WO2021114623A1 - Method, apparatus, computer device, and storage medium for identifying persons having deformed spinal columns - Google Patents

Abstract

Provided are a method, apparatus, computer device, and storage medium for identifying persons having deformed spinal columns, relating to the field of artificial intelligence, said method comprising: obtaining image data and non-image data associated with a target to be identified; extracting an image of a back area to be recognized in a back area in the image data; performing image enhancement processing on the image of the back area to be recognized to obtain an enhanced image of the area to be recognized; extracting the features of the spinal column in the enhanced image of the area to be recognized by means of a spinal column recognition model to obtain a first feature vector map, and performing normalization and edge weight processing of non-image data by means of a data standardization model to obtain a second feature vector map; edge-filling the second feature vector map to the first feature vector map to obtain a third feature vector map; according to the spectral domain method in GCN, extracting the frequency domain features of the spinal column by means of a spinal column image convolutional network model to obtain an identification result. The method can automatically identify the type of persons having a deformed spinal column.

Description

Method, device, computer equipment and storage medium for identifying people with deformed spine

This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on June 8, 2020, with the application number 202010513066.6, and the invention title "Methods, devices, computer equipment and storage media for identifying people with spinal deformities", all of which are approved The reference is incorporated in this application.

Technical field

This application relates to the field of artificial intelligence image classification, and in particular to a method, device, computer equipment and storage medium for identifying people with deformed spine.

Background technique

The spine is the central axis of the human body. When the spine is severely deformed, it will not only cause abnormal body appearance and motor dysfunction, but also cause cardiopulmonary dysfunction due to thoracic deformities, reduce the quality of life, and seriously affect the physical and mental health of young people. If it is not prevented and detected early, it will not only affect the patient's body shape and appearance, but also may cause abnormal heart and lung function, premature degeneration of the spine, pain, unbalanced trunk, and even death. Spine deformation refers to the horizontal and vertical deformation of the spine. It is usually said that the horizontal deformation becomes scoliosis and the vertical deformation becomes kyphosis.

There are many methods to check spinal deformation, mainly including Moiré image measurement method, X-ray film measurement method, Adams forward bending test, etc. The inventor realized that the existing technical solutions require manual physical measurement and detection, and The detection steps are cumbersome, resulting in low detection efficiency and high cost. In particular, X-ray film measurement can cause radiation damage to teenagers. Most of the existing solutions can only detect the patient population after the spinal deformation has occurred. Remind and prevent the crowd.

Summary of the invention

This application provides a method, device, computer equipment, and storage medium for identifying people with spinal deformity, which realizes the automatic identification of the types of people with spinal deformity through the spectral domain method in GCN, and reminds potential people to play a preventive role. Therefore, , Improve the accuracy and reliability of the identification of people with deformed spine, greatly reduce the cost of identification, and serve as a reminder to potential people.

A method for identifying people with deformed spine, including:

Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

Input the image data into a back region recognition model, and perform back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

Input the third feature vector map into the trained spine map convolutional network model;

According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.

A device for identifying people with deformed spine, including:

The receiving module is used to receive a target recognition instruction, and obtain image data and non-image data associated with the unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is related to the target to be recognized Information;

A recognition module, configured to input the image data into a back area recognition model, perform back area recognition on the image data through the back area recognition model, and obtain an image of the back area to be recognized that is intercepted by the back area recognition model; The back region recognition model is a deep convolutional neural network model based on the YOLO model;

An enhancement module, configured to perform image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

The acquiring module is configured to input the enhanced image of the region to be recognized into a spine recognition model, extract the features of the spine in the enhanced image of the region to be identified through the spine recognition model, and obtain the output of the spine recognition model according to the features of the spine A first feature vector map, and at the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

A filling module, configured to perform edge filling of the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

An input module, configured to input the third feature vector graph into the trained spine graph convolutional network model;

The output module is used to extract the frequency domain features of the spine in the third feature vector diagram through the spine map convolutional network model according to the spectral domain method in the GCN, and obtain the spine map convolutional network model according to the spine Frequency domain feature output recognition result; the recognition result characterizes the category of the spine deformed population of the target to be recognized, and the category of the spine deformed population includes the scoliosis population, the kyphotic population, the potential scoliosis population, and the potential spine People with hunchbacks and non-spine deformities.

A computer device includes a memory, a processor, and a computer program that is stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer program:

Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

Input the image data into the back region recognition model, and perform the back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

Input the third feature vector map into the trained spine map convolutional network model;

According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.

A computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

Input the image data into the back region recognition model, and perform the back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

Input the third feature vector map into the trained spine map convolutional network model;

According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.

This application realizes that according to the captured back image and related non-image information of the target to be recognized, the category of the spine deformed crowd corresponding to the target to be recognized (including potential scoliosis crowds, scoliosis crowds, etc.) can be automatically identified through the spectral domain method in the GCN. People with kyphosis, people with potential scoliosis, people with potential kyphosis, and people without spine deformation) can quickly and accurately identify the category of people with spine deformation corresponding to the target to be identified, and remind potential people to play a preventive role Therefore, the accuracy and reliability of the identification of people with deformed spine are improved, the identification cost is greatly reduced, and the potential crowd is reminded.

Description of the drawings

FIG. 1 is a schematic diagram of the application environment of the method for identifying people with deformed spine in an embodiment of the present application;

2 is a flowchart of a method for identifying people with deformed spine in an embodiment of the present application;

3 is a flowchart of step S20 of the method for identifying people with deformed spine in an embodiment of the present application;

4 is a flowchart of step S30 of the method for identifying people with deformed spine in an embodiment of the present application;

5 is a flowchart of step S40 of the method for identifying people with deformed spine in an embodiment of the present application;

6 is a flowchart of step S40 of the method for identifying people with deformed spine in another embodiment of the present application;

FIG. 7 is a flowchart of step S60 of the method for identifying people with deformed spine in an embodiment of the present application;

8 is a schematic block diagram of a device for identifying people with deformed spine in an embodiment of the present application;

Fig. 9 is a schematic diagram of a computer device in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

The technical solution of the present application can be applied to the fields of artificial intelligence, smart city and/or digital medical technology, and can automatically identify people with deformed spine and potential people, and perform disease risk assessment, so as to realize health management and realize smart medical treatment.

