Method for constructing medium activity intensity evaluation model for preschool children

Abstract

The invention provides a method for constructing a medium activity intensity evaluation model for preschool children. The method comprises the following steps: acquiring personalized indexes, three-axis acceleration signals and one-dimensional vectors and signals of the three-axis acceleration signals of the preschool children; and extracting statistical characteristic indexes of the three-axis acceleration signals and the one-dimensional vector sum signals in a plurality of preset time windows, wherein the plurality of preset time windows are non-overlapping time windows. And constructing a feature set based on the extracted statistical feature indexes and personalized indexes. According to the relation between the feature set and the motion energy consumption, motion energy consumption data are obtained through calculation. According to the relation between the exercise energy consumption data and the first moderate intensity threshold value and the second moderate intensity threshold value, the daily average moderate intensity activity index is determined. The exercise energy consumption monitoring model is used for performing exercise energy consumption monitoring on the preschool children, the average daily moderate and strong activity index for daily assessment activities of the preschool children can be obtained, and the accuracy of exercise energy consumption monitoring is improved.

Description

Technical Field

[0001] This application relates to the medical field, specifically to the field of measuring physiological parameters of children, and more specifically to a method for constructing a moderate physical activity assessment model for preschool children and related devices. Background Art

[0002] Physical activity (PA) refers to any bodily movement produced by skeletal muscle contractions that results in increased energy expenditure and is one of the important dimensions supporting early childhood growth and development. Physical activity is mainly divided into light physical activity (LPA), moderate physical activity (MPA), and vigorous physical activity (VPA). Increasing evidence shows that lower levels of sedentary behavior (SB) and appropriate physical activity intensity can effectively improve children's physical health. Since the preschool period (which can represent the group of children aged 3 to 6 years) is a critical period for forming good physical activity habits, the physical activity habits at this stage will affect the physical activity behavior in the later developmental stage. Objectively monitoring physical activity is an important prerequisite for evaluating the effectiveness of physical activity interventions and quantifying the relationship between physical activity and health outcomes. And energy consumption, as the basis for determining physical activity levels, is an important dimension for promoting early childhood growth and development. Therefore, it is necessary to objectively monitor the exercise energy consumption of preschool children and related indicators that quantify their physical activity intensity.

[0003] Under normal circumstances, the corresponding physical activity intensity can be determined by monitoring the exercise energy consumption of preschool children. Exercise energy consumption can be expressed by absolute energy (energy expenditure, EE) or relative energy consumption (metabolicequivalents, METs, also known as metabolic equivalents). That is, the physical activity intensity level can be determined according to the values of EE or METs. Currently, on the premise of determining the physical activity intensity level according to the values of METs, the exercise energy consumption range for moderate-intensity activities is generally recognized as 3 - 6 Mets. However, the objects for data collection and testing in this interval are only applicable to adolescent and adult populations. Affected by growth and development, the relationship between the acceleration signal synchronously recorded during exercise by preschool children and METs is completely different from that of children, adolescents, or adults. In addition, the exercise energy consumption threshold ranges corresponding to moderate-intensity activities for these preschool children aged 3, 4, 5, and 6 years are not given separately.

[0004] Currently, the movement energy consumption of testers is mainly monitored by methods based on accelerometer sensors to obtain the estimated movement energy consumption of the testers. And these movement energy consumption monitoring methods are all constructed based on accelerometer activity counts (AC). AC is the data obtained after processing the acceleration signal through a specific filter, which will lose a lot of movement-related information, thus affecting the accuracy of the movement energy consumption monitoring method.

[0005] Moreover, the accelerometer sensor device can also record the raw acceleration signal data of the testers. Existing evidence shows that compared with the movement energy consumption estimation model constructed based on AC, the estimation accuracy of the movement energy consumption estimation model constructed based on the characteristics of the raw acceleration signal is significantly increased. For example, a method of determining the movement type based on the waveform of the acceleration and thereby measuring the energy consumed by the movement; or a method of inputting the acceleration signal into a convolutional neural network to extract acceleration features and fusing them with other features to predict the energy consumption; or a method of substituting the filtered acceleration signal into a movement intensity estimation formula to calculate the movement intensity and then calculate the calories consumed; or a method of constructing a movement energy consumption estimation model for foreign preschool children based on the raw acceleration signals of foreign preschool children. However, these movement energy consumption estimation models for preschool children constructed based on the characteristics of the raw acceleration signal are all constructed with foreign children as the test subjects, and their reproducibility and accuracy are not yet ideal.

[0006] Secondly, in the research on the method for estimating the movement energy consumption of preschool children constructed based on the characteristics of the raw acceleration signal, there is no fusion or only fusion of personalized indicators, which will also lead to inaccurate estimation of the movement energy consumption value of preschool children by the constructed model. And during the process of constructing the model, generally, the feature contribution degrees of each feature that affect the prediction performance of the trained model are not analyzed. As a result, although a relevant model is constructed, since the key features with the main influencing effects are not found, the constructed relevant model may not be composed of all key features, resulting in room for improvement in the estimation accuracy of the model.

[0007] Finally, in previous methods, usually only the energy consumption value is output, and no model for estimating the moderate activity intensity of preschool children is proposed. And a model that outputs indicators related to quantifying their physical activity intensity, that is, a model that outputs the daily moderate activity index, is more in line with the needs of daily assessment of whether the time for preschool children to carry out moderate activity intensity meets the requirements. Summary of the Invention

[0008] In view of this, the present application is committed to providing a method and device for constructing a moderate activity intensity assessment model for preschool children, so as to use this model to monitor the movement energy consumption of preschool children and determine the compliance of the daily physical activity intensity and the corresponding activity duration of preschool children.

[0009] In a first aspect, the present application provides a method for constructing a medium activity intensity assessment model for preschool children, the method comprising: Obtain personalized indicators of preschool children, triaxial acceleration signals, and the one-dimensional vector sum signal of the triaxial acceleration signals; Extract statistical feature indicators of the triaxial acceleration signals and the one-dimensional vector sum signal within a preset time window, wherein a plurality of preset time windows are non-overlapping time windows; Construct a feature set based on the extracted statistical feature indicators and the personalized indicators; Calculate exercise energy consumption data according to the relationship between the feature set and exercise energy consumption; Determine the daily medium-intensity activity index according to the relationship between the exercise energy consumption data and the first medium-intensity threshold and the second medium-intensity threshold.

[0010] The daily medium-intensity activity index of preschool children represents the statistical information of daily activities that meet the medium intensity. This statistical information may include the total duration of daily medium-intensity activities, or may include the integral corresponding to the time period under certain set conditions of the curve of the metabolic equivalent of task (METs) value changing with time (i.e., the sum of the areas under the METs value curve within a certain time period), etc.

[0011] In a possible implementation manner, the daily medium-intensity activity index , where t is the length of the preset time window, and N is the total number of preset time windows that meet the requirements; the initial value of N is 0, and when the exercise energy consumption data corresponding to the preset time window is greater than or equal to the first medium-intensity threshold and less than or equal to the second medium-intensity threshold, let N be incremented by 1 for update.

[0012] In a possible implementation manner, the triaxial acceleration signals include an x-axis acceleration signal, a y-axis acceleration signal, and a z-axis acceleration signal. The statistical feature indicators of the x-axis acceleration signal at least include the maximum value of a plurality of x-axis acceleration signals within the preset time window; the statistical feature indicators of the y-axis acceleration signal at least include the peak values of a plurality of y-axis acceleration signals within the preset time window; the statistical feature indicators of the z-axis acceleration signal at least include the minimum values of a plurality of z-axis acceleration signals; the statistical feature indicators of the one-dimensional vector sum signal at least include the average value and variance of a plurality of one-dimensional vector sum signals within the preset time window.

[0013] In a possible implementation manner, the relationship between the feature set and exercise energy consumption is established by a trained machine learning model.

[0014] In a possible implementation, the trained machine learning model includes an extremely randomized tree regression model.

[0015] In a possible implementation, the exercise energy consumption data includes at least one of the average energy consumption and the average metabolic equivalent.

[0016] In a possible implementation, the personalized metrics include at least one of gender, age, height, weight, and body mass index.

[0017] In a possible implementation, the first moderate-intensity threshold is 3.2 Mets, and the second moderate-intensity threshold is 5.3 Mets.

[0018] In a possible implementation, the first moderate-intensity threshold corresponding to preschool children aged 3 is 3.2 Mets, and the second moderate-intensity threshold corresponding to preschool children aged 3 is 4.4 Mets.

[0019] In a possible implementation, the first moderate-intensity threshold corresponding to preschool children aged 4 is 3.4 Mets, and the second moderate-intensity threshold corresponding to preschool children aged 4 is 4.6 Mets.

