BMI prediction method based on convolutional neural network

Abstract

The invention discloses a BMI prediction method based on a convolutional neural network, and the method comprises the following steps: collecting preset exercise item data of a subject and a corresponding BMI classification result, constructing an exercise data set # imgabs0 #, carrying out the standardization processing of the exercise data set # imgabs1 #, and obtaining a standard data set D = {(xi, yi)}; performing feature extraction on a standard feature xi in a standard data set D = {(xi, yi)} based on a convolutional neural network, and obtaining an output feature hi corresponding to each motion; according to the method, a convolutional neural network and an attention mechanism are combined, an attention mechanism is introduced, according to contribution weights of motion data corresponding to different types of motions to BMI prediction, a personal weighted feature # imgabs2 # is obtained, the personal weighted feature # imgabs3 # is input into a preset LightGBM model, and a BMI prediction result # imgabs4 # is obtained. According to the method, the convolutional neural network and the attention mechanism are combined with the LightGBM model to process multiple types of motion data by adopting a mixed model; and the BMI prediction precision and stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of health informatics, and in particular to a BMI prediction method based on a convolutional neural network. Background Art

[0002] BMI (Body Mass Index) is an important standard commonly used internationally to measure the degree of human obesity and health. The reasonable range of BMI is usually closely related to the health status of an individual, and can effectively reflect whether the weight is normal or whether there are health problems such as obesity. Therefore, BMI prediction has important application value in individual health monitoring and fitness management.

[0003] At present, traditional BMI prediction methods mainly rely on factors such as physical examination data, diet and exercise habits, but these methods are often too simple and ignore the comprehensive impact of different exercise methods on BMI, thereby reducing the accuracy of BMI prediction.

[0004] Therefore, it is urgent to propose a BMI prediction method based on convolutional neural network to solve the problem raised. Summary of the invention

[0005] Based on this, it is necessary to provide a BMI prediction method based on convolutional neural networks to address the shortcomings of the existing technology, combine the attention mechanism to process the preset sports data, and improve the accuracy and stability of BMI prediction.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] The present invention provides a BMI prediction method based on a convolutional neural network, which comprises the following steps:

[0008] S110: Collect the subject's preset sports data and the corresponding BMI classification results to build a sports data set in, represents the eigenvalue corresponding to the preset sports test data of the i-th subject, y i represents the BMI classification result corresponding to the i-th subject;

[0009] S120, for motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )};

[0010] S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x iPerform feature extraction to obtain the output feature h corresponding to each preset sports item j ; Where 1≤j≤m, m represents the total number of sports types;

[0011] S140, introduce the attention mechanism, and obtain the personal weighted features according to the contribution weight of the sports data corresponding to different types of preset sports in BMI prediction

[0012] S150, weighting individual features Input into the preset LightGBM model to obtain the BMI prediction result

[0013] Furthermore, in step S110, the preset sports data of the subject and the corresponding BMI classification results are collected to construct a sports data set The method comprises the following steps:

[0014] Collect the physical test results corresponding to the subjects' preset sports data

[0015] Collect the BM1 classification results y corresponding to the subject i ; BMI classification includes four categories: underweight A, normal weight B, overweight C, and obese D.

[0016] Building a motion dataset

[0017] Furthermore, in step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps:

[0018] When m=1, obtain the physical test results corresponding to the subject in a single sport And the corresponding BMl classification result y i , then the corresponding feature set under this kind of sports Among them, m represents the total number of sports types, N represents the total number of subjects, and a single sport is one of the preset sports;

[0019] By Z-score normalization formula The corresponding feature set for this sport Processing is performed to obtain the physical test results corresponding to the subjects in this sport The corresponding standard feature x i , combined with the corresponding BMI classification result y i, and obtain the standard data set D of the subjects in this sport = {(x i ,y i )}.

[0020] Furthermore, in step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps:

[0021] When m≥2, obtain the corresponding physical test results of the subjects under various preset sports items And the corresponding BMI classification result y i , then the corresponding feature sets under different preset sports items Among them, m represents the total number of sports types, and N represents the total number of subjects;

[0022] By Z-score normalization formula Corresponding feature sets for different preset sports Processing is performed to obtain the corresponding physical test results of the subjects under each preset sports event The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and then obtain the standard data set D of the subjects under each preset sports event = {(x i ,y i )}.

