Wound assessment method and device based on machine learning, equipment and medium

Abstract

The invention relates to a wound assessment method and device based on machine learning, equipment and a medium, and belongs to the field of intelligent medical treatment. The method comprises the steps that a user uploads an original wound image; wound labeling information is obtained based on the original wound image and is transmitted back to the user side, the wound labeling information comprises a wound contour, a wound size and a wound recovery stage, the wound contour comprises a conventional wound contour and a beautified wound contour, and the wound recovery stage comprises a blood coagulation stage, an inflammation stage, a repair stage and a mature stage; the user side comprises a conventional user side and a pure user side, the conventional wound contour is transmitted back to the conventional user side, and the beautified wound contour is transmitted back to the pure user side; according to the method and the system, the wound is quickly and conveniently identified, the wound information is returned according to different requirements of the user, and the reasonable wound nursing suggestion is provided.

Description

Technical Field

[0001] The present invention belongs to the field of smart medical technology, and specifically relates to a wound assessment method, device, equipment and medium based on machine learning. Background Art

[0002] Usually, wound care requires one-on-one visits between professional nurses and patients, which is labor-intensive and takes up the time of both parties. It is easy for the wound to be deteriorated due to lack of timely treatment, especially for chronic wounds that require long-term treatment. Chronic wounds take a long time to heal and are prone to further deterioration during the healing process. Professional nurses are required to frequently observe the patient's wound and formulate the next treatment plan based on the feedback from the wound. This method requires a lot of time and labor for patients or medical staff to go back and forth.

[0003] Wound observation and analysis are the basis of wound treatment and care. With the development of machine learning technology and the continuous updating of medical equipment, research based on data-driven machine learning is widely used in the medical field, especially in the analysis and processing of medical images. Machine learning technology can assist doctors in reading films, analyzing valuable information in images, and realizing intelligent analysis of medical images through machine learning. However, the current wound assessment methods based on machine learning have the problem of incomplete extraction of wound features, and for people with poor psychological tolerance or groups such as young children and the elderly, they need to be constantly exposed to bloody images of wounds during the assessment process, which is not conducive to their physical and mental health. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a wound assessment method, device, equipment and medium based on machine learning.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A wound assessment method based on machine learning, the implementation of the wound assessment method comprising the following steps:

[0007] The user uploads the original image of the wound;

[0008] Based on the original wound image, wound annotation information is acquired and transmitted back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end;

[0009] Provide wound care suggestions to the user terminal according to the wound annotation information.

[0010] Preferably, the step of acquiring wound annotation information based on the original wound image and transmitting the information back to the user terminal includes:

[0011] Acquiring the conventional wound contour based on an improved contour recognition method;

[0012] Acquire an approximate beautified contour based on the conventional wound contour, fill the approximate beautified contour with color, and use the approximate beautified contour to replace the conventional wound contour to obtain the beautified wound contour;

[0013] Obtaining the wound size based on the conventional wound contour;

[0014] The wound recovery stage is obtained based on the improved wound recognition model.

[0015] Preferably, the obtaining of the conventional wound contour based on the improved contour recognition method comprises:

[0016] Uniformly collecting contour sampling points in the original wound image and obtaining the coordinates of the contour sampling points, and obtaining the contour centroid based on the coordinates of the contour sampling points;

[0017] Obtaining a centroid height value of the contour sampling point, wherein the centroid height value is the Euclidean distance between the contour sampling point and the contour centroid;

[0018] Calculating a centroid height increment based on the centroid height value;

[0019] Acquire a centroid height increment sequence of the i-th contour sampling point based on the centroid height increment;

[0020] Acquire a centroid height increment matrix based on the centroid height increment sequence;

[0021] The conventional wound contour is depicted based on the centroid height increment matrix.

[0022] Preferably, obtaining an approximate beautified contour based on the conventional wound contour comprises:

[0023] Randomly select a positive integer coefficient whose value range is in the open interval (1, N), divide the centroid height increment sequence into S non-overlapping intervals [1, d], [d+1, 2d], ... based on the positive integer coefficient, where S = N / d, d is the positive integer coefficient, and calculate the mean centroid height increment in the interval;

[0024] Obtaining a feature matrix of the i-th contour sampling point based on the mean of the centroid height increments within the interval;

[0025] Arranging the feature matrices of all the contour sampling points to obtain a contour feature matrix;

[0026] Preset a template image contour, and obtain a template contour feature matrix of the template image contour;

[0027] Calculating the regular path distance between the conventional wound contour and the template image contour;

[0028] Flipping the conventional wound contour to obtain a flipped conventional wound contour, and obtaining a flipped regular path distance between the flipped conventional wound contour and the template image contour;

[0029] Obtaining a conventional shape complexity of the conventional wound contour and a template shape complexity of the template image contour;

[0030] Calculating a similarity coefficient based on the regular path distance, the flipped regular path distance, the conventional shape complexity, and the template shape complexity;

[0031] The template image contours are traversed, and the template image contour with the smallest similarity coefficient is selected as the approximate beautified contour. The approximate beautified contour filled with color is used to cover the conventional wound contour to obtain the beautified wound contour.