The method for identifying people with deformed spine provided in this application can be applied in an application environment as shown in Fig. 1, where a client (computer device) communicates with a server through a network. Among them, the client (computer equipment) includes, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, cameras, and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers.

In one embodiment, as shown in FIG. 2, a method for identifying people with deformed spine is provided, and the technical solution mainly includes the following steps S10-S70:

S10: Receive a target recognition instruction, and obtain image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; and the non-image data is information related to the target to be recognized.

Understandably, the target recognition instruction is an instruction that is triggered when the target to be recognized needs to be recognized, the target to be recognized is a person who needs to recognize whether the spine is deformed, and the unique code is the only one of the target to be recognized. The image data is a photo or image related to the back of the target to be recognized, and the non-image data is information related to the target to be recognized, for example: the gender and age of the target to be recognized And occupations and other related information.

S20: Input the image data into a back area recognition model, and perform back area recognition on the image data through the back area recognition model, and obtain an image of the back area to be recognized that is intercepted by the back area recognition model; the back The region recognition model is a deep convolutional neural network model based on the YOLO model.

Understandably, the back area recognition model is a trained deep convolutional neural network model based on the YOLO model, and the back area recognition model is a model that recognizes and locates the target to be recognized in the image data. The model of the back area, the back area recognition model uses the YOLO (You Only Look Once) algorithm to identify the area of the back of the target to be recognized, and the YOLO algorithm uses a CNN (Convolutional Neural Networks, convolutional neural network) The algorithm directly predicts the categories and regions of different targets. The network structure of the YOLO model can be selected according to requirements. For example, the network structure of the YOLO model can be YOLO V1, YOLO V2, YOLO V3, or YOLO V4, etc. The image of the back area to be recognized is an image obtained by intercepting the recognized area after recognition by the back area recognition model. In this way, the back area recognition model can extract only the image of the effective area in the image data to remove interference Image information.

In an embodiment, as shown in FIG. 3, in the step S20, the image data is identified by the back region through the back region recognition model, and the back region to be identified extracted by the back region recognition model is obtained. Area image, including:

S201. Input the back image of the target back into the back back area recognition model in the back area recognition model, and input the back side image of the target into the back side area recognition model in the back area recognition model; the image data includes the An image of the back of the target and a side image of the back of the target.

Understandably, the back area recognition model includes the back back area recognition model and the back side area recognition model, and the image data includes the target back back image and the target back side image. The region recognition model is a trained deep convolutional neural network model based on the YOLO model, and is a model for identifying and locating the back area. The back side area recognition model is the depth of the training based on the YOLO model. The convolutional neural network model is a model that recognizes and locates the back side area. The back image of the target is a photo of the back back when the target to be recognized is wearing underwear or a naked back. The back side image of the target To take a photo of the back side of the target to be identified wearing underwear or naked back.

S202. According to the YOLO algorithm, perform recognition through the back back area recognition model, cut out the back back area image to be recognized that contains only the back back area of the target to be recognized, and at the same time perform recognition through the back side area recognition model, and intercept An image of the back side area to be recognized that only contains the back side of the target to be recognized is generated.

Understandably, according to the YOLO algorithm, the back key points in the back image of the target back are identified through the back back area recognition model. The back key points include the left back shoulder point, the right back shoulder point, the upper point of the spine, The midpoint of the spine, the lower point of the spine, the left waist point and the right waist point, locate the position and area of the back according to the identified key points of the back, and intercept the position and area of the located back to obtain the The image of the back area to be identified, the image of the back area to be identified is the back of the back that contains only the object to be identified, and does not include the neck and arms of the object to be identified; according to the YOLO algorithm, through the back side The region recognition model recognizes the side key points in the back side image of the target. The side key points include the upper point of the arm, the middle point of the arm, the lower point of the arm, the neck point, the upper point of the lateral back, the midpoint of the lateral back, and the side The lower point of the back and the front chest point, locate the position and area of the side according to the identified key points of the side, and intercept the position and area of the positioned side, so as to obtain the image of the side area of the back to be identified. The image of the back side area to be recognized is the back side only containing the target to be recognized.

S203: Determine the image of the back area to be recognized and the image of the back side area to be recognized as the image of the back area to be recognized.

Understandably, the image of the back area to be identified and the image of the side area of the back to be identified are marked as the image of the back area to be identified.

This application intercepts the back area image and the side area image of the back of the target to be identified, and can pass the dimensions corresponding to the back and the side of the back, that is, the back corresponds to the horizontal dimension and the side corresponds to the vertical dimension. Intercepting images in multiple dimensions provides an effective image for improving the accuracy and reliability of recognition, and improves the efficiency of recognition.

S30: Perform image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified.

Understandably, the image enhancement processing refers to image processing operations such as gray-scale processing, noise removal, and edge enhancement of the back image to be recognized, wherein the gray-scale processing is to process the three colors in the color image ( The brightness of the red, green, and blue) components is used as the gray value of the gray image. The noise removal algorithm can be selected according to requirements. For example, the noise removal algorithm can be selected as spatial domain filtering algorithm, transform domain filtering algorithm, and partial Differential equation algorithm, variational algorithm, morphological noise filtering algorithm, etc., as a preference, the noise removal algorithm is selected as a spatial domain filtering algorithm, and the edge enhancement processing is smoothing the image, and then performing edge points The process of detecting, locating edges, and sharpening the edges. The enhanced image of the area to be recognized is the image obtained after the image enhancement processing. In this way, the image enhancement processing can enhance and optimize the to-be-recognized area Features related to spinal deformation in the image of the back area, and can facilitate the recognition of spinal deformation, and improve the accuracy of recognition.