[0023] Furthermore, in step S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j The method comprises the following steps:

[0024] Through a filter w, convolution operation is performed with the time series data x(t) to generate feature y(t).

[0025]

[0026] Among them, x(t) is the motion data x i Corresponding time series data, w(k) is the parameter of the convolution kernel, b is the bias term, and K is the length of the filter;

[0027] The output of each convolutional layer is nonlinearly transformed by the activation function ReLU to obtain the linearly transformed feature ReLU(y(t)), ReLU(y(t))=max(0,y(t));

[0028] Perform the maximum pooling operation and select the motion data x in the pooling window i The maximum value of the corresponding time series data x(t) MaxPooling(x(t)), MaxPooling(x(t)) = max{x(t), x(t+1), ..., x(t+K-1)}, obtain the output feature h corresponding to each preset sports item j ,h j =f CNN (x i , W CNN ), where h j represents the output feature of the jth preset sports item, f cNN Function represents the processing of the entire convolutional network, W CNN are the trainable parameters of the convolutional network.

[0029] Furthermore, in step S140, the attention mechanism is introduced to obtain the personal weighted features according to the contribution weights of the sports data corresponding to different types of preset sports in the BMI prediction. The method comprises the following steps:

[0030] Get the motion data feature set H of each subject, H = {h1, h2, ..., h j}, h j is the output feature of the jth preset sports event, 1≤j≤m, m is the total number of preset sports events;

[0031] Get the output feature h through the fully connected layer j The corresponding preset sports item weight parameter α in BMI prediction j , Among them, f(h j ) is the Softmax function;

[0032] By formula For the output feature h j Perform weighted calculation to obtain weighted features Then all weighted features Sum and get the individual weighted features

[0033]

[0034] Furthermore, in step S150, the personal weighted features Input into the preset LightGBM model to obtain the BMI prediction result The method comprises the following steps:

[0035] Divide the dataset into training, validation and test sets;

[0036] Build multiple decision trees and input individual weighted features through the training set Use the preset LightGBM model for training;

[0037] In the preset LightGBM model, the input features of the preset LightGBM model are:

[0038]

[0039] The output features of the preset LightGBM model are BMI prediction results satisfy:

[0040]

[0041] Among them, f q () is the prediction function of the qth decision tree, W LightGBM is the parameter of the LightGBM model, and Q is the total number of decision trees.

[0042] Furthermore, the loss function L used for training using the preset LightGBM model satisfies:

[0043]

[0044] Among them, l() is the loss function and Ω() is the regularization term.

[0045] Further, Among them, y i is the actual BMI classification result, It is the BMI prediction result of the preset LightGBM model.

[0046] Furthermore, after step A150, the following step is further included:

[0047] A160, collect motion data of new subjects x new , the motion data of the new subject is normalized to obtain new personal weighted features Finally, the new personal weighted features Input into the preset LightGBM model and output the new BMI prediction result

[0048] In summary, the BMI prediction method based on convolutional neural network provided by the present invention combines convolutional neural network and attention mechanism, and uses a hybrid model to process multiple types of sports data with the LightGBM model to improve the accuracy and stability of BMI prediction; by taking the data of different sports as input, the convolutional neural network can extract the shallow features of each type of sports, and the attention mechanism weights the importance of each type of sports data, thereby optimizing the feature fusion process, and finally using the LightGBM model to comprehensively train and optimize the features, further improving the effect of BMI prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic flow chart of a first BMI prediction method based on a convolutional neural network provided in an embodiment of the present invention.

[0050] Figure 2 A schematic diagram of a flow chart of a second BMI prediction method based on a convolutional neural network provided in an embodiment of the present invention;

[0051] Figure 3 A flow chart of the first BMI prediction method based on convolutional neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] Some existing BMI prediction methods mainly rely on factors such as physical examination data, diet and exercise habits, but these methods often consider too simple aspects and ignore the comprehensive impact of different exercise methods on BMI prediction. The present invention collects data of preset sports such as running, strength training, etc., and uses the BMI prediction method based on convolutional neural network of the present invention to perform data analysis, thereby realizing the prediction of BMI, thereby providing more accurate decision support for personal biochemical health management and exercise intervention.