[0020] In a possible implementation, the first moderate-intensity threshold corresponding to preschool children aged 5 is 3.7 Mets, and the second moderate-intensity threshold corresponding to preschool children aged 5 is 4.8 Mets.

[0021] In a possible implementation, the first moderate-intensity threshold corresponding to preschool children aged 6 is 3.8 Mets, and the second moderate-intensity threshold corresponding to preschool children aged 6 is 5.1 Mets.

[0022] In a possible implementation, the age of preschool children is calculated according to the full years corresponding to the actual date of birth.

[0023] In a second aspect, the present application provides a device for constructing a moderate activity intensity evaluation model for preschool children, the device includes: A data acquisition unit, configured to obtain personalized metrics of preschool children, triaxial acceleration signals, and the one-dimensional vector sum signal of the triaxial acceleration signals; A feature extraction unit, configured to extract statistical feature metrics of the triaxial acceleration signals and the one-dimensional vector sum signal within a preset time window, where multiple preset time windows are non-overlapping time windows; A feature construction unit, configured to construct a feature set based on the extracted statistical feature metrics and the personalized metrics; A prediction unit, configured to calculate exercise energy consumption data according to the relationship between the feature set and exercise energy consumption; A judgment unit, configured to determine a daily average medium-intensity activity index according to the relationship between the exercise energy consumption data and a first medium-intensity threshold and a second medium-intensity threshold.

[0024] In a third aspect, the present application provides an electronic device, which includes: a memory and a processor; The memory is used to store relevant program codes; The processor is configured to call the program codes to execute the method for constructing the medium activity intensity evaluation model of preschool children according to any one of the implementation manners of the first aspect above.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method for constructing the medium activity intensity evaluation model of preschool children according to any one of the implementation manners of the first aspect above.

[0026] In a fifth aspect, the present application provides a computer program product, which includes computer programs / instructions, and when the computer programs / instructions are executed by a processor, the method for constructing the medium activity intensity evaluation model of preschool children according to any one of the implementation manners of the first aspect above is implemented.

[0027] The inventive point of the present invention lies in: 1. A medium activity intensity threshold range suitable for the characteristics of Chinese preschool children is given; further, medium activity intensity threshold ranges suitable for 3, 4, 5, and 6-year-old children are respectively given; 2. Personalized indicators that conform to the development characteristics of Chinese preschool children are integrated; 3. For the constructed model, important features are sorted to find the key features that play a major influencing role.

[0028] In the above implementation manner of the present application, in order to construct a medium activity intensity evaluation model for preschool children, first, personalized indicators, triaxial acceleration signals, one-dimensional vector sum signals of triaxial acceleration signals, and corresponding exercise energy consumption values of preschool children are obtained. Statistical feature indicators of triaxial acceleration signals and one-dimensional vector sum signals within a preset time window are extracted, where multiple preset time windows are non-overlapping time windows. A feature set is constructed based on the extracted statistical feature indicators and personalized indicators. According to the relationship between the feature set and exercise energy consumption, exercise energy consumption data is calculated. According to the relationship between the exercise energy consumption data and a first medium-intensity threshold and a second medium-intensity threshold, a daily average medium-intensity activity index is determined.

[0029] Among them, the personalized indicators of preschool children participating in model construction reflect certain development characteristics of Chinese preschool children. These seemingly conventional indicators actually have a certain impact on the accuracy of the preschool children's motion assessment model. Among the personalized indicators, the height indicator has the greatest impact on the model prediction accuracy. Moreover, during the model construction process, the feature contribution degrees of each feature that affect the prediction performance of the trained model are analyzed to find out the key features with the main influencing effects, so that the constructed relevant model is composed of all key features. These two aspects also help to improve the accuracy of motion energy consumption monitoring.

[0030] Moreover, the current globally applicable scope is for statistical analysis of foreign teenagers, while the data source of the threshold range we have counted is Chinese preschool children. Therefore, the obtained threshold range conforms to the development characteristics of Chinese preschool children. At the same time, the medium activity intensity threshold range we proposed is more accurate, which helps to improve the accuracy of the relevant assessment results of medium-intensity activities for preschool children. At the same time, the construction method of this model also helps to provide evaluation means and data support basis for the precise management of preschool children's sports, and has strong practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments provided in the present application. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings.

[0032] Figure 1 It is a flowchart of a method for constructing a medium activity intensity assessment model for preschool children provided by an embodiment of the present application.

[0033] Figure 2 It is a flowchart of a method for monitoring the motion energy consumption of preschool children provided by an embodiment of the present application.

[0034] Figure 3 It is a schematic diagram of a device for constructing a medium activity intensity assessment model for preschool children provided by an embodiment of the present application.

[0035] Figure 4 It is a schematic diagram of a device for monitoring the motion energy consumption of preschool children provided by an embodiment of the present application.

[0036] Figure 5 It is a schematic diagram of an electronic device provided by an embodiment of the present application.

[0037] Figure 6A box plot of the statistical range of the average Mets value obtained by averaging the measured Mets values synchronously recorded during moderate-intensity activities for a sample aged 4 to 6 years within a time window of a certain length. Among them, (a) is the box plot of the statistical range of the corresponding measured average mets value synchronously recorded during moderate-intensity activities for a sample all aged 4 years, (b) is the box plot of the statistical range of the corresponding measured average mets value synchronously recorded during moderate-intensity activities for a sample all aged 5 years, and (c) is the box plot of the statistical range of the corresponding measured average mets value synchronously recorded during moderate-intensity activities for a sample all aged 6 years. Detailed implementation manner

[0038] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. The described embodiments are only exemplary implementation manners of the present application, not all implementation manners. Those skilled in the art can obtain other embodiments without creative work in combination with the embodiments of the present application, and these embodiments are also within the protection scope of the present application.

[0039] Since the preschool period (which can represent the group aged 3 to 6 years) is a key period for forming good physical activity habits, the physical activity habits at this stage may affect the physical activity behavior in the later development stage. In order to promote the growth and development of children in the early stage, it is necessary to monitor the physical activity intensity of preschool children.

[0040] Under normal circumstances, the corresponding physical activity intensity can be determined by monitoring the exercise energy consumption of preschool children. Currently, the exercise energy consumption monitoring method based on the accelerometer sensor is mainly used to monitor the test personnel to estimate the exercise energy consumption of the test personnel. However, this exercise energy consumption monitoring method is mainly applicable to adolescents and adult groups, and the physical activity habits of preschool children are quite different from those of adolescents (which can represent the group aged 11 to 17 years) and adults. Therefore, the accuracy of using the existing exercise energy consumption monitoring model to monitor the energy consumption of preschool children is relatively low.

[0041] In addition, the current exercise energy consumption monitoring methods are all constructed based on the accelerometer activity counts (AC), and AC is the data obtained after processing the acceleration signal through a specific filter, which will lose a lot of information related to movement, thus affecting the accuracy of the exercise energy consumption monitoring method.

[0042] Currently, there are some studies on constructing an exercise energy consumption estimation model for preschool children based on the characteristics of the original acceleration signal. However, these exercise energy consumption estimation models are all constructed with foreign children as the test subjects, and the reproducibility and accuracy are not ideal.

[0043] Based on this, the embodiments of the present application provide a method for constructing a medium activity intensity evaluation model for preschool children, so as to improve the accuracy of exercise energy consumption monitoring. Specifically, when implemented, personalized indicators of preschool children, triaxial acceleration signals, and the one-dimensional vector sum signal of the triaxial acceleration signals are obtained. Statistical feature indicators of the triaxial acceleration signals and the one-dimensional vector sum signal within a preset time window are extracted, where multiple preset time windows are non-overlapping time windows. A feature set is constructed based on the extracted statistical feature indicators and personalized indicators. According to the relationship between the feature set and exercise energy consumption, exercise energy consumption data is calculated. According to the relationship between the exercise energy consumption data and the first medium intensity threshold and the second medium intensity threshold, the daily average medium-intensity activity index is determined. Subsequently, the exercise energy consumption monitoring model can be used to monitor the exercise energy consumption of preschool children. Since the exercise energy consumption monitoring model is constructed based on the exercise energy consumption data of preschool children, using the exercise energy consumption monitoring model to monitor the exercise energy consumption of preschool children can obtain the daily average medium-intensity activity index for the daily evaluation activities of preschool children, improving the accuracy of exercise energy consumption monitoring.

[0044] To facilitate understanding of the technical method provided by the embodiments of the present application, the following will be specifically introduced in conjunction with the accompanying drawings in the embodiments.

[0045] See Figure 1 as shown in Figure 1 is a flowchart of a method for constructing a medium activity intensity evaluation model for preschool children provided by the embodiments of the present application.

[0046] Optionally, this method can be executed by a processing device. The processing device can be an electronic device or other devices, and the embodiments of the present application do not limit this.

