diagram of an equivalent input circuit of a piezoelectric ceramic sensor voltage amplifier circuit provided by an embodiment of the present invention;

8 is a schematic diagram of a system framework of a method for sleeping posture recognition based on a piezoelectric ceramic sensor provided by the present invention;

9 is a schematic structural diagram of a sleeping posture recognition device based on a piezoelectric ceramic sensor provided by the present invention;

FIG. 10 is a schematic structural diagram of an electronic device provided by the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention. , not all examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

FIG. 1 is a schematic flowchart of a method for recognizing a sleeping posture based on a piezoelectric ceramic sensor provided by the present invention, as shown in FIG. 1 , including:

Step S1, when the user is in a stable sleep state, based on a piezoelectric ceramic sensor system, obtain a mixed cardiac shock signal of the user's chest and abdomen area, wherein the piezoelectric ceramic sensor system includes a plurality of piezoelectric ceramic sensors.

Optionally, the stable sleep state described in the present invention refers to a state in which the user has no obvious limb movements, and the entire body is in a stable and stable state.

The mixed cardiac impact (Ballistocardiogram; BCG) signal of the chest-abdominal region described in the present invention is obtained through signal acquisition by the piezoelectric ceramic sensor system under the chest-abdominal region of the user when the user is in a stable sleep state. Among them, the mixed cardiac shock signal is mainly the voltage change generated by cardiopulmonary activity, which contains a large amount of physiological information, including respiratory signal, cardiac shock signal, pulse signal and so on.

It should be noted that the traditional monitoring method uses a dense sensor matrix to collect the absolute pressure of the whole body. This method uses a large number of sensors, so the cost is high. At the same time, multiple sensors also greatly increase the cost of data processing and the response time of the algorithm. In addition, absolute pressure is easily disturbed by external forces. Different people's body shapes, body structures, and even their sleeping habits will also affect the pressure distribution of the mattress. If sleeping position recognition is performed by absolute pressure distribution, a large amount of training and testing needs to be performed for complex scenes of various groups of people, which will further increase the experimental cost, and it is difficult to ensure that the trained model has good adaptability and accuracy. Secondly, the use rate of absolute pressure data in sleep monitoring systems is low, and the monitoring of other data states such as human sleep stages, heart rate, and breathing rate requires data from other modalities.

In the embodiment of the present invention, the piezoelectric ceramic sensor system is not a dense sensor matrix system, but includes a plurality of piezoelectric ceramic sensors, wherein, in the piezoelectric ceramic sensor system, when the number of acquisition channels of the piezoelectric ceramic sensor is 10 When the number of routes is ~30, better technical effects can be achieved.

Step S2, based on the mixed cardiac shock signal, determine the distribution characteristics of the cardiopulmonary activity of the user.

Optionally, the distribution feature of the cardiopulmonary activity of the user described in the present invention refers to the distribution feature of the user's heartbeat motion amplitude and the distribution feature of the respiratory motion amplitude.

Further, by performing information extraction and processing on the obtained mixed cardiac shock signal, the distribution characteristics of the breathing motion amplitude and the distribution characteristics of the heartbeat motion amplitude can be obtained, so that the user's cardiopulmonary activity distribution characteristics can be determined.

Step S3, inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleep posture recognition classification network model to obtain the user's sleeping posture recognition result.

Optionally, the trained sleeping posture recognition and classification network model described in the present invention is obtained by training according to the model training samples, and is used to perform the user's sleeping posture recognition on the input cardiopulmonary activity distribution characteristics, and output the recognition result.

Among them, the model training samples are composed of multiple sets of cardiopulmonary activity distribution features and environmental vector feature samples carrying sleeping posture labels, which are used to improve the accuracy of various sleeping posture recognition and the environmental adaptability of the model.

In the embodiments of the present application, in order to adapt to the influence of the environment and consider more abundant scene characteristics, personalized environment vector features are also added, including the ratio of the amplitude of the heartbeat movement to the amplitude of the breathing movement, and the sensor occupied when the user is lying flat. The total number of channels, the total number of sensors occupied by the user while lying on the side, the current data state discrimination and the data state discrimination at the previous moment.

The sleeping position labels described in the present invention are supine, left, right and prone, which are predetermined according to the characteristic samples of cardiopulmonary activity distribution, and are in one-to-one correspondence with the characteristic samples of cardiopulmonary activity distribution. That is to say, each cardiopulmonary activity distribution feature and environment vector feature sample in the training sample is preset to carry a corresponding sleeping position label.

According to the method of the embodiment of the present invention, when the user is in a stable sleep state, based on the piezoelectric ceramic sensor system, the mixed cardiac shock signal of the user's chest and abdomen area is collected through the multi-channel piezoelectric ceramic sensor, and after signal processing, the user's heart and lungs are obtained. Activity distribution characteristics, in order to input the user's cardiopulmonary activity distribution characteristics and preset environment vector characteristics into the trained sleeping posture recognition classification network model, obtain the user's sleeping posture recognition results, and realize non-invasive real-time user sleeping posture monitoring, with strong performance. universality and environmental anti-interference ability.

Optionally, when the user is in a stable sleep state, based on the piezoelectric ceramic sensor system, the mixed cardiac shock signal of the thoracic and abdominal region of the user is obtained, specifically:

acquiring first pressure information applied to the piezoelectric ceramic sensor system by the user's thoracic and abdominal region;

determining the first voltage information corresponding to the first pressure information according to the mapping relationship between the pressure and the output voltage;

Judging the sleep state of the user according to the variation of the first voltage information, and determining the second voltage information of the user in a stable sleep state;

According to the variation of the second voltage information, the mixed cardiac shock signal of the user in a stable sleep state is determined.

Optionally, the first pressure information described in the present invention refers to the pressure information generated by the thoracic and abdominal area on the piezoelectric ceramic sensor system before the user goes to sleep. Here, the pressure information mainly comes from the user's own gravity and the user's breathing. The force produced by the movement and the force produced by the movement of the user's heartbeat.

The mapping relationship between pressure and output voltage described in the present invention refers to the acquisition of external pressure information by the signal acquisition module of the piezoelectric ceramic sensor system of the present invention, and the digital circuit processing to obtain the corresponding output digital voltage information, thereby obtaining the external pressure and the output voltage. The mapping relationship on the system output digital voltage.

The first voltage information described in the present invention refers to the pressure generated by the user's own gravity, the pressure generated by the user's breathing movement, and the pressure generated by the user's heartbeat movement in the first pressure information. Based on the mapping relationship between the pressure and the output voltage, the obtained user The voltage corresponding to self-gravity, the voltage of the user's breathing motion signal, and the voltage of the user's heartbeat motion signal.

Further, according to the mapping relationship between the pressure and the output voltage, the first voltage information corresponding to the first pressure information can be determined.

The second voltage information described in the present invention refers to that when the user is in a stable sleep state, since the user's own gravitational stress is almost unchanged or has a slow change, the output voltage of the system is hardly affected, so the output of the system is respiratory stress and heartbeat stress. Generated voltage change information.

The mixed cardiac shock signal when the user is in a stable sleep state described in the present invention refers to the collection of the mixed cardiac shock signal dominated by cardiopulmonary motion when the user has no obvious limb movement. According to the voltage change, energy entropy, approximate entropy and other information output by the system in the steady state, it can be judged whether the signal is a mixed cardiac shock signal in the steady state.

