CN113509176A - Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor - Google Patents
Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor Download PDFInfo
- Publication number
- CN113509176A CN113509176A CN202110836033.XA CN202110836033A CN113509176A CN 113509176 A CN113509176 A CN 113509176A CN 202110836033 A CN202110836033 A CN 202110836033A CN 113509176 A CN113509176 A CN 113509176A
- Authority
- CN
- China
- Prior art keywords
- signal
- sleeping posture
- data
- amplitude
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000036544 posture Effects 0.000 claims description 179
- 238000012549 training Methods 0.000 claims description 109
- 238000003066 decision tree Methods 0.000 claims description 78
- 238000012360 testing method Methods 0.000 claims description 42
- 238000005070 sampling Methods 0.000 claims description 20
- 238000012795 verification Methods 0.000 claims description 17
- 230000011218 segmentation Effects 0.000 claims description 16
- 230000036391 respiratory frequency Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 230000001121 heart beat frequency Effects 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 230000000241 respiratory effect Effects 0.000 claims description 5
- 201000010099 disease Diseases 0.000 abstract description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 10
- 238000007637 random forest analysis Methods 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 6
- 238000010276 construction Methods 0.000 description 5
- 210000002216 heart Anatomy 0.000 description 5
- 210000001015 abdomen Anatomy 0.000 description 3
- 210000001715 carotid artery Anatomy 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000029058 respiratory gaseous exchange Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 208000020401 Depressive disease Diseases 0.000 description 1
- 206010041591 Spinal osteoarthritis Diseases 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 208000036319 cervical spondylosis Diseases 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 201000002859 sleep apnea Diseases 0.000 description 1
- 208000019116 sleep disease Diseases 0.000 description 1
- 208000022925 sleep disturbance Diseases 0.000 description 1
- 208000005801 spondylosis Diseases 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1116—Determining posture transitions
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4809—Sleep detection, i.e. determining whether a subject is asleep or not
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4812—Detecting sleep stages or cycles
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4815—Sleep quality
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/7257—Details of waveform analysis characterised by using transforms using Fourier transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a sleeping posture identification method based on a multi-channel piezoelectric sensor, which comprises the following steps: acquire the multichannel signal data that multichannel piezoelectric sensor gathered, each way piezoelectric sensor among this multichannel piezoelectric sensor sets up respectively in the position department of predetermineeing of platform, no matter what kind of appearance of sleeping of disease like this, can both gather effectual BCG signal from multichannel piezoelectric sensor's signal, avoids sleeping the appearance because of the limitation of position and disease of piezoelectric sensor and changes and cause the influence to the collection of signal. Then, the signal characteristics of each path of signal data are extracted to obtain a target characteristic set, the target characteristic set is classified based on the trained sleeping posture identification model to determine the current sleeping posture of the target object, the sleeping posture identification model has strong classification capability on the input contents with characteristics, and the current sleeping posture can be well identified based on the sleeping posture identification model. In addition, a sleeping posture recognition device and equipment are also provided.
Description
Technical Field
The invention relates to the technical field of sleeping posture identification, in particular to a sleeping posture identification method, a sleeping posture identification device and sleeping posture identification equipment based on a multi-channel piezoelectric sensor.
Background
Sleep posture is an important index in Polysomnography (PSG), and is mainly used for sleep and dream studies and diagnosis of depression and sleep apnea syndrome. The selection of a suitable sleeping posture helps to alleviate sleep disturbance caused by a particular disease. For example, the cervical spondylosis patient can choose the lying sleeping position, and the snorer can choose the side sleeping position.
In the sleeping posture test, the non-contact piezoelectric sensor can measure the Ballistocardiogram (BCG) in a portable way without wearing complicated electrodes. The instantaneous impact force generated by the heart beat on the contact surface is collected by the piezoelectric sensor as a BCG signal, and the BCG signal contains abundant sleep-related physiological information such as sleeping posture/heart rate/respiration and the like.
However, in the process of acquiring BCG signals, the non-contact single piezoelectric sensor may cause misjudgment of the sleeping posture due to the limitation of the placement position of the piezoelectric sensor. For example, when the patient lies on his side or lies down, if the sensor is not disposed below the upper half of the body, the BCG signal collected by the single sensor will be weak, and the condition of being mistakenly identified as out of bed may also occur accordingly. In addition, because the sleeping postures of the person can not be controlled autonomously in the sleeping process, the single piezoelectric sensor can not monitor various sleeping postures perfectly due to the limitation of the placement position.
Disclosure of Invention
Based on the above, a method, a device and equipment for recognizing the sleeping posture based on multiple piezoelectric sensors are provided to solve the problem that a single piezoelectric sensor cannot perfectly monitor various sleeping postures due to the limitation of the placement position.
A sleeping posture identification method based on a multi-channel piezoelectric sensor comprises the following steps:
acquiring multi-channel signal data of a target object acquired by the multi-channel piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the multi-channel piezoelectric sensors is respectively arranged at a preset position of a platform, and the target object is laid on the platform;
extracting signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise signal amplitude characteristics and signal energy characteristics of the one path of signal data;
and inputting the target feature set into a trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
In one embodiment, the method further comprises:
acquiring n sets of training input data, one set of training input data comprising: the system comprises a plurality of training samples, a training sample and a sleep gesture recognition system, wherein the training sample comprises a sample characteristic set corresponding to a sample object and a sleep gesture mark corresponding to the sample characteristic set;
respectively calculating the kini coefficients under different signal characteristics for each group of training input data, and constructing a decision tree corresponding to each group of input data based on the minimum principle of the kini coefficients to obtain n decision trees, wherein each decision tree takes the signal characteristics as a judgment node, and the signal characteristics with small kini coefficients are father nodes of the signal characteristics with large kini coefficients;
and verifying the n decision trees, and combining the n decision trees after verification to obtain the sleep posture identification model.
In one embodiment, the calculating the kini coefficients under different signal characteristics for each set of training input data, and constructing the decision tree corresponding to each set of input data based on the minimum principle of the kini coefficients to obtain n decision trees includes:
determining the ith group of training input data in the n groups of training input data as the 1 st group of data, determining the dividing point of each signal feature in the jth layer of data, calculating the Kernel coefficient under each dividing point of the first signal feature, determining the Kernel coefficient of the target dividing point with the minimum Kernel coefficient as the Kernel coefficient of the first signal feature, wherein the first signal feature is any one of a plurality of signal features corresponding to the jth layer of training input data, i is more than or equal to 1 and is less than or equal to n, and the initial value of j is 1;
and taking the second signal characteristic with the minimum kini coefficient as a j-th layer judgment node, segmenting the i-th group of training input data into two groups of training input data according to a target segmentation point of the second signal characteristic, adding 1 to j, respectively determining the two groups of training input data as j-th layer data, and returning to the step of determining the segmentation point of each signal characteristic in the j-th layer data until any preset termination condition is met, wherein the termination condition comprises that j is greater than a preset threshold value or the j-th layer data is less than a preset number.