[0047] This method may include the following steps: S101: Obtain personalized indicators of preschool children, triaxial acceleration signals, and the one-dimensional vector sum signal of the triaxial acceleration signals.

[0048] Among them, the triaxial acceleration signal can be obtained by a preschool child wearing an acceleration measuring instrument, and the period of wearing the acceleration measuring instrument is at least 1 day. The triaxial acceleration signal may include an x-axis acceleration signal, a y-axis acceleration signal, and a z-axis acceleration signal. The personalized indicators may include at least one of gender, age, height, weight, and body mass index (BMI).

[0049] Optionally, the one-dimensional vector sum signal of the three-axis acceleration signal can be obtained by calculating the one-dimensional vector sum of the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal. For example, the sum of the squares of the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal can be calculated, and then the arithmetic square root of the sum of squares can be calculated as the one-dimensional vector sum signal.

[0050] S102: Extract the statistical feature indicators of the three-axis acceleration signal and the one-dimensional vector sum signal within a preset time window.

[0051] That is, the embodiments of the present application can use a preset time window to divide the three-axis acceleration signal and the one-dimensional vector sum signal, so as to obtain the three-axis acceleration signal and the one-dimensional vector sum signal corresponding to multiple preset time windows. Among them, the multiple preset time windows are non-overlapping time windows. That is, the times corresponding to the multiple preset time windows do not overlap with each other.

[0052] In a possible implementation manner, for each preset time window, there may be multiple three-axis acceleration signals and multiple one-dimensional vector sum signals. Based on this, the statistical feature indicators of the extracted x-axis acceleration signal at least include the maximum value of the multiple x-axis acceleration signals within the preset time window. The statistical feature indicators of the y-axis acceleration signal at least include the peak-to-peak value of the multiple y-axis acceleration signals within the preset time window. The statistical feature indicators of the z-axis acceleration signal at least include the minimum value of the multiple z-axis acceleration signals. The statistical feature indicators of the one-dimensional vector sum signal at least include the average value and variance of the multiple one-dimensional vector sum signals within the preset time window.

[0053] S103: Construct a feature set based on the extracted statistical feature indicators and personalized indicators.

[0054] After determining the statistical feature indicators, the statistical feature indicators and the personalized indicators of preschool children can be used to construct a feature set.

[0055] S104: Calculate the exercise energy consumption data according to the relationship between the feature set and the exercise energy consumption.

[0056] In a possible implementation manner, the relationship between the feature set and the exercise energy consumption is established by a trained machine learning model. Optionally, the trained machine learning model may include an extremely randomized tree regression model. Among them, the training process of the machine learning model can be seen in the subsequent embodiments and will not be elaborated here.

[0057] In a possible implementation, the exercise energy consumption data may include energy expenditure (EE) or metabolic equivalents (METs). When the exercise energy consumption data is different, the number of indicators included in the feature set corresponding to the exercise energy consumption data is also different. This process will be introduced in subsequent embodiments.

[0058] S105: Determine the daily average moderate-intensity activity index according to the relationship between the exercise energy consumption data and the first moderate-intensity threshold and the second moderate-intensity threshold.

[0059] Among them, the first moderate-intensity threshold is less than the second moderate-intensity threshold.

[0060] In a possible implementation, the level of physical activity intensity can be determined according to the value of metabolic equivalents (METs). When preschool children are at a moderate activity intensity, it is more conducive to promoting growth and development. Therefore, the daily average moderate-intensity activity index of preschool children can be determined to monitor the exercise energy consumption of preschool children.

[0061] Among them, the METs range corresponding to the moderate activity intensity can be determined by the first moderate-intensity threshold and the second moderate-intensity threshold. Optionally, the first moderate-intensity threshold and the second moderate-intensity threshold can be determined in advance through statistical experiments. Therefore, after calculating the exercise energy consumption data of preschool children, the daily average moderate-intensity activity index can be determined according to the relationship between the exercise energy consumption data and the first moderate-intensity threshold and the second moderate-intensity threshold.

[0062] In a possible implementation, the first moderate-intensity threshold can be 3.2 Mets, and the second moderate-intensity threshold can be 5.3 Mets. That is, the range corresponding to the moderate activity intensity of preschool children is [3.2, 5.3]. Among them, the threshold ranges corresponding to the moderate activity intensity of 4-year-old, 5-year-old, and 6-year-old preschool children are obtained by box plot statistics of the corresponding data. Among them, the box plot of the statistical range of the measured average mets value corresponding to the samples aged 4 to 6 years old during moderate-intensity activities is as Figure 6 shown. Therefore, according to the digital pattern corresponding to the moderate-intensity activity threshold of preschool children aged 4 - 6 years old, the moderate-intensity activity threshold of 3-year-old preschool children is estimated to be [3.2, 4.4] Mets; Among them, the ranges corresponding to the moderate activity intensity of preschool children of different ages are also different, which are specifically described as follows. The first moderate-intensity threshold corresponding to 3-year-old preschool children is 3.2 Mets, and the second moderate-intensity threshold corresponding to 3-year-old preschool children is 4.4 Mets. That is, the range of moderate activity intensity corresponding to 3-year-old preschool children is [3.2, 4.4], and the overall standard deviation is plus or minus 0.2. The value range of the standard deviation is [0, 0.3]. When specifically implemented, the value of the standard deviation is affected by factors such as the height value range of the measurement sample.

[0063] The first moderate-intensity threshold corresponding to 4-year-old preschool children is 3.4 Mets, and the second moderate-intensity threshold corresponding to 4-year-old preschool children is 4.6 Mets. That is, the range of moderate activity intensity corresponding to 4-year-old preschool children is [3.4, 4.6], and the overall standard deviation is plus or minus 0.2. The value range of the standard deviation is [0, 0.3]. When specifically implemented, the value of the standard deviation is affected by factors such as the height value range of the measurement sample.

[0064] The first moderate-intensity threshold corresponding to 5-year-old preschool children is 3.7 Mets, and the second moderate-intensity threshold corresponding to 5-year-old preschool children is 4.8 Mets. That is, the range of moderate activity intensity corresponding to 5-year-old preschool children is [3.7, 4.8], and the overall standard deviation is plus or minus 0.2. The value range of the standard deviation is [0, 0.4]. When specifically implemented, the value of the standard deviation is affected by factors such as the height value range of the measurement sample.

[0065] The first moderate-intensity threshold corresponding to 6-year-old preschool children is 3.8 Mets, and the second moderate-intensity threshold corresponding to 6-year-old preschool children is 5.1 Mets. That is, the range of moderate activity intensity corresponding to 6-year-old preschool children is [3.8, 5.1]. The overall standard deviation is plus or minus 0.2. The value range of the standard deviation is [0, 0.6]. When specifically implemented, the value of the standard deviation is affected by factors such as the height value range of the measurement sample. It should be noted that the age of preschool children is calculated according to the full years corresponding to the actual date of birth.

[0066] When specifically implemented, the daily average moderate-to-high activity index can be determined in the following way . Among them, t is the length of the preset time window, and N is the total number of preset time windows that meet the requirements. Set the initial value of N to 0. For the exercise energy consumption data determined for each preset time window, when the exercise energy consumption data corresponding to the preset time window is greater than or equal to the first moderate-intensity threshold and less than or equal to the second moderate-intensity threshold, it indicates that the current preset time window meets the requirements, and let N be incremented by 1 for update. Perform the above operations on multiple preset time windows, and then the total number N of preset time windows that meet the requirements can be finally determined. Multiply the length t of each preset time window by the total number N to obtain the daily average moderate-to-high activity index of preschool children.

[0067] Through the above steps, the execution process of the moderate activity intensity assessment model is determined, that is, the moderate activity intensity assessment model for preschool children is constructed.

[0068] To better understand the technical solutions in the above embodiments, the relationship between the sample features and the exercise energy consumption in step S104, that is, the trained machine learning model, will be introduced in detail below.

[0069] To obtain the trained machine learning model, it is first necessary to obtain the sample data for training, and use the sample data to train the machine learning algorithm to obtain the trained machine learning model. Since the moderate activity intensity assessment model is to accurately monitor the exercise energy consumption of preschool children and calculate the exercise energy consumption data, it is necessary to collect relevant data of preschool children. This method mainly includes the following steps A1 - A4: A1: Obtain the personalized indicators of preschool children, the triaxial acceleration signal, and the exercise energy consumption label value corresponding to the triaxial acceleration signal.

[0070] Among them, preschool children mainly include children between 3 and 6 years old, and the specific age range for data collection can be determined according to actual needs. The collected samples can cover as many age groups as possible to improve the diversity of the samples and make the trained machine learning model more accurate.