In one embodiment, as shown in FIG. 4, in the step S30, that is, performing image enhancement processing on the image of the back area to be identified to obtain an enhanced image of the area to be identified includes:

S301: Perform grayscale processing on the image of the back area to be recognized in the image of the back area to be recognized to obtain a grayscale image of the back back, and at the same time perform a grayscale process on the side of the back to be recognized in the back area image to be recognized. The area image is gray-scaled to obtain a gray-scale image of the back side.

Understandably, the image of the back area to be recognized is separated by channels, and the red channel image of the red channel, the green channel image of the green channel, and the blue channel image of the blue channel are separated, and the back area image to be recognized includes three Image with two channels (red channel, green channel and blue channel), that is, each pixel in the cropped image has three channel component values, which are the red component value, the green component value and the blue component value, Perform grayscale processing on the red channel image, green channel image, and blue channel image to obtain a grayscale image corresponding to the grayscale channel. The red ( The R) component value, the green (G) component value and the blue (B) component value are calculated by the weighted average method to obtain the gray component value of each pixel. The formula in the weighted average method can be set according to requirements, For example, the formula of the weighted average method is set as: Y=0.299R+0.587G+0.114B, where Y is the gray component value of each pixel, R is the red component value of each pixel, and G is each pixel. The green component value in each pixel, B is the blue component value in each pixel, so as to obtain the back back grayscale image of the back area image to be recognized. Similarly, the back area to be recognized The image of the back side area to be identified in the image is subjected to grayscale processing, so as to obtain the back side grayscale image.

S302: Perform image denoising and edge enhancement processing on the back and back grayscale image to obtain a back back enhanced image, and simultaneously perform image denoising and edge enhancement processing on the back side grayscale image to obtain a back side enhanced image.

Understandably, the spatial domain image of the back and back grayscale image is denoised (also known as noise reduction) through a spatial domain filtering algorithm. The spatial domain filtering algorithm can be selected according to requirements, for example, it can be a neighborhood average. Method, median filter, low-pass filter, etc. Preferably, the spatial domain filter algorithm is selected as the neighborhood average method, that is, the value of each pixel in the back and back grayscale image is adjacent to the periphery of the pixel Take the average of the values of the pixels to obtain the de-noised value corresponding to the pixel to obtain the de-noised back back grayscale image, and perform edge enhancement processing on the de-noised back back grayscale image , So as to obtain the back side enhanced image, the edge enhancement processing is a process of smoothing the image, then detecting edge points, locating the edge, and sharpening the edge. Similarly, the back side The grayscale image undergoes image denoising and edge enhancement processing to obtain the back side enhanced image.

S303: Determine the back enhanced image and the back side enhanced image as the to-be-recognized area enhanced image.

Understandably, the back back enhanced image and the back side enhanced image are marked as the to-be-identified area enhanced image.

This application performs grayscale processing, denoising and edge enhancement processing on both the back area image and the back side area image to obtain an optimized enhanced image of the area to be identified, which can enhance the deformation of the spine in the image of the back area to be identified. Related features, and can facilitate the recognition of spinal deformation, improve the accuracy of recognition.

S40. Input the enhanced image of the to-be-recognized area into a spine recognition model, extract the spine features in the enhanced image of the to-be-recognized area through the spine recognition model, and obtain a first feature output by the spine recognition model according to the spine feature A vector map is used to input the non-image data into a data standardization model, and the non-image data is normalized and edge weighted through the data standardization model to obtain a second feature vector map.

Understandably, the spine recognition model is a trained deep convolutional neural network that performs the spine feature extraction on the enhanced image of the to-be-recognized region, recognizes according to the spine feature, and outputs the first feature vector image Model, the network structure of the spine recognition model can be set according to requirements. For example, the network structure of the spine recognition model can be an Inception V4 network structure, or a VGG16 network structure, etc., and the spine features are related to the spine. Feature vectors related to shape, curvature, etc., the first feature vector map is a series of arrays containing feature vectors corresponding to the spine features output based on the spine features extracted from the enhanced image of the to-be-recognized region Matrix; the data standardization model is a collection model that determines the correlation function between each dimension in the non-image data and the degree of spine deformation and potential risks based on the collected historical non-image data, and the data standardization model can The non-image data is subjected to normalization processing and edge weighting processing, and the normalization processing is to normalize the data of each dimension in the non-image data according to the rules matching each dimension to obtain unified standard data The edge weight processing is to perform weighting processing on the values after the normalization processing according to the edge weight parameters matched with each dimension, that is, the values after the normalization processing are multiplied by the matched edge weight parameters, and the second feature The vector graph is an array matrix obtained after normalization and edge weight processing of the non-image data.

In one embodiment, as shown in FIG. 5, the spine features include scoliosis features and kyphosis features; in S40, that is, the spine features in the enhanced image of the region to be recognized are extracted through the spine recognition model , Acquiring the first feature vector diagram output by the spine recognition model according to the features of the spine includes:

S401. Perform the side curve feature extraction on the back enhanced image using a side curve recognition model, and simultaneously perform the kyphosis feature extraction on the back side enhanced image using a kyphosis recognition model; the spine recognition model includes the side A bend recognition model and the humpback recognition model.

Understandably, the spine recognition model includes the scoliosis recognition model and the kyphosis recognition model, and the scoliosis recognition model is trained and trained by extracting the scoliosis features from most of the back images containing the back. The completed neural network model. The network structure of the side curve recognition model can be set according to requirements. The hunchback recognition model is a neural network that is trained and trained by extracting the hunchback feature from most of the back side images. Model, the network structure of the humpback recognition model can be set according to requirements, the side bending feature is the asymmetrical two sides of the back trunk, the torso is curved, and the shoulders are uneven. The humpback feature is the convex back and the side back is present. Arch-shaped feature, the back back enhanced image is input into the side curve recognition model, the side curve feature in the back back enhanced image is extracted through the side curve recognition model, and the back side enhanced image Input into the hunchback recognition model, and extract the hunchback feature in the back side enhanced image through the hunchback recognition model.

S402: Obtain a side curve feature vector diagram output by the side curve recognition model according to the side curve feature, and simultaneously obtain a hunchback feature vector diagram output by the hunchback recognition model according to the hunchback feature.