The method of the embodiment of the present invention obtains corresponding voltage information based on the mapping relationship between the pressure and the output voltage by collecting the pressure information applied to the piezoelectric ceramic sensor system by the user's chest and abdomen area, so as to determine whether the user is not based on the change of the voltage information. The user is in a stable sleep state, and then according to the variation of the voltage information in the stable state, the mixed cardiac shock signal of the user in the stable sleep state is determined.

Optionally, based on the mixed cardiac shock signal, determine the distribution characteristics of the cardiopulmonary activity of the user, specifically:

Based on the mixed cardiac shock signal, calculate the cardiopulmonary activity intensity feature of the user;

According to the characteristics of the cardiopulmonary activity intensity of the user and the parameter elimination processing parameters in the piezoelectric ceramic sensor system, the distribution characteristics of the cardiopulmonary activity of the user are determined.

Optionally, the cardiopulmonary activity intensity feature of the user described in the present invention includes the intensity feature of the user's breathing motion and the intensity feature of the user's heartbeat motion.

The parameter elimination processing parameter described in the present invention refers to the processing parameter determined by calculation in order to eliminate the difference between the inherent parameters of various sensors. Through this processing method, the technical problem that the output voltage of the system is inconsistent due to the differences between the various sensors can be solved.

Further, according to the mixed cardiac shock signal when the user is in a stable sleep state, after signal analysis and processing, cardiopulmonary activity feature extraction is performed to obtain the cardiopulmonary activity intensity feature of the user.

According to the user's cardiorespiratory activity intensity characteristics and the parameter elimination processing parameters in the piezoelectric ceramic sensor system, the user's cardiorespiratory activity intensity characteristics are optimized and calculated, and the user's cardiorespiratory activity distribution characteristics are determined.

The method of the embodiment of the present invention obtains the user's cardiopulmonary activity intensity characteristics by performing signal analysis and feature extraction on the mixed cardiac shock signal when the user is in a stable sleep state, and then accurately calculates on the basis of eliminating the parameter difference between the sensors. The user's cardiopulmonary activity distribution characteristics.

Optionally, based on the mixed cardiac shock signal, calculate the cardiopulmonary activity intensity feature of the user, specifically:

extracting information from the mixed cardiac shock signal to obtain the cardiopulmonary activity signal voltage information of the user;

Calculate the cardiopulmonary activity stress amplitude information of the user according to the user's cardiorespiratory activity signal voltage information and approximate estimation based on signal differentiation and McLaughlin's formula;

The feature extraction of the preset signal length is performed on the cardiopulmonary activity stress amplitude information of the user to obtain the cardiopulmonary activity intensity feature of the user.

Optionally, the information extraction described in the present invention refers to separating and extracting the signal of the respiratory frequency band and the signal of the heartbeat frequency band from the mixed shock cardiac signal.

The user's cardiopulmonary activity signal voltage information described in the present invention refers to the signal in the breathing frequency band and the signal in the heartbeat frequency band. Based on the mapping relationship between pressure and output voltage, the relationship between the voltage of the breathing movement signal and the amplitude and frequency of the breathing movement stress is obtained. And the relationship between the heartbeat motion signal voltage and the heartbeat motion stress amplitude and frequency.

The cardiopulmonary activity stress magnitude information of the user described in the present invention includes the user's breathing motion stress magnitude and the user's heartbeat motion stress magnitude.

In the embodiment of the present invention, the feature extraction of the preset signal length refers to that due to the large difference between the breathing and heartbeat cycles of different users in different scenarios, the user's cardiopulmonary activity stress amplitude information is fixed according to this difference. Feature extraction of signal length.

Further, according to the stress amplitude information of the user's cardiopulmonary activity, the feature extraction formula of the fixed signal length is optimized by using the real-time heart rate and respiration rate of the signal, and the feature of the cardiopulmonary activity intensity of the user can be obtained.

In the embodiment of the present invention, the user's cardiopulmonary activity intensity feature includes the user's breathing exercise intensity feature and the user's heartbeat exercise intensity feature.

In the method of the embodiment of the present invention, by separating the mixed cardiac shock signal, the signal in the breathing frequency band and the signal in the heartbeat frequency band are extracted, and the feature extraction formula of the preset signal length is optimized by using the real-time heart rate and breathing rate of the signal, so as to obtain a more accurate signal. Accurate cardiorespiratory activity intensity profile.

Optionally, according to the cardiorespiratory activity intensity feature of the user and the parameter elimination processing parameters in the piezoelectric ceramic sensor system, determine the cardiorespiratory activity distribution feature of the user, specifically:

Calculate the cardiopulmonary activity amplitude information collected by each sensor according to the user's cardiorespiratory activity intensity feature and the parameter elimination processing parameters in the piezoelectric ceramic sensor system;

Wherein, the parameter elimination processing parameters are obtained by calculating the ratio of the parameters of the remaining sensors to the reference parameters;

Wherein, the reference parameter is set by selecting the sensor with the smallest parameter and working normally in the piezoelectric ceramic sensor system;

According to the cardiopulmonary activity amplitude information collected by the various sensors, the cardiorespiratory activity distribution characteristics of the user are determined.

Optionally, the cardiopulmonary activity amplitude information described in the present invention includes the breathing motion amplitude and the heartbeat motion amplitude.

The user's cardiopulmonary activity distribution feature described in the present invention refers to the real-time distribution feature of the user's cardiopulmonary motion amplitude monitored by various sensors, including the amplitude distribution feature of the user's breathing motion and the amplitude distribution feature of the user's heartbeat motion.

In the embodiment of the present invention, by selecting the sensor parameter with the smallest parameter and working normally in the piezoelectric ceramic sensor system as the reference parameter, and calculating the ratio of the parameters of the remaining sensors to the reference parameter, that is, the parameter elimination processing parameter, the same parameter can be obtained. The amplitude of the breathing movement and the amplitude of the heartbeat movement collected by different sensor channels at the moment are used to determine the amplitude distribution characteristics of the user's breathing movement and the amplitude distribution characteristics of the user's heartbeat movement.

The method of the embodiment of the present invention eliminates the influence of the difference between the sensor parameters by comparing the vibration test of different sensors, and then more accurately determines the user's cardiorespiratory activity distribution characteristics according to the user's cardiorespiratory activity intensity characteristics.

Optionally, before inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleep posture recognition and classification network model, the method further includes:

Obtain multiple samples of cardiopulmonary activity distribution characteristics and environmental vector characteristics carrying sleeping position labels;

Taking each cardiorespiratory activity distribution feature sample carrying the sleeping posture label as a set of training samples, multiple sets of training samples are obtained, and the sleeping posture recognition and classification network model is trained by using the multiple sets of training samples.

Optionally, before inputting the cardiopulmonary activity distribution feature and the preset environment vector feature into the sleeping posture recognition and classification network model, the sleeping posture recognition and classification network model needs to be trained, and the specific training process is as follows:

Take the cardiorespiratory activity distribution feature and environmental vector feature sample and the sleeping posture label carried by the cardiorespiratory activity distribution feature sample as a set of training samples, that is, each cardiorespiratory activity distribution feature and environmental vector feature sample with a sleeping posture label as a set of training samples , so that multiple sets of training samples can be obtained.

In the embodiment of the present invention, the cardiopulmonary activity distribution features and the environmental vector feature samples are in a one-to-one correspondence with the sleeping posture labels carried by the cardiopulmonary activity distribution feature samples.