[0053] In order to further understand the features, technical means, specific objectives and functions of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0054] Figure 1 is a flow chart of a BMI prediction method based on a convolutional neural network provided by an embodiment of the present invention, such as Figure 1 As shown, the BMI prediction method based on convolutional neural network is suitable for application scenarios such as physical fitness assessment and personalized health management of college students, including steps S110 to A150, which are as follows:

[0055] A110: Collect the subject's preset sports data and corresponding BMI classification results to build a sports data set in, represents the eigenvalue corresponding to the preset sports test data of the i-th subject, y i Indicates the BMI classification result corresponding to the i-th subject; in this embodiment, the BMI classification includes four categories: underweight A, normal weight B, overweight C, and obesity D; the age of the subjects is distributed between 14 and 25 years old, and the male-female ratio of the collected subjects is 1:1. The preset sports project data include physical fitness test data such as 3000-meter running, pull-ups, sit-ups, and 30*2 shuttle runs.

[0056] Specifically, the method of collecting the preset sports data of the subject and the corresponding BMI classification results and constructing the sports data set in step S110 includes the following steps:

[0057] Collect the physical test results corresponding to the preset sports data of the subjects Collect the subject's preset sports data such as 3000-meter running results, number of pull-ups, number of sit-ups, and 30*2 shuttle run results;

[0058] Collect the BMI classification results y corresponding to the subjects i In this embodiment, the BMI classification includes four categories: underweight A, normal weight B, overweight C, and obesity D. The BMI classification result corresponding to each subject can be known by detecting the height and weight of each subject. i This is a known technology and need not be elaborated here.

[0059] Building a motion dataset

[0060] S120, for motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}.

[0061] Specifically, in step A120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps:

[0062] When m=1, obtain the physical test results corresponding to the subject in a single sport And the corresponding BMI classification result y i , then the corresponding feature set under this kind of sports Wherein, m represents the total number of sports types, N represents the total number of subjects, and a single sport is one of the preset sports.

[0063] By Z-score normalization formula The corresponding feature set for this sport Processing is performed to obtain the physical test results corresponding to the subjects in this sport The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and then obtain the standard data set D of the subjects in this kind of sports event = {(x i ,y i )}; At this time, the corresponding physical test results of the subjects in this sport The standard dataset D = {(x i ,y i )}; where x i represents the normalized eigenvalue corresponding to the physical test result of the i-th subject in this sport, y i represents the BMI classification result corresponding to the i-th subject; μ and σ are the mean and standard deviation of the motion data, respectively. Standardization can eliminate data distribution differences and make data modeled at the same scale, which helps the gradient descent algorithm converge faster and improves model training efficiency.

[0064] The normalization processing of motion data sets corresponding to other types of motion can be obtained in the same way, as described below.

[0065] When m≥2, obtain the corresponding physical test results of the subjects under various preset sports items And the corresponding BMI classification result y i , then the corresponding feature sets under different preset sports items Wherein, m represents the total number of sports events, and N represents the total number of subjects. For example, when the multiple preset sports events include 3000-meter running, pull-ups, sit-ups, and 30*2 shuttle runs, m=4;

[0066] By Z-score normalization formula Corresponding feature sets for different preset sports Processing is performed to obtain the corresponding physical test results of the subjects under each preset sports event The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and then obtain the standard data set D of the subjects under each preset sports event = {(x i ,y i )}; At this time, the physical test results corresponding to each preset sports item The standard dataset D = {(x i ,yi )}; where x i represents the normalized eigenvalue corresponding to the physical fitness result of the preset sports event of the i-th subject, y i represents the BMI classification result corresponding to the i-th subject; μ and σ are the mean and standard deviation of the motion data, respectively. Standardization can eliminate data distribution differences. Different preset sports may have different dimensions and distribution ranges. Standardization can effectively eliminate these differences and make the motion data modeled at the same scale, which helps the gradient descent algorithm converge faster and improves the efficiency of model training.

[0067] like Figure 3 As shown, A130, based on convolutional neural network, is used to train the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j ; Among them, 1≤j≤m, m represents the total number of sports types.

[0068] Specifically, the step S130 is to use a convolutional neural network to perform a standard data set D={(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j The method comprises the following steps:

[0069] Through a filter w, convolution operation is performed with the time series data x(t) to generate feature y(t).