[0071] The personalized indicators of preschool children include: gender, age, height, weight, and body mass index, etc. The triaxial acceleration signal can be used to characterize the exercise state of preschool children. For example, preschool children can wear an acceleration meter to measure the triaxial acceleration signal in different exercise states. And preschool children wear a gas metabolism analyzer. While measuring the triaxial acceleration signal with the acceleration meter, use the gas metabolism analyzer to measure the exercise energy consumption in different exercise states as the exercise energy consumption label value, that is, the sample true value, so that each triaxial acceleration signal corresponds to an exercise energy consumption label value. Among them, the gas metabolism analyzer measures through the exhalation analysis method, and calculates the exercise energy consumption label value by measuring oxygen uptake, carbon dioxide output, etc.

[0072] Optionally, the exercise energy consumption label value can include EE or METs. When the obtained exercise energy consumption label value includes energy consumption, the trained machine learning model can output the energy consumption of preschool children. When the obtained exercise energy consumption label value includes metabolic equivalent, the trained machine learning model can monitor and output the metabolic equivalent of preschool children. That is, different machine learning models can be trained with different samples (exercise energy consumption label values) to monitor different exercise energy consumption data.

[0073] Among them, the exercise states of preschool children can include a variety of different states to improve the diversity of the samples. For example, they can include various exercise states such as sitting still, walking slowly, walking fast, running, and jumping. By collecting the triaxial acceleration signals and exercise energy consumption label values of preschool children in different exercise states, they are used as independent variables and target variables when training a machine learning model.

[0074] A2: Determine sample features based on personalized metrics and triaxial acceleration signals.

[0075] After obtaining the personalized metrics and triaxial acceleration signals of preschool children, the sample features for training can be determined.

[0076] Since the relevant data of the obtained preschool children may have incorrect records or missing data, in a possible implementation, the collected data can be preprocessed first to obtain preprocessed personalized metrics and preprocessed triaxial acceleration signals. Then, based on the preprocessed personalized metrics and preprocessed triaxial acceleration signals, sample features are determined.

[0077] In specific implementation, imputation can be performed on the missing values in the data. That is, when there are missing values in the personalized metrics or triaxial acceleration signals of the obtained preschool children, the missing values can be imputed to avoid affecting the subsequent data processing process. For example, the missing values in categorical data (such as boys or girls) are imputed with the value "NaN", and the missing values in numerical data (such as age) are imputed with the mean value.

[0078] Since most models cannot directly process categorical data, one-hot encoding processing needs to be performed on categorical data to convert categorical data into numerical data. Among them, one-hot encoding is also called one-hot encoding. Its processing method is to use an N-bit status register to encode N states. Each state has an independent register bit, and the value is 0 or 1. And at any time, only one bit is valid. For example, for the categorical data "boys" and "girls" for encoding, since there are only two types in this category, so take N = 2, then "01" and "10" can be used to represent "boys" and "girls".

[0079] For continuous data, in order to simplify the data processing process, the continuous data can be normalized. Or, the normalized continuous data can also be converted into data with an approximate normal distribution to improve the data processing efficiency.

[0080] In a possible implementation, the personalized metrics and triaxial acceleration signals can be fused as sample features.

[0081] In a possible implementation, the three-axis acceleration signal includes the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal. To improve the diversity and richness of the samples and enhance the ability of the training model, a one-dimensional vector sum signal can be determined based on the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal. That is, calculate the sum of the squares of the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal, and then calculate the arithmetic square root of the sum of squares as the one-dimensional vector sum signal. After obtaining the one-dimensional vector sum signal, since the three-axis acceleration signal is the three-axis acceleration signal collected from preschool children over a period of time, the statistical characteristic indexes of the x-axis acceleration signal, the y-axis acceleration signal, the z-axis acceleration signal, and the one-dimensional vector sum signal can be calculated respectively based on multiple x-axis acceleration signals, multiple y-axis acceleration signals, multiple z-axis acceleration signals, and multiple one-dimensional vector sum signals over a period of time.

[0082] Among them, the statistical characteristic index can represent mathematical statistical characteristics. For example, it can include the maximum value, the minimum value, the mean value, the standard deviation, the variance, the percentile, the coefficient of variation, and so on.

[0083] In a possible implementation, the statistical characteristic index of the x-axis acceleration signal includes at least the maximum value of multiple x-axis acceleration signals. The statistical characteristic index of the y-axis acceleration signal includes at least the peak-to-peak value of multiple y-axis acceleration signals, where the peak-to-peak value represents the difference between the maximum value and the minimum value. The statistical characteristic index of the z-axis acceleration signal includes at least the minimum value of multiple z-axis acceleration signals. The statistical characteristic index of the one-dimensional vector sum signal includes at least the average value and the variance of multiple one-dimensional vector sum signals.

[0084] Then, based on the statistical characteristic index and the personalized index, the sample characteristics are determined. For example, the statistical characteristic index and the personalized index can be fused to determine the sample characteristics.

[0085] It should be noted that the specific type and number of the statistical characteristic indexes are not limited in the embodiments of the present application and can be set according to actual needs. In a possible application scenario, for the x-axis acceleration signal, the y-axis acceleration signal, the z-axis acceleration signal, and the one-dimensional vector sum signal, 91 statistical characteristic indexes including the maximum value, the minimum value, the mean value, the standard deviation, the variance, the percentile, the coefficient of variation, etc. can be determined, and then combined with 5 personalized indexes of gender, age, height, weight, and body mass index, the statistical characteristic indexes and the personalized indexes are fused to obtain 96 indexes to determine the sample characteristics.

[0086] Although including more metrics in the sample features can improve the diversity of the samples, it will also increase the complexity of data processing and the difficulty of training a machine learning model, which may lead to overfitting during training and affect the accuracy of training the machine learning model. Based on this, in a possible implementation, a feature selection algorithm can be used to select a preset number of metrics from the above-determined statistical feature metrics and personalized metrics, and then determine the sample features based on the selected preset number of metrics. That is, some of the statistical feature metrics and personalized metrics can be selected as the sample features, which can not only ensure the diversity of the samples but also appropriately reduce the data volume (the number of metrics) of the samples, reduce the complexity of data processing, and improve the speed of model training.

[0087] For example, the feature selection algorithm can be the minimum redundancy maximum relevance method. The main idea of this method is to select features from the dataset that are highly correlated with the target variable but have low redundancy with each other. That is, select metrics that are highly correlated with the exercise energy consumption from the statistical feature metrics and personalized metrics and have low redundancy among the selected metrics.

[0088] Optionally, the preset number of selected metrics can be determined through multiple experiments. That is, multiple different numbers can be determined in advance. For each number, after selecting multiple metrics, a type of sample feature can be obtained. After training a machine learning model using the sample features, determine which number of sample features results in a machine learning model with higher accuracy, and then it can be determined as the preset number. Subsequently, when using the machine learning model to monitor preschool children to determine the exercise energy consumption data, the preset number of metrics can be selected to determine the sample features, making the exercise energy consumption data predicted by the machine learning model more accurate.

[0089] In a possible implementation, the three-axis acceleration signals of the preschool children collected may correspond to the motion states of multiple time periods, and the duration of each time period may be different. To unify the format of the sample features and facilitate the processing of the sample features, the obtained three-axis acceleration signals can be divided according to a preset time window to obtain the first acceleration signals corresponding to multiple preset time windows. That is, each preset time window corresponds to the divided first acceleration signal. For each preset time window, based on the first acceleration signal and personalized metrics of this preset time window, the first sample feature corresponding to this preset time window can be determined. Then, based on the first sample features corresponding to each preset time window, a machine learning model can be trained.

[0090] Among them, the process of determining the first sample feature can refer to the implementation process of determining the sample feature in the above embodiments. For any preset time window, based on multiple x-axis acceleration signals, multiple y-axis acceleration signals, multiple z-axis acceleration signals, and corresponding multiple one-dimensional vector sum signals within the preset time window, calculate the statistical feature indexes of the x-axis acceleration signals, the statistical feature indexes of the y-axis acceleration signals, the statistical feature indexes of the z-axis acceleration signals, and the statistical feature indexes of the one-dimensional vector sum signals respectively. Then, based on the statistical feature indexes and personalized indexes corresponding to the preset time window, determine the first sample feature corresponding to the preset time window. For example, the first sample feature can be determined by fusing each statistical feature index and personalized index.

[0091] In a possible implementation manner, a feature selection algorithm can be used to select a preset number of indexes from the determined statistical feature indexes and personalized indexes, and then determine the first sample feature based on the selected preset number of indexes. That is, some indexes in the statistical feature indexes and personalized indexes can be selected as the first sample feature.

[0092] Based on this, after determining the first sample feature corresponding to each preset time window, since the multiple three-axis acceleration signals within the preset time window correspond to multiple motion energy consumption label values, at this time, it can be determined that the sample true value corresponding to the first sample feature is the average value of the multiple motion energy consumption label values within the preset time window. That is, for each preset time window, obtain the multiple motion energy consumption label values within the preset time window, and calculate the average motion energy consumption label value of the multiple motion energy consumption label values as the sample true value corresponding to the first sample feature.