In one embodiment, the acquiring n sets of training input data includes:
acquiring sample multi-channel signal data acquired by a target sample object lying on the platform in a target sleeping posture, wherein the target sample object is any sample object, and the target sleeping posture is any sleeping posture;
extracting signal characteristics of the sample multipath signal data to obtain a sample characteristic set;
taking the target sleeping posture as the sleeping posture label of the sample characteristic set to obtain a training sample;
a plurality of training samples obtained by putting a plurality of sample objects on the platform in different sleeping postures are used as a sample set;
and segmenting the sample set according to a specific ratio to obtain a training set and a testing set, and randomly sampling the training set with place back n times to obtain n groups of training input data.
In one embodiment, the verifying the n decision trees, and combining the n decision trees after the verification to obtain the sleep posture identification model includes:
and inputting the test set into each decision tree to obtain test results output by n decision trees, verifying the test results according to sleeping posture marks in the test set, and if each test result passes verification, combining the n decision trees after passing verification to obtain the sleeping posture identification model.
In one embodiment, the sleep posture recognition model comprises n decision trees;
the step of inputting the target feature set into a trained sleeping posture recognition model for classification to obtain the current sleeping posture of the target object comprises the following steps:
inputting the target feature set into a trained sleeping posture identification model, obtaining sleeping posture classification results output by leaf nodes of n decision trees after the features in the target feature set reach the leaf nodes of the n decision trees, obtaining n sleeping posture classification results, and selecting the sleeping posture classification results with the largest number as the current sleeping posture of the target object.
In one embodiment, the extracting the set of target features from the multi-channel signal data includes:
acquiring a first amplitude of each sampling point of the multi-channel signal data in a first frequency band and a second amplitude of each sampling point in a second frequency band, wherein the first frequency band and the second frequency band are two different frequency bands in a preset frequency band, and the preset frequency band is a frequency band corresponding to a physiological signal;
calculating a first amplitude characteristic according to the first amplitude, wherein the first amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero-crossing point number of the first amplitude;
calculating the second amplitude characteristic according to the second amplitude, wherein the second amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero-crossing point number of the second amplitude;
determining the first amplitude characteristic and the second amplitude characteristic as signal amplitude characteristics of the multi-path signal data;
performing Fourier transform on the multi-channel signal data to obtain a first energy sum of the multi-channel signal data in a respiratory frequency band, a second energy sum of the multi-channel signal data in a heartbeat frequency band and third energy sums of the multi-channel signal data in all frequency bands, wherein the respiratory frequency band is the frequency band of the respiratory signal, and the heartbeat frequency band is the frequency band of the heartbeat signal;
and determining the ratio of the first energy to the third energy sum and the ratio of the second energy to the third energy sum as the signal energy characteristic of the multi-path signal data.
In one embodiment, before the extracting the set of target features from the multi-channel signal data, the method further includes:
filtering power frequency interference in the multi-channel signal data to obtain a first preprocessing signal;
denoising the first preprocessing signal through a Butterworth band-pass filter to obtain multi-channel signal data in the preset frequency band.
A sleeping posture identifying apparatus, the apparatus comprising:
the multi-channel signal data module is used for acquiring multi-channel signal data of a target object acquired by the multi-channel piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the multi-channel piezoelectric sensors is respectively arranged at a preset position of the platform, and the target object is laid on the platform;
the characteristic set extraction module is used for extracting the signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise the signal amplitude characteristics and the signal energy characteristics of the other path of signal data;
and the sleeping posture identification module is used for inputting the target feature set into a trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
A sleeping posture recognition device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:
acquiring multi-channel signal data of a target object acquired by the multi-channel piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the multi-channel piezoelectric sensors is respectively arranged at a preset position of a platform, and the target object is laid on the platform;
extracting signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise signal amplitude characteristics and signal energy characteristics of the one path of signal data;
and inputting the target feature set into a trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
The invention provides a sleeping posture identification method, a sleeping posture identification device and equipment based on multiple piezoelectric sensors, which are used for acquiring multiple paths of signal data acquired by the multiple piezoelectric sensors, wherein each piezoelectric sensor in the multiple paths of piezoelectric sensors is respectively arranged at a preset position of a platform, so that an effective BCG signal can be acquired from signals of the multiple paths of piezoelectric sensors no matter what sleeping posture a patient is, and the influence on the acquisition of the signals due to the limitation of the placement positions of the piezoelectric sensors and the change of the sleeping posture of the patient is avoided. Then, the signal characteristics of each path of signal data are extracted to obtain a target characteristic set, the target characteristic set is classified based on the trained sleeping posture identification model to determine the current sleeping posture of the target object, the sleeping posture identification model has strong classification capability on the input contents with characteristics, and the current sleeping posture can be well identified based on the sleeping posture identification model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Wherein:
FIG. 1 is a schematic diagram of a sleeping posture recognition process based on multiple piezoelectric sensors in one embodiment;
FIG. 2 is a schematic diagram of a multi-way piezoelectric sensor system in one embodiment;
FIG. 3 is a schematic diagram illustrating a process of training with a random forest model as a sleep recognition model in one embodiment;
FIG. 4 is a schematic diagram of a training process of a random forest model in an embodiment;
FIG. 5 is a schematic structural diagram of a sleeping posture identifying apparatus according to an embodiment;
fig. 6 is a block diagram of the sleeping posture identifying apparatus in one embodiment.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a sleep posture recognition method based on multiple piezoelectric sensors in one embodiment, and the multiple piezoelectric sensors are arranged in the multiple piezoelectric sensor system shown in fig. 2. Wherein S011 is a host unit, S012-S015 is a four-way piezoelectric sensor, and S016 is a platform. Each path of piezoelectric sensor and the host unit in the multiple paths of piezoelectric sensors are respectively arranged at preset positions on the platform, when sleeping posture identification is carried out, a target object directly lies on the platform, each path of piezoelectric sensor in S012-S015 respectively collects one path of BCG signal data and sends the collected one path of BCG signal data to the host unit, and the host unit receives the multiple paths of BCG signal data and carries out sleeping posture identification.