[0032] Preferably, the calculation formula of the similarity coefficient is: Among them, Q is the similarity coefficient, G is the conventional shape complexity, G Z is the template shape complexity, d DTM (F B ,F Z ) is the flip regularized path distance, d DTM (F,F Z ) is the regularized path distance.

[0033] Preferably, the improved wound recognition model includes a backbone extraction network CSPDarknet, an enhanced feature extraction network FPN, and a classification regressor YOLOhead.

[0034] Preferably, the backbone extraction network CSPDarknet includes a Focus block and a Resblock body block, wherein the Focus block is used to sample the pixel values ​​of the original wound image into four feature maps, with three RGB channels, forming a total of twelve feature maps, and the Resblock body block is used to perform feature extraction on the feature maps through four maximum pooling kernels;

[0035] The enhanced feature extraction network FPN is used to further extract feature layers and transform the feature layers of different sizes into feature layers of the same size, connect them in the channel dimension, complete feature fusion and obtain a multi-scale feature map, wherein the feature layer includes bottom-level features and top-level features;

[0036] The classification regressor YOLOhead includes a classification part and a regression part. The classification part uses two convolution calculations and then one convolution to realize the classification of the multi-scale feature map. The regression part uses two convolutions to extract features, and finally uses two convolutions to perform regression prediction, and finally obtains the wound recovery stage.

[0037] A wound assessment device based on machine learning, used to perform the wound assessment method described above, comprising an image uploading module, an information annotation module, and a suggestion providing module;

[0038] The image uploading module is used for users to upload original wound images;

[0039] The information annotation module is used to obtain wound annotation information based on the original wound image and transmit it back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end;

[0040] The suggestion providing module is used to provide wound care suggestions to the user terminal according to the wound annotation information.

[0041] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the above-mentioned wound assessment method is implemented when the processor executes the program.

[0042] A storage medium comprising computer executable instructions, wherein the computer executable instructions are used to perform the wound assessment method described above when executed by a computer processor.

[0043] The beneficial effects of the present invention are:

[0044] (1) Obtain wound annotation information through the original wound image and transmit it back to the user end, providing the user with reasonable wound care suggestions, saving time and medical resources for both doctors and patients, and reducing the situation where patients cannot seek medical treatment in time and cause the wound to deteriorate;

[0045] (2) Obtain an approximate beautified outline through the regular wound outline, fill the approximate beautified outline with color, and use the approximate beautified outline to replace the regular wound outline to obtain the beautified wound outline, and cover the bloody wound image with images similar to the wound, such as animals and cartoons, to reduce the visual impact on the user. The user can freely choose the regular user end or the pure user end, which has a high degree of freedom;

[0046] (3) Describing the conventional wound contour based on the centroid height increment matrix, obtaining the position of the wound contour point relative to the centroid and other wound contour points, and thus locating and delineating the conventional wound contour;

[0047] (4) The input wound original image will first be extracted by CSPDarknet, and then FPN will use the multi-layer feature map extracted by CSPDarknet to enhance feature extraction, that is, the multi-layer feature map of CPSDarknet will be fused to form a multi-scale feature map. Finally, YOLOhead will perform classification and prediction to achieve accurate judgment of the wound development stage;

[0048] (5) The pixel values ​​of the original wound image are sampled into four feature maps through the Focus block, which have three channels of RGB, forming a total of twelve feature maps. The number of channels is large, and the processing efficiency and accuracy are high. The feature layer is further extracted through FPN and the feature layers of different sizes are converted into feature layers of consistent size. The feature layer includes bottom-level features and top-level features, rather than only extracting top-level features, and the model accuracy is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0050] Figure 1 The present invention is a flowchart of the steps of a wound assessment method based on machine learning. DETAILED DESCRIPTION

[0051] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0052] The working principle and use process of the present invention:

[0053] See also Figure 1 , a wound assessment method based on machine learning, comprising:

[0054] S1: User uploads the original wound image;

[0055] S2: Based on the original wound image, wound annotation information is obtained and transmitted back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end;

[0056] S3: Providing wound care suggestions to the user terminal according to the wound annotation information.

