WO2020215566A1 - 基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备 - Google Patents

基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备 Download PDF

Info

Publication number
WO2020215566A1
WO2020215566A1 PCT/CN2019/103148 CN2019103148W WO2020215566A1 WO 2020215566 A1 WO2020215566 A1 WO 2020215566A1 CN 2019103148 W CN2019103148 W CN 2019103148W WO 2020215566 A1 WO2020215566 A1 WO 2020215566A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
limb movement
sample
single limb
tested
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.)
Ceased
Application number
PCT/CN2019/103148
Other languages
English (en)
French (fr)
Inventor
彭俊清
黄舒婷
王健宗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Priority to SG11202008420UA priority Critical patent/SG11202008420UA/en
Publication of WO2020215566A1 publication Critical patent/WO2020215566A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/103Measuring devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11Measuring movement of the entire body or parts thereof, e.g. head or hand tremor or mobility of a limb
    • A61B5/1113Local tracking of patients, e.g. in a hospital or private home
    • A61B5/1114Tracking parts of the body
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271Specific aspects of physiological measurement analysis
    • A61B5/7275Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2503/00Evaluating a particular growth phase or type of persons or animals
    • A61B2503/04Babies, e.g. for SIDS detection

Definitions

  • This application relates to the field of computer program technology, and in particular to a method, device, storage medium and electronic equipment for predicting infant development based on machine learning.
  • the prior art usually waits until the infant is 2 years old to determine whether the infant is stunted.
  • the inventor found that it was only at this time that infants and young children were diagnosed with stunting, and the intervention for infants and young children was relatively late, which is not conducive to the recovery of infants and young children with developmental delay. Therefore, there is a need for a method that can predict the development of infants, so as to provide a basis for intervention in the recovery of infants and young children with developmental delay.
  • the present application provides a method, device, storage medium and electronic equipment for predicting the development of infants based on machine learning, which can predict the development of infants and further obtain the possibility of infant developmental delay. According to this, It can provide a reliable basis for early intervention of infants with developmental delay, so that it is more suitable for practical use.
  • the method for predicting infant development based on machine learning includes the following steps:
  • the output result of the logical operation of the stacked extreme learning machine is used as the basis for the normal development or the developmental delay of the sample to be tested to obtain the prediction result of the infant's development.
  • the method for predicting infant development based on machine learning provided in this application can be further implemented by adopting the following technical measures.
  • the method for constructing the mathematical model of the stacked extreme learning machine includes the following steps:
  • L- is the number of nodes of a single hidden layer neural network
  • w i [W i1 ,w i2 ,...,w in ] T -is the input weight of the i-th hidden layer unit
  • bi- is the bias of the i-th hidden layer unit
  • ⁇ i [ ⁇ i1 , ⁇ i2 ,..., ⁇ im ] T -is the output weight of the i-th hidden layer unit
  • oj- is the logical operation output result after classification using the stacked extreme learning machine for the training sample, wherein the logical operation output result includes two major categories: normal development and developmental delay;
  • the expression of the activation function g(x), the input weight wi of the i-th hidden layer unit, and the value of the i-th hidden layer unit are determined Bias bi and the output weight ⁇ i of the i-th hidden layer unit;
  • the single limb movement data includes:
  • the duration t i of a single exercise is the difference T i1 -T i0 between the end time of the single exercise and the start time T i0 of the single exercise.
  • the method for performing data processing on the single limb movement data of the sample to be tested to obtain the feature value corresponding to the monthly age data for the single limb movement of the sample to be tested is selected from univariate feature selection, recursion One of feature selection and stepwise feature selection, wherein the data processing is performed on the single limb movement data of the sample to be tested to obtain the feature value corresponding to the monthly age data for the single limb movement of the sample to be tested Specifically:
  • the univariate feature selection is selected to obtain the feature value of the single limb movement through a statistical measurement method of a single variable in the single limb movement data;
  • the recursive feature selection is performed by normalizing data processing on each variable in the single limb movement data, and the obtained data is used as the characteristic value of the single limb movement; or
  • the single variables in the single limb movement data are selected one by one, and the data obtained by normalizing each variable in the single limb movement data is sequentially used as the characteristic value of the single limb movement. .
  • the device for predicting infant development based on machine learning includes:
  • the data acquisition unit is used to acquire the single limb movement data and the corresponding monthly age data of the sample to be tested;
  • a data processing unit configured to perform data processing on the single limb movement data of the sample to be tested to obtain a feature value corresponding to the monthly age data for the single limb movement of the sample to be tested;
  • An arithmetic unit used to bring a predetermined number of processed limb motion data and corresponding monthly age data corresponding to the sample to be tested into the mathematical model of the stacked extreme learning machine for logical operations;
  • the prediction result output unit is configured to use the output result of the logical operation of the stacked extreme learning machine as a basis for the normal development or developmental delay of the sample to be tested to obtain the prediction result of the infant's development.
  • the storage medium provided in the present application stores a machine learning-based infant development prediction program.
  • the machine learning-based infant development prediction program When executed by a processor, it implements the machine learning-based infant development prediction method provided in the present application. step.
  • the electronic equipment includes a motion sensor, a processor, a memory, and a machine learning-based infant development prediction program that is stored on the memory and can run on the processor, wherein:
  • the motion sensor is used to obtain single limb motion data of the sample to be tested
  • the method, device, storage medium, and electronic equipment for predicting infant development based on machine learning can obtain single limb movement data of the sample to be tested and the corresponding month when the stacked extreme learning and mathematical model are known.
  • Age data, the characteristic value of a single limb movement of the sample to be measured can be obtained through data processing, and the characteristic value can be substituted into the known stacked extreme learning and mathematical model, and the logical operation can be performed to obtain the output result.
  • the output results include two major categories of developmental delay and normal development.
  • the month-old data of the sample to be tested can be obtained according to the date of birth shown in the birth certificate, that is to say, the baby development prediction based on machine learning provided in this application
  • the data to be acquired is only a single limb movement data. In this case, it can be directly obtained through a motion sensor. Therefore, the application of the method for predicting infant development based on machine learning provided by this application can be more convenient for infants.
  • stacked extreme learning and mathematical models can cover babies of all months of age during the construction process. Therefore, even if the The sample is small in months, and relatively accurate output results can also be obtained through the stacked extreme learning and mathematical model. Therefore, it can provide a reliable basis for early intervention of infant developmental delay.
  • FIG. 1 is a schematic diagram of the structure of a machine learning-based infant development prediction device in a hardware operating environment involved in a solution of an embodiment of the application;
  • FIG. 2 is a flow chart of the steps of a method for predicting baby development based on machine learning involved in a scheme of an embodiment of the application;
  • FIG. 3 is a flow chart of the steps of a method for constructing a mathematical model of a stacked extreme learning machine used in a method for predicting infant development based on machine learning according to an embodiment of the application;
  • FIG. 4 is a schematic diagram of the signal flow relationship between the functional modules in the device for predicting infant development based on machine learning according to the embodiment of the application;
  • FIG. 5 is a schematic diagram of a stacked extreme learning machine used in a method for predicting infant development based on machine learning according to an embodiment of the application.
  • this application provides a method, device, storage medium and electronic equipment for predicting the development of infants based on machine learning, which can predict the development of infants and further obtain the possibility of infant developmental delay According to this, it can provide a reliable basis for early intervention of infants with developmental delay, which is more suitable for practical use.
  • FIG. 1 is a schematic structural diagram of a device for predicting baby development based on machine learning in a hardware operating environment involved in a solution of an embodiment of the application.
  • the device for predicting baby development based on machine learning may include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WI-FI) interface).
  • WI-FI wireless fidelity
  • the memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory.
  • RAM Random Access Memory
  • NVM Non-Volatile Memory
  • the memory 1005 may also be a storage device independent of the foregoing processor 1001.
  • FIG. 1 does not constitute a limitation on the infant development prediction device based on machine learning, and may include more or less components than shown, or combine certain components, or Different component arrangements.
  • the memory 1005 as a storage medium may include an operating system, a data storage module, a network communication module, a user interface module, and a baby development prediction program based on machine learning.
  • the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with users; this application is based on machine learning for baby development
  • the processor 1001 and the memory 1005 in the situation prediction device can be set in the baby development prediction device based on machine learning, and the baby development prediction device based on machine learning calls the baby development based on machine learning stored in the memory 1005 through the processor 1001
