WO2020103417A1

WO2020103417A1 - Bmi evaluation method and device, and computer readable storage medium

Info

Publication number: WO2020103417A1
Application number: PCT/CN2019/088637
Authority: WO
Inventors: 石磊; 马进; 王健宗; 肖京
Original assignee: 平安科技（深圳）有限公司
Priority date: 2018-11-20
Filing date: 2019-05-27
Publication date: 2020-05-28
Also published as: CN109637664A

Abstract

The present invention relates to the technical field of intelligent decision making, and provides a BMI evaluation method and device, and a computer readable storage medium. The method comprises: collecting sample face image data, the sample face image data comprising a BMI value (S10); training the sample face image data by adopting a convolutional neural network to obtain a training model (S20); obtaining a face image to be detected, positioning face key feature points of said face image, and obtaining the face key points (S30); performing face contour extraction according to the face key points to obtain a face contour (S40); performing equal-proportion stretching on the face contour according to a perspective view angle to obtain a preprocessed face image (S50); and performing category prediction on the preprocessed face image according to the training model to obtain an evaluation BMI value of said face image (S60). By means of the method, the BMI value can be rapidly detected in real time, and the BMI measurement difficulty of a measurer is reduced.

Description

BMI evaluation method, device and computer readable storage medium

This application requires the priority of the Chinese patent application submitted to the China Patent Office on November 20, 2018, with the application number 201811384493.8 and the invention titled "A BMI evaluation method, device, and computer-readable storage medium." The reference is incorporated in this application.

Technical field

The present application relates to the field of intelligent decision-making technology, and in particular, to a BMI evaluation method, device, and computer-readable storage medium.

Background technique

The BMI index (Body Mass Index, referred to as Body Mass Index, also known as Body Mass Index, in English Body Body Mass Index, referred to as BMI) is a parameter related to height and weight that can reflect the body mass index / obesity index, which is divided by the weight in kilograms divided by The number is derived from the square of height in meters. It is mainly used for statistics. When it is necessary to compare and analyze the health impact of a person ’s weight on people of different heights, the BMI value is a neutral and reliable indicator, which is commonly used in the world to measure the degree of body fatness and health A standard. BMI is an indicator closely related to total body fat, which takes into account two factors of weight and height. BMI is simple, practical, and can reflect systemic overweight and obesity. When measuring the body's risk of heart disease and high blood pressure due to being overweight, it is more accurate than simply weight. At present, there are two commonly used calculation methods for evaluating BMI: one is: adult: [height (cm) -100] × 0.9 = standard weight (kg) and the other is: male: height (cm) -105 = standard Weight (kg), female: height (cm)-100 = standard weight (kg).

The traditional BMI acquisition method needs to first measure the height and weight information of the person to be measured. This information needs to be measured using a height and weight tester or other measurement sensors, and then numerically calculated to obtain the final BMI index. Instruments (usually such instruments are not easy to carry) are used for measurement, and the use of procedures is cumbersome, time-consuming, and slow, and the measurement is slow. Effective measurement in real time, quickly and conveniently.

Summary of the invention

The present application provides a BMI evaluation method, device, and computer-readable storage medium. Its main purpose is to quickly detect BMI values in real time and reduce the difficulty of measuring BMI by a surveyor.

To achieve the above purpose, the present application also provides a BMI evaluation method, which includes:

Collect sample face image data, the sample face image data includes the BMI value;

Use convolutional neural network to train the sample face image data to obtain the training model;

Acquiring a face image to be detected, locating key feature points of the face in the face image to be detected, and acquiring key points of the face;

According to the key points of the face, extract the face contour to obtain the face contour;

Stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image;

Perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.

An embodiment of the present application further includes a BMI evaluation device. The device includes a memory and a processor. The memory stores a BMI evaluation program that can run on the processor. The BMI evaluation program is used by the processor. The following steps are implemented during execution:

An embodiment of the present application also provides a computer-readable storage medium, which stores a BMI evaluation program, which can be executed by one or more processors to implement the method described above step.