[0057] In this embodiment, obtaining wound annotation information based on the original wound image may be specifically implemented by the following steps:

[0058] S201: Acquire the conventional wound contour based on an improved contour recognition method;

[0059] S202: obtaining an approximate beautified outline based on the conventional wound outline, filling the approximate beautified outline with color, and using the approximate beautified outline to replace the conventional wound outline to obtain the beautified wound outline, covering the bloody wound picture with an image similar to the wound, such as an animal or cartoon, to reduce the visual impact on the user, and the user can freely choose a conventional user end or a pure user end, with a high degree of freedom;

[0060] S203: Obtaining the wound size based on the conventional wound contour;

[0061] S204: Obtaining the wound recovery stage based on an improved wound recognition model, wherein the improved wound recognition model includes a backbone extraction network CSPDarknet, an enhanced feature extraction network FPN, and a classification regressor YOLOhead.

[0062] In this embodiment, the conventional wound contour is obtained based on the improved contour recognition method, which can be implemented by the following steps:

[0063] S201-1: uniformly collect contour sampling points in the original wound image and obtain the coordinates of the contour sampling points, and obtain the contour centroid based on the contour sampling point coordinates. The expression of the contour centroid is: Among them, Q(x0,y0) is the contour centroid, N is the number of contour sampling points, (x i ,y i ) is the coordinate of the i-th contour sampling point;

[0064] S201-2: Obtain the centroid height value of the contour sampling point, where the centroid height value is the Euclidean distance between the contour sampling point and the contour centroid;

[0065] S201-3: Calculate the centroid height increment based on the centroid height value, and the calculation formula is h i，j =g j -g i , i, j = 1, 2, ..., N, where h i,j is the centroid height increment between the i-th contour sampling point and the j-th contour sampling point, g i is the centroid height value of the i-th contour sampling point, g j is the centroid height value of the jth contour sampling point;

[0066] S201-4: Obtain a centroid height increment sequence of the i-th contour sampling point based on the centroid height increment, wherein the centroid height increment sequence is expressed as: i =(h i，j ,h i，i+1 , ..., h i，N ,h i，1 , ..., h i，i-1 ) T , i = 1, 2, ..., N, where H i is the centroid height increment sequence of the i-th contour sampling point, h i,i is the height increment of the centroid between the ith contour sampling point and the ith contour sampling point, and its value is 0. i,i+1 is the centroid height increment between the i-th contour sampling point and the i+1-th contour sampling point, h i,N is the centroid height increment between the i-th contour sampling point and the N-th contour sampling point, h i,1 is the centroid height increment between the i-th contour sampling point and the first contour sampling point, h i,i-1 is the centroid height increment of the i-th contour sampling point and the i-1-th contour sampling point;

[0067] S201-5: Obtain a centroid height increment matrix based on the centroid height increment sequence. The centroid height increment matrix is ​​expressed as L(M)=(H1, H2, ..., H N-1 , H N ), where L(M) is the centroid height increment matrix, H1 is the centroid height increment sequence of the first contour sampling point, H2 is the centroid height increment sequence of the second contour sampling point, and H N-1 is the centroid height increment sequence of the N-1th contour sampling point, H N is the centroid height increment sequence of the Nth contour sampling point;

[0068] S201-6: Describe the conventional wound contour based on the centroid height increment matrix.

[0069] In this embodiment, obtaining an approximate beautified contour based on the conventional wound contour can be implemented by the following steps:

[0070] S202-1: Randomly select a positive integer coefficient whose value range is in the open interval (1, N), and divide the centroid height increment sequence into S non-overlapping intervals [1, d], [d+1, 2d], ... based on the positive integer coefficient, where S = N / d, d is the positive integer coefficient, and calculate the mean centroid height increment in the interval. The calculation formula is: t = 1, 2, ..., S, where c i,tis the mean value of the centroid height increment in the tth interval of the i-th contour sampling point, h i,j is the centroid height increment of the i-th contour sampling point and the j-th contour sampling point;

[0071] S202-2: Obtain the characteristic matrix of the i-th contour sampling point based on the mean value of the centroid height increment in the interval. The expression of the characteristic matrix is ​​C i =(c i，1 , c i，2 , ..., c i，S ) T , i = 1, 2, ..., N, where C i is the feature matrix of the i-th contour sampling point;