  • the situation prediction program and implements the infant development situation prediction method based on machine learning provided in the embodiments of the application.
  • the method for predicting infant development based on machine learning includes the following steps:
  • Step 101 Obtain single limb movement data and corresponding monthly age data of the sample to be tested.
  • the limb movement data includes the duration of a single movement, the average acceleration a v of a single movement, the peak acceleration a m of a single movement, the movement type S L of the left leg, and the movement type S R of the right leg.
  • the duration t i of a single exercise is the difference between the end time T i1 of the single exercise and the start time T i0 of the single exercise, that is, Ti1 ⁇ T i0 .
  • the left leg movement type S L and the right leg movement type S R can be displayed in the form of text description.
  • kick Q 1
  • Step 102 Perform data processing on the single limb movement data of the sample to be tested, and obtain the feature value corresponding to the monthly age data for the single limb movement of the sample to be tested.
  • the data processing method for the single limb movement data of the sample to be tested may include univariate feature selection, recursive feature selection, stepwise feature selection, etc., where:
  • Univariate feature selection refers to selecting the feature value X j of a single limb movement of the sample to be tested by statistical measurement methods based on some univariate.
  • Recursive feature selection refers to the normalized data processing of the above variables, and the obtained data is used as the feature value X j of a single limb movement of the sample to be tested.
  • X j ⁇ y 1 (T i1 -T i0 )+y 2 a v +y 3 a m , X L , X R , Y ⁇ , where y 1 is (T i1 -T i0 )
  • the characteristic coefficient, y 2 is the characteristic coefficient of a v
  • y 3 is the characteristic coefficient of a m .
  • the normalized data processing method adopted is linear normalized data processing.
  • some of the above three characteristics (T i1 -T i0 ), a v , and a m can also be performed.
  • the exponentiation operation can increase the influence level, and the square operation can also be performed on some of the above three characteristics (T i1 -T i0 ), a v , and a m to reduce the influence level.
  • Step 103 Bring a predetermined number of characteristic values of a single limb movement corresponding to the sample to be tested and the corresponding monthly age data into the mathematical model of the stacked extreme learning machine, and perform logical operations.
  • the characteristic value X j of the single limb movement of the sample to be tested is substituted into the mathematical model of the stacked extreme learning machine, namely
  • L- is the number of nodes of a single hidden layer neural network
  • w i [W i1 ,w i2 ,...,w in ] T -is the input weight of the i-th hidden layer unit
  • bi- is the bias of the i-th hidden layer unit
  • ⁇ i [ ⁇ i1 , ⁇ i2 ,..., ⁇ im ] T -is the output weight of the i-th hidden layer unit.
  • the expression of the activation function g(x), the input weight wi of the i-th hidden layer unit, the bias bi of the i-th hidden layer unit, and The output weight ⁇ i of the i-th hidden layer unit is known. Therefore, the output result oj can be obtained through the above-mentioned logical operation, which includes two major categories: normal development and developmental delay.
  • Step 104 Use the output result of the logical operation of the stacked extreme learning machine as a basis for the normal or developmental delay of the sample to be tested, and obtain the prediction result of the infant's development.
  • the sample to be tested is determined to be stunted after 5-10 sets of limb movement data, you can increase the number of limb movement groups obtained by 1-3 times, and more limb movement groups can be used to determine whether the infants and young children are stunted. Has a diagnostic role.
  • the method for predicting baby development based on machine learning is to obtain single limb movement data and corresponding monthly age data of the sample to be tested when stacked extreme learning and mathematical models are known.
  • the characteristic value of a single limb movement of the sample to be tested is obtained, and the characteristic value is substituted into the known mathematical model of stacked extreme learning and logical operation to obtain the output result.
  • the output result includes developmental delay
  • the month-old data of the sample to be tested can be obtained according to the date of birth shown in the birth certificate, that is to say, in the method for predicting the development of infants based on machine learning provided in this application, to be obtained
  • the data of is only a single limb movement data. In this case, it can be directly obtained through a motion sensor. Therefore, the application of the machine learning-based infant development prediction method provided by this application can more conveniently predict the infant development status.
  • stacked extreme learning and mathematical models can cover babies of all months of age during the construction process. Therefore, even if the sample to be tested is older than It is small, and relatively accurate output results can also be obtained through the stacked extreme learning and mathematical model. Therefore, it can provide a reliable basis for early intervention of infant developmental delay.
  • the method for constructing the mathematical model of the stacked extreme learning machine includes the following steps:
  • Step 201 Obtain single limb movement data and corresponding monthly age data of a training sample, where the training sample is a sample with a known developmental status.
  • the limb movement data includes the duration of a single movement, the average acceleration a v of a single movement, the peak acceleration a m of a single movement, the movement type S L of the left leg, and the movement type S R of the right leg.
  • the duration t i of a single exercise is the difference between the end time T i1 of the single exercise and the start time T i0 of the single exercise, that is, Ti1 ⁇ T i0 .
  • the left leg movement type S L and the right leg movement type S R can be displayed in the form of text description. For example, kicking, squatting, walking, crawling, bending, etc.
  • kick Q 1
  • infants and young children develop relatively rapidly in 24 months, especially within 12 months, their physical activities can change significantly even if they are only a few days apart. Therefore, months of age The data Y needs to be accurate to the day, for example, the age of the month data can be 2m+1, 8m+10, etc.
  • Step 202 Perform data processing on the single limb movement data of the training sample to obtain a feature value corresponding to the monthly age data for the single limb movement of the training sample.
  • the data processing method for the single limb movement data of the training sample may include univariate feature selection, recursive feature selection, stepwise feature selection, etc., and the single limb movement for the sample to be tested Performing data processing on the data to obtain the characteristic value corresponding to the monthly age data for the single limb movement of the sample to be tested specifically includes:
  • Univariate feature selection refers to selecting the feature value X j of a single limb movement as a training sample by a statistical measurement method based on some univariate.
  • Recursive feature selection refers to normalized data processing of the above variables, and the obtained data is used as the feature value X j of a single limb movement of the training sample.
  • X j ⁇ y 1 (T i1 -T i0 )+y 2 a v +y 3 a m , X L , X R , Y ⁇ , where y 1 is (T i1 -T i0 )
  • the characteristic coefficient, y 2 is the characteristic coefficient of a v
  • y 3 is the characteristic coefficient of a m .
  • the normalized data processing method adopted is linear normalized data processing.
  • some of the above three characteristics (T i1 -T i0 ), a v , and a m can also be performed.
  • the exponentiation operation can increase the influence level, and the square operation can also be performed on some of the above three characteristics (T i1 -T i0 ), a v , and a m to reduce the influence level. or
  • Step 203 Bring the characteristic value of the single limb movement of the training sample corresponding to the monthly age data into the logical operation formula: When there are multiple training samples, multiple expressions based on logical operation formulas are obtained;
  • L- is the number of nodes of a single hidden layer neural network
  • w i [W i1 ,w i2 ,...,w in ] T -is the input weight of the i-th hidden layer unit
  • bi- is the bias of the i-th hidden layer unit
  • ⁇ i [ ⁇ i1 , ⁇ i2 ,..., ⁇ im ] T -is the output weight of the i-th hidden layer unit
  • oj- is the output result of logical operation after classification using stacked extreme learning machine for training samples.
  • the output result of logical operation includes two major categories: normal development and developmental delay.
  • oj in the case of training samples, the development of the training samples is known, that is, in the case of training samples, oj is known. Under normal circumstances, oj includes two categories of developmental delay or normal development.
  • Step 204 Determine the expression of the activation function g(x), the input weight wi of the i-th hidden layer unit, the bias bi of the i-th hidden layer unit, and the i-th hidden layer unit according to multiple expressions based on logical operation formulas The output weight ⁇ i of the hidden unit.
  • a feature value X j that only needs to be input to the stacked trace learning machine for a single limb movement of the sample to be tested is obtained and the number of nodes L of the single hidden layer neural network is determined. , Which can output the logical operation of oj. That is to say, in this case, it is only necessary to obtain the characteristic value X j of a single limb movement of the sample to be tested to determine the development status of the sample to be tested.
  • Step 205 The expression of the determined activation function g(x), the input weight wi of the i-th hidden layer unit, the bias bi of the i-th hidden layer unit, and the output weight ⁇ i of the i-th hidden layer unit Bring back to the logical calculation formula and get the mathematical model of the stacked extreme learning machine.
  • the mathematical model of the stacked extreme learning machine can be obtained, namely
  • ⁇ i [ ⁇ i1 , ⁇ i2 ,..., ⁇ im ] T -the output weight of the i-th hidden layer unit.
  • the output result is unknown. It can be based on the expression of the known activation function g(x), the input weight wi of the i-th hidden unit, and the bias of the i-th hidden unit
  • the output weight ⁇ i of bi and the i-th hidden layer unit and the number of nodes L of the single hidden layer neural network are output by logical operations oj.
  • the single limb movement data includes: the beginning time T i0 of the single limb movement, the end time T i1 , the average acceleration a v , the peak acceleration am , the left leg movement type S L , and the right leg movement type S R.
  • the duration t i of a single exercise is the difference T i1 -T i0 between the end time of the single exercise and the start time T i0 of the single exercise.