[0070]

[0071] Where x(t)) is the motion data x i Corresponding time series data, w(k) is the parameter of the convolution kernel, b is the bias term, and K is the length of the filter; the convolution operation can capture local features in the time series data, such as the acceleration or deceleration pattern during running;

[0072] The output of each convolutional layer is transformed nonlinearly by the activation function ReLU to obtain the linearly transformed feature ReLU(y(t)), ReLU(y(t))=max(0, y(t)); the ReLU activation function helps the network learn nonlinear relationships and avoid the model falling into linear limitations;

[0073] Perform the maximum pooling operation and select the motion data x in the pooling window iThe maximum value of the corresponding time series data x(t) MaxPooling(x(t)), MaxPooling(x(t)) = max{x(t), x(t+1), ..., x(t+N-1)}, obtain the output feature h corresponding to each preset sports item j ,h j =f CNN (x i , W CNN ), where h j represents the output feature of the jth preset sport after processing by the convolutional network, which is for the input time series data x i The extracted local feature representation is the final feature after pooling, which contains information about key patterns in the input data, such as acceleration or deceleration patterns in time series; the pooling operation is used for downsampling to reduce the size of the feature map and extract more representative features. The pooling operation can effectively reduce the amount of calculation and improve the robustness of the feature; f CNN The function represents the entire convolutional network processing, including convolution, activation (ReLU) and pooling. The output of the maximum pooling is integrated into the feature representation as the input of the subsequent model; W CNN are the trainable parameters of the convolutional network.

[0074] S140: Introduce the attention mechanism to obtain personal weighted features based on the contribution weights of sports data corresponding to different types of preset sports in BMI prediction For example, a 3,000-meter run may have a greater impact on BMI, while pull-ups may have a relatively smaller impact. Through the attention mechanism, the model can adaptively learn the relative importance of each motion data.

[0075] Specifically, in step S140, the attention mechanism is introduced to obtain the personal weighted features according to the contribution weights of the sports data corresponding to different types of preset sports in the BMI prediction. The method comprises the following steps:

[0076] Get the motion data feature set H of each subject, H = {h1, h2, ..., h j}, h j is the output feature of the jth preset sports event, 1≤j≤m, m is the total number of preset sports events;

[0077] Get the output feature h through the fully connected layer j The corresponding preset sports item weight parameter α in BMI prediction j , Among them, f(h j) is a Softmax function, which is usually calculated through a fully connected layer to output the feature h j Importance of BMI prediction.

[0078] By formula For the output feature h j Perform weighted calculation to obtain weighted features Then all weighted features Sum and get the individual weighted features

[0079]

[0080] A150, weighting individual features Input into the preset LightGBM model to obtain the BMI prediction result The LightGBM model conducts a comprehensive analysis based on the data of different sports and predicts the most likely BMI category. LightGBM is an efficient gradient boosting tree (GBDT) model that improves prediction accuracy by integrating multiple decision trees.

[0081] Specifically, in step A150, the personal weighted features Input into the preset LightGBM model to obtain the BMI prediction result The method comprises the following steps:

[0082] Divide the dataset into training set, validation set and test set; usually the training set occupies 70% to 80% of the data, the validation set is used to adjust the hyperparameters during the training process, and the test set is used to evaluate the final effect of the model;

[0083] Build multiple decision trees and input individual weighted features through the training set Use the preset LightGBM model for training; during the training process, the preset LightGBM model will continuously adjust the model parameters based on the error between the predicted result and the true value through optimization algorithms such as gradient descent to minimize the error.

[0084] In the preset LightGBM model, the input features of the preset LightGBM model are:

[0085]

[0086] The output features of the preset LightGBM model are BMI prediction results satisfy:

[0087]

[0088] Among them, fq () is the prediction function of the qth decision tree, W LightGBM is the parameter of the LightGBM model, and Q is the total number of decision trees.

[0089] Furthermore, the loss function L used for training using the preset LightGBM model satisfies:

[0090]

[0091] Where l() is the loss function, such as mean squared error or cross entropy, Ω() is the regularization term used to control model complexity, and N is the total number of subjects.

[0092] In this embodiment, Among them, y i is the actual BMI classification result, It is the BMI prediction result of the preset LightGBM model.

[0093] Furthermore, during the training process using the preset LightGBM model, the preset LightGBM model is verified using the validation set, and the hyperparameters of the preset LightGBM model are adjusted through cross-validation and grid search; for example, the hyperparameters of the preset LightGBM model such as tree depth, learning rate, etc. are adjusted to improve the accuracy and generalization ability of the preset LightGBM model.