The preset position is based on pre-testing, requiring four-way sensors to cover the carotid artery, heart and abdomen. Firstly, 50 adults (age 38.4 +/-8.8, height 169.6cm +/-12.5 cm) are recruited as test objects, when each test object lies on a platform in a preset sleeping posture (out of bed, lying on the face, lying on the left side, lying on the right side, lying on the face, lying on the upper side, lying on the lower side and lying on the upside down), the central positions of the carotid artery, the heart and the abdomen of the test object are marked, and the positions of the marks are counted to obtain a marked position set. And then, determining the preset position of each path of piezoelectric sensor on the platform based on the marked position set, wherein the maximum percentage of the marks of one part between the two piezoelectric sensors in a certain sleeping posture is required. For example, in a sleeping posture such as lying on a face, it is required that the largest part of the marked points of the carotid artery is between the piezoelectric sensor 1 and the piezoelectric sensor 2, the largest part of the marked points of the heart is between the piezoelectric sensor 2 and the piezoelectric sensor 3, and the largest part of the marked points of the abdomen is between the piezoelectric sensor 3 and the piezoelectric sensor 4.
The sleeping posture identification method based on the multiple piezoelectric sensors in the embodiment comprises the following steps:
and 102, acquiring multi-channel signal data of the target object acquired by the multi-channel piezoelectric sensor.
When a target object to be subjected to sleeping posture identification lies on the platform, each path of piezoelectric sensor in S012-S015 respectively collects one path of BCG signal data and sends the collected one path of BCG signal data to the host unit, the BCG signal data is preset for 5 minutes in the embodiment, and the piezoelectric sensor performs data transmission once after the requirement of 5 minutes of collection is met. The host unit acquires the multiple BCG signal data and performs the subsequent steps.
And 104, extracting the signal characteristics of each path of signal data to obtain a target characteristic set.
The signal characteristics of one path of signal data comprise signal amplitude characteristics and signal energy characteristics of one path of signal data. In this embodiment, the signal amplitude characteristic and the signal energy characteristic of the four-path signal data are respectively extracted.
Before extracting the signal characteristics, in order to obtain better characteristic effect, still carry out data preprocessing to signal data, include: 50Hz power frequency interference in the filtering way signal data is filtered through the wave trap, the power frequency interference can cause interference to electrical equipment and electronic equipment, the equipment is caused to operate abnormally, sine waves or other signals appearing in the signal measurement process are overlapped with the sine waves, and the problem can be effectively avoided through filtering power frequency interference. Obtaining a first preprocessed signal after filtering, denoising the first preprocessed signal through a Butterworth band-pass filter, wherein a second-order Butterworth filter is selected for denoising, and the first-order Butterworth band-pass filter is opposite to the second-order Butterworth filterThe denoising effect of the Butterworth filter is better, the Butterworth band-pass filter enables the frequency response curve of the first preprocessing signal in the preset pass band to be flat to the maximum extent, and the frequency response curve of the first preprocessing signal in the preset stop band gradually drops to zero to obtain multi-channel signal data within 0.1-10Hz, and therefore a signal section which mainly needs to be analyzed can be obtained. And then segmenting 5-minute signal data acquired by the four piezoelectric sensors by taking 30 seconds as one frame, wherein all 30-second data segments in the four piezoelectric sensors are taken as data samples of the target object and are expressed as follows: [ x ] of1,,x2,,x3,,x4,]Wherein x ism,Representing the nth data segment acquired by the mth sensor. After the segmentation, the signal data in a short time can be identified, and the sleeping posture identification is favorably carried out according to the changing condition of the sleeping posture.
When the characteristics are extracted, in the time domain of each data sample, a first amplitude value of each sampling point of the multi-channel signal data in a first frequency band and a second amplitude value of each sampling point in a second frequency band are obtained, the first frequency band and the second frequency band are two different frequency bands in a preset frequency band, and the preset frequency band is a frequency band corresponding to the preprocessed physiological signal.
And then calculating a first amplitude characteristic according to the first amplitude, wherein the first amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero-crossing point number of the first amplitude.
Illustratively, the first frequency band is 0.1-0.3Hz, firstly obtaining the amplitude of each sampling point within 0.1-0.3Hz, and calculating the following 5 types of signal amplitude characteristics within 0.1-0.3Hz according to the amplitudes: amplitude mean, amplitude standard deviation std, first order difference mean of amplitude diffmean, first order difference standard deviation diffstd of amplitude and the number of zero-crossing points. Wherein:
in the above formula, N is the total number of sampling points of the data sample within 0.1-0.3Hz, xiThe amplitude of the ith sample point.
Further, the second amplitude characteristic is calculated according to the second amplitude, and the second amplitude characteristic includes an amplitude mean value, an amplitude standard deviation, an amplitude first-order difference mean value, an amplitude first-order difference standard deviation and a zero-crossing point number of the second amplitude.
For example, the second frequency band is 2-10Hz, the amplitude of each sampling point within 2-10Hz is also obtained, 5 signal amplitude characteristics, namely an amplitude mean value mean, an amplitude standard deviation std, an amplitude first-order difference mean value diffmean, an amplitude first-order difference standard deviation diffstd and the number of zero-crossing points within 2-10Hz, are calculated according to the amplitudes, and since the calculation formulas are the same, the difference is only that N is the total number of sampling points of the data sample within 2-10Hz, and thus the description is omitted. Under different sleeping postures, the amplitude characteristics of the 10 types of signals in the first frequency band and the second frequency band have obvious difference, and the sleeping posture identification can be well completed based on the characteristics in the frequency bands.
In addition, each data sample is subjected to fast fourier transform, and 0.1-0.3Hz is taken as a respiratory frequency band in which a respiratory signal is located in a frequency domain, because signals in the frequency band are mainly caused by respiration. 0.67-5Hz is taken as the frequency band of the heartbeat in which the heartbeat signal is located, because the signal in this frequency band is mainly caused by the heartbeat. A first energy sum in the respiratory frequency band, a second energy sum in the heartbeat frequency band, and a third energy sum in the 0.1-10Hz frequency band are calculated. And determining the ratio of the first energy to the third energy sum and the ratio of the second energy to the third energy sum as the 2-type signal energy characteristics of the data samples.
As described above, for the multi-channel piezoelectric sensor, 48 features (i.e., 4 × 12 classes) can be extracted from data samples in every 30 seconds, the 48 features in different sleeping postures can have significant differences, and the signals in different sleeping postures can be identified according to the 48 features, so that the distinction is good. Finally, 480 characteristics of 10 data samples in total are used as a target characteristic set.
And 106, inputting the target feature set into the trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
The sleeping posture identification model is a model with classification capability, such as a random forest model, a KNN (K-Nearest Neighbor, K-neighborhood) model, a naive bayes model, and the like.