The BMI evaluation method, device and computer-readable storage medium proposed in this application collect the face image and BMI value in advance, train the network model through a learning algorithm, and predict the BMI value of the face image to be detected according to the trained network model. BMI detection can be completed automatically without complicated measurement equipment, which not only greatly reduces the measurement difficulty of BMI measurers, but also enables real-time measurement for the measurers to monitor their own health status.

BRIEF DESCRIPTION

1 is a schematic flowchart of a BMI evaluation method provided by an embodiment of this application;

FIG. 2 is a schematic diagram of a human face labeled with facial feature points according to an embodiment of the present application;

3 is an effect diagram of edge contour extraction using a sobel operator provided by an embodiment of the present application;

4 is a schematic diagram of a bilinear interpolation model provided by an embodiment of this application;

5 is a schematic diagram of an internal structure of a BMI evaluation device provided by an embodiment of the present application;

6 is a schematic block diagram of a BMI evaluation program in a BMI evaluation device provided by an embodiment of the present application.

The implementation, functional characteristics and advantages of the present application will be further described in conjunction with the embodiments and with reference to the drawings.

detailed description

It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

This application provides a BMI evaluation method. Referring to FIG. 1, FIG. 1 is a schematic flowchart of a BMI evaluation method provided by an embodiment of the present application. The method may be executed by an apparatus, and the apparatus may be implemented by software and / or hardware.

In this embodiment, the BMI evaluation method includes:

Step S10: Collect sample face image data, the sample face image data includes a BMI value.

In step S20, a convolutional neural network is used to train sample face image data to obtain a training model.

Specifically, the online server can collect 10,000 sample face images with BMI values, and train the convolutional neural network under the caffe deep learning framework.

The training of the convolutional neural network includes: first, uniformly cropping the face image into an image with a size of 224 * 224, and then converting into a uniform leveldb format, and finally training the convolutional neural network VGG-16 with these images. The convolutional neural network used includes 1 data input layer, 13 convolutional layers, and 3 fully connected layers. The number of convolution kernels of the 13 convolutional layers is 64, 64, 128, 128, 256, 256, 256, 512, 512, 512, 512, 512, and 512, respectively. Between the second convolution layer and the third convolution layer, between the fourth convolution layer and the fifth convolution layer, between the seventh convolution layer and the eighth convolution layer, the fourth One convolutional layer and the fifth convolutional layer, the tenth convolutional layer and the eleventh convolutional layer, the thirteenth convolutional layer and the first fully connected layer are all connected Pooling layer, the above 13 convolutional layers and 3 fully connected layers are all processed with ReLU (Nonlinear Activation Function). The last layer of the VGG-16 network model is removed and retrained, and finally the softmax function is used to output the probability values of three categories of values: extremely low BMI, normal BMI, and extremely high BMI. The output value is the category value corresponding to the maximum probability value. The softmax function can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ (z), so that the range of each element is between (0, 1), and the sum of all elements Is 1. For example, if the input vector [BMI is extremely low, BMI is normal, and BMI is extremely high], the corresponding Softmax function value is [0.2, 0.5, 0.3], then the item with the largest weight in the output vector corresponds to the maximum value in the input vector "BMI normal".

Step S30: Acquire a face image to be detected, locate key feature points of the face in the face image to be detected, and obtain key points of the face.

Specifically, the image collection unit of the offline terminal, such as the camera of the mobile phone, can collect the face image of the user to be detected, and send the face image to the online server through the network transmission unit, and simultaneously send the image The camera parameters of the collection unit are also sent to the online server, and the effect image of the collected face image is displayed synchronously on the mobile phone.

Specifically, the method for locating the key feature points of the face in the face image is as follows. Optionally, in this embodiment, an active shape model (Active Shape Model, ASM) algorithm is used to locate key feature points of the face.

The basic idea of the ASM algorithm is to combine the texture features of human faces with the position constraints between each feature point. The ASM algorithm is divided into training and searching steps. During training, the position constraints of each feature point are established, and the local features of each specific point are constructed. When searching, match iteratively.