[0093] A3: Input the sample feature and the motion energy consumption label value corresponding to the sample feature into a machine learning algorithm to obtain a motion energy consumption prediction value.

[0094] After determining the sample feature for training, the sample feature can be input into a machine learning algorithm, and the machine learning algorithm outputs a motion energy consumption prediction value.

[0095] In a possible implementation manner, the machine learning algorithm can include an extremely randomized tree regression algorithm, a random forest regression algorithm, a light gradient boosting machine (LGBM) algorithm, a CatBoost algorithm, or a linear regression algorithm.

[0096] A4: Train the machine learning algorithm based on the motion energy consumption prediction value and the motion energy consumption label value until the training cut-off condition is met, and obtain a trained machine learning model.

[0097] After obtaining the predicted value of exercise energy consumption, since there is a corresponding exercise energy consumption label value, i.e., the true value of the sample, for the three-axis acceleration signal in the sample features, and the predicted value of exercise energy consumption represents the predicted value of the machine learning algorithm, the machine learning algorithm can be trained based on the predicted value of exercise energy consumption and the exercise energy consumption label value, and the iterative training is performed until the training cut-off condition is met. Among them, the goal of training is to reduce the error between the predicted value of exercise energy consumption and the exercise energy consumption label value.

[0098] In specific implementation, the estimation error can be determined based on the predicted value of exercise energy consumption and the exercise energy consumption label value, and this estimation error is used to represent the error between the predicted value of exercise energy consumption and the exercise energy consumption label value. When the estimation error is larger, it indicates that the error between the predicted value of exercise energy consumption and the exercise energy consumption label value is also larger, and thus the accuracy of the machine learning algorithm is lower. When the estimation error is greater than or equal to the preset value, it indicates that the error between the predicted value of exercise energy consumption and the exercise energy consumption label value is large and does not meet the requirements. It is necessary to readjust the parameters of the machine learning algorithm and re-execute step A3 and step A4, that is, re-execute the process of inputting the sample features into the machine learning algorithm with adjusted parameters and subsequent determination of the estimation error until the training cut-off condition is met.

[0099] In a possible implementation manner, the training cut-off condition can be that the estimation error is less than the preset value, or the number of iterative training reaches the preset number (the number of times of adjusting the parameters of the machine learning algorithm reaches the preset number). At this time, the training process can be stopped to obtain the trained machine learning model.

[0100] In a possible implementation manner, the machine learning model can be trained using the k-fold cross-validation method. Among them, the main principle of the k-fold cross-validation method is to randomly divide the sample data into k parts, where k - 1 parts are used as the training set and the remaining 1 part is used as the test set. When training and testing, the training set and the test set corresponding to the training set are selected in turn. For example, the ten-fold cross-validation method is to divide the sample data into 10 sub-samples, a single sub-sample is used as the test set to verify the test model, and the remaining 9 sub-samples are used for training. The cross-validation is repeated 10 times, and each sub-sample is used as the test set to verify once. That is, the sample features can be divided into a training set and a test set, the training set is used to train the machine learning model, and the test set is used to test the machine learning model. Then, the mean value of the results obtained from the 10 tests is calculated as the final test result.

[0101] When testing a machine learning model, the test set is input into the machine learning model to obtain the test value of the exercise energy consumption. Based on the test value of the exercise energy consumption and the label value of the exercise energy consumption corresponding to the test set, the test error is determined. Optionally, the root mean squared error (RMSE) between the test value of the exercise energy consumption and the label value of the exercise energy consumption can be calculated as the test error. For example, when training using the ten-fold cross-validation method, after the test value of the exercise energy consumption is output in each test process, the difference between the test value of the exercise energy consumption and the label value of the exercise energy consumption can be calculated. Then, the mean of the squares of the ten differences is calculated, and the arithmetic square root of the mean of the squares is calculated as the test error. The ten-fold cross-validation method can obtain 10 test errors, and the average of these 10 test errors is calculated as the final test result of the machine learning model.

[0102] According to the above embodiments, in order to enable the trained machine learning model to more accurately monitor the exercise energy consumption of preschool children, multiple machine learning algorithms can be used to train the machine learning model. For example, the multiple machine learning algorithms can include the extremely randomized tree regression algorithm, the random forest regression algorithm, LGBM, the CatBoost algorithm, or the linear regression algorithm. By testing the accuracy of multiple machine learning models, the most accurate machine learning model is selected.

[0103] In specific implementation, the sample features and the label values of the exercise energy consumption corresponding to the sample features can be fused to obtain a data set, and the data set is divided into a training set and a test set. Among them, the training set includes multiple sample features and the label values of the exercise energy consumption corresponding to each sample feature, and the test set also includes multiple sample features and the label values of the exercise energy consumption corresponding to each sample feature. For each machine learning algorithm, the training set is used to train the machine learning algorithm to obtain a trained machine learning model. Among them, the specific training process can refer to the above embodiments and will not be elaborated here. After obtaining the trained machine learning model, the test set is input into the machine learning model to obtain the test value of the exercise energy consumption. Based on the test value of the exercise energy consumption and the label value of the exercise energy consumption corresponding to the test set, the test error is determined. Since each machine learning model corresponds to a test error as the accuracy test result of the machine learning model, the minimum test error can be determined based on the test errors corresponding to multiple machine learning models respectively, and the machine learning model corresponding to the minimum test error is determined as the preferred machine learning model. That is, the machine learning model corresponding to the minimum test error is used as the machine learning model for subsequent monitoring of preschool children.

[0104] Among them, the method for determining the test error can be to calculate the root mean square error between the measured value of the exercise energy consumption and the labeled value of the exercise energy consumption, or to calculate the mean absolute error between the measured value of the exercise energy consumption and the labeled value of the exercise energy consumption, etc. The embodiments of the present application do not limit this.

[0105] In the embodiments of the present application, training and testing are respectively performed on the above-mentioned 5 machine learning algorithms. It can be known that among the 5 machine learning algorithms, the machine learning model trained based on the extremely randomized tree regression algorithm has the highest test accuracy. Therefore, when subsequently monitoring the exercise energy consumption of preschool children, a machine learning model trained based on the extremely randomized tree regression algorithm can be selected for monitoring.

[0106] Based on the embodiments of determining sample features introduced in the above step A2, it can be known that the obtained three-axis acceleration signal can be divided according to a preset time window to obtain a first acceleration signal corresponding to each of the multiple preset time windows. For each preset time window, based on the first acceleration signal and the personalized index of this preset time window, the first sample feature corresponding to this preset time window can be determined. And the multiple exercise energy consumption labeled values within this preset time window are obtained, and the average exercise energy consumption labeled value of the multiple exercise energy consumption labeled values is calculated as the sample true value corresponding to the first sample feature. Then, the first sample feature corresponding to each preset time window and the average exercise energy consumption labeled value corresponding to the first sample feature of each preset time window are input into the machine learning algorithm to obtain the first exercise energy consumption prediction value corresponding to each preset time window. Based on the first exercise energy consumption prediction value corresponding to each preset time window and the average exercise energy consumption labeled value, the machine learning algorithm is trained to obtain a trained machine learning model.

[0107] Among them, the length of dividing the preset time window may affect the accuracy of the trained machine learning model. Therefore, multiple different time windows can be determined in advance. For each time window, a process of training the machine learning model is completed, and the accuracy of the machine learning model is tested. The time window corresponding to the machine learning model with the highest accuracy is determined as the preset time window. Subsequently, when applying the machine learning model to monitor the exercise of preschool children, the three-axis acceleration signal is divided according to the preset time window.

[0108] In specific implementation, multiple different time windows are first determined. Among them, the different lengths of the multiple time windows can be divided in combination with experience and actual requirements. For example, the length of the time window can be set to 15 seconds, 30 seconds, 60 seconds, etc. For the first time window, the obtained three-axis acceleration signals are divided according to the first time window, so as to obtain second acceleration signals corresponding to multiple first time windows respectively. Among them, the first time window represents any one of the multiple different time windows. Based on the second acceleration signal and the personalized index, the second sample feature corresponding to the first time window is determined. Based on the second sample features corresponding to multiple first time windows and the average motion energy consumption label values corresponding to the second sample features of each first time window, a training set and a test set are determined. For example, the second sample features of all first time windows and the average motion energy consumption label values corresponding to the second sample features of each first time window can be fused to obtain a data set, and then the data set is divided into a training set and a test set according to a preset ratio. For example, the fused data set can be divided into a training set and a test set according to 8:2. Among them, the training set includes multiple second sample features and the average motion energy consumption label values corresponding to each second sample feature, and the test set also includes multiple second sample features and the average motion energy consumption label values corresponding to each second sample feature.