  • the reason why these indicators are selected as the data of a single limb movement the first consideration is the availability of motion sensor data. Then, we need to consider the correlation between the data and the baby’s developmental status. Among them, the average acceleration a v , peak acceleration am , left leg movement type S L , right leg movement type S R z these indicators can indicate the baby's gross motor ability Therefore, they are used as the basic data for obtaining the characteristic value X j of a single limb movement. Among them, the single limb movement data of the sample to be tested and the single limb movement data of the training sample are measured by the same movement sensor, or at least the same type of sensor produced by the same manufacturer. The reason why the same sensor is at least the same type of sensor produced by the same manufacturer is because this choice can reduce the judgment error caused by the sensor detection error.
  • the method of performing data processing on the single limb movement data of the sample to be tested to obtain the feature value corresponding to the monthly age data for the single limb movement of the sample to be tested is selected from univariate feature selection, recursive feature selection, and stepwise feature One of the choices. among them:
  • Univariate feature selection uses a statistical measurement method for a single variable in a single limb movement data to select and obtain the characteristic value of a single limb movement.
  • the selected feature is a single variable
  • the calculation efficiency is high.
  • a single variable usually considers less influencing factors, the possibility of error in the judgment result is relatively large. .
  • Recursive feature selection is through the normalized data processing of each variable in the single limb movement data, and the obtained data is used as the characteristic value of the single limb movement.
  • the recursive feature selection comprehensively considers a single variable, and also considers the association between each single variable by quoting feature coefficients, it can reduce the possibility of error in the judgment result .
  • a single variable in the single limb movement data is selected one by one, and the data obtained by normalizing each variable in the single limb movement data is sequentially used as the characteristic value of a single limb movement.
  • the degree of retardation of grade 3 is the mildest, and the degree of retardation of grade 3 is the heaviest; normal-excellent, normal-good, and normal-medium.
  • the three detection data of (T i1 -T i0 ), a v , and a m can also be combined. Values and the categories of X L and X R are further classified as normal or slow.
  • the device for predicting infant development based on machine learning includes:
  • the data acquisition unit 301 is used to acquire single limb movement data and corresponding monthly age data of the sample to be tested.
  • the limb movement data includes the duration of a single movement, the average acceleration a v of a single movement, the peak acceleration a m of a single movement, the movement type S L of the left leg, and the movement type S R of the right leg.
  • the duration t i of a single exercise is the difference between the end time T i1 of the single exercise and the start time T i0 of the single exercise, that is, Ti1 ⁇ T i0 .
  • the left leg movement type S L and the right leg movement type S R can be displayed in the form of text description.
  • kick Q 1
  • the data processing unit 302 is configured to perform data processing on the single limb movement data of the sample to be tested, to obtain the feature value corresponding to the monthly age data for the single limb movement of the sample to be tested.
  • the data processing method for the single limb movement data of the sample to be tested may include univariate feature selection, recursive feature selection, stepwise feature selection, etc., where:
  • Univariate feature selection refers to selecting the feature value X j of a single limb movement of the sample to be tested by statistical measurement methods based on some univariate.
  • Recursive feature selection refers to the normalized data processing of the above variables, and the obtained data is used as the feature value X j of a single limb movement of the sample to be tested.
  • X j ⁇ y 1 (T i1 -T i0 )+y 2 a v +y 3 a m , X L , X R , Y ⁇ , where y 1 is (T i1 -T i0 )
  • the characteristic coefficient, y 2 is the characteristic coefficient of a v
  • y 3 is the characteristic coefficient of a m .
  • the normalized data processing method adopted is linear normalized data processing.
  • some of the above three characteristics (T i1 -T i0 ), a v , and a m can also be performed.
  • the exponentiation operation can increase the influence level, and the square operation can also be performed on some of the above three characteristics (T i1 -T i0 ), a v , and a m to reduce the influence level.
  • the arithmetic unit 303 is used to bring a predetermined number of processed limb motion data and corresponding monthly age data corresponding to the sample to be tested into the mathematical model of the stacked extreme learning machine for logical operations.
  • the characteristic value X j of the single limb movement of the sample to be tested is substituted into the mathematical model of the stacked extreme learning machine, namely
  • L- is the number of nodes of a single hidden layer neural network
  • w i [W i1 ,w i2 ,...,w in ] T -is the input weight of the i-th hidden layer unit
  • bi- is the bias of the i-th hidden layer unit
  • ⁇ i [ ⁇ i1 , ⁇ i2 ,..., ⁇ im ] T -is the output weight of the i-th hidden layer unit.
  • the expression of the activation function g(x), the input weight wi of the i-th hidden layer unit, the bias bi of the i-th hidden layer unit, and The output weight ⁇ i of the i-th hidden layer unit is known. Therefore, the output result oj can be obtained through the above-mentioned logical operation, which includes two major categories: normal development and developmental delay.
  • the prediction result output unit 304 is configured to use the output result of the logical operation of the stacked extreme learning machine as a basis for the normal development or developmental delay of the sample to be tested to obtain the prediction result of the baby's development.
  • the sample to be tested is determined to be stunted after 5-10 sets of limb movement data, you can increase the number of limb movement groups obtained by 1-3 times, and more limb movement groups can be used to determine whether the infants and young children are stunted. Has a diagnostic role.
  • the device for predicting baby development based on machine learning obtained in the second embodiment of the application obtains single limb movement data and corresponding monthly age data of the sample to be tested when the stacked extreme learning and mathematical models are known. Through data processing, the characteristic value of a single limb movement of the sample to be tested is obtained, and the characteristic value is substituted into the known mathematical model of stacked extreme learning and logical operation to obtain the output result. Among them, the output result includes developmental delay There are two categories of normal development and normal development.
  • the month-old data of the sample to be tested can be obtained according to the date of birth shown in the birth certificate, that is to say, in the method for predicting the development of infants based on machine learning provided in this application, to be obtained
  • the data of is only a single limb movement data. In this case, it can be directly obtained through a motion sensor. Therefore, the application of the machine learning-based infant development prediction method provided by this application can more conveniently predict the infant development status.
  • stacked extreme learning and mathematical models can cover babies of all months of age during the construction process. Therefore, even if the sample to be tested is older than It is small, and relatively accurate output results can also be obtained through the stacked extreme learning and mathematical model. Therefore, it can provide a reliable basis for early intervention of infant developmental delay.
  • the storage medium provided in the third embodiment of the application stores a machine learning-based baby development prediction program, and the machine learning-based baby development prediction program is executed by a processor to implement the machine learning-based baby development prediction method provided in this application A step of.
  • the storage medium provided in the third embodiment of the application obtains the single limb movement data and the corresponding monthly age data of the sample to be tested when the stacked extreme learning and mathematical model are known, and the sample to be tested can be obtained through data processing
  • the output results include developmental delay and normal development.
  • the monthly age data of the sample to be tested can be obtained according to the date of birth shown in the birth certificate, that is to say, in the method for predicting baby development based on machine learning provided in this application, the data to be obtained is only a single limb In this case, the motion data can be directly obtained through the motion sensor. Therefore, the application of the method for predicting infant development based on machine learning provided in this application can more conveniently predict the development of infants.
  • stacking extreme learning and mathematical models can cover babies of all months of age during the construction process. Therefore, even if the sample to be tested is of a relatively small month, it can be used by stacking The formula extreme learning and mathematical model get relatively accurate output results, therefore, it can provide a reliable basis for early intervention of infant developmental delay.
  • the electronic device provided in the fourth embodiment of the present application includes a motion sensor, a processor, a memory, and a machine learning-based infant development prediction program stored on the memory and running on the processor, wherein:
  • Motion sensor used to obtain single limb movement data of the sample to be tested
  • the electronic device provided in the fourth embodiment of the application obtains the single limb movement data and the corresponding monthly age data of the sample to be tested when the stacked extreme learning and the mathematical model are known, and the sample to be tested can be obtained through data processing
  • the output results include developmental delay and normal development.
  • the monthly age data of the sample to be tested can be obtained according to the date of birth shown in the birth certificate, that is to say, in the method for predicting baby development based on machine learning provided in this application, the data to be obtained is only a single limb In this case, the motion data can be directly obtained through the motion sensor. Therefore, the application of the method for predicting infant development based on machine learning provided in this application can more conveniently predict the development of infants.
  • stacking extreme learning and mathematical models can cover babies of all months of age during the construction process. Therefore, even if the sample to be tested is of a relatively small month, it can be used by stacking The formula extreme learning and mathematical model get relatively accurate output results, therefore, it can provide a reliable basis for early intervention of infant developmental delay.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Physiology (AREA)
  • Surgery (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Dentistry (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备,属于计算机程序技术领域。方法包括:获取待测样本的单次肢体运动数据及对应的月龄数据(101);针对待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对待测样本单次肢体运动的特征值(102);将预定次数的与待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算(103);以堆叠式极限学习机的逻辑运算的输出结果作为待测样本发育正常或者发育迟缓的依据,得到婴儿发育情况的预测结果(104)。装置、存储介质及电子设备能够用于实现婴儿发育情况预测方法,通过其能够对婴儿进行发育情况预测,进一步得到婴儿发育迟缓的可能性,据此,能够为婴儿发育迟缓的早期干预提供可靠的依据。