[0072] S202-3: Arrange the feature matrices of all the contour sampling points to obtain a contour feature matrix, expressed as F=(C1, C2, ..., C N ), where F is the profile feature matrix;

[0073] S202-4: Preset a template image contour, the template image contour is the contour of an aesthetically pleasing template image of an animal, plant, cartoon, etc., and obtain a template contour feature matrix of the template image contour. The expression of the template contour feature matrix is ​​F Z =(C Z1 , C Z2 , ..., C ZN ), where F Z is the template contour feature matrix, C Zl is the feature matrix of the lth contour sampling point in the template image, l = 1, 2, ..., N;

[0074] S202-5: Calculate the regular path distance between the conventional wound contour and the template image contour, the calculation formula is d DTM (F,F Z )=min∑|C i -C Zl | 2 , i, l = 1, 2, ..., N, where d DTM (F,F Z ) is the regularized path distance;

[0075] S202-6: Flipping the conventional wound contour to obtain a flipped conventional wound contour, and obtaining a flipped regular path distance d between the flipped conventional wound contour and the template image contour DTM (F B ,F Z );

[0076] S202-7: Obtain the conventional shape complexity of the conventional wound contour and the template shape complexity of the template image contour. The expression of the conventional shape complexity is: Among them, G is the general shape complexity, std is the standard deviation;

[0077] S202-8: Calculate the similarity coefficient based on the regular path distance, the flipped regular path distance, the conventional shape complexity, and the template shape complexity. The calculation formula is: Among them, Q is the similarity coefficient, G Z is the complexity of the template shape;

[0078] S202-9: traversing the template image contours, selecting the template image contour with the smallest similarity coefficient as the approximate beautified contour, and using the approximate beautified contour filled with color to cover the conventional wound contour to obtain the beautified wound contour.

[0079] In this embodiment, the wound recovery stage is obtained based on the improved wound recognition model, which can be implemented by the following steps:

[0080] S204-1: The backbone extraction network CSPDarknet includes a Focus block and a Resblock body block. The Focus block is used to sample the pixel values ​​of the original wound image into four feature maps, with three channels of RGB, forming a total of twelve feature maps, with a large number of channels, high processing efficiency and accuracy, and the Resblock body block is used to extract features from the feature maps through four maximum pooling kernels;

[0081] S204-2: The enhanced feature extraction network FPN is used to further extract feature layers and transform the feature layers of different sizes into feature layers of the same size, connect them in the channel dimension, complete feature fusion to obtain a multi-scale feature map, the feature layer includes bottom-level features and top-level features, instead of only extracting top-level features, and the model accuracy is further improved;

[0082] S204-3: The classification regressor YOLOhead includes a classification part and a regression part. The classification part uses two convolution calculations and then one convolution to realize the classification of the multi-scale feature map. The regression part uses two convolutions to extract features, and finally uses two convolutions to perform regression prediction to finally obtain the wound recovery stage.

[0083] In this embodiment, wound care suggestions are provided to the user terminal according to the wound annotation information, which can be implemented by the following steps:

[0084] The wound care recommendations are obtained by crawling online resources. During the coagulation phase, it is recommended to stop bleeding and disinfect in time. If the wound is large, medical treatment should be sought in time. During the inflammation phase, it is recommended to destroy bacteria and remove necrotic tissue to lay the foundation for tissue regeneration and repair. During the repair phase, it is recommended to prevent wound proliferation. During the maturation phase, it is recommended to improve the appearance and function of the injured part.

[0085] A wound assessment device based on machine learning, comprising an image uploading module, an information annotation module, and a suggestion providing module;

[0086] The image uploading module is used for users to upload original wound images;

[0087] The information annotation module is used to obtain wound annotation information based on the original wound image and transmit it back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end;

[0088] The suggestion providing module is used to provide wound care suggestions to the user terminal according to the wound annotation information.

[0089] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0090] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0091] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0092] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A wound assessment method based on machine learning, characterized in that: The implementation of the wound assessment method comprises the following steps: The user uploads the original image of the wound; Based on the original wound image, wound annotation information is acquired and transmitted back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end; Provide wound care suggestions to the user terminal according to the wound annotation information.

2. The wound assessment method according to claim 1, characterized in that: The step of acquiring wound annotation information based on the original wound image and transmitting the information back to the user terminal includes: Acquiring the conventional wound contour based on an improved contour recognition method; Acquire an approximate beautified contour based on the conventional wound contour, fill the approximate beautified contour with color, and use the approximate beautified contour to replace the conventional wound contour to obtain the beautified wound contour; Obtaining the wound size based on the conventional wound contour; The wound recovery stage is obtained based on the improved wound recognition model.