In a specific application scenario, classification is performed through a trained random forest model, and a random forest is composed of n decision trees. Inputting the target feature set into a random forest model, obtaining the sleeping posture classification results output by leaf nodes of n decision trees after the features in the target feature set reach the leaf nodes of the n decision trees, obtaining n sleeping posture classification results, and selecting the sleeping posture classification results with the largest number as the current sleeping posture of the target object. The random forest model is used as an integrated learning algorithm and comprises a plurality of classifiers, and the overall generalization capability can be improved through integrated learning.
In a specific application scene, classifying through a KNN model, inputting a target feature set into the KNN model, representing the target feature set as a target point on a coordinate according to feature values of all feature points in the target feature set, calculating distances between the target point and all sample points in the model, sequencing the distances, finding K sample points closest to the target point, voting by using labels of the sample points, and regarding as the label of the target point, namely the current sleeping posture of a target object. The calculation formula of the distance can adopt an Euclidean distance calculation formula:
wherein x isiIs the i-th feature of the target point, yiThe ith feature of the sample point, n, 48. The KNN model is suitable for multi-classification problems, and is simple in calculation and relatively high in accuracy.
According to the sleeping posture identification method based on the multiple piezoelectric sensors, the multiple paths of signal data collected by the multiple piezoelectric sensors are obtained, the multiple paths of piezoelectric sensors in the multiple paths of piezoelectric sensors are respectively arranged at the preset positions of the platform, so that no matter what sleeping posture a patient is, effective BCG signals can be collected from the signals of the multiple paths of piezoelectric sensors, and the influence on the collection of the signals due to the limitation of the positions where the piezoelectric sensors are placed and the change of the sleeping postures of the patient is avoided. Then, the signal characteristics of each path of signal data are extracted to obtain a target characteristic set, the target characteristic set is classified based on the trained sleeping posture identification model to determine the current sleeping posture of the target object, the sleeping posture identification model has strong classification capability on the input contents with characteristics, and the current sleeping posture can be well identified based on the sleeping posture identification model.
Referring to fig. 3, fig. 3 is a schematic diagram of a training process of an embodiment in which a random forest model is used as a sleep posture recognition model, and referring to fig. 4, fig. 4 is a schematic diagram of a training process of the random forest model, n sets of training input data are obtained by sampling original samples back n times, the n sets of training input data train n decision trees, each decision tree outputs a classification result to obtain n classification results, and a final classification result is obtained based on voting of the n classification results.
Specifically, the steps provided in this embodiment include:
Wherein the set of training data input data comprises: the system comprises a plurality of training samples, wherein one training sample comprises a sample feature set corresponding to a sample object and a sleeping posture label corresponding to the sample feature set, and the sample feature set corresponding to one sample object comprises a plurality of signal features corresponding to multi-channel signal data.
Illustratively, 100 adults (50 men and women, age 38.4 + -12.9, height 168.6cm + -13.5 cm) may be recruited in advance when performing training. One of the 100 adults is selected as a target sample object, and the target sample object is laid on a platform of the multi-channel piezoelectric sensor system in preset 8 sleeping postures (out of bed, lying on the face, lying on the left side, lying on the right side, lying on the face, lying on the upper side, lying on the lower side, lying on the upside down). When a target sample object lies on a platform in a certain sleeping posture, for example, when the surface lies on the platform, the surface lies on the platform as the target sleeping posture, sample multi-path signal data of the target sample object in the state of lying on the surface for 5 minutes is recorded, the signal characteristics of the sample multi-path signal data in the sleeping posture are extracted, the signal characteristics of the sample multi-path signal data are extracted based on the implementation principle of the same step 104 to obtain a sample characteristic set, the 'lying on the surface' is used as the sleeping posture label of the sample characteristic set, and finally the 1 sample characteristic set and the sleeping posture label are used as 1 training sample. In the same way, other training samples in 7 sleeping postures can be obtained. The repeated signal acquisition and feature extraction work is also carried out on other adults, and 3 abnormal data are removed, for example, data acquired under the conditions that the acquisition time is incomplete, the equipment acquisition is wrong, the manual recording is wrong, and the volunteers are not completely matched are removed as abnormal data, and 7760 training samples (97 (people) × 10 (frames) × 8 (sleeping postures)) are finally obtained as a sample set.
Then, the sample set is cut by a specific ratio, for example, sample set 8: 2 to divide into training set and testing set. And randomly sampling m training samples with replacement in the training set for n times to obtain n groups of training input data. For example, n may be set to 150 during actual training. Therefore, the data can be ensured to be different, the training samples are used as much as possible, each constructed decision tree has difference, and the overall generalization capability of the random forest model can be improved. The test set is used for carrying out accuracy test after the construction of the decision tree is completed, and training samples in the training sets are not added into the decision tree in the construction process, so that the uncertainty requirement of the test is met.
And 304, respectively calculating the kini coefficients under different signal characteristics for each group of training input data, and constructing a decision tree corresponding to each group of input data based on the minimum principle of the kini coefficients to obtain n decision trees.
The decision tree is composed of a plurality of judgment nodes, and the judgment node of the upper layer is a father node of the judgment node of the lower layer. In this embodiment, each decision tree uses the signal feature as a judgment node, and uses the signal feature with a small kini coefficient as a parent node of the signal feature with a large kini coefficient.
Specifically, the ith group of training input data in the n groups of training input data is determined as the 1 st group of data, and i is more than or equal to 1 and less than or equal to n. Starting from the root node of the decision tree, where the number of layers j of the root node is 1, the signal characteristic a in the layer of data is determined1-A48By dividing points, e.g. obtaining features AiDetermining n equally divided segmentation points to obtain the feature A according to the difference value between the maximum feature value and the minimum feature valueiCutting into n +1 parts; or all the characteristic values are sequenced to determine n segmentation points to obtain the characteristic AiThe cut is n +1 parts, the number in each feature being the same. Illustratively, signal characteristic A1: the mean value mean of the amplitudes of the piezoelectric sensor 1 in the range from 0.1 to 0.3Hz comprises the cut points: a is1、a2。a1The ith set of training input data may be divided into two parts, "a 1" and "a 2, A3"; a is2Dividing the ith group of training input data into two parts, namely 'A3' and 'A1 and A2'; a is1And a2The ith set of training input data was divided into two parts, "A2" and "A1, A3". Can be calculated at A based on the following calculation formula of the kini coefficient1The calculation formula of the kini coefficient of each part under the characteristics is as follows:
where | D | is the total number of samples of training input data D, | D1L is at aiFor the first sample D at the point of sectioning1Total number of samples, | D2L is at aiIs a second sample D at the point of sectioning2Total number of samples, | D | ═ D1|+|D2L. Calculating the signal characteristic A based on the above formula1The kini coefficient in three cases.