Description

基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备
本申请要求与2019年4月26日提交中国专利局、申请号为201910346364.8、申请名称为“基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及计算机程序技术领域,特别是涉及一种基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备。
背景技术
现有技术通常是要等到婴幼儿2岁以后才能确诊婴幼儿是否发育迟缓。然而,发明人发现时至此时才确诊婴幼儿发育迟缓,对婴幼儿的干预已经相对较晚,不利于发育迟缓的婴幼儿恢复。因此,就需要一种能够针对婴儿发育情况进行预测的方法,从而能够为发育迟缓的婴幼儿的恢复进行干预提供依据。
发明内容
有鉴于此,本申请提供了一种基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备,通过其能够对婴儿进行发育情况预测,进一步得到婴儿发育迟缓的可能性,据此,能够为婴儿发育迟缓的早期干预提供可靠的依据,从而更加适于实用。
为了达到上述第一个目的,本申请提供的基于机器学习的婴儿发育情况预测方法的技术方案如下:
本申请提供的基于机器学习的婴儿发育情况预测方法包括以下步骤:
获取待测样本的单次肢体运动数据及对应的月龄数据;
针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
将预定次数的与所述待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
本申请提供的基于机器学习的婴儿发育情况预测方法还可采用以下技术措施进一步实现。
作为优选,所述堆叠式极限学习机的数学模型的构建方法包括以下步骤:
获取训练样本的单次肢体运动数据及对应的月龄数据,其中,所述训练样本为已知发育状况的样本;
针对所述训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述训练样本单次肢体运动的特征值;
将所述与月龄数据相对应的针对所述训练样本单次肢体运动的特征值带入至逻辑运算公式:
Figure PCTCN2019103148-appb-000001
当所述训练样本为多个时,得到多个基于所述逻辑运算公式的表达式;
其中,
L-为单隐层神经网络的节点数,
g(x)-为激活函数,
w i=[W i1,w i2,...,w in] T-为第i个隐层单元的输入权重,
bi-为第i个隐层单元的偏置,
β i=[β i1i2,...,β im] T-为第i个隐层单元的输出权重,
oj-为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,所述逻辑运算输出结果包括发育正常和发育迟缓2个大类;
根据所述多个基于所述逻辑运算公式的表达式,确定所述激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi;
再将确定后的激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi带回至所述逻辑运算公式,得到所述堆叠式极限学习机的数学模型。
作为优选,所述单次肢体运动数据包括:
单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R
其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
作为优选,针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
所述单变量特征选择通过对所述单次肢体运动数据中的单一变量的统计度量方法,选取得到作为所述单次肢体运动的特征值;或
所述递归特征选择通过对所述单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为所述单次肢体运动的特征值;或
所述逐步特征选择通过逐一选取所述单次肢体运动数据中的单一变量、对所述单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
作为优选,在所述逻辑运算输出结果中,
根据所述堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
根据所述堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
为了达到上述第二个目的,本申请提供的基于机器学习的婴儿发育情况预测装置的技术方案如下:
本申请提供的基于机器学习的婴儿发育情况预测装置包括:
数据获取单元,用于获取待测样本的单次肢体运动数据及对应的月龄数据;
数据处理单元,用于针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
运算单元,用于将预定次数的与所述待测样本相对应的处理后的肢体运动数据及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
预测结果输出单元,用于以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
为了达到上述第三个目的,本申请提供的存储介质的技术方案如下:
本申请提供的存储介质上存储有基于机器学习的婴儿发育情况预测程序,所述基于机器学习的婴儿发育情况预测程序被处理器执行时实现本申请提供的基于机器学习的婴儿发育情况预测方法的步骤。
为了达到上述第四个目的,本申请提供的电子设备的技术方案如下:
本申请提供的电子设备包括运动传感器、处理器、存储器及存储在所述存储器上并可在所述处理器上运行的基于机器学习的婴儿发育情况预测程序,其中,
所述运动传感器,用于获取待测样本的单次肢体运动数据;
所述基于机器学习的婴儿发育情况预测程序被所述处理器执行时实现本申请提供的基于机器学习的婴儿发育情况预测方法的步骤。
本申请提供的基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备在堆叠式极限学习及的数学模型已知的情况下,获取待测样本的单次肢体运动数据及对应的月龄数据,即可通过数据处理得到待测样本单次肢体运动的特征值,将该特征值代入至已知的堆叠式极限学习及的数学模型,进行逻辑运算,即可得到输出结果,其中,输出结果包括发育迟缓和发育正常2个大类,其中,由于待测样本的月龄数据可以根据出生证明显示的出生日期获得,也就是说,在本申请提供的基于机器学习的婴儿发育情况预测方法中,待获取的数据仅为单次肢体运动数据,在这种情况下,可以通过运动传感器直接获得,因此,应用本申请提供的基于机器学习的婴儿发育情况预测方法能够较为便捷地对婴儿发育状况进行预测,此外,在本申请提供的基于机器学习的婴儿发育情况预测方法中,堆叠式极限学习及的数学模型在构建的过程中,可以涵盖所有月龄的婴儿,因此,即使待测样本的月龄较小,也可以通过该堆叠式极限学习及的数学模型 得到相对准确的输出结果,因此,其能够为婴儿发育迟缓的早期干预提供可靠的依据。
附图说明
通过阅读下文优选实施方式的详细描述,各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的,而并不认为是对本申请的限制。而且在整个附图中,用相同的参考符号表示相同的部件。在附图中:
图1为本申请实施例方案涉及的硬件运行环境的基于机器学习的婴儿发育情况预测设备结构示意图;
图2为本申请实施例方案涉及的基于机器学习的婴儿发育情况预测方法的步骤流程图;
图3为本申请实施例方案涉及的基于机器学习的婴儿发育情况预测方法中应用的堆叠式极限学习机的数学模型的构建方法的步骤流程图;
图4为本申请实施例方案涉及的基于机器学习的婴儿发育情况预测装置中各功能模块之间的信号流向关系示意图;
图5为本申请实施例方案涉及的基于机器学习的婴儿发育情况预测方法中应用的堆叠式极限学习机的原理示意图。
具体实施方式
本申请为解决现有技术存在的问题,提供一种基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备,通过其能够对婴儿进行发育情况预测,进一步得到婴儿发育迟缓的可能性,据此,能够为婴儿发育迟缓的早期干预提供可靠的依据,从而更加适于实用。
为更进一步阐述本申请为达成预定发明目的所采取的技术手段及功效,以下结合附图及较佳实施例,对依据本申请提出的基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备,其具体实施方式、结构、特征及其功效,详细说明如后。在下述说明中,不同的“一实施例”或“实施例”指的不一定是同一实施例。此外,一或多个实施例中的特征、结构、或特点可由任何合适形式组合。
本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/ 或B,具体的理解为:可以同时包含有A与B,可以单独存在A,也可以单独存在B,能够具备上述三种任一种情况。
参照图1,图1为本申请实施例方案涉及的硬件运行环境的基于机器学习的婴儿发育情况预测设备结构示意图。
如图1所示,该基于机器学习的婴儿发育情况预测设备可以包括:处理器1001,例如中央处理器(Central Processing Unit,CPU),通信总线1002、用户接口1003,网络接口1004,存储器1005。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如无线保真(WIreless-FIdelity,WI-FI)接口)。存储器1005可以是高速的随机存取存储器(Random Access Memory,RAM)存储器,也可以是稳定的非易失性存储器(Non-Volatile Memory,NVM),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解,图1中示出的结构并不构成对基于机器学习的婴儿发育情况预测设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种存储介质的存储器1005中可以包括操作系统、数据存储模块、网络通信模块、用户接口模块以及基于机器学习的婴儿发育情况预测程序。
在图1所示的基于机器学习的婴儿发育情况预测设备中,网络接口1004主要用于与网络服务器进行数据通信;用户接口1003主要用于与用户进行数据交互;本申请基于机器学习的婴儿发育情况预测设备中的处理器1001、存储器1005可以设置在基于机器学习的婴儿发育情况预测设备中,基于机器学习的婴儿发育情况预测设备通过处理器1001调用存储器1005中存储的基于机器学习的婴儿发育情况预测程序,并执行本申请实施例提供的基于机器学习的婴儿发育情况预测方法。
实施例一
参见附图1,本申请实施例一提供的基于机器学习的婴儿发育情况预测方法包括以下步骤:
步骤101:获取待测样本的单次肢体运动数据及对应的月龄数据。
具体而言,此处是通过传感器获取待测样本的肢体运动数据。其中,肢体运动数据包括单次运动持续时间、单次运动平均加速度a v、单次运动峰值加速度a m、左腿运动类型S L、右腿运动类型S R。其中,单次运动持续时间t i是单次运动末时刻T i1与单次运动始时刻T i0的差值即T i1-T i0。其中,左腿运动类型S L、右腿运动类型S R可以应用文字描述的方式展示,例如,可以定性地描述踢腿、下蹲、行走、爬行、弯曲等,还可以给各运动类型定义一符号,例如踢腿=Q 1,下蹲=Q 2,行走=Q 3,爬行=Q 4,弯曲=Q 5,然后,以符号的方式进行描述。另外,由于婴幼儿在24个月内特别是12个月内时,肢体动作能力发展较为迅速,有时,即使仅相差数天,婴幼儿的肢体动作即可发生较大的变化,因此,月龄数据Y需要精确到天,例如,月龄数据可以为2m+1、8m+10等。