Understandably, the convolution layer, the pooling layer, and the fully connected layer are processed on the back and back enhanced image through the side curve recognition model to extract the side curve feature, and the extracted side curve feature Arrange and output into a eigenvector diagram, which is the side curve eigenvector diagram. The side curve eigenvector diagram is a matrix diagram containing multiple eigenvectors. For example, the side curve eigenvector diagram is a 100×100 matrix. , At the same time, perform convolutional layer, pooling layer, and fully connected layer processing on the back side enhanced image through the humpback recognition model to extract the humpback feature, and then output the extracted humpback feature into a feature vector The graph is the humpback feature vector graph, and the humpback feature vector graph is a matrix graph containing multiple feature vectors. For example, the humpback feature vector graph is a 100×100 matrix.

S403, splicing the side curve feature vector graph and the humpback feature vector graph to obtain the first feature vector graph.

Understandably, the side curve eigenvector diagram and the hunchback eigenvector diagram are connected up and down in a matrix to obtain the first eigenvector diagram. For example, the side curve eigenvector diagram is a 100×100 matrix, and the The humpback feature vector graph is a 100×100 matrix, and the first feature vector graph is a 100×200 matrix.

This application uses the scoliosis recognition model in the spine recognition model to extract the features of the back enhanced image and identify the scoliosis feature vector image, and at the same time, the kyphosis recognition model in the spine recognition model extracts the kyphosis feature of the back side enhanced image And the recognition of the humpback feature vector map can more specifically identify the side curvature and the humpback in the spinal deformation, which improves the accuracy and reliability of the recognition.

In an embodiment, as shown in FIG. 6, in the step S40, that is, the normalization and edge weight processing of the non-image data through the data standardization model to obtain the second feature vector graph includes:

S404: Acquire each dimension in the non-image data and dimension data corresponding to each of the dimensions.

Understandably, the non-image data is information related to the target to be identified, and the non-image data contains multiple dimensions. These dimensions can be set according to requirements, and the corresponding dimensions of the non-image data are obtained. The dimensional data of the dimensional data is the corresponding input content of the target to be identified for the dimensionality.

In an embodiment, the dimensions in the non-image data include target gender, target age, target occupation, and target information.

Understandably, the target gender is the gender of the target to be recognized, the target age is the age of the target to be recognized, the target occupation is the occupation of the target to be recognized, and the target information is The family information related to the target to be identified may be a family map, and the target information can be set according to requirements.

S405: Obtain normalization rules and edge weight parameters that match each of the dimensions.

Understandably, the normalization rule is a rule for unified processing of the dimension data that matches the dimension, that is, the dimension data of the same dimension is converted in a unified rule, for example: the dimension is male in the target age The dimensional data is converted to 1, and the dimensional data of the target age is converted to 0 and other rules. The edge weight parameter is a weighting parameter preset according to the dimension and matched with the dimension, and the edge weight parameter is based on historical data. Obtained from statistical analysis, the edge weight parameter indicates a measurement index of the degree of potential correlation between the dimension and the spine feature.

S406: Perform a normalization process on all the dimension data according to the normalization rule matched with each of the dimensions to obtain a dimension standard value corresponding to each of the dimensions.

Understandably, the normalization process is to convert dimensional data into a unified format according to a normalization rule, and perform unified conversion on all the dimensional data according to the normalization rule matching each dimension to obtain The dimension standard value corresponding to each of the dimensions.

S407: Perform edge weighting processing on all standard values of the dimensions according to the edge weight parameters matched with each of the dimensions to obtain a weight value corresponding to each of the dimensions.

Understandably, the dimension standard value and the edge weight parameter matching the dimension are multiplied to obtain the weight value, and the edge weighting is processed as multiplying the dimension standard value by the corresponding edge weight The parameter, that is, the dimension standard value is expanded according to the edge weight parameter, and the potential correlation degree between each dimension and the spine feature is opened, so that it can be identified objectively and scientifically.

S408: Expand all the weight values to obtain the second feature vector graph.

Understandably, the expansion is to copy and fill all the weights to reach a preset matrix size, and the copy and fill is to copy all the weights in one dimension as a whole to the preset matrix size. Finally, the elements that are insufficient to be copied are filled with zeros, and then all the elements of the row are copied to the longitudinal length of the preset matrix size, thereby obtaining the second eigenvector map.

This application performs normalization and edge weighting of non-image data and generates a second feature vector diagram related to the features of spine deformation, which can better establish the third feature vector diagram of the relationship between each feature vector, and improve recognition Reliability.

S50: Perform edge filling of the second feature vector graph to the first feature vector graph to obtain a third feature vector graph.

Understandably, the edge filling refers to filling on the edge of the array on the basis of the first eigenvector image, that is, filling the second eigenvector image above and below the matrix of the first eigenvector image to the preset value. Set the size to obtain the third feature vector diagram.

S60: Input the third feature vector map into the trained spine map convolutional network model.

Understandably, the spine diagram convolutional network model is a neural network model that is identified and trained based on the network structure of the diagram convolution, and the spine diagram convolutional network model can be based on the elements in the third feature vector diagram. The spine frequency domain features are extracted from the feature vector and the association relationship between each element, and the extracted spine frequency domain features are classified and recognized, and finally the category of the corresponding spine deformed crowd is output.

In an embodiment, as shown in FIG. 7, before step S60, that is, before inputting the third feature vector image into the trained spine graph convolutional network model, the method includes:

S601. Obtain a sample data set; the sample data set includes sample data and sample labels corresponding to the sample data one-to-one; the sample data is a historical third feature vector diagram; the sample labels include scoliosis people and kyphotic people , Potential scoliosis crowd and potential spine kyphosis crowd.

Understandably, the historical third feature vector graph is obtained after the sample image data and sample non-image data associated with the sample data are processed through the above-mentioned steps S20 to S50, and the sample data set contains a plurality of the samples. Data, one sample data is associated with one sample label, and the sample label includes a scoliosis crowd, a kyphosis crowd, a potential scoliosis crowd, and a potential kyphosis crowd.