Similarly, determine A2-A48And calculating the kini coefficients of different features at different cut points based on the formula. And finding out the minimum one of all the finally obtained basic coefficients, determining the signal characteristic corresponding to the minimum basic coefficient as a judgment node of the layer j 1, determining the segmentation point corresponding to the minimum basic coefficient as a target segmentation point, segmenting the ith group of training input data into two groups of training input data according to the target segmentation point, adding 1 to j, and determining the two groups of input data as the data of two sub-nodes in the layer j respectively. And determining a judgment node and a target segmentation point for the child nodes of the two root nodes and the nodes extending subsequently based on the same method as the root nodes, and repeating the steps until any preset termination condition is met, and stopping the construction of the decision tree. Wherein one of the termination conditions is that the current j layer is greater than a preset threshold d. The second termination condition is that the data of the j-th layer is less than the preset quantity, because the data quantity at the deep node is reduced along with the halving of the data, when the number q of the sample data of the node is 1 or the data in the node are all in the same sleeping posture, the node is a leaf node, and the splitting is stopped.
Illustratively, referring to table 1 below, the construction of the decision tree by the feature A, B, C is merely used as an example for illustration.
Table 1:
wherein, the characteristic A comprises A1 and A2, and the cutting point of A is only a1Only need to calculate a1The kinson coefficient at the point of tangency.
The feature B comprises B1, B2 and B3 based on the cutting point B1Data were split into "B1" and "B2, B3". Based on b2Data were split into "B3" and "B1, B2". Based on b1And b2Data were split into "B2" and "B1, B3". B1 includes 1 "and 2", B2 includes 2 "1" and 1 "2", and B3 includes 1 "and 3".
The characteristic C comprises C1 and C2, so that the division point of C is only one C1Only need to calculate c1The kinson coefficient at the point of tangency.
It can be seen that the minimum kini coefficient is 0.17, the corresponding optimal characteristic is B, and the target division point is B1. Based on the optimal characteristics as B and the target segmentation point as B1Table 1 is split into data as in tables 2 and 3 below as child nodes.
Table 2:
| A | C | categories |
| A2 | C1 | 1 |
| A1 | C1 | 2 |
| A2 | C1 | 2 |
Table 3:
| A | C | categories |
| A1 | C1 | 1 |
| A1 | C1 | 2 |
| A2 | C2 | 1 |
| A1 | C2 | 1 |
| A2 | C2 | 3 |
| A2 | C2 | 3 |
| A1 | C2 | 3 |
Similarly, for tables 2 and 3, further divisions are made based on the same kuney coefficient minimization algorithm described above according to feature A, C.
And step 306, verifying the n decision trees, and combining the n verified decision trees to obtain a sleep posture identification model.
After the n decision trees are constructed, the n decision data are verified based on the test set so as to judge whether the sleeping posture recognition model is trained.
In a specific embodiment, a test set is input into each decision tree to obtain test results output by n decision trees, the test results are verified according to sleeping posture labels in the test set, and if the test results are consistent with the sleeping posture labels, the verification is considered to be passed; and if the test result is inconsistent with the sleeping posture mark, the verification is not passed. And if each test result is verified to be passed or the percentage of the verified decision trees is greater than a preset value P%, combining the n verified decision trees to obtain the sleep posture identification model. And because the training samples in the training set are not added into the decision tree in the construction process, the uncertainty requirement of the test is met.
In one embodiment, as shown in fig. 5, a sleeping posture identifying apparatus is proposed, which includes:
the multi-channel signal data module 502 is configured to acquire multi-channel signal data of a target object acquired by a plurality of channels of piezoelectric sensors, where one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the plurality of channels of piezoelectric sensors is respectively disposed at a preset position of the platform, and the target object lies on the platform;
the feature set extraction module 504 is configured to extract signal features of each channel of signal data to obtain a target feature set, where the signal features of one channel of signal data include a signal amplitude feature and a signal energy feature of one channel of signal data;
and the sleeping posture identifying module 506 is configured to input the target feature set into the trained sleeping posture identifying model for classification, so as to obtain the current sleeping posture of the target object.
In one embodiment, the sleeping posture identifying apparatus further comprises: a training module for obtaining n sets of training input data, a set of training input data comprising: the system comprises a plurality of training samples, a training sample comprises a sample characteristic set corresponding to a sample object and a sleeping posture mark corresponding to the sample characteristic set, and the sample characteristic set corresponding to the sample object comprises a plurality of signal characteristics corresponding to multi-channel signal data; respectively calculating the kini coefficients under different signal characteristics for each group of training input data, and constructing a decision tree corresponding to each group of input data based on the minimum principle of the kini coefficients to obtain n decision trees, wherein each decision tree takes the signal characteristics as a judgment node, and the signal characteristics with small kini coefficients are father nodes of the signal characteristics with large kini coefficients; and verifying the n decision trees, and combining the n decision trees after verification to obtain the sleep posture identification model.
In one embodiment, the training module is specifically configured to: determining the ith group of training input data in the n groups of training input data as the 1 st group of data, determining the dividing point of each signal feature in the jth layer of data, calculating the Kernel coefficient under each dividing point of the first signal feature, determining the Kernel coefficient of the target dividing point with the minimum Kernel coefficient as the Kernel coefficient of the first signal feature, wherein the first signal feature is any one of a plurality of signal features corresponding to the jth layer of training input data, i is more than or equal to 1 and is less than or equal to n, and the initial value of j is 1; and taking the second signal characteristic with the minimum kini coefficient as a j-th layer judgment node, segmenting the i-th group of training input data into two groups of training input data according to a target segmentation point of the second signal characteristic, adding 1 to j, respectively determining the two groups of training input data as j-th layer data, and returning to the step of determining the segmentation point of each signal characteristic in the j-th layer data until any one preset termination condition is met, wherein the termination condition comprises that j is greater than a preset threshold value or that the j-th layer data is less than a preset number.
In one embodiment, the training module is further specifically configured to: acquiring sample multi-channel signal data acquired by a target sample object lying on a platform in a target sleeping posture, wherein the target sample object is any sample object, and the target sleeping posture is any sleeping posture; extracting signal characteristics of sample multipath signal data to obtain a sample characteristic set; taking the target sleeping posture as the sleeping posture label of the sample characteristic set to obtain a training sample; a plurality of training samples obtained by putting a plurality of sample objects on a platform in different sleeping postures are used as a sample set; and segmenting the sample set according to a specific ratio to obtain a training set and a testing set, and randomly sampling the training set with place back n times to obtain n groups of training input data.
In one embodiment, the training module is further specifically configured to: and inputting the test set into each decision tree to obtain test results output by the n decision trees, verifying the test results according to the sleeping posture labels in the test set, and if each test result passes the verification, combining the n decision trees after the verification passes to obtain a sleeping posture identification model.