Then, after obtaining multiple sets of training samples, the multiple sets of training samples are input into the sleep posture recognition and classification network model in turn, that is, the cardiopulmonary activity distribution characteristics and the environmental vector feature samples and the corresponding sleep posture labels in each set of training samples are input at the same time. The sleeping posture recognition and classification network model adjusts the parameters of the sleeping posture recognition and classification network model by calculating the loss function value according to each output result of the sleeping posture recognition and classification network model, and finally completes the training process of the sleeping posture recognition and classification network model.

With the method of the embodiment of the present invention, the cardiopulmonary activity distribution feature, the environmental vector feature sample and the sleeping posture label carried by the cardiopulmonary activity distribution feature sample are taken as a group of training samples, and the sleeping posture recognition and classification network model is trained by using multiple groups of training samples. .

Optionally, the use of multiple sets of training samples to train the sleeping posture recognition and classification network model is specifically:

For any set of training samples, input the training samples into the sleeping posture recognition and classification network model, and output the predicted probability corresponding to the training samples;

Using a preset loss function, calculate the loss value according to the predicted probability corresponding to the training sample and the sleeping position label in the training sample;

If the loss value is less than the preset threshold, the training of the sleeping posture recognition and classification network model is completed.

Optionally, in the embodiment of the present invention, the preset loss function refers to a loss function preset in the sleeping posture recognition and classification network model, which is used for model evaluation; the preset threshold refers to the threshold preset by the model, Used to obtain the minimum loss value to complete the model training

After obtaining multiple sets of training samples, for any set of training samples, the cardiopulmonary activity distribution features and environmental vector feature samples in the training samples and the corresponding sleeping posture labels are input into the sleeping posture recognition and classification network model, and the training samples are output. The predicted probability corresponding to the sample, where the predicted probability refers to the predicted probability corresponding to the training sample for different sleeping posture recognition results.

On this basis, a preset loss function is used to calculate the loss value according to the prediction probability corresponding to the training sample and the sleeping position label in the training sample. The preset loss function may be a cross-entropy loss function. In other embodiments, the representation of the sleeping posture label and the preset loss function may be set according to actual requirements, which are not specifically limited here.

Further, after calculating and obtaining the loss value, the training process ends, and the error back-propagation algorithm can be used to update the parameters of the preset sleeping posture recognition and classification network model, and then the next training can be performed. During the training process, if the loss value calculated for a certain group of training samples is less than the preset threshold, the training of the network model for sleeping posture recognition and classification is completed.

The method of the embodiment of the present invention controls the loss value of the sleeping posture recognition and classification network model within a preset range by training the sleeping posture recognition and classification network model, thereby helping to improve the sleeping posture output by the sleeping posture recognition and classification network model. Accuracy of recognition results.

FIG. 2 is a schematic flowchart of steps of a method for recognizing a sleeping posture based on a piezoelectric ceramic sensor provided by an embodiment of the present invention. As shown in FIG. 2 , the method includes the following steps: analyzing the force and working state of the sensor, and calculating an equivalent output circuit , output voltage and stress mapping formula, sleep stable state discrimination, cardiopulmonary activity intensity characteristics in steady state, optimization of cardiopulmonary activity feature extraction environmental features, elimination of the influence of hyperparameters between sensors, real-time distribution characteristics of cardiopulmonary exercise and recognition of sleeping posture in steady state .

Optionally, in the embodiment of the present invention, the steps of analyzing the force and working state of the sensor are as follows:

In the piezoelectric ceramic sensor system of the present invention, the piezoelectric ceramic sensor has a piezoelectric effect, and the present invention mainly utilizes its positive piezoelectric effect, that is, when an external force acts and deforms along a certain direction, the internal electric polarization phenomenon occurs, and the sensor Charges of opposite sign will be generated on the surface. When the external force is removed, it will return to an uncharged state, and within a certain range, the amount of charge generated by the piezoelectric ceramic sensor is proportional to the magnitude of the external force.

The external force on the piezoelectric ceramic sensor will deform in the three-dimensional directions of the three-dimensional space, and generate stress. According to Hooke's law of isomorphism, the internal stress is proportional to the amount of strain. Since the compliance coefficient and stiffness coefficient are different on the three planes, the three planes need to be discussed separately.

The stress state at any point should be decomposed into 3 planes, and the force of each plane can be decomposed into 3 coordinate systems of x-axis, y-axis, and z-axis, so the stress state at any point is composed of 9 stress components T _xx , T _xy , T _xz , T _yx , T _yy , T _yz , T _zx , T _zy , T _{zz are} decided. According to the shear stress reciprocity law, T _yz =T _zy , T _xy =T _yx , T _zx =T _xz , so the nine stress components are simplified to six. Among them, T _xx , T _yy , T _zz represent normal stress respectively, represented by T ₁ , T ₂ , T ₃ ; T _yz , T _zx , T _xy represent shear stress respectively, represented by T ₄ , T ₅ , T ₆ .

的公式为Electrical Displacement of Piezoelectric Ceramic Sensors

The formula is

为各个方向的电场强度。Among them, ε ₀ is the dielectric constant in vacuum, and the piezoelectric ceramic does not have polarization in vacuum.

is the electric field strength in all directions.

与外应力

的比值：Piezoelectric coefficient is polarization strength

with external stress

The ratio of:

表示极化强度，其在3个面的分量为P₁、P₂、P₃；[d]表示压电系数，

表示外应力。in,

represents the polarization intensity, and its components on the three surfaces are P ₁ , P ₂ , and P ₃ ; [d] represents the piezoelectric coefficient,

represents the external stress.

为6个应力分量，所以，压电系数[d]是一个3*6的系数矩阵，压电系数第一个下标表示极化方向，也即电场方向，第二个下标是分解力的方向。because

is 6 stress components, so the piezoelectric coefficient [d] is a 3*6 coefficient matrix, the first subscript of the piezoelectric coefficient represents the polarization direction, that is, the electric field direction, and the second subscript is the decomposition force direction.

Vertical short-circuit analysis of the Z-axis of the piezoelectric ceramic sensor system:

The subscript of the piezoelectric ceramic sensor in the vertical direction of the Z axis is set to 3, and the electric field intensity is 0 when the polarized electrodes are short-circuited in the vertical direction, that is:

E ₃ = 0;

Only the direction of the polarized electric field is considered, i.e.

方向的极化强度由正应力压电效应产生，即According to the piezoelectric coefficient formula, at this time

The polarization in the direction is produced by the normal stress piezoelectric effect, namely

In the piezoelectric coefficient matrix, d ₃₁ =d ₃₃ ,

Piezoelectric ceramic sensor vertical open circuit analysis:

When the polarized electrodes are interrupted in the vertical direction, the charge between the plates cannot be displaced, that is,

D ₃ = 0;

Only the direction of the polarized electric field is considered, i.e.

方向的极化强度由正应力压电效应产生。According to the piezoelectric coefficient formula, at this time,

The directional polarization is produced by the normal stress piezoelectric effect.

同样也为压电系数，此时的电场强度与应力的关系为Assume

It is also the piezoelectric coefficient, and the relationship between the electric field strength and the stress at this time is

Fig. 3 is a schematic diagram comparing the output voltage curves of the user and the equal weight object on the piezoelectric ceramic sensor system in the embodiment of the present invention. As shown in Fig. 3, the abscissa represents the number of sampling points, and each sampling point is 10ms, that is, The abscissa also represents the sampling time; the ordinate represents the digital voltage output by the analog-to-digital conversion circuit. It should be noted that in this application, the ordinate is similar to the output voltage value of the oscilloscope. In theory, the working voltage of the analog-to-digital converter is -3.3V to 3.3V, and when the digital scale on the ordinate is 4096, the corresponding working voltage is 3.3V, when the digital scale on the ordinate is 2048, the corresponding working voltage is 0V; when there is no voltage output, 2048 is output.