The training steps of ASM are as follows: First, build a shape model: collect n training samples of face (n = 400); manually mark facial feature points, as shown in Figure 2, and Figure 2 is the face labeled with facial feature points Schematic diagram; the coordinate string of the feature points in the training set is transformed into a feature vector; the shape is normalized and aligned (alignment uses Procrustes method); PCA processing is performed on the aligned shape features. The basic principle of the PCA processing is as follows: m pieces of n-dimensional data are provided, 1) the original data is composed of n rows and m columns of matrix X; 2) each row of X (representing an attribute field) is zero-averaged, That is, subtract the mean of this row; 3) find the covariance matrix; 4) find the eigenvalues of the covariance matrix and the corresponding eigenvectors; 5) arrange the eigenvectors in rows from top to bottom according to the size of the corresponding eigenvalues Form a matrix, take the first k rows to form a matrix P; 6) is the data after dimension reduction to k dimension.

Next, construct local features for each feature point. The purpose is that each feature point can find a new location during each iteration of the search process. Gradient features are generally used for local features to prevent changes in illumination. Some methods extract along the normal direction of the edge, and some methods extract in the rectangular area near the feature point.

Then perform the ASM search steps as follows: first, calculate the position of the eyes (or eyes and mouth), make simple scale and rotation changes, and align the face; then, match each local feature point (often using Mahalanobis distance) , Calculate the new position; get the parameters of the affine transformation, and iterate until convergence. In addition, multi-scale methods are often used to accelerate. The search process eventually converges to the original high-resolution image.

Step S40: Perform face contour extraction based on key points of the face to obtain the face contour.

After obtaining the key points of the face that can identify the eyebrows and jaws, determine the relative coordinates of the eyes, nose and mouth based on this, and then extract the face contour. Optionally, the sobel operator is used for face contour extraction, and the background outside the face area is removed. Sobel operator is a discrete differential operator, which combines Gaussian smoothing and differential differentiation to calculate the approximate gradient of the image gray function. The basic principle is to convolve the incoming image pixels. The essence of convolution is to find the gradient value, or to give a weighted average, where the weight is the so-called convolution kernel; Threshold value is used to determine the edge information. At the edge of the image, the pixel value will change significantly. One way to express this change is to use derivatives. A large change in gradient value indicates a significant change in the content of the image. If G _x is the convolution of the original image in the x direction, and G _y is the convolution of the original image in the y direction, the pixel value of the action point in the original image after passing the convolution is:

After obtaining the new pixel value of the pixel, given a threshold, the image edge calculated by the sobel operator can be obtained. As shown in Fig. 3, Fig. 3 is an effect diagram of edge contour extraction using the sobel operator.

In step S50, the contour of the human face is stretched in proportion to the perspective perspective to obtain the pre-processed human face image.

Optionally, a two-dimensional linear interpolation algorithm is used to proportionally stretch the contour of the human face according to the perspective. Assume that the source image size is m × n and the target image is a × b. Then the side length ratios of the two images are: m / a and n / b. Note that usually this ratio is not an integer. When programming storage, use floating point. The (i, j) th pixel (i row and j column) of the target image can be returned to the source image by the side length ratio. The corresponding coordinates are (i × m / a, j × n / b). Obviously, this corresponding coordinate is generally not an integer, and non-integer coordinates cannot be used on discrete data such as images. Bilinear interpolation calculates the value (gray value or RGB value) of this point by looking for the four pixels closest to the corresponding coordinate. If the image is a gray-scale image, then the mathematical calculation model of the gray value of (i, j) point is: f (x, y) = b ₁ + b ₂ x + b ₃ y + b ₄ xy. Where b ₁ , b ₂ , b ₃ , and b ₄ are related coefficients. The calculation process is as follows: As shown in Figure 4, Figure 4 is a schematic diagram of a bilinear interpolation model. Q ₁₂ , Q ₂₂ , Q ₁₁ and Q _{21 are known} , but the point to be interpolated is point P, which requires bilinear interpolation. First, in the x-axis direction, perform two points on R ₁ and R ₂ Interpolate, and then interpolate P points according to R ₁ and R ₂ , which is bilinear interpolation.