步骤102:针对待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对待测样本单次肢体运动的特征值。
具体而言,针对所述待测样本的单次肢体运动数据进行数据处理的方式可以包括单变量特征选择、递归特征选择、逐步特征选择等方式,其中,
单变量特征选择指的是通过基于一些单变量的统计度量方法选取到作为待测样本单次肢体运动的特征值X j。此时,第一实施例可以为X j={T i1-T i0,m};第二实施例可以为X j={a v,Y};第三实施例可以为X j={a m,Y};第四实施例可以为X j={X L,X R,Y}。
递归特征选择指的是通过对上述各变量进行归一化数据处理,以得到的数据作为待测样本单次肢体运动的特征值X j。例如,此时X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y},其中,y 1为(T i1-T i0)的特征系数,y 2为a v的特征系数,y 3为a m的特征系数。本实施例中,采取的归一化数据处理方式是线性归一化数据处理,根据实际需要还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行乘方运算,从而增加其影响等级,还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行开方运算,从而降低影响等级。
逐步特征选择指的是,第一次选取X j={T i1-T i0,Y},第二次选取X j={a v,Y},第三次选取X j={a m,Y},第四次选取X j={X L,X R,Y};然后,还可以第五次选取X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y}。
步骤103:将预定次数的与待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算。
具体而言,将上述待测样本单次肢体运动的特征值X j代入至堆叠式极限学习机的数学模型即
Figure PCTCN2019103148-appb-000002
其中,
L-为单隐层神经网络的节点数,
g(x)-为激活函数,
w i=[W i1,w i2,...,w in] T-为第i个隐层单元的输入权重,
bi-为第i个隐层单元的偏置,
β i=[β i1i2,...,β im] T-为第i个隐层单元的输出权重。
在经过训练的堆叠式极限学习机的数学模型中,激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi均为已知的,因此,通过上述逻辑运算,即可得到输出结果oj,其包括发育正常和发育迟缓2个大类。
步骤104:以堆叠式极限学习机的逻辑运算的输出结果作为待测样本发育正常或者发育迟缓的依据,得到婴儿发育情况的预测结果。
此处需要说明的是,在同一时间段内,获取针对同一待测样本的运动数据的过程中,至少需要获取5-10组肢体运动数据才进行判断,其原因在于,若在获取待测样本的肢体运动数据时,获取的肢体运动数据的组数过少,则有可能由于待测样本由于偶然因素导致运动过于激烈或者运动过于缓慢而导致判断错误,但是,若在同一时间段内,至少获取5-10组肢体运动数据才进行判断时,则能够尽量减少由于偶然因素导致的判断错误。如若获取的肢体运动组数过多,则需要耗费较多的时间,对判断婴幼儿究竟是发育正常还 是发育迟缓的意义也不大。若待测样本经过5-10组肢体运动数据判断的结果是发育迟缓,则可以再增加获取1-3倍的肢体运动组数获取数量,更多的肢体运动组数对确诊婴幼儿是否发育迟缓具有确诊的作用。
本申请实施例一提供的基于机器学习的婴儿发育情况预测方法在堆叠式极限学习及的数学模型已知的情况下,获取待测样本的单次肢体运动数据及对应的月龄数据,即可通过数据处理得到待测样本单次肢体运动的特征值,将该特征值代入至已知的堆叠式极限学习及的数学模型,进行逻辑运算,即可得到输出结果,其中,输出结果包括发育迟缓和发育正常2个大类,其中,由于待测样本的月龄数据可以根据出生证明显示的出生日期获得,也就是说,在本申请提供的基于机器学习的婴儿发育情况预测方法中,待获取的数据仅为单次肢体运动数据,在这种情况下,可以通过运动传感器直接获得,因此,应用本申请提供的基于机器学习的婴儿发育情况预测方法能够较为便捷地对婴儿发育状况进行预测,此外,在本申请提供的基于机器学习的婴儿发育情况预测方法中,堆叠式极限学习及的数学模型在构建的过程中,可以涵盖所有月龄的婴儿,因此,即使待测样本的月龄较小,也可以通过该堆叠式极限学习及的数学模型得到相对准确的输出结果,因此,其能够为婴儿发育迟缓的早期干预提供可靠的依据。
参见附图3和附图5,堆叠式极限学习机的数学模型的构建方法包括以下步骤:
步骤201:获取训练样本的单次肢体运动数据及对应的月龄数据,其中,训练样本为已知发育状况的样本。
具体而言,此处是通过传感器获取训练样本的肢体运动数据。其中,肢体运动数据包括单次运动持续时间、单次运动平均加速度a v、单次运动峰值加速度a m、左腿运动类型S L、右腿运动类型S R。其中,单次运动持续时间t i是单次运动末时刻T i1与单次运动始时刻T i0的差值即T i1-T i0。其中,左腿运动类型S L、右腿运动类型S R可以应用文字描述的方式展示,例如,可以定性地描述踢腿、下蹲、行走、爬行、弯曲等,还可以给各运动类型定义一符号,例如踢腿=Q 1,下蹲=Q 2,行走=Q 3,爬行=Q 4,弯曲=Q 5,然后,以符号的方式进行描述。另外,由于婴幼儿在24个月内特别是12个月内时,肢体动作能力发展较为迅速,有时,即使仅相差数天,婴幼儿的肢体动作即可发生较大的变化,因此,月龄数据Y需要精确到天,例如,月龄数 据可以为2m+1、8m+10等。
步骤202:针对训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对训练样本单次肢体运动的特征值。
具体而言,针对所述训练样本的单次肢体运动数据进行数据处理的方式可以包括单变量特征选择、递归特征选择、逐步特征选择等方式,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
单变量特征选择指的是通过基于一些单变量的统计度量方法选取到作为训练样本单次肢体运动的特征值X j。此时,第一实施例可以为X j={T i1-T i0,m};第二实施例可以为X j={a v,Y};第三实施例可以为X j={a m,Y};第四实施例可以为X j={X L,X R,Y}。或
递归特征选择指的是通过对上述各变量进行归一化数据处理,以得到的数据作为训练样本单次肢体运动的特征值X j。例如,此时X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y},其中,y 1为(T i1-T i0)的特征系数,y 2为a v的特征系数,y 3为a m的特征系数。本实施例中,采取的归一化数据处理方式是线性归一化数据处理,根据实际需要还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行乘方运算,从而增加其影响等级,还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行开方运算,从而降低影响等级。或
逐步特征选择指的是,第一次选取X j={T i1-T i0,Y},第二次选取X j={a v,Y},第三次选取X j={a m,Y},第四次选取X j={X L,X R,Y};然后,还可以第五次选取X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y}。
步骤203:将与月龄数据相对应的针对训练样本单次肢体运动的特征值带入至逻辑运算公式:
Figure PCTCN2019103148-appb-000003
当训练样本为多个时,得到多个基于逻辑运算公式的表达式;
其中,
L-为单隐层神经网络的节点数,
g(x)-为激活函数,
w i=[W i1,w i2,...,w in] T-为第i个隐层单元的输入权重,
bi-为第i个隐层单元的偏置,
β i=[β i1i2,...,β im] T-为第i个隐层单元的输出权重,
oj-为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,逻辑运算输出结果包括发育正常和发育迟缓2个大类。
此处需要说明的是,在训练样本的情形下,训练样本的发育情况是已知的,也就是说,在训练样本的情形下,oj是已知的。通常情况下,oj包括发育迟缓或者发育正常2个分类,
步骤204:根据多个基于逻辑运算公式的表达式,确定激活函数g(x)的表达式、第i个隐层单元的输入权重wi、第i个隐层单元的偏置bi以及第i个隐层单元的输出权重βi。
在这种情况下,在堆叠式极限学习完成后,得到一个只需要向堆叠式迹线学习机输入待测样本单次肢体运动的特征值X j并确定单隐层神经网络的节点数L后,即可以输出oj的逻辑运算。也就是说,在这种情况下,只需要获取待测样本单次肢体运动的特征值X j,即可判断出待测样本的发育状况。
步骤205:再将确定后的激活函数g(x)的表达式、第i个隐层单元的输入权重wi、第i个隐层单元的偏置bi以及第i个隐层单元的输出权重βi带回至逻辑运算公式,得到堆叠式极限学习机的数学模型。
具体而言,在这种情况下,可以得到堆叠式极限学习机的数学模型即
Figure PCTCN2019103148-appb-000004
其中,
L-单隐层神经网络的节点数,
g(x)-激活函数,
w i=[W i1,w i2,...,w in] T-第i个隐层单元的输入权重,
bi-第i个隐层单元的偏置,
β i=[β i1i2,...,β im] T-第i个隐层单元的输出权重。
其中,针对待测样本而言,输出结果是未知的,可以根据已知的激活函数g(x)的表达式、第i个隐层单元的输入权重wi、第i个隐层单元的偏置bi以及第i个隐层单元的输出权重βi以及单隐层神经网 络的节点数L,通过逻辑运算输出oj。
其中,单次肢体运动数据包括:单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R。其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
此处需要解释的是,之所以选取这些指标作为单次肢体运动的数据,首先考虑到的是运动传感器数据的可获得性。然后,还要考虑数据与婴儿发育状况的关联性,其中平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S Rz这几个指标能够表示婴儿的大运动能力,因此,将它们作为获取单次肢体运动特征值X j的基础数据。其中,待测样本的单次肢体运动数据、训练样本的单次肢体运动数据是由同一运动传感器测得,或者,至少是有同一厂商生产的相同型号的传感器得的。之所以选用同一传感器至少是同一厂商生产的相同型号的传感器,是因为这样的选择能够降低由于传感器检测误差导致的判断误差的情况。