当应力增加时，电压快速抬升。但实际等效电阻并非是理想无穷大，实际电路并非断路一直存在电位移，所以一段时间后压电陶瓷两极平衡，电场与电压归零。As shown in (a) of Figure 3, it is the output voltage change curve of the process of placing equal weight objects. In piezoelectric ceramics, the leakage resistance of the sensor and the input resistance of the amplifier are large, and the electric displacement is weak in the working state. Under ideal conditions, the moment of pressure change can be regarded as a circuit breakage. At this time, the surface electric field strength of the piezoelectric ceramic sensor is close to

When the stress increases, the voltage rises rapidly. However, the actual equivalent resistance is not ideal and infinite, and the actual circuit is not open circuit and there is always electrical displacement, so after a period of time, the two poles of the piezoelectric ceramic are balanced, and the electric field and voltage return to zero.

As shown in (b) in Figure 3, it is the output voltage change curve of the process of picking up an equal weight object. When the object leaves, the sensor stress decreases, the voltage drops rapidly, and the electric field and voltage also return to equilibrium after a period of time.

As shown in (c) of FIG. 3 , it is the output voltage change curve of the user lying down process.

As shown in (d) in Figure 3, it is the output voltage change curve of the user getting up.

When the user is lying down and getting up, although the voltage change is consistent with placing and picking up an equal weight object, after a period of time, the collected voltage change is a periodic sinusoidal signal, which is a mixed cardiac shock signal in medicine. . According to the medical meaning of the mixed cardiac shock signal, the signal is the voltage change generated by the cardiopulmonary activity. According to the normal breathing motion amplitude is much larger than the motion amplitude generated by the heartbeat, and in the open circuit state, the stress is proportional to the electric field strength, that is, the surface voltage. From this, it can be inferred that when the user is in a stable sleep state, in addition to relatively stable gravity, it also has a sine-like force generated by the breathing movement. Moreover, the stress change caused by the breathing movement should be much smaller than the user's own gravity, so the pressure at the moment when the user gets up and lies down is consistent with picking up and placing an equal-weight object. After a period of time, the charge displacement caused by gravity is completed, that is, when it returns to the short-circuit state, the pressure change of the breathing movement at this time leads the output voltage change to produce a sinusoidal piezoelectric change on the surface of the piezoelectric ceramic sensor.

In the embodiment of the present application, the mixed cardioplegia signal also contains a shockcardiogram in addition to the respiration signal.

FIG. 4 is a schematic diagram of the respiratory frequency band signal and the original mixed signal in the mixed cardiac shock signal provided by an embodiment of the present invention. As shown in FIG. 4 , the thick solid line is the respiratory frequency band signal, and the thin solid line is the mixed cardiac shock signal, wherein, The abscissa represents the number of sampling points, and each sampling point is 10ms, that is, the abscissa also represents the sampling time; the ordinate represents the digital voltage output by the analog-to-digital conversion circuit.

FIG. 5 is a schematic diagram of the envelope of the shock cardiac signal and its heartbeat cycle in the mixed shock cardiac signal provided by an embodiment of the present invention. As shown in FIG. 5 , the thin solid line represents the shock cardiac signal, and the thick solid line represents the heartbeat of the shock cardiac signal. The envelope of the cycle, where the abscissa represents the number of sampling points, and each sampling point is 10ms, that is, the abscissa also represents the sampling time; the ordinate represents the digital voltage output by the analog-to-digital conversion circuit.

According to the difference between the frequency of the respiratory wave (0.1-0.5 Hz) and the frequency of the cardiac shock signal (5-20 Hz), the cardiac shock signal can be separated from the respiratory signal. Like the respiration signal, the cardiac shock signal is generated by the pressure change during the heartbeat. Although its signal is not a sinusoidal signal, it is also a periodic signal. In the medical definition, a cycle is a complete heartbeat process.

Further, the force model on the surface of the piezoelectric ceramic sensor is established:

表示压电陶瓷传感器表面受到的总应力，

表示用户重力产生的应力，

表示用户呼吸运动产生的应力，

表示用户心跳运动产生的应力。in,

represents the total stress on the surface of the piezoelectric ceramic sensor,

represents the stress caused by the user's gravity,

represents the stress generated by the user's breathing movement,

Indicates the stress caused by the user's heartbeat movement.

Although the three acting forces are aliased together in the time domain to affect the output voltage, their respective frequencies and voltage change periods do not interfere with each other.

Further, in the embodiment of the present invention, the steps of calculating the equivalent output circuit are as follows:

In the circuit, the piezoelectric ceramic is equivalent to a voltage source with its own capacitance C ₁ in series, and the sensor leakage resistance R ₁ in parallel. Because the voltage generated by the sensor itself is very weak, the surface voltage of the sensor needs to be amplified by an operational amplifier.

FIG. 6 is a schematic diagram of a piezoelectric ceramic sensor voltage amplifying circuit provided by an embodiment of the present invention. As shown in FIG. 6 , C ₂ represents the input capacitance of the operational amplifier, C ₃ represents the equivalent capacitance of the line, and R ₁ represents the piezoelectric ceramic _The leakage resistance of the sensor, R2 represents the input resistance of the op amp.

FIG. 7 is a schematic diagram of an equivalent input circuit of a piezoelectric ceramic sensor voltage amplifier circuit provided by an embodiment of the present invention. As shown in FIG. 7 , the equivalent input resistance is

The equivalent input capacitance is

C=C ₁ +C ₂ ;

Further, in the embodiment of the present invention, the steps of outputting the voltage and stress mapping formula are as follows:

The stress T ₃ >>T ₁ , T ₂ of the user lying on the pressure sensor in the vertical direction of the sensor,

According to U=Ed, d is the equipotential surface distance, that is, the thickness _of the piezoelectric ceramic, and T3 is the vertical force per unit area, that is,

Among them, F3 is the vertical direction force _of the Z axis, S is the surface area of the piezoelectric ceramic, and ε is the dielectric constant.

Assuming that the current user is in a stable sleep state, that is,

F ₃ ≈F _N3 +F _M3 +F _H3 ;

Among them, the F _N3 gravitational stress is a constant force, and the F _N3 component is equivalent to a short-circuit circuit in a steady state, and the resulting voltage is 0. F _M3 respiratory stress can be equivalent to a sinusoidal signal with an amplitude of F _m , where f _m is the respiratory rate.

F _M3 ≈F _m sinωt;

ω=2πf _m ;

Respiratory motion component of sensor surface voltage _U3

The input voltage of the op amp

The input voltage amplitude value of the operational amplifier

The leakage resistance R ₁ in the equivalent input resistance R, the amplifier input resistance R ₂ is relatively large, that is, ωR(C ₁ +C ₂ +C ₃ )>>1;

After the operational amplifier voltage U _3Min is amplified, it undergoes low-pass filtering, voltage boosting, and digital-to-analog conversion in sequence. The finally collected digital voltage U _3Mout is just the digitization of the analog voltage U _3Mout , that is, U′ _3Mout ≈ U _3Mout . The low-pass filter mainly filters out the hardware noise above 50Hz, which hardly affects the amplitude value of the collected voltage after filtering.