Optionally, while stretching the face contour, it also includes the RGB chroma component of each pixel of the face contour according to the camera parameters, chromatic correction and brightness correction point by point to reduce the image acquisition environment The effects of light, camera parameters, etc. Face image after contour extraction and distortion correction,

Step S60: Perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.

Optionally, the evaluation BMI value of the face image may be extremely low BMI, normal BMI or extremely high BMI, and the evaluation BMI value is returned to the offline terminal in real time, such as a smart phone, a tablet computer, or a portable computer.

The BMI evaluation method proposed in this embodiment, further, in another embodiment of the method of the present application, the method further includes the following steps after step S60:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the learning model.

Specifically, after obtaining the evaluation BMI value of the learning model evaluation, the user may choose to upload the actual measured BMI value of the user to perform fine-tuning training on the learning model to achieve rapid update iteration of the model. When fine-tuning the learning model, taking the aforementioned 10,000 face sample as an example, the learning rate of the first 8000 small batch samples is set to 0.001, the learning rate of the last 2,000 small batch samples is set to 0.0001, and the small batch size of each iteration is 300, the momentum value is set to 0.9, and the weight attenuation value is 0.0005.

For different users, according to the measured BMI value uploaded by them, increase the weight of their face pictures in the training process to enhance the generalization of the learning model and better match the actual physical condition of the user.

This application also provides a BMI evaluation device. Referring to FIG. 2, it is a schematic diagram of the internal structure of the device provided by an embodiment of the present application.

In this embodiment, the device 1 may be a personal computer (Personal Computer, PC), or may be a terminal device such as a smartphone, tablet computer, or portable computer. The BMI evaluation device 1 includes at least a memory 11, a processor 12, a communication bus 13, and a network interface 14.

Wherein, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, and the like. In some embodiments, the memory 11 may be an internal storage unit of the BMI evaluation device 1, such as a hard disk of the BMI evaluation device 1. In other embodiments, the memory 11 may also be an external storage device of the BMI evaluation device 1, for example, a plug-in hard disk equipped on the BMI evaluation device 1, a smart memory card (Smart Media, Card, SMC), and a secure digital (Secure Digital, SD) card, flash card (Flash Card), etc. Further, the memory 11 may also include both the internal storage unit of the BMI evaluation device 1 and the external storage device. The memory 11 can be used not only to store application software and various types of data installed in the BMI evaluation device 1, such as codes of the BMI evaluation program 01, but also to temporarily store data that has been or will be output.

In some embodiments, the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip for running the program code or processing stored in the memory 11 Data, for example, the BMI evaluation program 01 is executed.

The communication bus 13 is used to realize connection and communication between these components.

The network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is generally used to establish a communication connection between the device 1 and other electronic devices.

Optionally, the device 1 may further include a user interface. The user interface may include a display and an input unit such as a keyboard. The optional user interface may also include a standard wired interface and a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an organic light-emitting diode (OLED) touch device, or the like. Among them, the display may also be appropriately referred to as a display screen or a display unit, for displaying information processed in the BMI evaluation device 1 and for displaying a visual user interface.

FIG. 2 only shows the BMI evaluation device 1 having the components 11-14 and the BMI evaluation program 01. Those skilled in the art can understand that the structure shown in FIG. 1 does not constitute a limitation on the BMI evaluation device 1, and may include There are fewer or more components than shown, or some components are combined, or different components are arranged.

In the embodiment of the device 1 shown in FIG. 2, the BMI evaluation program 01 is stored in the memory 11; the processor 12 implements the following steps when executing the BMI evaluation program 01 stored in the memory 11:

The training of the convolutional neural network includes: first, uniformly cropping the face image into an image with a size of 224 * 224, and then converting into a uniform leveldb format, and finally training the convolutional neural network VGG-16 with these images. The convolutional neural network used includes 1 data input layer, 13 convolutional layers, and 3 fully connected layers. The number of convolution kernels of the 13 convolutional layers is 64, 64, 128, 128, 256, 256, 256, 512, 512, 512, 512, 512, and 512, respectively. Between the second convolution layer and the third convolution layer, between the fourth convolution layer and the fifth convolution layer, between the seventh convolution layer and the eighth convolution layer, the fourth Between the convolutional layer and the fifth convolutional layer, between the tenth convolutional layer and the eleventh convolutional layer, and between the thirteenth convolutional layer and the first fully connected layer Pooling layer, the above 13 convolutional layers and 3 fully connected layers are all processed with ReLU (Nonlinear Activation Function). The last layer of the VGG-16 network model is removed and retrained, and finally the softmax function is used to output the probability values of three categories of values: extremely low BMI, normal BMI, and extremely high BMI. The output value is the category value corresponding to the maximum probability value. The softmax function can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ (z), so that the range of each element is between (0, 1), and the sum of all elements Is 1. For example, if the input vector [BMI is extremely low, BMI is normal, and BMI is extremely high], the corresponding Softmax function value is [0.2, 0.5, 0.3], then the item with the largest weight in the output vector corresponds to the maximum value in the input vector "BMI normal".

The training steps of ASM are as follows: First, build a shape model: collect n training samples of face (n = 400); manually mark facial feature points, as shown in Figure 2, and Figure 2 is the face labeled with facial feature points Schematic diagram; the coordinate string of the feature points in the training set is transformed into a feature vector; the shape is normalized and aligned (alignment uses Procrustes method); PCA processing is performed on the aligned shape features. The basic principle of the PCA processing is as follows: m pieces of n-dimensional data are provided, 1) the original data is composed of n rows and m columns of matrix X; 2) each row of X (representing an attribute field) is zero-averaged, That is, subtract the mean of this row; 3) find the covariance matrix; 4) find the eigenvalues of the covariance matrix and the corresponding eigenvectors; 5) arrange the eigenvectors in rows from top to bottom according to the corresponding eigenvalues Form a matrix, take the first k rows to form a matrix P; 6) is the data after dimension reduction to k dimension.

Optionally, the evaluation BMI value of the face image may be extremely low BMI, normal BMI, or extremely high BMI, and the evaluation BMI value is returned to an offline terminal in real time, such as a smartphone, tablet computer, or portable computer.

Obtain the measured BMI value uploaded by the user;

Update and iterate the learning model.

Optionally, in other embodiments, the BMI evaluation program may also be divided into one or more modules, and the one or more modules are stored in the memory 11 and are processed by one or more processors (in this embodiment, processing 12) is executed to complete this application. The module referred to in this application refers to a series of computer program instruction segments capable of performing specific functions, and is used to describe the execution process of the BMI evaluation program in the BMI evaluation device.

For example, referring to FIG. 3, it is a schematic diagram of a program module of a BMI evaluation program in an embodiment of a BMI evaluation device of the present application. In this embodiment, the BMI evaluation program may be divided into a sample data collection module 10 and a sample data model training module 20. Face key point positioning module 30, face contour extraction module 40, face contour stretching module 50, and face image BMI value prediction module 60.

Exemplarily:

The sample data collection module 10 is used to: collect sample face image data, the sample face image data includes a BMI value;

The sample data model training module 20 is used to: train the sample face image data using a convolutional neural network to obtain the training model;

The face key point positioning module 30 is used to: acquire a face image to be detected, locate key feature points of the face to be detected, and obtain face key points;

The face contour extraction module 40 is used to: extract the face contour according to the key points of the face to obtain the face contour;

Face contour stretching module 50: used to perform proportional stretching on the face contour according to perspective perspective to obtain pre-processed face image;

Face image BMI value prediction module 60: used to perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.

The above-mentioned sample data acquisition module 10, sample data model training module 20, face key point positioning module 30, face contour extraction module 40, face contour stretching module 50, face image BMI value prediction module 60 and other program modules are executed The functions or operation steps implemented at the time are substantially the same as those in the above embodiment, and will not be repeated here.