[0109] Then, the training set is input into the machine learning algorithm to obtain the predicted motion energy consumption value output by the machine learning algorithm. The machine learning algorithm is trained based on the predicted motion energy consumption value and the average motion energy consumption label value corresponding to the first time window until the training cut-off condition is met, and a trained machine learning model is obtained. Among them, the process of determining the average motion energy consumption label value corresponding to the first time window can refer to the above embodiments and will not be elaborated here.

[0110] After obtaining the trained machine learning model, the accuracy of the machine learning model can be verified by using the test set. Specifically, based on the second sample features corresponding to multiple first time windows, a test set is determined. For example, a certain proportion is divided from the second sample features of multiple first time windows as the test set. The test set is input into the machine learning model to obtain the test value of motion energy consumption. Based on the test value of motion energy consumption and the motion energy consumption label value corresponding to the test set, the test error is determined.

[0111] The above process is executed for multiple different time windows, and then the test error corresponding to each time window can be determined. Based on the test errors corresponding to multiple different time windows, the minimum test error is determined, so that the time window corresponding to the minimum test error can be determined as the preset time window. Subsequently, when using the machine learning model to monitor the motion energy consumption of preschool children, the three-axis acceleration signals can be divided according to the preset time window.

[0112] After multiple training and testing processes, when the preset time window is determined to be 15 seconds in the embodiments of the present application, the accuracy of the trained machine learning model is relatively high.

[0113] According to the above embodiments, the three-axis acceleration signal may include an x-axis acceleration signal, a y-axis acceleration signal, and a z-axis acceleration signal, and their one-dimensional vector sum signal is determined. Specifically, when the three-axis acceleration signal is divided according to the preset time window, for the x-axis acceleration signal, y-axis acceleration signal, z-axis acceleration signal, and one-dimensional vector sum signal of each preset time window, the statistical feature indexes of the x-axis acceleration signal, the statistical feature indexes of the y-axis acceleration signal, the statistical feature indexes of the z-axis acceleration signal, and the statistical feature indexes of the one-dimensional vector sum signal are respectively determined. When determining the sample features through each statistical feature index and the personalized index, since there are a relatively large number of indexes in the sample features, it will increase the complexity of data processing and affect the accuracy of training the machine learning model. Therefore, some of the indexes can be selected from each statistical feature index and the personalized index to train the machine learning model. The accuracy of the trained machine learning model is different when the number of selected indexes is different. Based on this, multiple different numbers can be pre-determined to select indexes, the process of training the machine learning model is completed for each number of indexes, and the accuracy of the machine learning model is tested, and the number of indexes corresponding to the machine learning model with the highest accuracy is selected as the preset number.

[0114] Specifically, when implementing, using the feature selection algorithm, multiple different numbers of indexes are selected from the statistical feature indexes and the personalized index, and the sample features corresponding to multiple different numbers of indexes are determined. For the third sample feature corresponding to the first number of indexes, a training set and a test set are determined. Wherein, the first number represents any one of the multiple different numbers. That is, for any first number, the feature selection algorithm can be used to select the first number of indexes from the statistical feature indexes and the personalized index, and the selected indexes are fused to determine the third sample feature. The third sample feature and the motion energy consumption label value corresponding to the third sample feature are fused, and then the fused data set is divided into a training set and a test set. The machine learning algorithm is trained using the training set to obtain a machine learning model. Among them, the specific training process will not be elaborated here.

[0115] Then, the test set is input into the trained machine learning model to obtain the test value of exercise energy consumption. Based on the test value of exercise energy consumption and the exercise energy consumption label value corresponding to the test set, the test error is determined. Based on the test errors corresponding to multiple different numbers of metrics respectively, the minimum test error is determined. The number corresponding to the minimum test error is determined as the preset number. In subsequent applications of the machine learning model to monitor exercise energy consumption, among the statistical feature metrics and personalized metrics, select the preset number of metrics as the features input into the machine learning model.

[0116] The exercise energy consumption label value in the embodiments of the present application includes energy consumption or metabolic equivalent, and two machine learning models can be trained for energy consumption and metabolic equivalent respectively. In order to obtain a machine learning model for monitoring energy consumption, after obtaining the personalized metrics, triaxial acceleration signals, and energy consumption label values of preschool children, multiple numbers of metrics can be determined to form sample features for training the machine learning model. After completing the training and testing processes using multiple numbers of metrics, the embodiments of the present application determine that when the sample features include 80 metrics, the accuracy of the obtained machine learning model in monitoring energy consumption is relatively high.

[0117] Similarly, in order to obtain a machine learning model for monitoring metabolic equivalent, after obtaining the personalized metrics, triaxial acceleration signals, and metabolic equivalent label values of preschool children, multiple numbers of metrics can be determined to form sample features for training the corresponding machine learning model. After completing the training and testing processes using multiple numbers of metrics, the embodiments of the present application determine that when the sample features include 96 metrics, the accuracy of the obtained machine learning model in monitoring metabolic equivalent is relatively high.

[0118] Specifically, as shown in Table 1, it is an example of determining sample features. Among them, the total number of samples represents the number of sample features. Whether feature selection is performed indicates whether only some metrics are selected from all the statistical feature metrics and personalized metrics to form sample features.

[0119] Table 1

[0120] By collecting the acceleration signals, personalized metrics, and exercise energy consumption data of preschool children, after obtaining the trained machine learning model, in subsequent application scenarios, the machine learning model can be used to determine the exercise energy consumption data of the preschool children to be monitored. Thus, according to the relationship between the exercise energy consumption data and the first moderate-intensity threshold and the second moderate-intensity threshold, the daily average moderate-activity index can be determined, that is, a moderate-activity intensity assessment model for preschool children is constructed, realizing the monitoring of the exercise energy consumption of preschool children and improving the accuracy of exercise energy consumption monitoring.

[0121] In a possible implementation, after obtaining the personalized indicators, triaxial acceleration signals, and one-dimensional vector sum signals of preschool children, the triaxial acceleration signals and one-dimensional vector sum signals can be divided according to a preset time window of 15 seconds. For each preset time window, statistical feature indicators of the triaxial acceleration signals and one-dimensional vector sum signals within the preset time window are extracted.

[0122] In a possible implementation, the determined statistical feature indicators of the triaxial acceleration signals and one-dimensional vector sum signals include a total of 91 indicators. The personalized indicators include 5 indicators: gender, age, height, weight, and body mass index.

[0123] In a possible implementation, when the motion energy consumption data to be obtained is energy consumption, 80 indicators from the statistical feature indicators and personalized indicators can be extracted to construct a feature set. Then, the feature set is input into the trained extremely randomized tree regression model to calculate the energy consumption.

[0124] In a possible implementation, when the motion energy consumption data to be obtained is metabolic equivalent, 96 indicators from the statistical feature indicators and personalized indicators can be extracted to construct a feature set. Then, the feature set is input into the trained extremely randomized tree regression model to calculate the metabolic equivalent. According to the relationship between the metabolic equivalent and the first moderate-intensity threshold and the second moderate-intensity threshold, the daily average moderate-intensity activity index is determined.

[0125] Based on the method introduced in the above embodiments, the embodiments of the present application further provide a method for monitoring the motion energy consumption of preschool children. Refer to Figure 2 As shown, it is a flowchart of a method for monitoring the motion energy consumption of preschool children provided by the embodiments of the present application.

[0126] This method may include the following steps: S201: Obtain the personalized indicators of the preschool children to be monitored and the acceleration signals in the state to be monitored.

[0127] S202: Input the personalized indicators and acceleration signals into the motion energy consumption monitoring model to obtain the motion energy consumption indicators of the preschool children to be monitored.

[0128] Among them, the motion energy consumption monitoring model can be expressed as the machine learning model trained in the above method embodiments. The motion energy consumption indicators may include energy consumption or metabolic equivalent.

[0129] Specifically, when implementing, input features can be determined based on the personalized indicators and acceleration signals. The input features are input into the motion energy consumption monitoring model to obtain the motion energy consumption indicators. Among them, the motion energy consumption indicators may include energy consumption or metabolic equivalent.

[0130] Among them, after obtaining the personalized index and the acceleration signal, the process of determining the input feature based on the personalized index and the acceleration signal can refer to the process of determining the sample feature in the training process of the above-mentioned embodiment. The principles of the two are the same and will not be elaborated here.

[0131] Since the motion energy consumption monitoring model is trained by collecting the acceleration signal, personalized index and its motion energy consumption data of preschool children, using this motion energy consumption monitoring model to monitor the motion energy consumption of preschool children can improve the accuracy of motion energy consumption monitoring.

[0132] In a possible implementation manner, when the motion energy consumption index includes metabolic equivalent, the daily average moderate-intensity activity index can be determined according to the relationship between the metabolic equivalent and the first moderate-intensity threshold and the second moderate-intensity threshold.