S602: Input the sample data into a spinal diagram convolutional neural network model containing initial parameters.

Understandably, the initial parameters of the spine graph convolutional neural network model can be set according to requirements. For example, the initial parameters can be obtained through a transfer learning method to obtain all the parameters of other graph convolution models related to the back recognition, or all Set to a preset value.

S603, according to the spectral domain method in the GCN, extract the spine frequency domain features in the sample data through the spine map convolutional neural network model, and obtain the spine map convolutional neural network model to output according to the spine frequency domain features Sample results.

Understandably, the GCN is Graph Convolutional Network, that is, to identify the existence or potential classification by extracting relevant spatial features of a topological graph that uses vertices and edges to establish a corresponding relationship. The spectral domain method is by relying on topological The eigenvalues and eigenvectors of the Laplacian matrix corresponding to the graph are used to study the properties of the topological graph and the GCN classification result obtained by Fourier transforming the eigenvalues and eigenvectors of the Laplacian matrix. The spine The frequency domain feature is a feature that is transformed by using the frequency domain method and is related to spinal deformation, and the spine map convolutional neural network model identifies the sample result of the sample data according to the extracted frequency domain feature of the spine, The sample results include people with scoliosis, people with kyphosis, people with potential scoliosis, people with potential kyphosis, and people without spinal deformity.

S604: Determine a loss value according to the sample result and the sample label corresponding to the sample data.

Understandably, the sample result and the sample label are input into the loss function in the spine map convolutional neural network model, and the loss value corresponding to the sample data is calculated. The loss function can be set according to requirements The loss function is the logarithm of the difference between the sample result and the sample label, and indicates the difference between the sample result and the sample label.

S605: When the loss value reaches a preset convergence condition, record the spine diagram convolutional neural network model after convergence as a trained spine diagram convolutional network model.

Understandably, the convergence condition may be a condition that the loss value is less than the set threshold, that is, when the loss value is less than the set threshold, the spine map convolutional neural network model after convergence is recorded as training completed The convolutional network model of the spine diagram.

In an embodiment, after step S604, that is, after determining the loss value according to the sample result and the sample label corresponding to the sample data, the method further includes:

S606: When the loss value does not reach the preset convergence condition, iteratively update the initial parameters of the spine map convolutional neural network model, until the loss value reaches the preset convergence condition, converge The spine map convolutional neural network model is recorded as a trained spine map convolutional network model.

Understandably, the convergence condition may also be a condition that the loss value is small and will not decrease after 10,000 calculations, that is, the loss value is small and does not fall after 10,000 calculations. When it drops again, the training is stopped, and the spine map convolutional neural network model after convergence is recorded as the spine map convolutional network model that has been trained.

In this way, when the loss value does not reach the preset convergence condition, the initial parameters of the iterative convolutional neural network model of the spine diagram can be continuously updated, which can continuously move closer to the accurate recognition result, so that the accuracy of the recognition result becomes more and more accurate. high.

S70. According to the spectral domain method in the GCN, extract the spine frequency domain features in the third feature vector map through the spine map convolution network model, and obtain the spine map convolution network model according to the spine frequency domain features The output recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes the scoliosis crowd, the kyphotic crowd, the potential scoliosis crowd, the potential kyphotic crowd, and People with non-spine deformities.

Understandably, the GCN is Graph Convolutional Network, that is, to identify the existence or potential classification by extracting relevant spatial features of a topological graph that uses vertices and edges to establish a corresponding relationship. The spectral domain method is by means of topology The eigenvalues and eigenvectors of the Laplacian matrix corresponding to the graph are used to study the properties of the topological graph and the method of performing the Fourier transform on the eigenvalues and eigenvectors of the Laplacian matrix to obtain the result of GCN classification. The spine map convolutional network model recognizes the back photos of the target to be recognized, determines the category of the spine deformed crowd where the target to be recognized is located, can remind potential crowds, and can target the spine deformed after determination The corresponding reminder or prevention reminder is output for the category of the crowd.

This application obtains the image data and non-image data associated with the unique code corresponding to the target to be recognized by receiving the target recognition instruction; inputs the image data into the back area recognition model, and obtains the back to be recognized cut out by the back area recognition model Regional image; perform image enhancement processing on the image of the back region to be recognized to obtain an enhanced image of the region to be recognized; input the enhanced image of the region to be recognized into a spine recognition model, and extract the enhanced image of the region to be recognized through the spine recognition model The spine feature in the spine, acquiring the first feature vector diagram output by the spine recognition model according to the spine feature, and inputting the non-image data into the data standardization model, and classifying the non-image data through the data standardization model Unification and edge weight processing to obtain the second eigenvector image; edge-filling the second eigenvector image to the first eigenvector image to obtain the third eigenvector image; according to the spectral domain method in GCN, pass The spine map convolution network model extracts the spine frequency domain features in the third feature vector map, and obtains the recognition result output by the spine map convolution network model according to the spine frequency domain features.

This application realizes that by acquiring the image data and non-image data associated with the target to be recognized; intercepting the image of the back area to be recognized in the back area of the image data; performing image enhancement processing on the image of the back area to be recognized to obtain the to be recognized Region-enhanced image; extract the features of the spine in the enhanced image of the region to be identified through the spine recognition model to obtain the first feature vector diagram, and at the same time normalize the non-image data and process the edge weights through the data standardization model to obtain the first Second feature vector graph; edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph; according to the spectral domain method in GCN, the spine graph convolutional network model is used to extract the The frequency domain features of the spine in the third feature vector diagram are obtained, and the recognition results output by the convolutional network model of the spine diagram according to the frequency domain features of the spine are obtained. Related non-image information, through the spectrum domain method in the GCN, automatically identify the category of the spine deformed population corresponding to the target to be identified (including the potential crowd scoliosis crowd, spine kyphosis crowd, potential scoliosis crowd, potential spine kyphosis crowd and potential scoliosis crowd). People with non-spine deformities) can quickly and accurately identify the categories of people with spine deformities corresponding to the target to be identified, and remind potential people to prevent them. Therefore, the accuracy and reliability of the recognition of people with spine deformities are improved. , Which greatly reduces the cost of identification and serves as a reminder to potential crowds.