Set the voltage amplification factor B, and the voltage rise is A. At this time, the digital voltage U′ _3Mout collected by the breathing motion is

生成图5的包络信号，β为当前心跳角速度。所以，等效心跳

是心跳运动应力F_h的周期包络，

与心跳应力的幅值F_h具有一致性。The shock cardiac signal is a periodic signal, such as a thin solid line signal as shown in Fig. 5. According to the medical properties, the envelope of the shock cardiac signal is a chirp-like signal, and the frequency is determined by the current heart rate. According to the change of heart rate in a short time (30 seconds), generally less than 0.3 Hz, the envelope signal of the heartbeat frequency band of the short-term shock signal can be equivalent to a sinusoidal signal, such as the thick solid line signal shown in Figure 5. The equivalent sinusoidal signal contains the peak amplitude and periodic characteristics of the original cardiac shock signal. According to the sensor surface output voltage is proportional to the stress that generates the voltage, the equivalent periodic motion can be used.

The envelope signal of Fig. 5 is generated, and β is the current heartbeat angular velocity. So, the equivalent heartbeat

is the periodic envelope of the heartbeat motion stress F _h ,

It is consistent with the amplitude F _h of the heartbeat stress.

Similar to the breathing exercise, the relationship between the cardiac impulse envelope voltage signal and the equivalent heartbeat stress can be obtained as follows:

Further, in the embodiment of the present invention, the steps of determining the sleep stable state are as follows:

According to the previous calculation process, the collected data is the mapping of the external stress on the digital voltage, but the present invention only considers the ideal stationary state, and the different force structure will cause the state of the data to change. After excluding paragraphs with abnormal data, normal data can be divided into non-stationary state and stationary state, where non-stationary state includes states with obvious body movements such as getting up, lying down, and body movement, and also including getting out of bed and no-load state

According to the force analysis of the above state, the vertical force of the Z axis can be divided into gravitational stress, respiratory motion stress and heartbeat motion stress, among which, the gravitational stress is much greater than the respiratory motion stress, and the respiratory motion stress is much greater than the heartbeat motion stress.

The process of getting up is a transition from a steady state of being in bed to getting out of bed. During getting up, the gravitational stress decreases rapidly, and the change speed is much greater than the stress of the heartbeat, respiration, and movement. At this time, the output voltage drops rapidly.

The state of leaving the bed, that is, the state of the mattress without load, at this time, there is no stress, and the output is 0.

The process of lying down is the transition from the state of getting out of bed to the state of being stable in bed. During the period of lying down, the gravitational stress increases rapidly, and the change speed is much greater than the stress of the heartbeat, respiration, and movement. At this time, the output voltage rises rapidly.

The body movement state is the state in which the user's limbs move in a stable state. Before and after the body movement is in a stable sleep state, the gravitational stress may increase or decrease during the body movement, and the change speed is much greater than that of the heartbeat and breathing movement. At this time, the output voltage will increase. Rapid changes occur, but sleep is stable before and after body movement occurs.

In the steady state process, the gravitational stress hardly changes or changes slowly, hardly affects the output voltage, but only affects the output trend wave, and because the respiratory cycle stress is much larger than the heartbeat cycle stress, the output in the steady state is mainly the respiratory stress. The resulting voltage change U' _3Mout .

Further, in the embodiment of the present invention, the steps of the cardiopulmonary activity intensity feature in a steady state are as follows:

During normal sleep, the stable state accounts for most of the time. In the stable state, the mixed cardiac shock signal can be well observed, and the breathing motion change signal and the heartbeat motion change signal are extracted respectively. The present invention focuses on the signal in the stable state. The distribution of cardiorespiratory activity was processed. For the states of getting out of bed, getting up, lying down and body movement, all are unstable or data-free states, and the present invention only determines them for monitoring the sleep state without further processing.

By extracting and processing the collected mixed cardiac shock signals, the respiratory motion frequency signal U' _3Mout and the envelope signal U' _3Hout of the heartbeat motion frequency can be obtained. According to the mapping relationship between the pressure and the output voltage, the real-time signal and lift can be obtained. Voltage, the current signal related to the phase ωt and the boosted voltage A can be eliminated through signal differentiation, that is,

Among them, Δt is a very small time interval, and the smallest time interval in the implementation process is a sampling time difference; ω represents the respiration angular velocity; f represents the respiration rate, which is 0.1Hz-0.5Hz, Δtω<<2π tends to 0 .

From this it can be seen that:

sinω(t+Δt)-sinωt=sinωt cosωΔt+sinωΔt cosωt-sinωt;

According to McLaughlin's formula:

It is known that:

sinω(t+Δt)-sinωt≈sinωt(1)+(ωΔt)cosωt-sinωt

=(ωΔt)cosωt;

Among them, U _3Mout (t) represents the digital signal collected, Δt represents the sampling interval of the selected signal, d ₃₃ represents the piezoelectric coefficient, B represents the operational amplifier amplification factor, C represents the equivalent input internal capacitance, ω represents the angular velocity of breathing motion, F _m represents the magnitude of respiratory exercise stress.

选取N次周期NT＞＞Δt，电压输出的采样间隔为Δt,则The collected data U _3Mout (t) is a periodic signal, so it is related to the current time, that is, the instantaneous phase ωt, and the phase interference is reduced by the characteristic that the integral value is constant under the complete cycle of the sinusoidal signal. Let the period of one breath be

Select N cycles NT >> Δt, and the sampling interval of voltage output is Δt, then

Wherein, d ₃₃ , C, and B respectively represent the hyperparameters of the piezoelectric ceramic sensor, and N represents the number of cycles of the preset signal.

Thus, the relationship between the collected respiratory motion voltage U _3Mout (t) and the respiratory motion stress amplitude F _m is established.

的关系,如下：Similarly, the envelope signal U _3Hout (t) of the heartbeat motion voltage and the stress amplitude of the heartbeat motion can be obtained

relationship, as follows:

表示一次呼吸的周期，M表示预设的心跳处理周期数。Among them, d ₃₃ , C and B respectively represent the hyperparameters of the piezoelectric ceramic sensor, β represents the current heartbeat acceleration,

Represents the cycle of one breath, and M represents the preset number of heartbeat processing cycles.

Further, in the embodiment of the present invention, the steps of optimizing the cardiopulmonary activity feature to extract the environmental feature are as follows:

Through frequency division and signal processing of the mixed cardiac impulse signal, the respiratory motion voltage signal U _3Mout (t) and the heartbeat motion voltage signal U _3Hout (t) can be obtained.

In the characteristic formula obtained according to the above steps, the length of the integral in the formula corresponding to the breathing motion is NT ₁ , and the length of the integral in the formula corresponding to the heartbeat motion is MT ₂ . _The period _T1 is quite different from the breathing period T2. Therefore, the present invention presets a feature extraction formula with a fixed signal length, and selects the best signal length for processing heart rate and respiration to be 30 seconds to 3 minutes per frame. The longer the signal processing time, the larger the N and M in each frame of signal, the more stable the feature, and the smaller the impact on the phase, but the greater the possibility of changes in the heart rate and respiration rate per unit time. Therefore, taking 30 seconds as an example:

Number of breathing cycles in 30 seconds

Number of heartbeat cycles within 30 seconds

的相位的积分都为1，所以对120f₁、120f₂取整，即The integral of the absolute value of the sinusoidal signal, each

The integrals of the phases are all 1, so 120f ₁ and 120f ₂ are rounded, that is

At this point, signals with incomplete cycles can be estimated with an accuracy of 0.25 cycles and a small error, namely:

Further, in the embodiment of the present invention, the steps of eliminating the influence of hyperparameters between sensors are as follows:

According to the above steps, the relationship between the eigenvalue and the output voltage under a single sensor can be obtained, d ₃₃ , C, and B are the hyperparameters of the sensor, that is, the inherent parameters of the sensor. Although the hyperparameter of the same sensor is constant, when calculating the motion force distribution of multiple sensors, there are differences between different sensors. Even if the same batch of components is used, the consistency of the output circuit cannot be guaranteed. Therefore, it is necessary to eliminate the Hyperparameters between sensors.