其中,针对待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种。其中:
单变量特征选择通过对单次肢体运动数据中的单一变量的统计度量方法,选取得到作为单次肢体运动的特征值。
具体而言,在这种情况下,由于选取的特征为单一变量,因此,计算效率高,但是,由于单一变量考虑的影响因子通常较少,因此,判断结果出现误差的可能性也相对较大。
递归特征选择通过对单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为单次肢体运动的特征值。
具体而言,在这种情况下,由于递归特征选择综合考虑了单一变量,并且还通过引用特征系数的方式,考虑了各单一变量之间的关联,因此,能够减少判断结果出现误差的可能性。
逐步特征选择通过逐一选取单次肢体运动数据中的单一变量、对单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
具体而言,在这种情况下,结合了单一变量以及归一化变量进行综合判断,能够更进一步减少判断结果出现误差的可能性。
其中,在逻辑运算输出结果中,
根据堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
根据堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
具体而言,在一些需要具体细分的情况下,还可以分别针对发育迟缓和发育正常分类进行更加具体的分类,例如迟缓-1级,迟缓-2级,迟缓-3级,其中迟缓-1级的迟缓程度最轻,迟缓-3级的迟缓程度最重;正常-优,正常-良,正常-中等。具体而言,在上述点变量特征选择、递归特征选择、逐步特征选择的判断结果输出oj的基础上,还可以结合(T i1-T i0)、a v、a m这三个检测数据的取值以及X L,X R这两个数据的类别,对正常或者迟缓进一步进行分级,其中,当oj为正常时,(T i1-T i0)、a v、a m这三个数据数值越大,说明在发育正常的情况下越优秀,此时,还可以进一步针对(T i1-T i0)、a v、a m这三个数据设定优、良、中各自对应的阈值;X L,X R的类别越多,说明在发育正常的情况下越优秀,此时,还可以进一步针对X L,X R的类别设置优、良、中各自对应的阈值。当oj为迟缓时,(T i1-T i0)、a v、a m这三个数据数值越小,说明在发育迟缓的情况下越严重,此时,还可以进一步针对(T i1-T i0)、a v、a m这三个数据设定1级、2级、3级各自对应的阈值;X L,X R的类别越少,说明在发育迟缓的情况下越严重,此时,还可以进一步针对X L,X R的类别设置1级、2级、3级各自对应的阈值。
实施例二
参见附图4,本申请实施例二提供的基于机器学习的婴儿发育情况预测装置包括:
数据获取单元301,用于获取待测样本的单次肢体运动数据及对应的月龄数据。
具体而言,此处是通过传感器获取待测样本的肢体运动数据。其中,肢体运动数据包括单次运动持续时间、单次运动平均加速度a v、单次运动峰值加速度a m、左腿运动类型S L、右腿运动类型S R。其中,单次运动持续时间t i是单次运动末时刻T i1与单次运动始时刻T i0的差值即T i1-T i0。其中,左腿运动类型S L、右 腿运动类型S R可以应用文字描述的方式展示,例如,可以定性地描述踢腿、下蹲、行走、爬行、弯曲等,还可以给各运动类型定义一符号,例如踢腿=Q 1,下蹲=Q 2,行走=Q 3,爬行=Q 4,弯曲=Q 5,然后,以符号的方式进行描述。另外,由于婴幼儿在24个月内特别是12个月内时,肢体动作能力发展较为迅速,有时,即使仅相差数天,婴幼儿的肢体动作即可发生较大的变化,因此,月龄数据Y需要精确到天,例如,月龄数据可以为2m+1、8m+10等。
数据处理单元302,用于针对待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对待测样本单次肢体运动的特征值。
具体而言,针对所述待测样本的单次肢体运动数据进行数据处理的方式可以包括单变量特征选择、递归特征选择、逐步特征选择等方式,其中,
单变量特征选择指的是通过基于一些单变量的统计度量方法选取到作为待测样本单次肢体运动的特征值X j。此时,第一实施例可以为X j={T i1-T i0,m};第二实施例可以为X j={a v,Y};第三实施例可以为X j={a m,Y};第四实施例可以为X j={X L,X R,Y};
递归特征选择指的是通过对上述各变量进行归一化数据处理,以得到的数据作为待测样本单次肢体运动的特征值X j。例如,此时X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y},其中,y 1为(T i1-T i0)的特征系数,y 2为a v的特征系数,y 3为a m的特征系数。本实施例中,采取的归一化数据处理方式是线性归一化数据处理,根据实际需要还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行乘方运算,从而增加其影响等级,还可以针对上述(T i1-T i0)、a v、a m这三个特征中的某些进行开方运算,从而降低影响等级。
逐步特征选择指的是,第一次选取X j={T i1-T i0,Y},第二次选取X j={a v,Y},第三次选取X j={a m,Y},第四次选取X j={X L,X R,Y};然后,还可以第五次选取X j={y 1(T i1-T i0)+y 2a v+y 3a m,X L,X R,Y}。
运算单元303,用于将预定次数的与待测样本相对应的处理后的肢体运动数据及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算。
具体而言,将上述待测样本单次肢体运动的特征值X j代入至堆叠式极限学习机的数学模型即
Figure PCTCN2019103148-appb-000005
其中,
L-为单隐层神经网络的节点数,
g(x)-为激活函数,
w i=[W i1,w i2,...,w in] T-为第i个隐层单元的输入权重,
bi-为第i个隐层单元的偏置,
β i=[β i1i2,...,β im] T-为第i个隐层单元的输出权重。
在经过训练的堆叠式极限学习机的数学模型中,激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi均为已知的,因此,通过上述逻辑运算,即可得到输出结果oj,其包括发育正常和发育迟缓2个大类。
预测结果输出单元304,用于以堆叠式极限学习机的逻辑运算的输出结果作为待测样本发育正常或者发育迟缓的依据,得到婴儿发育情况的预测结果。
此处需要说明的是,在同一时间段内,获取针对同一待测样本的运动数据的过程中,至少需要获取5-10组肢体运动数据才进行判断,其原因在于,若在获取待测样本的肢体运动数据时,获取的肢体运动数据的组数过少,则有可能由于待测样本由于偶然因素导致运动过于激烈或者运动过于缓慢而导致判断错误,但是,若在同一时间段内,至少获取5-10组肢体运动数据才进行判断时,则能够尽量减少由于偶然因素导致的判断错误。如若获取的肢体运动组数过多,则需要耗费较多的时间,对判断婴幼儿究竟是发育正常还是发育迟缓的意义也不大。若待测样本经过5-10组肢体运动数据判断的结果是发育迟缓,则可以再增加获取1-3倍的肢体运动组数获取数量,更多的肢体运动组数对确诊婴幼儿是否发育迟缓具有确诊的作用。
本申请实施例二提供的基于机器学习的婴儿发育情况预测装置在堆叠式极限学习及的数学模型已知的情况下,获取待测样本的单次肢体运动数据及对应的月龄数据,即可通过数据处理得到待测样本单次肢体运动的特征值,将该特征值代入至已知的堆叠式极限学习及的数学模型,进行逻辑运算,即可得到输出 结果,其中,输出结果包括发育迟缓和发育正常2个大类,其中,由于待测样本的月龄数据可以根据出生证明显示的出生日期获得,也就是说,在本申请提供的基于机器学习的婴儿发育情况预测方法中,待获取的数据仅为单次肢体运动数据,在这种情况下,可以通过运动传感器直接获得,因此,应用本申请提供的基于机器学习的婴儿发育情况预测方法能够较为便捷地对婴儿发育状况进行预测,此外,在本申请提供的基于机器学习的婴儿发育情况预测方法中,堆叠式极限学习及的数学模型在构建的过程中,可以涵盖所有月龄的婴儿,因此,即使待测样本的月龄较小,也可以通过该堆叠式极限学习及的数学模型得到相对准确的输出结果,因此,其能够为婴儿发育迟缓的早期干预提供可靠的依据。
实施例三
本申请实施例三提供的存储介质上存储有基于机器学习的婴儿发育情况预测程序,基于机器学习的婴儿发育情况预测程序被处理器执行时实现本申请提供的基于机器学习的婴儿发育情况预测方法的步骤。
本申请实施例三提供的存储介质在堆叠式极限学习及的数学模型已知的情况下,获取待测样本的单次肢体运动数据及对应的月龄数据,即可通过数据处理得到待测样本单次肢体运动的特征值,将该特征值代入至已知的堆叠式极限学习及的数学模型,进行逻辑运算,即可得到输出结果,其中,输出结果包括发育迟缓和发育正常2个大类,其中,由于待测样本的月龄数据可以根据出生证明显示的出生日期获得,也就是说,在本申请提供的基于机器学习的婴儿发育情况预测方法中,待获取的数据仅为单次肢体运动数据,在这种情况下,可以通过运动传感器直接获得,因此,应用本申请提供的基于机器学习的婴儿发育情况预测方法能够较为便捷地对婴儿发育状况进行预测,此外,在本申请提供的基于机器学习的婴儿发育情况预测方法中,堆叠式极限学习及的数学模型在构建的过程中,可以涵盖所有月龄的婴儿,因此,即使待测样本的月龄较小,也可以通过该堆叠式极限学习及的数学模型得到相对准确的输出结果,因此,其能够为婴儿发育迟缓的早期干预提供可靠的依据。
实施例四
本申请实施例四提供的电子设备包括运动传感器、处理器、存储器及存储在存储器上并可在处理器上 运行的基于机器学习的婴儿发育情况预测程序,其中,
运动传感器,用于获取待测样本的单次肢体运动数据;
基于机器学习的婴儿发育情况预测程序被处理器执行时实现本申请实施例一提供的基于机器学习的婴儿发育情况预测方法的步骤。
本申请实施例四提供的电子设备在堆叠式极限学习及的数学模型已知的情况下,获取待测样本的单次肢体运动数据及对应的月龄数据,即可通过数据处理得到待测样本单次肢体运动的特征值,将该特征值代入至已知的堆叠式极限学习及的数学模型,进行逻辑运算,即可得到输出结果,其中,输出结果包括发育迟缓和发育正常2个大类,其中,由于待测样本的月龄数据可以根据出生证明显示的出生日期获得,也就是说,在本申请提供的基于机器学习的婴儿发育情况预测方法中,待获取的数据仅为单次肢体运动数据,在这种情况下,可以通过运动传感器直接获得,因此,应用本申请提供的基于机器学习的婴儿发育情况预测方法能够较为便捷地对婴儿发育状况进行预测,此外,在本申请提供的基于机器学习的婴儿发育情况预测方法中,堆叠式极限学习及的数学模型在构建的过程中,可以涵盖所有月龄的婴儿,因此,即使待测样本的月龄较小,也可以通过该堆叠式极限学习及的数学模型得到相对准确的输出结果,因此,其能够为婴儿发育迟缓的早期干预提供可靠的依据。
尽管已描述了本申请的优选实施例,但本领域内的技术人员一旦得知了基本创造性概念,则可对这些实施例作出另外的变更和修改。所以,所附权利要求意欲解释为包括优选实施例以及落入本申请范围的所有变更和修改。
显然,本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的精神和范围。这样,倘若本申请的这些修改和变型属于本申请权利要求及其等同技术的范围之内,则本申请也意图包含这些改动和变型在内。