In one embodiment, a device for identifying people with deformed spine is provided, and the device for identifying people with deformed spine corresponds one-to-one with the method for identifying people with deformed spine in the above-mentioned embodiment. As shown in FIG. 8, the device for identifying people with deformed spine includes a receiving module 11, a recognition module 12, an enhancement module 13, an acquisition module 14, a filling module 15, an input module 16 and an output module 17. The detailed description of each functional module is as follows:

The receiving module 11 is configured to receive a target recognition instruction, and obtain image data and non-image data associated with the unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is the target to be recognized Related information;

The recognition module 12 is configured to input the image data into a back area recognition model, and perform back area recognition on the image data through the back area recognition model, and obtain the back area image to be recognized cut out by the back area recognition model ; The back region recognition model is a deep convolutional neural network model based on the YOLO model;

The enhancement module 13 is configured to perform image enhancement processing on the image of the back area to be identified to obtain an enhanced image of the area to be identified;

The acquiring module 14 is configured to input the enhanced image of the region to be recognized into a spine recognition model, extract the features of the spine in the enhanced image of the region to be identified through the spine recognition model, and obtain the spine recognition model and output according to the spine features At the same time, the non-image data is input into a data standardization model, and the non-image data is normalized and edge weighted through the data standardization model to obtain a second feature vector diagram;

The filling module 15 is used for edge filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

The input module 16 is configured to input the third feature vector map into the trained spine map convolutional network model;

The output module 17 is configured to extract the spine frequency domain features in the third feature vector diagram through the spine map convolutional network model according to the spectral domain method in the GCN, and obtain the spine map convolutional network model according to the spine map convolutional network model. The recognition result of the spine frequency domain feature output; the recognition result characterizes the category of the spine deformed population of the target to be recognized, and the category of the spine deformed population includes the scoliosis population, the kyphotic population, the potential scoliosis population, and the potential scoliosis population. People with kyphosis and non-spine deformities.

For the specific definition of the device for identifying people with deformed spine, please refer to the above definition of the method for identifying people with deformed spine, which will not be repeated here. The various modules in the device for identifying people with deformed spine can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 9. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to realize a method for identifying people with deformed spine.

In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor executes the computer program to implement the method for identifying people with deformed spine in the above-mentioned embodiment. .

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method for identifying people with deformed spine in the above-mentioned embodiment is realized. Optionally, the storage medium involved in this application, such as a computer-readable storage medium, may be non-volatile or volatile.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage. In the medium, when the computer program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the division of the above functional units and modules is used as an example. In practical applications, the above functions can be allocated to different functional units and modules as needed. Module completion, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still implement the foregoing The technical solutions recorded in the examples are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the application, and should be included in Within the scope of protection of this application.

Claims (20)

1. A method for identifying people with deformed spine, which includes:

   Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

   Input the image data into a back region recognition model, and perform back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

   Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

   The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

   Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

   Input the third feature vector map into the trained spine map convolutional network model;

   According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.
2. The method for recognizing people with deformed spine according to claim 1, wherein the image data is recognized by the back area recognition model to obtain the back area image to be recognized cut out by the back area recognition model ,include:

   Input the back image of the target back into the back back area recognition model in the back area recognition model, and input the back side image of the target into the back side area recognition model in the back area recognition model; the image data includes the back of the target The back image and the back side image of the target;

   According to the YOLO algorithm, the recognition is performed through the back area recognition model, and the image of the back back area to be recognized that contains only the back back of the target to be recognized is cut out, and the back side area recognition model is used for recognition. An image of the back side area to be recognized containing the back side of the target to be recognized;

   The image of the back area to be recognized and the image of the back side area to be recognized are determined as the image of the back area to be recognized.
3. The method for identifying people with deformed spine as claimed in claim 1, wherein said performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified comprises:

   Perform gray-scale processing on the image of the back area to be recognized in the image of the back area to be recognized to obtain a gray-scale image of the back back, and at the same time perform the gray-scale processing on the image of the back side area to be recognized in the back area image to be recognized Perform gray-scale processing to obtain a gray-scale image of the back side;

   Performing image denoising and edge enhancement processing on the back and back grayscale image to obtain a back back enhanced image, and simultaneously performing image denoising and edge enhancement processing on the back side grayscale image to obtain a back side enhanced image;

   The back back enhanced image and the back side enhanced image are determined as the enhanced image of the region to be recognized.
4. The method for identifying people with deformed spine according to claim 1, wherein the spine features include scoliosis features and kyphosis features;

   The extracting the spine features in the enhanced image of the region to be recognized through the spine recognition model, and obtaining the first feature vector diagram output by the spine recognition model according to the spine features, includes:

   The scoliosis feature extraction is performed on the back enhanced image through a scoliosis recognition model, and the kyphosis feature extraction is performed on the back side enhanced image through a kyphosis recognition model; the spine recognition model includes the scoliosis recognition Model and said humpback recognition model;

   Acquiring a side curve feature vector diagram output by the side curve recognition model according to the side curve feature, and simultaneously acquiring a hunchback feature vector diagram output by the hunchback recognition model according to the hunchback feature;

   The side curve feature vector graph and the humpback feature vector graph are spliced together to obtain the first feature vector graph.
5. The method for identifying people with deformed spine according to claim 1, wherein said normalizing and edge weighting said non-image data through a data standardization model to obtain a second feature vector map comprises:

   Acquiring each dimension in the non-image data and dimension data corresponding to each of the dimensions;

   Obtaining normalization rules and edge weight parameters that match each of the dimensions;

   Performing normalization processing on all the dimensional data according to the normalization rule matching each of the dimensions to obtain the dimension standard value corresponding to each of the dimensions;

   Performing edge weighting processing on all standard values of the dimensions according to the edge weight parameters matched with each of the dimensions to obtain weighted values corresponding to each of the dimensions;