In the embodiment of the present invention, it is assumed that the superscripts of d ⁱ ₃₃ , C ⁱ , B ⁱ are the numbers of the sensors, and the output breathing exercise intensity feature K ⁱ , that is,

A constant vibration source is placed on each sensor, and the output voltage is tested for many times ^{, and F im and ω i} _remain ^constant . The ratio between the eigenvalues of the nth sensor and the eigenvalues of the jth sensor can be obtained, that is, the ratio between the hyperparameters of the sensors can be obtained.

其余传感器得出与其超参的比值Dⁱ，则同一时刻下不同传感器采集到呼吸运动强度特征Kⁱ与呼吸运动压力幅度Fⁱ _m之间的关系：Set the sensor with the smallest parameters and working properly as the unit base

The other sensors obtain the ratio D ⁱ to their hyper-parameters, then the relationship between the breathing exercise intensity feature K ⁱ and the breathing exercise pressure amplitude F ⁱ _m collected by different sensors at the same time:

where ω is the breathing frequency. Different sensors have different breathing motion pressure amplitude F ⁱ _m , and the breathing frequency is consistent.

For the distribution of breathing motion intensity, that is, through the feature K ⁱ , the relationship between the breathing motion amplitude F ⁱ _m under different sensors is obtained, that is,

之间的关系，在此不再赘述。In the same way, the heartbeat movement amplitude under different sensors can be calculated

The relationship between them will not be repeated here.

Further, in the embodiment of the present invention, the steps of the real-time distribution characteristics of cardiopulmonary exercise are as follows:

同时也能得到此时的心率、呼吸、心率变异性(HeartRate Variability；HRV)等睡眠信息。Through the above steps, the intensity distribution of the heartbeat motion and the breathing motion can be calculated respectively through the collected voltage feature K ⁱ and the tested sensor feature D ⁱ .

At the same time, sleep information such as heart rate, respiration, and heart rate variability (HeartRate Variability; HRV) at this time can also be obtained.

Further, in the embodiment of the present invention, the steps of recognizing the sleeping posture in a stable state are as follows:

Through the above steps, the data is discriminated to identify getting up, getting out of bed, lying down, body movement and steady state, and further, when the user sleeps in a steady state, the sleeping posture is identified.

In order to adapt to the influence of the environment and consider more abundant scene characteristics, the present invention also adds a personalized environment vector feature, which includes the ratio of the amplitude of the heartbeat movement to the amplitude of the breathing movement, the total number of sensors occupied when the user is lying flat, The total number of sensors occupied by the user lying on the side, the current data state discrimination and the data state discrimination at the previous moment.

In the embodiment of the present invention, the training sample used is 4000case, wherein the characteristic matrix is the amplitude distribution characteristic of breathing motion 2*8, the characteristic matrix is the amplitude distribution characteristic of heartbeat motion 2*8, the environmental vector characteristic 1*5, the sleeping posture Labels lie flat, left side, right side, prone. In the present invention, a compression and excitation network (Squeeze-and-Excitation Networks; SENet) classification network framework is used for classification, and the feature map (feature map) obtained after convolution is processed to obtain a one-dimensional vector with the same number of channels as each The evaluation scores of each channel are then applied to the corresponding channels to obtain the results. Thereby, the interdependence between channels is explicitly modeled, the feature responses of channels are adaptively recalibrated, and the accuracy of model classification is improved.

FIG. 8 is a schematic diagram of the system framework of the sleep posture recognition method based on the piezoelectric ceramic sensor provided by the present invention. As shown in FIG. 8 , when the user is in a stable sleep state, the user’s chest is collected through the signal acquisition module of the piezoelectric ceramic sensor system. The abdominal area acts on the pressure of the piezoelectric ceramic sensor system to obtain the corresponding output voltage. After the operational amplifier, voltage boost, low-pass filter and digital-to-analog conversion, the mixed cardiac shock signal of the user's chest and abdomen area is obtained.

Further, through sensor state analysis, sensor stress analysis and output equivalent circuit calculation, the mapping relationship between output voltage and stress is obtained.

Further, the mixed cardiac shock signal of the user's chest and abdomen area can be obtained through signal analysis and processing to obtain the cardiopulmonary exercise signal. By feature extraction, the distribution features of cardiopulmonary activity can be determined.

Further, according to the user's cardiorespiratory activity distribution characteristics and state environment vector characteristics, based on the Squeeze-and-Excitation Networks (SENet) structural model, the sleeping posture of the user is recognized.

The method of the embodiment of the present invention provides a distribution method for a multi-channel sensor device to identify the combined cardiopulmonary activity of a user during sleep. The present invention firstly uses piezoelectric ceramic sensor and MSP430 chip to design a 16-channel pressure acquisition mattress, and acquires the mixed cardiac shock signal of the user's chest and abdomen during sleep. According to the equivalent circuit principle, the piezoelectric ceramic sensor can be equivalent to a voltage source and a capacitor. The change of external pressure will change the voltage value of the voltage source, and then the voltage signal will pass through the op amp circuit, filter circuit, boost circuit and analog-digital Convert the circuit, and finally upload the digital signal to the server. The data collected by the invention is also the function change of the piezoelectric ceramic surface voltage, and the operation relationship between the piezoelectric ceramic surface voltage and the output voltage can be calculated through the equivalent circuit of the piezoelectric ceramic and the operational amplifier. Ideally, the leakage resistance R1 _of the piezoelectric sensor and the input resistance R2 of the operational amplifier are much larger _than the equivalent capacitance C of the circuit, and the equivalent resistance R can be regarded as an open circuit with infinite resistance. Therefore, almost no electrical displacement occurs at the moment when the piezoelectric ceramic is stressed, and only the voltage of the two poles is changed. But in practice, the equivalent circuit is not really open, and there is a small amount of electrical displacement at the moment of force. With the passage of time, if the force remains unchanged, the charges on the two electrodes of the piezoelectric ceramic will eventually balance for a long time, and the internal electric field strength and voltage will return to zero.

越大，则该路传感器监测到的呼吸、心跳运动强度越大。The invention divides the voltage change into a voltage fast changing part and a voltage stable part by comparing the output digital voltage in the user's lying down, getting up and stable state and the placing, picking up and stable state of an equal weight object. For the part where the voltage changes rapidly, the voltage changes of the user and the object of equal weight are basically the same. At this time, the output circuit is equivalent to an open circuit, and the charge generated by the rapid increase or decrease of the force on the sensor has almost no electrical displacement, and a large voltage change occurs up and down the sensor. . For the stable part of the voltage, the supporting force of the equal weight object remains constant at this time, and the voltage in the stable state is almost the same as that in the no-load state. Because the actual output equivalent resistance is not infinite, there is a weak electrical displacement in the circuit. After a period of time, the displacement reaches a balance, and the equivalent short-circuit circuit voltage of the output circuit returns to zero. However, the steady state of the user still has periodic pressure changes. According to the medical significance of the mixed cardiac shock signal, the mixed cardiac shock signal collected by the user in the steady state is mainly generated by the breathing motion, and also includes the cardiopulmonary cycle of the heartbeat and pulse vibration. When the mixed cardiac shock signal is collected in a steady state, it changes in a quasi-sinusoidal period. By calculating that the amplified voltage is proportional to the instantaneous force, it can be obtained that the pressure of the cardiopulmonary activity is a quasi-sinusoidal change with an amplitude of F _m in a period of time. The change period is related to the respiration. the same frequency. Therefore, the distribution of cardiopulmonary activity is identified, that is, the real-time distribution of cardiopulmonary motion amplitudes monitored by various sensors is analyzed, and the motion amplitudes F _m ,

The larger the value, the greater the breathing and heartbeat movement intensity monitored by the sensor.