In addition, an embodiment of the present application further proposes a computer-readable storage medium on which a BMI evaluation program is stored. The BMI evaluation program may be executed by one or more processors to implement the following operations:

Step S10: Collect sample face image data, the sample face image data includes a BMI value;

Step S20, the convolutional neural network is used to train the sample face image data to obtain the training model;

Step S30: Acquire a face image to be detected, locate key feature points of the face to be detected, and obtain key points of the face;

Step S40: Perform face contour extraction based on key points of the face to obtain the face contour;

Step S50, the contour of the human face is stretched in proportion to the perspective perspective to obtain a pre-processed human face image;

The specific implementation of the computer-readable storage medium of the present application is basically the same as the above embodiments of the BMI evaluation device and method, and will not be described here.

It should be noted that the serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments. And the terms "include", "include", or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, device, article, or method that includes a series of elements includes not only those elements, but also unclear The other elements listed may also include elements inherent to this process, device, article, or method. Without more restrictions, the element defined by the sentence "include one ..." does not exclude that there are other identical elements in the process, device, article or method that includes the element.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods in the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware, but in many cases the former is better Implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence or part that contributes to the existing technology, and the computer software product is stored in a storage medium (such as ROM / RAM as described above) , Magnetic disks, optical disks), including several instructions to enable a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to perform the method described in each embodiment of the present application.

The above are only the preferred embodiments of the present application, and do not limit the scope of the patent of the present application. Any equivalent structure or equivalent process transformation made by the description and drawings of this application, or directly or indirectly used in other related technical fields The same reason is included in the patent protection scope of this application.

Claims

A BMI evaluation method, characterized in that the method includes:

Collect sample face image data, the sample face image data includes the BMI value;

Use convolutional neural network to train the sample face image data to obtain the training model;

Acquiring a face image to be detected, locating key feature points of the face in the face image to be detected, and acquiring key points of the face;

According to the key points of the face, extract the face contour to obtain the face contour;

Stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image;

Perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.
The BMI evaluation method according to claim 1, wherein the step of proportionally stretching the contour of the face according to perspective perspective to obtain a pre-processed face image, further comprising the steps of:

According to the camera parameters, the RGB chroma component of each pixel in the contour area of the human face is subjected to chroma correction and brightness correction point by point.
The BMI evaluation method according to claim 1, wherein the step uses a convolutional neural network to train sample face image data to obtain a training model, further comprising the steps of:

Cut the face image into an image of size 224 * 224;

Convert the cropped image to leveldb format;

Training the convolutional neural network VGG-16 with the image in the leveldb format;

The softmax function is used to output the probability values of three categories: BMI extremely low, BMI normal, and BMI extremely high. The output value is the category value corresponding to the maximum probability value.
The BMI evaluation method according to claim 1, wherein the step of obtaining a face image to be detected, locates key feature points of the face to be detected, and obtains key points of the face; according to the face The key point is to extract the face contour to obtain the face contour; perform the same proportion stretching on the face contour according to the perspective perspective to obtain the pre-processed face image; further including the steps:

Acquiring a face image to be detected, and using an active shape model algorithm to locate key feature points of the face to be detected, to obtain key points of the face;

According to the key points of the face, the sobel operator is used to extract the face contour, and the background outside the face area is removed to obtain the face contour;

The two-dimensional linear interpolation algorithm is used to proportionally stretch the contour of the human face according to a perspective perspective to obtain a pre-processed human face image.
The BMI evaluation method according to claim 1, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected, further Including steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
The BMI evaluation method according to claim 2, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected, further Including steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
The BMI evaluation method according to claim 3 or 4, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected , Also includes steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
A BMI evaluation device, characterized in that the device includes a memory and a processor, and the memory stores a BMI evaluation program that can be run on the processor, when the BMI evaluation program is executed by the processor Implement the following steps:

Collect sample face image data, the sample face image data includes the BMI value;

Use convolutional neural network to train the sample face image data to obtain the training model;

Acquiring a face image to be detected, locating key feature points of the face in the face image to be detected, and acquiring key points of the face;

According to the key points of the face, extract the face contour to obtain the face contour;

Stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image;

Perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.
The BMI evaluation device according to claim 8, characterized in that the step of proportionally stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image, further comprising the steps of:

According to the camera parameters, the RGB chroma component of each pixel in the contour area of the human face is subjected to chroma correction and brightness correction point by point.
The BMI evaluation device according to claim 8, characterized in that the step uses a convolutional neural network to train sample face image data to obtain a training model, further comprising the steps of:

Cut the face image into an image of size 224 * 224;

Convert the cropped image to leveldb format;

Training the convolutional neural network VGG-16 with the image in the leveldb format;

The softmax function is used to output the probability values of three categories: BMI extremely low, BMI normal, and BMI extremely high. The output value is the category value corresponding to the maximum probability value.
The BMI evaluation device according to claim 8, wherein the step of obtaining a face image to be detected, locates key feature points of the face to be detected, and obtains key points of the face; according to the face The key point is to extract the face contour to obtain the face contour; perform the same proportion stretching on the face contour according to the perspective perspective to obtain the pre-processed face image; further including the steps:

Acquiring a face image to be detected, and using an active shape model algorithm to locate key feature points of the face to be detected, to obtain key points of the face;

According to the key points of the face, the sobel operator is used to extract the face contour, and the background outside the face area is removed to obtain the face contour;

The two-dimensional linear interpolation algorithm is used to proportionally stretch the contour of the human face according to a perspective perspective to obtain a pre-processed human face image.
The BMI evaluation device according to claim 8, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected, further Including steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
The BMI evaluation device according to claim 9, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected, further Including steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
The BMI evaluation device according to claim 10 or 11, wherein after the step of performing category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected , Also includes steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
A computer-readable storage medium, characterized in that a BMI evaluation program is stored on the computer-readable storage medium, and the program can be executed by one or more processors to implement the following steps:

Collect sample face image data, the sample face image data includes the BMI value;

Use convolutional neural network to train the sample face image data to obtain the training model;

Acquiring a face image to be detected, locating key feature points of the face in the face image to be detected, and acquiring key points of the face;

According to the key points of the face, extract the face contour to obtain the face contour;

Stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image;

Perform category prediction on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected.
The computer-readable storage medium according to claim 15, wherein the step of proportionally stretching the contour of the face according to a perspective perspective to obtain a pre-processed face image, further comprising the steps of:

According to the camera parameters, the RGB chroma component of each pixel in the contour area of the human face is subjected to chroma correction and brightness correction point by point.
The computer-readable storage medium according to claim 15, wherein the step uses a convolutional neural network to train the sample face image data to obtain the training model, further comprising the steps of:

Cut the face image into an image of size 224 * 224;

Convert the cropped image to leveldb format;

Training the convolutional neural network VGG-16 with the image in the leveldb format;

The softmax function is used to output the probability values of three categories: BMI extremely low, BMI normal, and BMI extremely high. The output value is the category value corresponding to the maximum probability value.
The computer-readable storage medium according to claim 15, wherein the step of obtaining a face image to be detected, locates key feature points of the face of the face image to be detected, and obtains key points of the face; Face key points, extract the face contour to obtain the face contour; stretch the face contour according to the perspective perspective to obtain the pre-processed face image; further include the steps of:

Acquiring a face image to be detected, and using an active shape model algorithm to locate key feature points of the face to be detected, to obtain key points of the face;

According to the key points of the face, the sobel operator is used to extract the face contour, and the background outside the face area is removed to obtain the face contour;

The two-dimensional linear interpolation algorithm is used to proportionally stretch the contour of the human face according to a perspective perspective to obtain a pre-processed human face image.
The computer-readable storage medium according to claim 15 or 16, wherein the category prediction is performed on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected After the steps, there are also steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.
The computer-readable storage medium according to claim 17 or 18, wherein the class prediction is performed on the pre-processed face image according to the training model to obtain the evaluation BMI value of the face image to be detected After the steps, there are also steps:

Obtain the measured BMI value uploaded by the user;

Use the convolutional network model to fine-tune the evaluation BMI value and the measured BMI value;

Update and iterate the training model.