[0094] In one embodiment, after step A150, the following step is further included:

[0095] S160, collect motion data of new subject x new , the motion data of the new subject is normalized to obtain new personal weighted features Finally, the new personal weighted features Input into the preset LightGBM model and output the new BMI prediction result

[0096] After the preset LightGBM model training and verification are completed, the test set is finally used to make predictions for the preset LightGBM model; new motion data is input, and after passing through the convolutional neural network, attention mechanism calculation and the preset LightGBM model respectively, the new BMI prediction result is output.

[0097] Assume that the new input data is x new , firstly extract the feature H through the convolutional neural network new ={h 1new ,h 2new ,……,h jnew}, then introduce the attention mechanism weighted features to obtain personal weighted features Finally, these weighted features are input into the trained LightGBM model to obtain the BMI prediction results.

[0098] Output It is the predicted BMI value, or the predicted category such as normal, overweight, etc.

[0099] like Figure 2 As shown, in order to make the technical solution of the present invention clearer, a preferred embodiment is described below.

[0100] S110: Collect the subject's preset sports data and the corresponding BMI classification results to build a sports data set in, represents the eigenvalue corresponding to the preset sports test data of the i-th subject, y i represents the BMTI classification result corresponding to the i-th subject;

[0101] S120, for motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )};

[0102] S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j ;

[0103] A140, introduce the attention mechanism, obtain personal weighted features according to the contribution weight of sports data corresponding to different types of preset sports in BMI prediction

[0104] S150, weighting individual features Input into the preset LightGBM model to obtain the BMI prediction result

[0105] S160, collect motion data of new subject x new , the motion data of the new subject is normalized to obtain new personal weighted features Finally, the new personal weighted features Input into the preset LightGBM model and output the new BMI prediction result

[0106] In summary, the BMI prediction method based on convolutional neural network of the present invention combines convolutional neural network and attention mechanism, and uses a hybrid model to process multiple types of sports data with the LightGBM model to improve the accuracy and stability of BMI prediction; by taking the data of different sports as input, the convolutional neural network can extract the shallow features of each type of sports, and the attention mechanism weights the importance of each type of sports data, thereby optimizing the feature fusion process, and finally using the LightGBM model to comprehensively train and optimize the features, further improving the effect of BMI prediction.

[0107] The present invention not only makes up for the shortcomings of traditional BMI prediction methods, but also innovatively solves the problems of multi-type sports data fusion and feature weighting by introducing multi-level data processing and feature extraction mechanisms, providing a more accurate and reliable BMI prediction method, and providing strong technical support for personal biochemical health management and exercise intervention.

[0108] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0109] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0110] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a terminal, or a network device, etc.) to perform all or part of the steps of the method described in the various embodiments of the present invention.

[0111] The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A BMI prediction method based on convolutional neural network, characterized in that: The steps include: S110: Collect the subject's preset sports data and the corresponding BMI classification results to build a sports data set in, represents the eigenvalue corresponding to the preset sports test data of the i-th subject, y i represents the BMI classification result corresponding to the i-th subject; S120, for motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}; S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j ; Where 1≤j≤m, m represents the total number of sports types; S140, introduce the attention mechanism, and obtain the personal weighted features according to the contribution weight of the sports data corresponding to different types of preset sports in BMI prediction S150, weighting individual features Input into the preset LightGBM model to obtain the BMI prediction result 2. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: Step S110: collecting the subject's preset sports data and the corresponding BMI classification results to construct a sports data set The method comprises the following steps: Collect the physical test results corresponding to the subjects' preset sports data Collect the BMI classification results y corresponding to the subjects i ; BMI classification includes four categories: underweight A, normal weight B, overweight C, and obese D. Building a motion dataset 3. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: In step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps: When m=1, obtain the physical test results corresponding to the subject in a single sport And the corresponding BMI classification result y i , then the corresponding feature set under this kind of sports Among them, m represents the total number of sports types, N represents the total number of subjects, and a single sport is one of the preset sports; By Z-score normalization formula The corresponding feature set for this sport Processing is performed to obtain the physical test results corresponding to the subjects in this sport The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and obtain the standard data set D of the subjects in this sport = {(x i ,y i )}.

4. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: In step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps: When m≥2, obtain the corresponding physical test results of the subjects under various preset sports items And the corresponding BMI classification result y i , then the corresponding feature sets under different preset sports items Among them, m represents the total number of sports types, and N represents the total number of subjects; By Z-score normalization formula Corresponding feature sets for different preset sports Processing is performed to obtain the corresponding physical test results of the subjects under each preset sports event The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and then obtain the standard data set D of the subjects under each preset sports event = {(x i ,y i )}.

5. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: In step S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j The method comprises the following steps: Through a filter w and time series data x(t)) convolution operation, generate feature y(t), Where x(t)) is the motion data x i Corresponding time series data, w(k) is the parameter of the convolution kernel, b is the bias term, and K is the length of the filter; The output of each convolutional layer is nonlinearly transformed by the activation function ReLU to obtain the linearly transformed feature ReLU(y(t)), ReLU(y(t))=max(0,y(t)); Perform the maximum pooling operation and select the motion data x in the pooling window i The maximum value of the corresponding time series data x(t) MaxPooling(x(t)), MaxPooling(x(t)) = max{x(t), x(t+1), ..., x(t+K-1)}, obtain the output feature h corresponding to each preset sports item j ,h j =f CNN (x i , W CNN ), where h j represents the output feature of the jth preset sports item, f cNN Function represents the processing of the entire convolutional network, W CNN are the trainable parameters of the convolutional network.

6. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: The step S140 introduces an attention mechanism to obtain individual weighted features according to the contribution weights of the exercise data corresponding to different types of preset sports in BMI prediction. The method comprises the following steps: Get the motion data feature set H of each subject, H = {h1, h2, ..., h j }, h j is the output feature of the jth preset sports event, 1≤j≤m, m is the total number of preset sports events; Obtain the output feature h through the fully connected layer j The corresponding preset sports item weight parameter α in BMI prediction j , Among them, f(h j ) is the Softmax function; By formula For the output feature h j Perform weighted calculation to obtain weighted features Then all weighted features Sum and get the individual weighted features 7. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: In step S150, the personal weighted features Input into the preset LightGBM model to obtain the BMI prediction result The method comprises the following steps: Divide the dataset into training, validation and test sets; Build multiple decision trees and input individual weighted features through the training set Use the preset LightGBM model for training; In the preset LightGBM model, the input features of the preset LightGBM model are: The output features of the preset LightGBM model are BMI prediction results satisfy: Among them, f q () is the prediction function of the qth decision tree, W LightGBM is the parameter of the LightGBM model, and Q is the total number of decision trees.

8. The BMI prediction method based on convolutional neural network according to claim 7, characterized in that: The loss function L used for training using the preset LightGBM model satisfies: Among them, l() is the loss function and Ω() is the regularization term.

9. The BMI prediction method based on convolutional neural network according to claim 8, characterized in that: Among them, y i is the actual BMI classification result, It is the BMI prediction result of the preset LightGBM model.