In an embodiment, the sleeping posture identifying module 506 is specifically configured to input the target feature set into the trained sleeping posture identifying model, obtain the sleeping posture classification results output by the leaf nodes of the n decision trees after the features in the target feature set reach the leaf nodes of the n decision trees, obtain the n sleeping posture classification results, and select the sleeping posture classification result with the largest number as the current sleeping posture of the target object.
In one embodiment, the feature set extraction module 504 is specifically configured to: acquiring a first amplitude of each sampling point of multi-channel signal data in a first frequency band and a second amplitude of each sampling point in a second frequency band, wherein the first frequency band and the second frequency band are two different frequency bands in a preset frequency band, and the preset frequency band is a frequency band corresponding to a physiological signal; calculating a first amplitude characteristic according to the first amplitude, wherein the first amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero crossing point number of the first amplitude; calculating a second amplitude characteristic according to the second amplitude, wherein the second amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, an amplitude first-order difference mean value, an amplitude first-order difference standard deviation and the number of zero-crossing points of the second amplitude; determining the first amplitude characteristic and the second amplitude characteristic as signal amplitude characteristics of the multi-channel signal data; carrying out Fourier transform on the multi-channel signal data to obtain a first energy sum of the multi-channel signal data in a respiratory frequency band, a second energy sum in a heartbeat frequency band and a third energy sum in all frequency bands, wherein the respiratory frequency band is the frequency band of the respiratory signal, and the heartbeat frequency band is the frequency band of the heartbeat signal; and determining the ratio of the first energy to the third energy sum and the ratio of the second energy to the third energy sum as the signal energy characteristic of the multi-path signal data.
In an embodiment, the feature set extraction module 504 is further specifically configured to: filtering power frequency interference in the multi-channel signal data to obtain a first preprocessing signal; and denoising the first preprocessed signal through a Butterworth band-pass filter to obtain multi-channel signal data in a preset frequency band.
Fig. 6 shows an internal structure diagram of the sleeping posture identifying apparatus in one embodiment. As shown in fig. 6, the sleeping posture identifying apparatus includes a processor, a memory, and a network interface connected through a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the sleeping posture identification device stores an operating system and also stores a computer program, and when the computer program is executed by the processor, the processor can realize the sleeping posture identification method based on the multi-channel piezoelectric sensor. The internal memory may also store a computer program, and when the computer program is executed by the processor, the processor may execute a sleep posture recognition method based on the plurality of piezoelectric sensors. It will be appreciated by those skilled in the art that the configuration shown in fig. 6 is a block diagram of only a portion of the configuration relevant to the present application, and does not constitute a limitation of the sleep posture identifying apparatus to which the present application is applied, and a particular sleep posture identifying apparatus may include more or fewer components than those shown in the drawings, or combine some components, or have a different arrangement of components.
A sleeping posture identification device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the following steps when executing the computer program: acquiring multi-channel signal data of a target object acquired by a plurality of piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the plurality of channels of piezoelectric sensors is respectively arranged at a preset position of a platform, and the target object is laid on the platform; extracting the signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise the signal amplitude characteristics and the signal energy characteristics of one path of signal data; and inputting the target feature set into the trained sleeping posture recognition model for classification so as to obtain the current sleeping posture of the target object.
In one embodiment, the method further comprises: acquiring n sets of training input data, one set of training input data comprising: the system comprises a plurality of training samples, a training sample comprises a sample characteristic set corresponding to a sample object and a sleeping posture mark corresponding to the sample characteristic set, and the sample characteristic set corresponding to the sample object comprises a plurality of signal characteristics corresponding to multi-channel signal data; respectively calculating the kini coefficients under different signal characteristics for each group of training input data, and constructing a decision tree corresponding to each group of input data based on the minimum principle of the kini coefficients to obtain n decision trees, wherein each decision tree takes the signal characteristics as a judgment node, and the signal characteristics with small kini coefficients are father nodes of the signal characteristics with large kini coefficients; and verifying the n decision trees, and combining the n decision trees after verification to obtain the sleep posture identification model.
In one embodiment, the calculating a kini coefficient under different signal characteristics for each set of training input data, and constructing a decision tree corresponding to each set of input data based on a minimum principle of the kini coefficient to obtain n decision trees includes: determining the ith group of training input data in the n groups of training input data as the 1 st group of data, determining the dividing point of each signal feature in the jth layer of data, calculating the Kernel coefficient under each dividing point of the first signal feature, determining the Kernel coefficient of the target dividing point with the minimum Kernel coefficient as the Kernel coefficient of the first signal feature, wherein the first signal feature is any one of a plurality of signal features corresponding to the jth layer of training input data, i is more than or equal to 1 and is less than or equal to n, and the initial value of j is 1; and taking the second signal characteristic with the minimum kini coefficient as a j-th layer judgment node, segmenting the i-th group of training input data into two groups of training input data according to a target segmentation point of the second signal characteristic, adding 1 to j, respectively determining the two groups of training input data as j-th layer data, and returning to the step of determining the segmentation point of each signal characteristic in the j-th layer data until any one preset termination condition is met, wherein the termination condition comprises that j is greater than a preset threshold value or that the j-th layer data is less than a preset number.
In one embodiment, obtaining n sets of training input data comprises: acquiring sample multi-channel signal data acquired by a target sample object lying on a platform in a target sleeping posture, wherein the target sample object is any sample object, and the target sleeping posture is any sleeping posture; extracting signal characteristics of sample multipath signal data to obtain a sample characteristic set; taking the target sleeping posture as the sleeping posture label of the sample characteristic set to obtain a training sample; a plurality of training samples obtained by putting a plurality of sample objects on a platform in different sleeping postures are used as a sample set; and segmenting the sample set according to a specific ratio to obtain a training set and a testing set, and randomly sampling the training set with place back n times to obtain n groups of training input data.
In one embodiment, the verifying n decision trees, and combining the n decision trees after the verification to obtain the sleep posture recognition model includes: and inputting the test set into each decision tree to obtain test results output by the n decision trees, verifying the test results according to the sleeping posture labels in the test set, and if each test result passes the verification, combining the n decision trees after the verification passes to obtain a sleeping posture identification model.
In one embodiment, the sleep posture recognition model comprises n decision trees; inputting the target feature set into the trained sleep posture recognition model for classification to obtain the current sleep posture of the target object, wherein the method comprises the steps of inputting the target feature set into the trained sleep posture recognition model, obtaining the sleep posture classification results output by leaf nodes of n decision trees after the features in the target feature set reach the leaf nodes of the n decision trees, obtaining n sleep posture classification results, and selecting the sleep posture classification results with the largest number as the current sleep posture of the target object.