[0133] Based on the above method embodiments, the embodiments of the present application further provide a construction device for a moderate activity intensity evaluation model of preschool children. Refer to Figure 3 as shown Figure 3 which is a schematic diagram of a construction device for a moderate activity intensity evaluation model of preschool children provided by the embodiments of the present application.

[0134] The device 300 includes: A data acquisition unit 301, configured to obtain the personalized index, three-axis acceleration signal, and one-dimensional vector sum signal of the three-axis acceleration signal of the preschool children; A feature extraction unit 302, configured to extract statistical feature indexes of the three-axis acceleration signal and the one-dimensional vector sum signal within a preset time window, where multiple preset time windows are non-overlapping time windows; A feature construction unit 303, configured to construct a feature set based on the extracted statistical feature indexes and the personalized index; A prediction unit 304, configured to calculate motion energy consumption data according to the relationship between the feature set and the motion energy consumption; A judgment unit 305, configured to determine the daily average moderate-intensity activity index according to the relationship between the motion energy consumption data and the first moderate-intensity threshold and the second moderate-intensity threshold.

[0135] In a possible implementation manner, the daily average moderate-intensity activity index , where t is the length of the preset time window, and N is the total number of preset time windows that meet the requirements; the initial value of N is 0. When the motion energy consumption data corresponding to the preset time window is greater than or equal to the first moderate-intensity threshold and less than or equal to the second moderate-intensity threshold, let N be incremented by 1 for update.

[0136] In a possible implementation, the triaxial acceleration signal includes an x-axis acceleration signal, a y-axis acceleration signal, and a z-axis acceleration signal. The statistical characteristic index of the x-axis acceleration signal includes at least the maximum value of multiple x-axis acceleration signals within the preset time window; the statistical characteristic index of the y-axis acceleration signal includes at least the peak value of multiple y-axis acceleration signals within the preset time window; the statistical characteristic index of the z-axis acceleration signal includes at least the minimum value of multiple z-axis acceleration signals; the statistical characteristic index of the one-dimensional vector sum signal includes at least the average value and variance of multiple one-dimensional vector sum signals within the preset time window.

[0137] In a possible implementation, the relationship between the feature set and the exercise energy consumption is established by a trained machine learning model.

[0138] In a possible implementation, the trained machine learning model includes an extremely randomized tree regression model.

[0139] In a possible implementation, the exercise energy consumption data includes at least one of the average energy consumption and the average metabolic equivalent.

[0140] In a possible implementation, the personalized index includes at least one of gender, age, height, weight, and body mass index.

[0141] In a possible implementation, the first moderate intensity threshold is 3.2 Mets, and the second moderate intensity threshold is 5.3 Mets.

[0142] In a possible implementation, the first moderate intensity threshold corresponding to 3-year-old preschool children is 3.2 Mets, and the second moderate intensity threshold corresponding to 3-year-old preschool children is 4.4 Mets. That is, the range of moderate activity intensity corresponding to 3-year-old preschool children is [3.2, 4.4], and the overall standard deviation is plus or minus 0.2, where the value range of the standard deviation is [0, 0.3]. In specific implementation, the value of this standard deviation is affected by factors such as the height value range of the measurement sample.

[0143] In a possible implementation, the first moderate intensity threshold corresponding to 4-year-old preschool children is 3.4 Mets, and the second moderate intensity threshold corresponding to 4-year-old preschool children is 4.6 Mets. That is, the range of moderate activity intensity corresponding to 4-year-old preschool children is [3.4, 4.6], and the overall standard deviation is plus or minus 0.2, where the value range of the standard deviation is [0, 0.3]. In specific implementation, the value of this standard deviation is affected by factors such as the height value range of the measurement sample.

[0144] In a possible implementation, the first moderate-intensity threshold corresponding to 5-year-old preschool children is 3.7 Mets, and the second moderate-intensity threshold corresponding to 5-year-old preschool children is 4.8 Mets. That is, the range of moderate activity intensity corresponding to 5-year-old preschool children is [3.7, 4.8], and the overall standard deviation is plus or minus 0.2, where the value range of the standard deviation is [0, 0.4]. When specifically implemented, the value of this standard deviation is affected by factors such as the height value range of the measurement sample.

[0145] In a possible implementation, the first moderate-intensity threshold corresponding to 6-year-old preschool children is 3.8 Mets, and the second moderate-intensity threshold corresponding to 6-year-old preschool children is 5.1 Mets. That is, the range of moderate activity intensity corresponding to 6-year-old preschool children is [3.8, 5.1]. The overall standard deviation is plus or minus 0.2, where the value range of the standard deviation is [0, 0.6]. When specifically implemented, the value of this standard deviation is affected by factors such as the height value range of the measurement sample.

[0146] In a possible implementation, the age of preschool children is calculated according to the full years corresponding to the actual date of birth.

[0147] In addition, the embodiments of the present application also provide a device for monitoring the exercise energy consumption of preschool children. Refer to Figure 4 as shown, Figure 4 which is a schematic diagram of a device for monitoring the exercise energy consumption of preschool children provided by the embodiments of the present application.

[0148] The device 400 includes: An acquisition unit 401, configured to acquire the personalized index of the preschool child to be monitored and the acceleration signal in the state to be monitored; A monitoring unit 402, configured to input the personalized index and the acceleration signal into the exercise energy consumption monitoring model to obtain the exercise energy consumption index of the preschool child to be monitored; Among them, the exercise energy consumption monitoring model can be expressed as a machine learning model trained by the above method embodiments.

[0149] Based on the above method embodiments and device embodiments, the embodiments of the present application also provide an electronic device. It will be introduced below with reference to the accompanying drawings.

[0150] Refer to Figure 5 , Figure 5 which is a schematic diagram of an electronic device provided by the embodiments of the present application.

[0151] The device 500 includes: a memory 501 and a processor 502; The memory 501 is used to store relevant program codes; The processor 502 is used to call the program code to execute the method for constructing a moderate activity intensity evaluation model for preschool children or the method for monitoring the exercise energy consumption of preschool children described in the above method embodiments.

[0152] In addition, an embodiment of the present application further provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method for constructing a moderate activity intensity evaluation model for preschool children or the method for monitoring the exercise energy consumption of preschool children described in the above method embodiments.

[0153] An embodiment of the present application further provides a computer program product, which includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the method for constructing a moderate activity intensity evaluation model for preschool children or the method for monitoring the exercise energy consumption of preschool children described in the above method embodiments is implemented.

[0154] In addition, the present application further provides a method for determining an exercise energy consumption monitoring model, and the method includes: Obtain the personalized indicators, acceleration signals of preschool children, and the exercise energy consumption label values corresponding to the acceleration signals; Based on the personalized indicators and the acceleration signals, determine sample features; Input the sample features and the exercise energy consumption label values corresponding to the sample features into a machine learning algorithm to obtain exercise energy consumption prediction values; Based on the exercise energy consumption prediction values and the exercise energy consumption label values, train the machine learning algorithm until the training cut-off condition is met to obtain a trained exercise energy consumption monitoring model.

[0155] In a possible implementation manner, the determining sample features based on the personalized indicators and the acceleration signals includes: Divide the acceleration signals according to a preset time window to obtain first acceleration signals corresponding to multiple preset time windows; Based on the first acceleration signals of each preset time window and the personalized indicators, determine the first sample features corresponding to each preset time window.

[0156] In a possible implementation manner, the inputting the sample features into a machine learning algorithm to obtain exercise energy consumption prediction values includes: Input the first sample features corresponding to each preset time window and the exercise energy consumption label values corresponding to the first sample features of each preset time window into the machine learning algorithm to obtain first exercise energy consumption prediction values corresponding to each preset time window; Training the machine learning algorithm based on the predicted value of exercise energy consumption and the labeled value of exercise energy consumption includes: For each preset time window, obtain multiple labeled values of exercise energy consumption within the preset time window, and calculate the average labeled value of exercise energy consumption of the multiple labeled values of exercise energy consumption; Train the machine learning algorithm based on the first predicted value of exercise energy consumption corresponding to each preset time window and the average labeled value of exercise energy consumption.