10. The BMI prediction method based on convolutional neural network according to claim 1, characterized in that: After step S150, the following steps are also included: S160, collect motion data of new subject x new , the motion data of the new subject is normalized to obtain new personal weighted features Finally, the new personal weighted features Input into the preset LightGBM model and output the new BMI prediction result BMI prediction method based on convolutional neural network Technical Field The present invention relates to the technical field of health informatics, and in particular to a BMI prediction method based on a convolutional neural network. Background Art BMI (Body Mass Index) is an important standard commonly used internationally to measure the degree of human obesity and health. The reasonable range of BMI is usually closely related to the health status of an individual, and can effectively reflect whether the weight is normal or whether there are health problems such as obesity. Therefore, BMI prediction has important application value in individual health monitoring and fitness management. At present, traditional BMI prediction methods mainly rely on factors such as physical examination data, diet and exercise habits, but these methods are often too simple and ignore the comprehensive impact of different exercise methods on BMI, thereby reducing the accuracy of BMI prediction. Therefore, it is urgent to propose a BMI prediction method based on convolutional neural network to solve the problem raised. Summary of the invention Based on this, it is necessary to provide a BMI prediction method based on convolutional neural networks to address the shortcomings of the existing technology, combine the attention mechanism to process the preset sports data, and improve the accuracy and stability of BMI prediction. In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides a BMI prediction method based on a convolutional neural network, which comprises the following steps: S110: Collect the subject's preset sports data and the corresponding BMI classification results to construct a sports data set in, represents the eigenvalue corresponding to the preset sports test data of the i-th subject, y i represents the BMI classification result corresponding to the i-th subject; S120, for motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}; S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j ; Where 1≤j≤m, m represents the total number of sports types; S140, introduce the attention mechanism, and obtain the personal weighted features according to the contribution weight of the sports data corresponding to different types of preset sports in BMI prediction S150, weighting individual features Input into the preset LightGBM model to obtain the BMI prediction result Furthermore, in step S110, the preset sports data of the subject and the corresponding BMI classification results are collected to construct a sports data set The method comprises the following steps: Collect the physical test results corresponding to the subjects' preset sports data Collect the BMI classification results y corresponding to the subjects i ; BMI classification includes four categories: underweight A, normal weight B, overweight C, and obese D. Building a motion dataset Furthermore, in step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps: When m=1, obtain the physical test results corresponding to the subject in a single sport And the corresponding BMI classification result y i , then the corresponding feature set under this kind of sports Among them, m represents the total number of sports types, N represents the total number of subjects, and a single sport is one of the preset sports; By Z-score normalization formula The corresponding feature set for this sport Processing is performed to obtain the physical test results corresponding to the subjects in this sport The corresponding standard feature x i , combined with the corresponding BMI classification result y i , and obtain the standard data set D of the subjects in this sport = {(x i ,y i )}. Furthermore, in step S120, the motion data set Perform standardization to obtain the standard data set D = {(x i ,y i )}, comprising the following steps: When m≥2, obtain the corresponding physical test results of the subjects under various preset sports items And the corresponding BMI classification result y i , then the corresponding feature sets under different preset sports items Among them, m represents the total number of sports types, and N represents the total number of subjects; By Z-score normalization formula Corresponding feature sets for different preset sports Processing is performed to obtain the corresponding physical test results of the subjects under each preset sports event The corresponding standard feature x i , combined with the corresponding BMI classification result yi, we can then obtain the standard data set D = {(x i ,y i )}. Furthermore, in step S130, based on the convolutional neural network, the standard data set D = {(x i ,y i )} in the standard feature x i Perform feature extraction to obtain the output feature h corresponding to each preset sports item j The method comprises the following steps: Through a filter w, convolution operation is performed with the time series data x(t) to generate feature y(t). Where x(t)) is the motion data x i Corresponding time series data, w(k) is the parameter of the convolution kernel, b is the bias term, and K is the length of the filter; The output of each convolutional layer is nonlinearly transformed by the activation function ReLU to obtain the linearly transformed feature ReLU(y(t)), ReLU(y(t))=max(0,y(t)); Perform the maximum pooling operation and select the motion data x in the pooling window i The maximum value of the corresponding time series data x(t) MaxPooling(x(t)), MaxPooling(x(t)) = max{x(t), x(t+1), ..., x(t+K-1)}, obtain the output feature h corresponding to each preset sports item j ,h j =f CNN (x i , W CNN ), where h j represents the output feature of the jth preset sports item, f CNN Function represents the processing of the entire convolutional network, W CNN are the trainable parameters of the convolutional network. Furthermore, in step S140, the attention mechanism is introduced to obtain the personal weighted features according to the contribution weights of the sports data corresponding to different types of preset sports in the BMI prediction. The method comprises the following steps: Get the motion data feature set H of each subject, H = {h1, h2, ..., h j }, h j is the output feature of the jth preset sports event, 1≤j≤m, m is the total number of preset sports events; Obtain the output feature h through the fully connected layer j The corresponding preset sports item weight parameter α in BMI prediction j , Among them, f(h j ) is the Softmax function; By formula For the output feature h j Perform weighted calculation to obtain weighted features Then all weighted features Sum and get the individual weighted features Furthermore, in step S150, the personal weighted features Input into the preset LightGBM model to obtain the BMI prediction result The method comprises the following steps: Divide the dataset into training, validation and test sets; Build multiple decision trees and input individual weighted features through the training set Use the preset LightGBM model for training; In the preset LightGBM model, the input features of the preset LightGBM model are: The output features of the preset LightGBM model are BMI prediction results satisfy: Among them, f q () is the prediction function of the qth decision tree, W LightGBM is the parameter of the LightGBM model, and Q is the total number of decision trees. Furthermore, the loss function L used for training using the preset LightGBM model satisfies: Among them, l() is the loss function and Ω() is the regularization term.

Images