线性相关的特征映射，心跳信号虽然具有周期性，但不是标准的正弦信号，结合心率的变化能得到心跳平均幅度特征

从而得到睡眠期间心跳运动均幅的实时分布。The invention combines calculus and McLaughlin formula to design a feature extraction method for adaptive breathing frequency. The method first uses the collected voltage data to determine whether the current data state is getting up, getting out of bed, moving or lying down. Still stable. When it is in a steady state at this time, it conforms to the working conditions of the model. Using the voltage and breathing frequency output by the sensor, a feature map linearly related to the breathing amplitude F _m is established, and the sampling length and phase difference are adjusted according to the breathing frequency of the signal. A more accurate respiration amplitude characteristic of the piezoelectric ceramic sensor is obtained. By comparing the vibration test of the stable vibration source in different sensors, the influence of the hyperparameters between the sensors is eliminated, and the real-time distribution of the breathing motion amplitude during sleep is obtained. Similarly, the average amplitude of the heartbeat can be calculated from the voltage signal in the heartbeat frequency band.

Linear correlation feature mapping. Although the heartbeat signal is periodic, it is not a standard sinusoidal signal. Combined with the change of heart rate, the average amplitude of the heartbeat can be obtained.

Thereby, the real-time distribution of the average amplitude of heartbeat motion during sleep is obtained.

The invention combines the piezoelectric effect, the equivalent circuit, the medical characteristics of the mixed cardiac shock signal and the McLaughlin limit idea, and establishes a mapping relationship between the amplitude of the breathing movement and the mean amplitude of the heartbeat movement and the output voltage signal. The signal features extracted by the present invention have strong anti-interference ability, are easy to distinguish data states, solve the problem of differences in absolute pressure of different test scenarios and different test groups, overcome the large demand for sensors by conventional discriminant models, and reduce the amount of data collection. And using the multimodal fusion of breathing signal and heartbeat signal, based on the method of adjusting the period and phase of heart rate and breathing rate, the distribution characteristics are optimized, and the signal length corresponding to the distribution characteristics is standardized, and the two key physiological activities of heartbeat and breathing are obtained. joint features. Compared with the traditional absolute pressure distribution feature, the distribution feature of the present invention has better ductility, and has great advantages in sleeping posture recognition and joint monitoring of data status.

FIG. 9 is a schematic structural diagram of a sleeping posture recognition device based on a piezoelectric ceramic sensor provided by the present invention, as shown in FIG. 9 , including:

The hybrid cardiac shock signal acquisition module 910 is configured to acquire the hybrid cardiac shock signal of the user's chest and abdomen area based on the piezoelectric ceramic sensor system when the user is in a stable sleep state, wherein the piezoelectric ceramic sensor system includes multiple Piezoelectric ceramic sensor;

a cardiopulmonary activity distribution feature generation module 920, configured to determine the user's cardiopulmonary activity distribution feature based on the mixed cardiac shock signal;

The sleeping posture recognition result generation module 930 is used to input the cardiopulmonary activity distribution feature and the preset environment vector feature into the trained sleeping posture recognition classification network model to obtain the user's sleeping posture recognition result;

Wherein, the trained sleeping posture recognition and classification network model is obtained by training according to the cardiopulmonary activity distribution features and the environmental vector feature samples carrying the sleeping posture labels.

The sleeping posture recognition device based on the piezoelectric ceramic sensor provided by the embodiment of the present invention obtains the mixed cardiac shock signal of the user's chest and abdomen area based on the piezoelectric ceramic sensor system when the user is in a stable sleep state; The extraction and processing of the impact signal, to obtain the user's cardiorespiratory activity distribution characteristics, so as to input the user's cardiorespiratory activity distribution characteristics and the preset environment vector characteristics into the trained sleeping posture recognition classification network model, and obtain the user's sleeping posture recognition results. The intrusive real-time monitoring of user sleeping posture has strong universality and environmental anti-interference ability.

The sleeping posture recognition device based on the piezoelectric ceramic sensor described in this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, and details are not described herein again.

FIG. 10 is a schematic structural diagram of an electronic device provided by the present invention. As shown in FIG. 10 , the electronic device may include: a processor (processor) 1010, a communication interface (Communications Interface) 1020, a memory (memory) 1030 and a communication bus 1040, The processor 1010 , the communication interface 1020 , and the memory 1030 communicate with each other through the communication bus 1040 . The processor 1010 can call the logic instructions in the memory 1030 to execute the method for recognizing the sleeping posture based on the piezoelectric ceramic sensor. Mixed cardiac shock signals in the thoracic and abdominal region, wherein the piezoelectric ceramic sensor system includes a plurality of piezoelectric ceramic sensors; based on the mixed cardiac shock signals, the user's cardiopulmonary activity distribution characteristics are determined; the cardiopulmonary activity distribution characteristics are determined Input the trained sleeping posture recognition and classification network model with the preset environment vector feature to obtain the user's sleeping posture recognition result; wherein, the trained sleeping posture recognition and classification network model is based on the cardiopulmonary activity distribution features carrying the sleeping posture label. It is obtained by training with the environment vector feature samples.

In addition, the above-mentioned logic instructions in the memory 1030 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

In another aspect, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer During execution, the computer can execute the piezoelectric ceramic sensor-based sleeping posture recognition method provided by the above methods, and the method includes: when the user is in a stable sleep state, based on the piezoelectric ceramic sensor system, obtain the user's chest and abdomen. The hybrid cardiac shock signal of the region, wherein the piezoelectric ceramic sensor system includes a plurality of piezoelectric ceramic sensors; based on the hybrid cardiac shock signal, determine the user's cardiopulmonary activity distribution characteristics; Suppose the environment vector feature is input into the trained sleeping posture recognition and classification network model, and the user's sleeping posture recognition result is obtained; wherein, the trained sleeping posture recognition and classification network model is based on the cardiopulmonary activity distribution characteristics and the environment carrying the sleeping posture label. The vector feature samples are obtained by training.

In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored, the computer program being implemented by a processor to execute the piezoelectric ceramic sensor-based sensor provided by the above methods. A method for recognizing a sleeping posture, the method comprising: when a user is in a stable sleep state, based on a piezoelectric ceramic sensor system, acquiring a mixed cardiac shock signal in the thoracic and abdominal region of the user, wherein the piezoelectric ceramic sensor system includes a plurality of Piezoelectric ceramic sensor; based on the mixed cardiac shock signal, determine the user's cardiopulmonary activity distribution characteristics; input the cardiopulmonary activity distribution characteristics and the preset environment vector characteristics into the trained sleeping posture recognition and classification network model to obtain the user's sleeping posture Recognition results; wherein, the trained sleeping posture recognition and classification network model is obtained by training according to the cardiopulmonary activity distribution features and environmental vector feature samples carrying the sleeping posture labels.