[0157] In a possible implementation, determining sample features based on the personalized metric and the acceleration signal includes: Determine multiple different time windows; For the first time window, divide the acceleration signal according to the first time window to obtain multiple second acceleration signals respectively corresponding to the first time window, where the first time window represents any one of the multiple different time windows; Determine the second sample features corresponding to the first time window based on the second acceleration signal and the personalized metric; Inputting the sample features and the labeled value of exercise energy consumption corresponding to the sample features into a machine learning algorithm to obtain a predicted value of exercise energy consumption includes: Determine a training set based on the second sample features corresponding to multiple first time windows and the labeled value of exercise energy consumption corresponding to the second sample features of the first time window; Input the training set into the machine learning algorithm to obtain the predicted value of exercise energy consumption; The method further includes: Determine a test set based on the second sample features corresponding to multiple first time windows and the labeled value of exercise energy consumption corresponding to the second sample features of the first time window; Input the test set into the exercise energy consumption monitoring model to obtain a test value of exercise energy consumption; Determine a test error based on the test value of exercise energy consumption and the labeled value of exercise energy consumption corresponding to the test set; Determine a minimum test error based on the test errors corresponding to multiple different time windows; Determine the time window corresponding to the minimum test error as the preset time window.

[0158] In a possible implementation, the machine learning algorithm includes multiple different machine learning models, and each machine learning model corresponds to a trained exercise energy consumption monitoring model. The method further includes: Determine a test set based on the sample features and the labeled value of exercise energy consumption corresponding to the sample features; For any machine learning model, input the test set into the motion energy consumption monitoring model corresponding to the machine learning model to obtain motion energy consumption test values; Based on the motion energy consumption test values and the motion energy consumption label values corresponding to the test set, determine the test error; Based on the test errors corresponding to multiple machine learning models respectively, determine the minimum test error; Determine the machine learning model corresponding to the minimum test error as the preferred machine learning algorithm.

[0159] In a possible implementation, the acceleration signal includes an x-axis acceleration signal, a y-axis acceleration signal, and a z-axis acceleration signal. The determining of the sample features based on the personalized index and the acceleration signal includes: Determine the vector sum signal of the x-axis acceleration signal, the y-axis acceleration signal, and the z-axis acceleration signal; Respectively determine the statistical feature indexes of the x-axis acceleration signal, the y-axis acceleration signal, the z-axis acceleration signal, and the vector sum signal; Based on the statistical feature indexes and the personalized index, determine the sample features.

[0160] In a possible implementation, the determining of the sample features based on the statistical feature indexes and the personalized index includes: Use a feature selection algorithm to select a preset number of indexes from the statistical feature indexes and the personalized index; Based on the preset number of indexes, determine the sample features.

[0161] In a possible implementation, the determining process of the preset number includes: Use a feature selection algorithm to select various different numbers of indexes from the statistical feature indexes and the personalized index, and determine the sample features corresponding to the various different numbers of indexes respectively; For the third sample features corresponding to the first number of indexes and the motion energy consumption label values corresponding to the third sample features, determine the training set and the test set, where the first number represents any one of the various different numbers; Use the training set to train the machine learning algorithm to obtain the motion energy consumption monitoring model; Input the test set into the motion energy consumption monitoring model to obtain motion energy consumption test values; Based on the motion energy consumption test values and the motion energy consumption label values corresponding to the test set, determine the test error; Based on the test errors corresponding to the various different numbers of indexes respectively, determine the minimum test error; Determine that the number corresponding to the minimum test error is the preset number.

[0162] In a possible implementation, the determining of the sample features based on the personalized metrics and the acceleration signal includes: Preprocess the personalized metrics and the acceleration signal to obtain preprocessed personalized metrics and a preprocessed acceleration signal; Determine the sample features based on the preprocessed personalized metrics and the preprocessed acceleration signal; Wherein, the preprocessing process includes one or more of the following: Impute missing values; or, Perform one-hot encoding on categorical data; or, Normalize continuous data; or, Normalize continuous data and convert the normalized data into data with an approximate normal distribution.

[0163] In a possible implementation, train the machine learning algorithm using the sample features and the k-fold cross-validation method; The machine learning algorithm includes an extremely randomized tree regression algorithm.

[0164] In a possible implementation, the personalized metrics include at least one of gender, age, height, weight, and body mass index; The exercise energy consumption label value includes at least one of energy consumption and metabolic equivalent.

[0165] In a possible implementation, the acceleration signal is measured by an acceleration meter worn by the preschool child; The exercise energy consumption label value is measured by a gas metabolism analyzer worn by the preschool child.

[0166] It should be noted that the computer-readable medium in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0167] The computer program product can be written in any combination of one or more programming languages for programming code to perform the operations of the embodiments of the present application. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The programming code can be executed entirely on the user's computing device, partially on the user's device, executed as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0168] It should be noted that the various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. In particular, for system or device embodiments, since they are basically similar to method embodiments, they are described relatively simply. The relevant parts can refer to the partial descriptions of the method embodiments. The device embodiments described above are merely illustrative. The units or modules described as separate components may or may not be physically separated. The components shown as units or modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network units. Some or all of the units or modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative work.

[0169] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of methods, devices, and equipment according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0170] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Here, A and B can be singular or plural. The character " / " generally indicates an "or" relationship between the related objects before and after. "At least one (item) of the following" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or plural.

[0171] It should also be noted that in this application, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0172] The steps of the methods or algorithms described in connection with the embodiments disclosed in this application can be implemented directly by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0173] The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined in this application can be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application will not be limited to these embodiments shown in this application, but will be accorded the widest scope consistent with the principles and novel features disclosed in this application.

Claims

1. A method for constructing a moderate activity intensity assessment model for preschool children, characterized in that: The method comprises: Acquire personalized indicators of preschool children, three-axis acceleration signals, and one-dimensional vector sum signals of the three-axis acceleration signals; Extracting statistical characteristic indicators of the three-axis acceleration signal and the one-dimensional vector sum signal within a preset time window, wherein the multiple preset time windows are non-overlapping time windows; Constructing a feature set based on the extracted statistical feature indicators and the personalized indicators; According to the relationship between the feature set and the exercise energy consumption, the exercise energy consumption data is calculated; Based on the relationship between exercise energy consumption data and the first moderate intensity threshold and the second moderate intensity threshold, the average daily moderate to strong activity index is determined.

2. The method according to claim 1, characterized in that The daily average moderate to strong activity index , t is the length of the preset time window, N is the total number of preset time windows that meet the requirements; the initial value of N is 0, and when the exercise energy consumption data corresponding to the preset time window is greater than or equal to the first moderate intensity threshold and less than or equal to the second moderate intensity threshold, N is updated by adding 1.

3. The method according to claim 1, characterized in that The relationship between the feature set and exercise energy consumption is established through a trained machine learning model.

4. The method according to claim 3, characterized in that The trained machine learning model includes an extreme random tree regression model.

5. The method according to claim 1, characterized in that The exercise energy consumption data includes at least one of energy consumption and metabolic equivalent.

6. The method according to claim 1, characterized in that The personalized indicator includes at least one of gender, age, height, weight and body mass index.

7. The method according to claim 1, characterized in that The first medium-intensity threshold is 3.2 Mets, and the second medium-intensity threshold is 5.3 Mets.

8. The method according to claim 1, characterized in that The first moderate-intensity threshold for 3-year-old preschoolers is 3.2 Mets, and the second moderate-intensity threshold for 3-year-old preschoolers is 4.4 Mets.

9. The method according to claim 1, characterized in that: The first moderate-intensity threshold for 4-year-old preschoolers is 3.4 Mets, and the second moderate-intensity threshold for 4-year-old preschoolers is 4.6 Mets.

10. The method according to claim 1, characterized in that The first moderate-intensity threshold for 5-year-old preschoolers is 3.7 Mets, and the second moderate-intensity threshold for 5-year-old preschoolers is 4.8 Mets.

11. The method according to claim 1, characterized in that: The first moderate-intensity threshold for 6-year-old preschoolers is 3.8 Mets, and the second moderate-intensity threshold for 6-year-old preschoolers is 5.1 Mets.

12. The method according to claim 1, characterized in that The age of preschool children is calculated based on the age corresponding to their actual date of birth.

13. A device for constructing a model for evaluating the intensity of moderate activities for preschool children, characterized in that: The device comprises: A data acquisition unit, used to obtain personalized indicators of preschool children, three-axis acceleration signals, and one-dimensional vector sum signals of the three-axis acceleration signals; A feature extraction unit, used to extract statistical feature indicators of the three-axis acceleration signal and the one-dimensional vector sum signal within a preset time window, wherein the multiple preset time windows are non-overlapping time windows; A feature construction unit, used to construct a feature set based on the extracted statistical feature indicators and the personalized indicators; A prediction unit, used for calculating and obtaining exercise energy consumption data according to the relationship between the feature set and exercise energy consumption; The judgment unit is used to determine the daily average moderate-intensity activity index according to the relationship between the exercise energy consumption data and the first moderate-intensity threshold and the second moderate-intensity threshold.

14. An electronic device, characterized in that: The device comprises: a memory and a processor; The memory is used to store relevant program codes; The processor is used to call the program code to execute the method for constructing a moderate activity intensity assessment model for preschool children as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method for constructing a moderate activity intensity assessment model for preschool children as described in any one of claims 1 to 12.

Images