   All the weighted values are expanded to obtain the second feature vector graph.
6. 5. The method for identifying people with deformed spine, wherein the dimensions in the non-image data include target gender, target age, target occupation, and target information.
7. The method for identifying people with deformed spine according to claim 1, wherein, before inputting the third feature vector map into the trained spine map convolutional network model, the method comprises:

   Obtain a sample data set; the sample data set includes sample data and sample labels corresponding to the sample data one-to-one; the sample data is a historical third feature vector diagram; the sample labels include scoliosis crowds, kyphotic crowds, and potential People with scoliosis and people with potential kyphosis;

   Inputting the sample data into a spine diagram convolutional neural network model containing initial parameters;

   According to the spectral domain method in GCN, the spine frequency domain features in the sample data are extracted through the spine map convolutional neural network model, and the samples output by the spine map convolutional neural network model according to the spine frequency domain features are obtained result;

   Determine a loss value according to the sample result and the sample label corresponding to the sample data;

   When the loss value reaches a preset convergence condition, the spine diagram convolutional neural network model after convergence is recorded as the trained spine diagram convolutional network model.
8. A device for identifying people with deformed spine, which includes:

   The receiving module is used to receive a target recognition instruction, and obtain image data and non-image data associated with the unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is related to the target to be recognized Information;

   A recognition module, configured to input the image data into a back area recognition model, perform back area recognition on the image data through the back area recognition model, and obtain an image of the back area to be recognized that is intercepted by the back area recognition model; The back region recognition model is a deep convolutional neural network model based on the YOLO model;

   An enhancement module, configured to perform image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

   The acquiring module is configured to input the enhanced image of the region to be recognized into a spine recognition model, extract the features of the spine in the enhanced image of the region to be identified through the spine recognition model, and obtain the output of the spine recognition model according to the features of the spine A first feature vector map, and at the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

   A filling module, configured to perform edge filling of the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

   An input module, configured to input the third feature vector graph into the trained spine graph convolutional network model;

   The output module is used to extract the frequency domain features of the spine in the third feature vector diagram through the spine map convolutional network model according to the spectral domain method in the GCN, and obtain the spine map convolutional network model according to the spine Frequency domain feature output recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be recognized, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, and potential spine crowd People with hunchbacks and non-spine deformities.
9. A computer device includes a memory, a processor, and a computer program that is stored in the memory and can run on the processor, wherein the processor implements the following steps when the processor executes the computer program:

   Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

   Input the image data into a back region recognition model, and perform back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

   Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

   The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

   Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

   Input the third feature vector map into the trained spine map convolutional network model;

   According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.
10. The computer device according to claim 9, wherein, when the image data is recognized by the back area recognition model to recognize the back area, and the back area image to be recognized extracted by the back area recognition model is obtained, specifically Implement the following steps:

    Input the back image of the target back into the back back area recognition model in the back area recognition model, and input the back side image of the target into the back side area recognition model in the back area recognition model; the image data includes the back of the target The back image and the back side image of the target;

    According to the YOLO algorithm, the recognition is performed by the back back area recognition model, and the back back area image to be recognized that contains only the back back of the target to be recognized is cut out, and at the same time, the back side area recognition model is used for recognition, and only the back back area recognition model is cut out. An image of the back side area to be recognized containing the back side of the target to be recognized;

    The image of the back area to be recognized and the image of the back side area to be recognized are determined as the image of the back area to be recognized.
11. 9. The computer device according to claim 9, wherein when the image enhancement processing is performed on the image of the back region to be recognized to obtain the enhanced image of the region to be recognized, the following steps are specifically implemented:

    Perform gray-scale processing on the image of the back area to be recognized in the image of the back area to be recognized to obtain a gray-scale image of the back back, and at the same time perform the gray-scale processing on the image of the back side area to be recognized in the back area image to be recognized Perform gray-scale processing to obtain a gray-scale image of the back side;

    Performing image denoising and edge enhancement processing on the back and back grayscale image to obtain a back back enhanced image, and simultaneously performing image denoising and edge enhancement processing on the back side grayscale image to obtain a back side enhanced image;

    The back enhanced image and the back side enhanced image are determined as the enhanced image of the region to be recognized.
12. 9. The computer device of claim 9, wherein the spine features include side curvature features and kyphosis features;

    When extracting the spine features in the enhanced image of the region to be recognized through the spine recognition model, and obtaining the first feature vector map output by the spine recognition model according to the spine features, the following steps are specifically implemented:

    The scoliosis feature extraction is performed on the back enhanced image through a scoliosis recognition model, and the kyphosis feature extraction is performed on the back side enhanced image through a kyphosis recognition model; the spine recognition model includes the scoliosis recognition Model and said humpback recognition model;

    Acquiring a side curve feature vector diagram output by the side curve recognition model according to the side curve feature, and simultaneously acquiring a hunchback feature vector diagram output by the hunchback recognition model according to the hunchback feature;

    The side curve feature vector graph and the humpback feature vector graph are spliced together to obtain the first feature vector graph.
13. 9. The computer device according to claim 9, wherein, when the non-image data is normalized and edge weighted by the data standardization model to obtain the second feature vector map, the following steps are specifically implemented:

    Acquiring each dimension in the non-image data and dimension data corresponding to each of the dimensions;

    Obtaining normalization rules and edge weight parameters that match each of the dimensions;

    Performing normalization processing on all the dimensional data according to the normalization rule matching each of the dimensions to obtain the dimension standard value corresponding to each of the dimensions;

    Performing edge weighting processing on all standard values of the dimensions according to the edge weight parameters matched with each of the dimensions to obtain weighted values corresponding to each of the dimensions;

    All the weighted values are expanded to obtain the second feature vector graph.
14. 9. The computer device according to claim 9, wherein before the input of the third feature vector map into the trained spine map convolutional network model, the processor further implements the following steps when executing the computer program:

    Obtain a sample data set; the sample data set includes sample data and sample labels corresponding to the sample data one-to-one; the sample data is a historical third feature vector diagram; the sample labels include scoliosis crowds, kyphotic crowds, and potential People with scoliosis and people with potential kyphosis;