Claims (20)

  1. 一种基于机器学习的婴儿发育情况预测方法,包括以下步骤:
    获取待测样本的单次肢体运动数据及对应的月龄数据;
    针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
    将预定次数的与所述待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
    以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
  2. 根据权利要求1所述的基于机器学习的婴儿发育情况预测方法,所述堆叠式极限学习机的数学模型的构建方法包括以下步骤:
    获取训练样本的单次肢体运动数据及对应的月龄数据,其中,所述训练样本为已知发育状况的样本;
    针对所述训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述训练样本单次肢体运动的特征值;
    将所述与月龄数据相对应的针对所述训练样本单次肢体运动的特征值带入至逻辑运算公式:
    Figure PCTCN2019103148-appb-100001
    当所述训练样本为多个时,得到多个基于所述逻辑运算公式的表达式;
    其中,
    L为单隐层神经网络的节点数,
    g(x)为激活函数,
    w i=[W i1,w i2,...,w in] T为第i个隐层单元的输入权重,
    Bi为第i个隐层单元的偏置,
    β i=[β i1i2,...,β im] T为第i个隐层单元的输出权重,
    Oj为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,所述逻辑运算输出结果包括发育正常和发育迟缓2个大类;
    根据所述多个基于所述逻辑运算公式的表达式,确定所述激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi;
    再将确定后的激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi带回至所述逻辑运算公式,得到所述堆叠式极限学习机的数学模型。
  3. 根据权利要求1或2所述的基于机器学习的婴儿发育情况预测方法,所述单次肢体运动数据包括:
    单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R
    其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
  4. 根据权利要求1或2所述的基于机器学习的婴儿发育情况预测方法,针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
    所述单变量特征选择通过对所述单次肢体运动数据中的单一变量的统计度量方法,选取得到作为所述单次肢体运动的特征值;或
    所述递归特征选择通过对所述单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为所述单次肢体运动的特征值;或
    所述逐步特征选择通过逐一选取所述单次肢体运动数据中的单一变量、对所述单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
  5. 根据权利要求2所述的基于机器学习的婴儿发育情况预测方法,在所述逻辑运算输出结果中,
    根据所述堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
    根据所述堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
  6. 一种基于机器学习的婴儿发育情况预测装置,包括:
    数据获取单元,用于获取待测样本的单次肢体运动数据及对应的月龄数据;
    数据处理单元,用于针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
    运算单元,用于将预定次数的与所述待测样本相对应的处理后的肢体运动数据及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
    预测结果输出单元,用于以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
  7. 根据权利要求6所述的基于机器学习的婴儿发育情况预测装置,所述预测结果输出单元包括:
    单次肢体运动数据获取模块,用于获取训练样本的单次肢体运动数据及对应的月龄数据,其中,所述训练样本为已知发育状况的样本;
    第一数据处理模块,用于针对所述训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述训练样本单次肢体运动的特征值;
    第二数据处理模块,将所述与月龄数据相对应的针对所述训练样本单次肢体运动的特征值带入至逻辑运算公式:
    Figure PCTCN2019103148-appb-100002
    当所述训练样本为多个时,得到多个基于所述逻辑运算公式的表达式;
    其中,
    L为单隐层神经网络的节点数,
    g(x)为激活函数,
    w i=[W i1,w i2,...,w in] T为第i个隐层单元的输入权重,
    Bi为第i个隐层单元的偏置,
    β i=[β i1i2,...,β im] T为第i个隐层单元的输出权重,
    Oj为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,所述逻辑运算输出结果包括发育正常和发育迟缓2个大类;
    激活函数确定模块,用于根据所述多个基于所述逻辑运算公式的表达式,确定所述激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi;
    第三数据处理模块,用于再将确定后的激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi带回至所述逻辑运算公式,得到所述堆叠式极限学习机的数学模型。
  8. 根据权利要求6或7所述的基于机器学习的婴儿发育情况预测方法,所述单次肢体运动数据包括:
    单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R
    其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
  9. 根据权利要求6或7所述的基于机器学习的婴儿发育情况预测方法,针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
    所述单变量特征选择通过对所述单次肢体运动数据中的单一变量的统计度量方法,选取得到作为所述单次肢体运动的特征值;或
    所述递归特征选择通过对所述单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为所述单次肢体运动的特征值;或
    所述逐步特征选择通过逐一选取所述单次肢体运动数据中的单一变量、对所述单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
  10. 根据权利要求7所述的基于机器学习的婴儿发育情况预测方法,在所述逻辑运算输出结果中,
    根据所述堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
    根据所述堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
  11. 一种存储介质,所述存储介质上存储有基于机器学习的婴儿发育情况预测程序,所述基于机器学习的婴儿发育情况预测程序被处理器执行时实现基于机器学习的婴儿发育情况预测方法,包括:
    获取待测样本的单次肢体运动数据及对应的月龄数据;
    针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
    将预定次数的与所述待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
    以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
  12. 根据权利要求11所述的存储介质,所述堆叠式极限学习机的数学模型的构建方法包括以下步骤:
    获取训练样本的单次肢体运动数据及对应的月龄数据,其中,所述训练样本为已知发育状况的样本;
    针对所述训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述训练样本单次肢体运动的特征值;
    将所述与月龄数据相对应的针对所述训练样本单次肢体运动的特征值带入至逻辑运算公式:
    Figure PCTCN2019103148-appb-100003
    当所述训练样本为多个时,得到多个基于所述逻辑运算公式的表达式;
    其中,
    L为单隐层神经网络的节点数,
    g(x)为激活函数,
    w i=[W i1,w i2,...,w in] T为第i个隐层单元的输入权重,
    Bi为第i个隐层单元的偏置,
    β i=[β i1i2,...,β im] T为第i个隐层单元的输出权重,
    Oj为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,所述逻辑运算输出结果包括发育正常和发育迟缓2个大类;
    根据所述多个基于所述逻辑运算公式的表达式,确定所述激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi;
    再将确定后的激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi带回至所述逻辑运算公式,得到所述堆叠式极限学习机的数学模型。
  13. 根据权利要求11或12所述的存储介质,所述单次肢体运动数据包括:
    单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R
    其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
  14. 根据权利要求11或12所述的存储介质,针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
    所述单变量特征选择通过对所述单次肢体运动数据中的单一变量的统计度量方法,选取得到作为所述单次肢体运动的特征值;或
    所述递归特征选择通过对所述单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为 所述单次肢体运动的特征值;或
    所述逐步特征选择通过逐一选取所述单次肢体运动数据中的单一变量、对所述单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
  15. 根据权利要求12所述的存储介质,在所述逻辑运算输出结果中,
    根据所述堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
    根据所述堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
  16. 一种电子设备,包括运动传感器、处理器、存储器及存储在所述存储器上并可在所述处理器上运行的基于机器学习的婴儿发育情况预测程序,其中,
    所述运动传感器,用于获取待测样本的单次肢体运动数据;
    所述基于机器学习的婴儿发育情况预测程序被所述处理器执行时实现基于机器学习的婴儿发育情况预测方法,包括:
    获取待测样本的单次肢体运动数据及对应的月龄数据;
    针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值;
    将预定次数的与所述待测样本相对应的单次肢体运动的特征值及对应的月龄数据带入至堆叠式极限学习机的数学模型,进行逻辑运算;
    以所述堆叠式极限学习机的逻辑运算的输出结果作为所述待测样本发育正常或者发育迟缓的依据,得到所述婴儿发育情况的预测结果。
  17. 根据权利要求16所述的电子设备,所述堆叠式极限学习机的数学模型的构建方法包括以下步骤:
    获取训练样本的单次肢体运动数据及对应的月龄数据,其中,所述训练样本为已知发育状况的样本;
    针对所述训练样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述训练样本单次肢体运动的特征值;
    将所述与月龄数据相对应的针对所述训练样本单次肢体运动的特征值带入至逻辑运算公式:
    Figure PCTCN2019103148-appb-100004
    当所述训练样本为多个时,得到多个基于所述逻辑运算公式的表达式;
    其中,
    L为单隐层神经网络的节点数,
    g(x)为激活函数,
    w i=[W i1,w i2,...,w in] T为第i个隐层单元的输入权重,
    Bi为第i个隐层单元的偏置,
    β i=[β i1i2,...,β im] T为第i个隐层单元的输出权重,
    Oj为针对训练样本,应用堆叠式极限学习机进行分类后的逻辑运算输出结果,其中,所述逻辑运算输出结果包括发育正常和发育迟缓2个大类;
    根据所述多个基于所述逻辑运算公式的表达式,确定所述激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi;
    再将确定后的激活函数g(x)的表达式、所述第i个隐层单元的输入权重wi、所述第i个隐层单元的偏置bi以及所述第i个隐层单元的输出权重βi带回至所述逻辑运算公式,得到所述堆叠式极限学习机的数学模型。
  18. 根据权利要求16或17所述的电子设备,所述单次肢体运动数据包括:
    单次肢体运动的始时刻T i0、末时刻T i1、平均加速度a v、峰值加速度a m、左腿运动类型S L、右腿运动类型S R
    其中,单次运动持续时间t i是单次运动末时刻与单次运动始时刻T i0的差值T i1-T i0
  19. 根据权利要求16或17所述的电子设备,针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数据相对应的针对所述待测样本单次肢体运动的特征值的方法选自单变量特征选择、递归特征选择、逐步特征选择中的一种,所述针对所述待测样本的单次肢体运动数据进行数据处理,得到与月龄数 据相对应的针对所述待测样本单次肢体运动的特征值具体包括:
    所述单变量特征选择通过对所述单次肢体运动数据中的单一变量的统计度量方法,选取得到作为所述单次肢体运动的特征值;或
    所述递归特征选择通过对所述单次肢体运动数据中的各变量进行归一化数据处理,以得到的数据作为所述单次肢体运动的特征值;或
    所述逐步特征选择通过逐一选取所述单次肢体运动数据中的单一变量、对所述单次肢体运动数据中的各变量进行归一化数据处理得到的数据依次作为单次肢体运动的特征值。
  20. 根据权利要求17所述的电子设备,在所述逻辑运算输出结果中,
    根据所述堆叠式极限学习机的数学模型,对发育正常进行不同的分级;
    根据所述堆叠式极限学习机的数学模型,对发育迟缓进行不同的分级。
PCT/CN2019/103148 2019-04-26 2019-08-28 基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备 Ceased WO2020215566A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
SG11202008420UA SG11202008420UA (en) 2019-04-26 2019-08-28 Machine learning based method and device for infant development prediction, storage medium, and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910346364.8A CN110236558A (zh) 2019-04-26 2019-04-26 婴儿发育情况预测方法、装置、存储介质及电子设备
CN201910346364.8 2019-04-26