The device embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic A disc, an optical disc, etc., includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or some parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. A sleeping posture identification method based on a piezoelectric ceramic sensor is characterized by comprising the following steps:

under the condition that a user is in a stable sleep state, acquiring a mixed cardiac shock signal of a chest and abdomen area of the user based on a piezoelectric ceramic sensor system, wherein the piezoelectric ceramic sensor system comprises a plurality of piezoelectric ceramic sensors; the number of the piezoelectric ceramic sensor acquisition channels is 10-30;

determining a cardiorespiratory activity distribution characteristic of the user based on the blended ballistocardiographic signal;

wherein, the determining the distribution characteristics of the cardiopulmonary activity of the user based on the mixed heart impact signal specifically comprises:

calculating cardiorespiratory activity intensity characteristics of the user based on the mixed heart impact signal;

determining the cardiorespiratory activity distribution characteristics of the user according to the cardiorespiratory activity intensity characteristics of the user and the parameter eliminating processing parameters in the piezoelectric ceramic sensor system;

wherein, the calculating the cardiorespiratory activity intensity characteristics of the user based on the mixed heart impact signal specifically comprises:

extracting information of the mixed heart impact signal to obtain the voltage information of the cardiopulmonary activity signal of the user;

calculating the stress amplitude information of the cardiopulmonary activity of the user based on signal differentiation and Maclaurin formula approximate estimation according to the voltage information of the cardiopulmonary activity signal of the user;

performing feature extraction of preset signal length on the cardiorespiratory activity stress amplitude information of the user to obtain cardiorespiratory activity intensity features of the user;

inputting the cardiopulmonary activity distribution characteristics and preset environment vector characteristics into a trained sleeping posture recognition classification network model to obtain a sleeping posture recognition result of the user;

the sleep posture identification and classification network model is constructed based on a compression and excitation network classification network framework, and the trained sleep posture identification and classification network model is obtained by training according to cardiopulmonary activity distribution characteristics carrying sleep posture labels and environment vector characteristic samples; the preset environment vector characteristics comprise the strength ratio of the heartbeat motion amplitude to the respiratory motion amplitude, the total number of the sensors occupied by the user when the user lies down in a standard manner, the total number of the sensors occupied by the user when the user lies down on one side, the judgment of the current data state and the judgment of the data state at the previous moment.

2. The sleeping posture identification method based on the piezoceramic sensor according to claim 1, wherein under the condition that the user is in a stable sleep state, the mixed cardiac shock signal of the chest and abdomen area of the user is acquired based on the piezoceramic sensor system, and specifically comprises the following steps:

acquiring first pressure information applied to the piezoelectric ceramic sensor system by the chest and abdomen area of a user;

determining first voltage information corresponding to the first pressure information according to a mapping relation between pressure and output voltage;

judging the sleep state of the user according to the variable quantity of the first voltage information, and determining second voltage information of the user in a stable sleep state;

and determining the mixed heart impact signal of the user in the stable sleep state according to the variable quantity of the second voltage information.

3. The sleeping posture identification method based on the piezoceramic sensor according to claim 1, wherein the cardiopulmonary activity distribution characteristics of the user are determined according to the cardiopulmonary activity intensity characteristics of the user and the parameter elimination processing parameters in the piezoceramic sensor system, and specifically are as follows:

calculating the cardio-pulmonary activity amplitude information acquired by each sensor according to the cardio-pulmonary activity intensity characteristics of the user and the parameter eliminating processing parameters in the piezoelectric ceramic sensor system;

wherein, the parameter eliminating processing parameter is obtained by calculating the ratio of the parameters of the rest sensors to the reference parameter;

the reference parameters are set by selecting the sensor parameters with the minimum parameters and normal operation in the piezoelectric ceramic sensor system;

and determining the cardio-pulmonary activity distribution characteristics of the user according to the cardio-pulmonary activity amplitude information acquired by the sensors.

4. The sleeping posture recognition method based on the piezoceramic sensor according to claim 1, wherein before inputting the cardiopulmonary activity distribution characteristics and the preset environment vector characteristics into the trained sleeping posture recognition classification network model, the method further comprises:

obtaining a plurality of cardio-pulmonary activity distribution characteristic and environment vector characteristic samples carrying sleeping posture labels; taking each cardiopulmonary activity distribution characteristic and environment vector characteristic sample carrying a sleeping posture label as a group of training samples to obtain a plurality of groups of training samples, and training the sleeping posture identification classification network model by using the plurality of groups of training samples.

5. The sleeping posture identification method based on the piezoelectric ceramic sensor, according to claim 4, characterized in that the sleeping posture identification classification network model is trained by using a plurality of groups of training samples, specifically:

for any group of training samples, inputting the training samples into the sleep posture recognition classification network model, and outputting the prediction probability corresponding to the training samples;

calculating a loss value according to the prediction probability corresponding to the training sample and the sleeping posture label in the training sample by using a preset loss function;

and if the loss value is smaller than a preset threshold value, finishing the training of the sleep posture identification and classification network model.

6. The utility model provides a sleep appearance recognition device based on piezoceramics sensor which characterized in that includes:

the mixed cardiac shock signal acquisition module is used for acquiring a mixed cardiac shock signal of a chest and abdomen area of a user based on a piezoelectric ceramic sensor system under the condition that the user is in a stable sleep state, wherein the piezoelectric ceramic sensor system comprises a plurality of piezoelectric ceramic sensors; the number of the piezoelectric ceramic sensor acquisition channels is 10-30;

the heart-lung activity distribution characteristic generating module is used for determining the heart-lung activity distribution characteristic of the user based on the mixed heart impact signal;

wherein, the determining the distribution characteristics of the cardiopulmonary activity of the user based on the mixed heart impact signal specifically comprises:

calculating cardiorespiratory activity intensity characteristics of the user based on the mixed heart impact signal;

determining the cardiorespiratory activity distribution characteristics of the user according to the cardiorespiratory activity intensity characteristics of the user and the parameter eliminating processing parameters in the piezoelectric ceramic sensor system;

wherein, the calculating the cardiorespiratory activity intensity characteristics of the user based on the mixed heart impact signal specifically comprises:

extracting information of the mixed heart impact signal to obtain the voltage information of the cardiopulmonary activity signal of the user;

calculating the stress amplitude information of the cardiopulmonary activity of the user based on signal differentiation and Maclaurin formula approximate estimation according to the voltage information of the cardiopulmonary activity signal of the user;

performing feature extraction of preset signal length on the cardiorespiratory activity stress amplitude information of the user to obtain cardiorespiratory activity intensity features of the user;

the sleeping posture recognition result generation module is used for inputting the cardiopulmonary activity distribution characteristics and preset environment vector characteristics into a trained sleeping posture recognition classification network model to obtain a sleeping posture recognition result of the user;

the sleep posture identification and classification network model is constructed based on a compression and excitation network classification network framework, and the trained sleep posture identification and classification network model is obtained by training according to cardiopulmonary activity distribution characteristics carrying sleep posture labels and environment vector characteristic samples; the preset environment vector characteristics comprise the strength ratio of the heartbeat motion amplitude to the respiratory motion amplitude, the total number of the sensors occupied by the user when the user lies down in a standard manner, the total number of the sensors occupied by the user when the user lies down on one side, the judgment of the current data state and the judgment of the data state at the previous moment.

7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the piezoceramic sensor-based sleep posture identification method according to any one of claims 1 to 5.

8. A non-transitory computer readable storage medium, on which a computer program is stored, wherein the computer program, when being executed by a processor, implements the steps of the piezoceramic sensor-based sleep posture identification method according to any one of claims 1 to 5.

Images