3. The wound assessment method according to claim 2, characterized in that: The obtaining of the conventional wound contour based on the improved contour recognition method comprises: Uniformly collecting contour sampling points in the original wound image and obtaining contour sampling point coordinates, and obtaining a contour centroid based on the contour sampling point coordinates; Obtaining a centroid height value of the contour sampling point, wherein the centroid height value is the Euclidean distance between the contour sampling point and the contour centroid; Calculating a centroid height increment based on the centroid height value; Acquire a centroid height increment sequence of the i-th contour sampling point based on the centroid height increment; Acquire a centroid height increment matrix based on the centroid height increment sequence; The conventional wound contour is depicted based on the centroid height increment matrix.

4. The wound assessment method according to claim 3, characterized in that: The obtaining of an approximate beautified contour based on the conventional wound contour comprises: Randomly select a positive integer coefficient whose value range is in the open interval (1, N), divide the centroid height increment sequence into S non-overlapping intervals [1, d], [d+1, 2d], ... based on the positive integer coefficient, where S = N / d, d is the positive integer coefficient, and calculate the mean centroid height increment in the interval; Obtaining a feature matrix of the i-th contour sampling point based on the mean of the centroid height increments within the interval; Arranging the feature matrices of all the contour sampling points to obtain a contour feature matrix; Preset a template image contour, and obtain a template contour feature matrix of the template image contour; Calculating the regular path distance between the conventional wound contour and the template image contour; Flipping the conventional wound contour to obtain a flipped conventional wound contour, and obtaining a flipped regular path distance between the flipped conventional wound contour and the template image contour; Obtaining a conventional shape complexity of the conventional wound contour and a template shape complexity of the template image contour; Calculating a similarity coefficient based on the regular path distance, the flipped regular path distance, the conventional shape complexity, and the template shape complexity; The template image contours are traversed, and the template image contour with the smallest similarity coefficient is selected as the approximate beautified contour. The approximate beautified contour filled with color is used to cover the conventional wound contour to obtain the beautified wound contour.

5. The wound assessment method according to claim 4, characterized in that: The calculation formula of the similarity coefficient is: Among them, Q is the similarity coefficient, G is the conventional shape complexity, G Z is the template shape complexity, d DTM (F B ,F Z ) is the flip regularized path distance, d DTM (F,F Z ) is the regularized path distance.

6. The wound assessment method according to claim 2, characterized in that: The improved wound recognition model includes a backbone extraction network CSPDarknet, an enhanced feature extraction network FPN, and a classification regressor YOLOhead.

7. The wound assessment method according to claim 6, characterized in that: The backbone extraction network CSPDarknet includes a Focus block and a Resblock body block, wherein the Focus block is used to sample the pixel values ​​of the original wound image into four feature maps, with three RGB channels, forming a total of twelve feature maps, and the Resblock body block is used to extract features from the feature maps through four maximum pooling kernels; The enhanced feature extraction network FPN is used to further extract feature layers and transform the feature layers of different sizes into feature layers of the same size, connect them in the channel dimension, complete feature fusion and obtain a multi-scale feature map, wherein the feature layer includes bottom-level features and top-level features; The classification regressor YOLOhead includes a classification part and a regression part. The classification part uses two convolution calculations and then one convolution to realize the classification of the multi-scale feature map. The regression part uses two convolutions to extract features, and finally uses two convolutions to perform regression prediction, and finally obtains the wound recovery stage.

8. A wound assessment device based on machine learning, characterized in that: The device is applied to the wound assessment method as described in any one of claims 1 to 7, including an image uploading module, an information annotation module, and a suggestion providing module; The image uploading module is used for users to upload original wound images; The information annotation module is used to obtain wound annotation information based on the original wound image and transmit it back to the user end, the wound annotation information includes wound contour, wound size, and wound recovery stage, the wound contour includes a conventional wound contour and a beautified wound contour, the wound recovery stage includes a coagulation stage, an inflammation stage, a repair stage, and a maturation stage, the user end includes a conventional user end and a pure user end, the conventional wound contour is transmitted back to the conventional user end, and the beautified wound contour is transmitted back to the pure user end; The suggestion providing module is used to provide wound care suggestions to the user terminal according to the wound annotation information.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the wound assessment method according to any one of claims 1 to 7 is implemented.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions are used to perform the wound assessment method according to any one of claims 1 to 7 when executed by a computer processor.

Images