Publications (1)

Publication Number Publication Date
WO2020215566A1 true WO2020215566A1 (zh) 2020-10-29

Family

ID=67883539

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103148 Ceased WO2020215566A1 (zh) 2019-04-26 2019-08-28 基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备

Country Status (3)

Country Link
CN (1) CN110236558A (zh)
SG (1) SG11202008420UA (zh)
WO (1) WO2020215566A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114469072B (zh) * 2021-12-08 2023-08-08 四川大学华西第二医院 用摄像机自动预测婴儿心理发育的方法
CN115177239A (zh) * 2022-06-27 2022-10-14 永康国科康复工程技术有限公司 一种婴儿姿态运动发育检测方法、介质及设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150094544A1 (en) * 2013-09-12 2015-04-02 Sproutling, Inc. Infant monitoring system and methods
CN104765959A (zh) * 2015-03-30 2015-07-08 燕山大学 基于计算机视觉的婴儿全身运动评估方法
CN107851356A (zh) * 2015-04-05 2018-03-27 斯米拉布莱斯有限公司 确定婴幼儿的姿势和运动的可穿戴婴幼儿监测装置和系统
CN107924643A (zh) * 2015-04-05 2018-04-17 斯米拉布莱斯有限公司 远程地聚集和分析来自多个婴幼儿监测系统的测量数据
CN108351862A (zh) * 2015-08-11 2018-07-31 科格诺亚公司 利用人工智能和用户输入来确定发育进展的方法和装置
CN109171682A (zh) * 2012-08-25 2019-01-11 奥丽特婴儿保健公司 无线的婴儿健康监控器

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4189298B2 (ja) * 2003-10-31 2008-12-03 農工大ティー・エル・オー株式会社 乳児運動解析システム
US8961438B2 (en) * 2010-08-26 2015-02-24 The Regents Of The University Of California System for evaluating infant movement
US9232912B2 (en) * 2010-08-26 2016-01-12 The Regents Of The University Of California System for evaluating infant movement using gesture recognition
CN105310695B (zh) * 2015-11-03 2019-09-06 苏州景昱医疗器械有限公司 异动症评估设备
CN106388831B (zh) * 2016-11-04 2020-01-10 郑州航空工业管理学院 基于样本加权算法的检测摔倒动作的方法
CN109009143B (zh) * 2018-07-12 2021-01-29 杭州电子科技大学 一种通过人体步态预测心电信息的方法
CN109620244B (zh) * 2018-12-07 2021-07-30 吉林大学 基于条件生成对抗网络和svm的婴儿异常行为检测方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109171682A (zh) * 2012-08-25 2019-01-11 奥丽特婴儿保健公司 无线的婴儿健康监控器
US20150094544A1 (en) * 2013-09-12 2015-04-02 Sproutling, Inc. Infant monitoring system and methods
CN104765959A (zh) * 2015-03-30 2015-07-08 燕山大学 基于计算机视觉的婴儿全身运动评估方法
CN107851356A (zh) * 2015-04-05 2018-03-27 斯米拉布莱斯有限公司 确定婴幼儿的姿势和运动的可穿戴婴幼儿监测装置和系统
CN107924643A (zh) * 2015-04-05 2018-04-17 斯米拉布莱斯有限公司 远程地聚集和分析来自多个婴幼儿监测系统的测量数据
CN108351862A (zh) * 2015-08-11 2018-07-31 科格诺亚公司 利用人工智能和用户输入来确定发育进展的方法和装置

Also Published As

Publication number Publication date
SG11202008420UA (en) 2020-11-27
CN110236558A (zh) 2019-09-17

Similar Documents

Publication Publication Date Title
Reither et al. Should age-period-cohort studies return to the methodologies of the 1970s?
US11544281B2 (en) Query-oriented approximate query processing based on machine learning techniques
Duan et al. The predictive performance and stability of six species distribution models
WO2022068435A1 (zh) 对知识点掌握状态进行测量的方法
CN114840375B (zh) 一种半导体存储产品的老化性能测试方法及系统
CN110210244B (zh) 检测社交媒体用户隐私泄露的方法和系统
Abdel Majeed et al. Regression techniques employing feature selection to predict clinical outcomes in stroke
CN114487976A (zh) 一种mcm电子式互感器校验仪溯源不确定度评定方法及系统
WO2020215566A1 (zh) 基于机器学习的婴儿发育情况预测方法、装置、存储介质及电子设备
Kolbe et al. Using restricted factor analysis to select anchor items and detect differential item functioning
Dai et al. Minimum distance quantile regression for spatial autoregressive panel data models with fixed effects
Fellows et al. A new method for estimating HIV incidence from a single cross-sectional survey
CN120929316A (zh) 芯片测试方法、装置、计算机设备及存储介质
CN111401563A (zh) 机器学习模型更新方法和装置
Wolock et al. Assessing variable importance in survival analysis using machine learning
US12405875B2 (en) Managing integration of new data for data pipelines
CN114490412A (zh) 基于自减逆向云发生器的三维cad软件性能度量方法及装置
TWM656833U (zh) 用於機器學習模型的學習樣本資料篩選系統
CN118313974A (zh) 一种基于大数据的学前教育资源配置方法及系统
CN116415686A (zh) 温升模型校准方法、装置、超声成像设备及存储介质
Jang et al. Diagnostic evaluation of pharmacokinetic features of functional markers
CN109003680B (zh) 癫痫数据统计方法及装置
CN107562697A (zh) 认知诊断方法和系统
Steyn et al. A Bayesian model for repeated cross-sectional epidemic prevalence survey data
CN118690086B (zh) 一种学生个性化内容推荐方法、装置、设备及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19926238

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19926238

Country of ref document: EP

Kind code of ref document: A1