In one embodiment, extracting a set of target features from the multiplexed signal data comprises: acquiring a first amplitude of each sampling point of multi-channel signal data in a first frequency band and a second amplitude of each sampling point in a second frequency band, wherein the first frequency band and the second frequency band are two different frequency bands in a preset frequency band, and the preset frequency band is a frequency band corresponding to a physiological signal; calculating a first amplitude characteristic according to the first amplitude, wherein the first amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero crossing point number of the first amplitude; calculating a second amplitude characteristic according to the second amplitude, wherein the second amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, an amplitude first-order difference mean value, an amplitude first-order difference standard deviation and the number of zero-crossing points of the second amplitude; determining the first amplitude characteristic and the second amplitude characteristic as signal amplitude characteristics of the multi-channel signal data; carrying out Fourier transform on the multi-channel signal data to obtain a first energy sum of the multi-channel signal data in a respiratory frequency band, a second energy sum in a heartbeat frequency band and a third energy sum in all frequency bands, wherein the respiratory frequency band is the frequency band of the respiratory signal, and the heartbeat frequency band is the frequency band of the heartbeat signal; and determining the ratio of the first energy to the third energy sum and the ratio of the second energy to the third energy sum as the signal energy characteristic of the multi-path signal data.
In one embodiment, before extracting the set of target features from the multi-channel signal data, the method further comprises: filtering power frequency interference in the multi-channel signal data to obtain a first preprocessing signal; and denoising the first preprocessed signal through a Butterworth band-pass filter to obtain multi-channel signal data in a preset frequency band.
It should be noted that the sleeping posture identification method, device and equipment based on the multi-channel piezoelectric sensor belong to a general inventive concept, and the contents in the embodiments of the sleeping posture identification method, device and equipment based on the multi-channel piezoelectric sensor are mutually applicable.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above examples only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A sleeping posture identification method based on a multi-channel piezoelectric sensor is characterized by comprising the following steps:
acquiring multi-channel signal data of a target object acquired by the multi-channel piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the multi-channel piezoelectric sensors is respectively arranged at a preset position of a platform, and the target object is laid on the platform;
extracting signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise signal amplitude characteristics and signal energy characteristics of the one path of signal data;
and inputting the target feature set into a trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
2. The method of claim 1, further comprising:
acquiring n sets of training input data, one set of training input data comprising: the system comprises a plurality of training samples, a training sample and a sleep gesture recognition system, wherein the training sample comprises a sample characteristic set corresponding to a sample object and a sleep gesture mark corresponding to the sample characteristic set;
respectively calculating the kini coefficients under different signal characteristics for each group of training input data, and constructing a decision tree corresponding to each group of input data based on the minimum principle of the kini coefficients to obtain n decision trees, wherein each decision tree takes the signal characteristics as a judgment node, and the signal characteristics with small kini coefficients are father nodes of the signal characteristics with large kini coefficients;
and verifying the n decision trees, and combining the n decision trees after verification to obtain the sleep posture identification model.
3. The method according to claim 2, wherein the step of calculating the kini coefficients under different signal characteristics for each set of training input data, and constructing the decision tree corresponding to each set of input data based on the principle of minimum kini coefficients to obtain n decision trees comprises:
determining the ith group of training input data in the n groups of training input data as the 1 st group of data, determining the dividing point of each signal feature in the jth layer of data, calculating the Kernel coefficient under each dividing point of the first signal feature, determining the Kernel coefficient of the target dividing point with the minimum Kernel coefficient as the Kernel coefficient of the first signal feature, wherein the first signal feature is any one of a plurality of signal features corresponding to the jth layer of training input data, i is more than or equal to 1 and is less than or equal to n, and the initial value of j is 1;
and taking the second signal characteristic with the minimum kini coefficient as a j-th layer judgment node, segmenting the i-th group of training input data into two groups of training input data according to a target segmentation point of the second signal characteristic, adding 1 to j, respectively determining the two groups of training input data as j-th layer data, and returning to the step of determining the segmentation point of each signal characteristic in the j-th layer data until any preset termination condition is met, wherein the termination condition comprises that j is greater than a preset threshold value or the j-th layer data is less than a preset number.
4. The method of claim 2, wherein the obtaining n sets of training input data comprises:
acquiring sample multi-channel signal data acquired by a target sample object lying on the platform in a target sleeping posture, wherein the target sample object is any sample object, and the target sleeping posture is any sleeping posture;
extracting signal characteristics of the sample multipath signal data to obtain a sample characteristic set;
taking the target sleeping posture as the sleeping posture label of the sample characteristic set to obtain a training sample;
a plurality of training samples obtained by putting a plurality of sample objects on the platform in different sleeping postures are used as a sample set;
and segmenting the sample set according to a specific ratio to obtain a training set and a testing set, and randomly sampling the training set with place back n times to obtain n groups of training input data.
5. The method according to claim 4, wherein the verifying the n decision trees, and combining the n decision trees after verification to obtain the sleep posture recognition model comprises:
and inputting the test set into each decision tree to obtain test results output by n decision trees, verifying the test results according to sleeping posture marks in the test set, and if each test result passes verification, combining the n decision trees after passing verification to obtain the sleeping posture identification model.
6. The method of claim 1, wherein the sleep posture recognition model comprises n decision trees;
the step of inputting the target feature set into a trained sleeping posture recognition model for classification to obtain the current sleeping posture of the target object comprises the following steps:
inputting the target feature set into a trained sleeping posture identification model, obtaining sleeping posture classification results output by leaf nodes of n decision trees after the features in the target feature set reach the leaf nodes of the n decision trees, obtaining n sleeping posture classification results, and selecting the sleeping posture classification results with the largest number as the current sleeping posture of the target object.
7. The method of claim 1, wherein said extracting a set of target features from the multi-channel signal data comprises:
acquiring a first amplitude of each sampling point of the multi-channel signal data in a first frequency band and a second amplitude of each sampling point in a second frequency band, wherein the first frequency band and the second frequency band are two different frequency bands in a preset frequency band, and the preset frequency band is a frequency band corresponding to a physiological signal;
calculating a first amplitude characteristic according to the first amplitude, wherein the first amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero-crossing point number of the first amplitude;
calculating the second amplitude characteristic according to the second amplitude, wherein the second amplitude characteristic comprises an amplitude mean value, an amplitude standard deviation, a first-order difference mean value, a first-order difference standard deviation and a zero-crossing point number of the second amplitude;
determining the first amplitude characteristic and the second amplitude characteristic as signal amplitude characteristics of the multi-path signal data;
performing Fourier transform on the multi-channel signal data to obtain a first energy sum of the multi-channel signal data in a respiratory frequency band, a second energy sum of the multi-channel signal data in a heartbeat frequency band and third energy sums of the multi-channel signal data in all frequency bands, wherein the respiratory frequency band is the frequency band of the respiratory signal, and the heartbeat frequency band is the frequency band of the heartbeat signal;
and determining the ratio of the first energy to the third energy sum and the ratio of the second energy to the third energy sum as the signal energy characteristic of the multi-path signal data.