    Inputting the sample data into a spine diagram convolutional neural network model containing initial parameters;

    According to the spectral domain method in GCN, the spine frequency domain features in the sample data are extracted through the spine map convolutional neural network model, and the samples output by the spine map convolutional neural network model according to the spine frequency domain features are obtained result;

    Determine a loss value according to the sample result and the sample label corresponding to the sample data;

    When the loss value reaches a preset convergence condition, the spine diagram convolutional neural network model after convergence is recorded as the trained spine diagram convolutional network model.
15. A computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the following steps:

    Receiving a target recognition instruction, and obtaining image data and non-image data associated with a unique code corresponding to the target to be recognized; the image data is an image related to the back; the non-image data is information related to the target to be recognized;

    Input the image data into a back region recognition model, and perform back region recognition on the image data through the back region recognition model, and obtain the back region image to be recognized cut out by the back region recognition model; the back region recognition The model is a deep convolutional neural network model based on the YOLO model;

    Performing image enhancement processing on the image of the back region to be identified to obtain an enhanced image of the region to be identified;

    The enhanced image of the to-be-recognized area is input into a spine recognition model, the spine features in the enhanced image of the to-be-recognized area are extracted through the spine recognition model, and a first feature vector image output by the spine recognition model according to the spine feature is obtained At the same time input the non-image data into a data standardization model, and perform normalization and edge weight processing on the non-image data through the data standardization model to obtain a second feature vector map;

    Edge-filling the second feature vector graph to the first feature vector graph to obtain a third feature vector graph;

    Input the third feature vector map into the trained spine map convolutional network model;

    According to the spectral domain method in GCN, the spine frequency domain features in the third feature vector diagram are extracted through the spine map convolutional network model, and the output of the spine map convolution network model according to the spine frequency domain features is obtained. Recognition result; the recognition result characterizes the category of the spine deformed crowd of the target to be identified, and the category of the spine deformed crowd includes scoliosis crowd, kyphotic crowd, potential scoliosis crowd, potential kyphotic crowd, and non-spine Deformed crowd.
16. The computer-readable storage medium according to claim 15, wherein the back area recognition is performed on the image data through the back area recognition model to obtain the back area image to be recognized cut by the back area recognition model When, specifically implement the following steps:

    Input the back image of the target back into the back back area recognition model in the back area recognition model, and input the back side image of the target into the back side area recognition model in the back area recognition model; the image data includes the back of the target The back image and the back side image of the target;

    According to the YOLO algorithm, the recognition is performed by the back back area recognition model, and the back back area image to be recognized that contains only the back back of the target to be recognized is cut out, and the back side area recognition model is used for recognition at the same time, and only An image of the back side area to be recognized containing the back side of the target to be recognized;

    The image of the back area to be recognized and the image of the back side area to be recognized are determined as the image of the back area to be recognized.
17. 15. The computer-readable storage medium according to claim 15, wherein when the image enhancement processing is performed on the image of the back region to be recognized to obtain the enhanced image of the region to be recognized, the following steps are specifically implemented:

    Perform gray-scale processing on the image of the back area to be recognized in the image of the back area to be recognized to obtain a gray-scale image of the back back, and at the same time perform the gray-scale processing on the image of the back side area to be recognized in the back area image Perform gray-scale processing to obtain a gray-scale image of the back side;

    Performing image denoising and edge enhancement processing on the back and back grayscale image to obtain a back back enhanced image, and simultaneously performing image denoising and edge enhancement processing on the back side grayscale image to obtain a back side enhanced image;

    The back enhanced image and the back side enhanced image are determined as the enhanced image of the region to be recognized.
18. 15. The computer-readable storage medium of claim 15, wherein the spinal features include lateral curvature and kyphosis;

    When extracting the spine features in the enhanced image of the region to be recognized through the spine recognition model, and obtaining the first feature vector map output by the spine recognition model according to the spine features, the following steps are specifically implemented:

    The scoliosis feature extraction is performed on the back enhanced image through a scoliosis recognition model, and the kyphosis feature extraction is performed on the back side enhanced image through a kyphosis recognition model; the spine recognition model includes the scoliosis recognition Model and said humpback recognition model;

    Obtaining a side curve feature vector diagram output by the side curve recognition model according to the side curve feature, and simultaneously obtaining a hunchback feature vector diagram output by the hunchback recognition model according to the hunchback feature;

    The side curve feature vector graph and the humpback feature vector graph are spliced together to obtain the first feature vector graph.
19. 15. The computer-readable storage medium according to claim 15, wherein when the non-image data is normalized and edge weighted by the data normalization model to obtain the second feature vector map, the following steps are specifically implemented:

    Acquiring each dimension in the non-image data and dimension data corresponding to each of the dimensions;

    Obtaining normalization rules and edge weight parameters that match each of the dimensions;

    Performing a normalization process on all the dimension data according to the normalization rule matching each dimension to obtain the dimension standard value corresponding to each dimension;

    Performing edge weighting processing on all standard values of the dimensions according to the edge weight parameters matched with each of the dimensions to obtain weighted values corresponding to each of the dimensions;

    All the weighted values are expanded to obtain the second feature vector graph.
20. The computer-readable storage medium according to claim 15, wherein before the input of the third feature vector map into the trained spine map convolutional network model, the computer program further implements the following steps when being executed by the processor :

    Obtain a sample data set; the sample data set includes sample data and sample labels corresponding to the sample data one-to-one; the sample data is a historical third feature vector diagram; the sample labels include scoliosis crowds, kyphotic crowds, and potential People with scoliosis and people with potential kyphosis;

    Inputting the sample data into a spine diagram convolutional neural network model containing initial parameters;

    According to the spectral domain method in GCN, the spine frequency domain features in the sample data are extracted through the spine map convolutional neural network model, and the samples output by the spine map convolutional neural network model according to the spine frequency domain features are obtained result;

    Determine a loss value according to the sample result and the sample label corresponding to the sample data;

    When the loss value reaches a preset convergence condition, the spine map convolutional neural network model after convergence is recorded as the trained spine map convolutional network model.

Images