8. The method of claim 6, further comprising, prior to said extracting a set of target features from said multiplexed signal data:
filtering power frequency interference in the multi-channel signal data to obtain a first preprocessing signal;
denoising the first preprocessing signal through a Butterworth band-pass filter to obtain multi-channel signal data in the preset frequency band.
9. A sleeping posture identifying device, characterized in that the device comprises:
the multi-channel signal data module is used for acquiring multi-channel signal data of a target object acquired by the multi-channel piezoelectric sensors, wherein one channel of signal data is acquired by one channel of piezoelectric sensor, each channel of piezoelectric sensor in the multi-channel piezoelectric sensors is respectively arranged at a preset position of the platform, and the target object is laid on the platform;
the characteristic set extraction module is used for extracting the signal characteristics of each path of signal data to obtain a target characteristic set, wherein the signal characteristics of one path of signal data comprise the signal amplitude characteristics and the signal energy characteristics of the other path of signal data;
and the sleeping posture identification module is used for inputting the target feature set into a trained sleeping posture identification model for classification so as to obtain the current sleeping posture of the target object.
10. A sleeping posture recognition device comprising a memory and a processor, characterized in that the memory stores a computer program which, when executed by the processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110836033.XA CN113509176A (en) | 2021-07-23 | 2021-07-23 | Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110836033.XA CN113509176A (en) | 2021-07-23 | 2021-07-23 | Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113509176A true CN113509176A (en) | 2021-10-19 |
Family
ID=78067538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110836033.XA Pending CN113509176A (en) | 2021-07-23 | 2021-07-23 | Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113509176A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016102184A1 (en) * | 2014-12-23 | 2016-06-30 | Koninklijke Philips N.V. | Method and sleep monitoring device for determining a body posture of a person in a bed |
| CN107595242A (en) * | 2017-07-26 | 2018-01-19 | 来邦科技股份公司 | A kind of sleep physiology signal monitoring method, device, electronic equipment and storage medium |
| CN108309328A (en) * | 2018-01-31 | 2018-07-24 | 南京邮电大学 | A kind of Emotion identification method based on adaptive fuzzy support vector machines |
| CN110236517A (en) * | 2019-04-02 | 2019-09-17 | 复旦大学 | The perception of cardiopulmonary signal and acquisition system for sleep monitor |
| CN111481207A (en) * | 2020-03-24 | 2020-08-04 | 南京润楠医疗电子研究院有限公司 | Sleep posture recognition device and method based on cardiac shock signal |
| CN112200293A (en) * | 2020-11-02 | 2021-01-08 | 吉林大学 | CART-AMV improved random forest algorithm |
| CN112971770A (en) * | 2021-02-10 | 2021-06-18 | 北京邮电大学 | Method and system for controlling and processing quality of cardiac shock signal |
-
2021
- 2021-07-23 CN CN202110836033.XA patent/CN113509176A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016102184A1 (en) * | 2014-12-23 | 2016-06-30 | Koninklijke Philips N.V. | Method and sleep monitoring device for determining a body posture of a person in a bed |
| CN107595242A (en) * | 2017-07-26 | 2018-01-19 | 来邦科技股份公司 | A kind of sleep physiology signal monitoring method, device, electronic equipment and storage medium |
| CN108309328A (en) * | 2018-01-31 | 2018-07-24 | 南京邮电大学 | A kind of Emotion identification method based on adaptive fuzzy support vector machines |
| CN110236517A (en) * | 2019-04-02 | 2019-09-17 | 复旦大学 | The perception of cardiopulmonary signal and acquisition system for sleep monitor |
| CN111481207A (en) * | 2020-03-24 | 2020-08-04 | 南京润楠医疗电子研究院有限公司 | Sleep posture recognition device and method based on cardiac shock signal |
| CN112200293A (en) * | 2020-11-02 | 2021-01-08 | 吉林大学 | CART-AMV improved random forest algorithm |
| CN112971770A (en) * | 2021-02-10 | 2021-06-18 | 北京邮电大学 | Method and system for controlling and processing quality of cardiac shock signal |
Non-Patent Citations (1)
| Title |
|---|
| 农天振: ""智慧医疗系统中基于心冲击信号的睡姿监测及睡眠质量分析"", 《中国优秀博硕士学位论文全文数据库(硕士)医药卫生科技辑》, no. 5 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416367B (en) | Sleep staging method based on multi-sensor data decision-level fusion | |
| CN110811609B (en) | Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion | |
| CN111035367B (en) | Signal detection system for judging sleep apnea | |
| Singh et al. | Correlation-based classification of heartbeats for individual identification | |
| Aziz et al. | An automated system towards diagnosis of pneumonia using pulmonary auscultations | |
| CN109480783B (en) | Apnea detection method and device and computing equipment | |
| CN101773394A (en) | Identification method and identification system using identification method | |
| Zhang et al. | A new ECG identification method using bayes' teorem | |
| CN108681685A (en) | A kind of body work intension recognizing method based on human body surface myoelectric signal | |
| Pal et al. | ECG biometric recognition | |
| Shen et al. | The PLR-DTW method for ECG based biometric identification | |
| Choi et al. | User Authentication System Based on Baseline-corrected ECG for Biometrics. | |
| Gürkan et al. | A novel biometric authentication approach using electrocardiogram signals | |
| CN111783715B (en) | Identity recognition method based on pulse signal feature extraction | |
| CN114638272B (en) | Identity recognition method and device based on fingertip pulse wave signals | |
| Wu et al. | ECG identification based on neural networks | |
| Canento et al. | Review and comparison of real time electrocardiogram segmentation algorithms for biometric applications | |
| Chandrashekhar et al. | Pulse ID: the case for robustness of ECG as a biometric identifier | |
| CN113509176A (en) | Sleeping posture identification method, device and equipment based on multi-channel piezoelectric sensor | |
| Biran et al. | Automatic qrs detection and segmentation using short time fourier transform and feature fusion | |
| KR101483218B1 (en) | Activity diagnosis apparatus | |
| CN111345815B (en) | Method, device, equipment and storage medium for detecting QRS wave in electrocardiosignal | |
| Sancho et al. | Photoplethysmographic authentication in long-term scenarios: A preliminary assessment | |
| Komeili et al. | Feature selection from multisession electrocardiogram signals for identity verification | |
| Tashiro et al. | Individual identification with high frequency ECG: preprocessing and classification by neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |