Skin barrier layering detection system based on AI image recognition

Abstract

The invention relates to a skin barrier layering detection system based on AI image recognition, and belongs to the technical field of artificial intelligence and skin detection. The system comprises an image acquisition module, an image adaptive zooming module, a data preprocessing module, an image clustering analysis module, a blue cluster discrimination module, a mask generation module and a grading module. According to the method, the interference of human subjective factors is avoided, the objectivity and accuracy of an image analysis result are ensured, reliable data support is provided for subsequent related research and application, doctors can accurately judge the damage condition of the skin barrier, and the accuracy of the skin barrier damage is improved. And a more accurate basis is provided for research and development of skin care products.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and skin detection technology, and specifically relates to a skin barrier layer detection system based on AI image recognition. Background Art

[0002] The skin barrier is crucial to human health, protecting against harmful external factors, maintaining moisture, and regulating inflammation. Accurately detecting damage at different levels of the barrier is crucial for the diagnosis and treatment of skin diseases and the development of skincare products.

[0003] Currently, clinical testing for damaged skin barrier levels faces numerous limitations. While non-invasive skin physiological function tests, such as those measuring transepidermal water loss (TEWL), water content, pH, and oil content, can determine whether the skin barrier is damaged, they cannot accurately reflect the extent of damage. This is because these tests primarily measure relevant physiological parameters using specific instruments, providing only a holistic assessment of skin barrier function and lacking targeted testing of damaged levels. For example, a skin evaporation meter can only reflect the skin's barrier function against water.

[0004] Lactic acid and capsaicin tests can detect skin barrier damage, but they have significant drawbacks. They cannot determine the extent of damage, are invasive, and can cause discomfort and even skin damage to the test subject, hindering their widespread clinical application. Furthermore, these tests are highly subjective, with responses varying widely between individuals, making it impossible to quantify the extent of damage.

[0005] The fundamental problem with existing technologies is that they focus solely on the overall function of the skin barrier, failing to delve deeper into the different layers of the skin to examine microscopic aspects such as cellular metabolism, keratinocyte maturity, and desquamation. This makes it difficult for clinicians to accurately assess a patient's skin barrier damage and develop precise treatment plans. Skin care product development also lacks a design basis tailored to the different damaged layers of the skin.

[0006] Therefore, how to overcome the shortcomings of the existing technology is a problem that needs to be solved urgently in the current technical field. Summary of the Invention

[0007] The purpose of the present invention is to address the deficiencies of the prior art and provide a skin barrier layer detection system based on AI image recognition.

[0008] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A skin barrier layer detection system based on AI image recognition, including:

[0010] An image acquisition module, used for acquiring skin images to be detected;

[0011] An image adaptive scaling module is connected to the image acquisition module and is used to perform scaling operations of various ratios on the images acquired by the image acquisition module;

[0012] The data preprocessing module is connected to the image adaptive scaling module and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating-point type;

[0013] The image clustering analysis module is connected to the data preprocessing module and is used to perform unsupervised clustering on the preprocessed floating-point pixel array to obtain the cluster label to which each pixel belongs and determine the center color value of each cluster;

[0014] The blue cluster discrimination module is connected to the image cluster analysis module and is used to determine which cluster represents blue by comparing the R component in the color value of each cluster obtained by the image cluster analysis module;

[0015] The mask generation module is connected to the blue cluster discrimination module and is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 and the values ​​corresponding to other pixels to 0, thereby generating an image mask;

[0016] The grade classification module is connected to the mask generation module and is used to count the blue pixel ratios of each scaled image based on the image mask generated by the mask generation module (106), and take the average of the blue pixel ratios of all scaled ratios as the final average blue pixel ratio; then, the grade is divided according to the final average blue pixel ratio to obtain the grade corresponding to the skin to be detected and the detection result.

[0017] Furthermore, the scaling ratios in the image adaptive scaling module include 1 / 7, 1 / 5, 1 / 3 and 1 / 1.

[0018] Furthermore, in the data preprocessing module, the specific method for extracting the pixel values ​​and converting them into a one-dimensional array is as follows: the H×W×3 image matrix is ​​converted into an N×3 two-dimensional array using the img.reshape((-1,3)) method of the OpenCV library, where N = H×W, H is the height of the image, and W is the width of the image;

[0019] The specific method to convert the data type in the array to floating point type is: use the np.float32() function to convert the pixel value from 8-bit integer to 32-bit floating point type.

[0020] Furthermore, in the image cluster analysis module, K-means clustering algorithm is used for clustering;

[0021] The label of each cluster is obtained through the labels matrix returned by cv2.kmeans(). The cluster label refers to the category number to which each pixel is assigned; the center color value of the cluster refers to the representative color BGR value of each category, which is obtained through the centers matrix returned by cv2.kmeans().

[0022] Furthermore, the number of clusters K=2; the category number is 0 or 1.

[0023] Furthermore, the clustering stopping condition is when the change of the cluster center is less than a preset minimum value, or after multiple consecutive iterations.

[0024] Furthermore, the minimum value is 0.01.

[0025] Furthermore, in the blue cluster discrimination module, the R channel value of the central color value of each cluster is compared, and the cluster with the smaller R value is determined to be a blue cluster.

[0026] Furthermore, in the level division module, the specific method of level division is as follows:

[0027] When the final average blue pixel ratio is ≤25%, it is Level 1, indicating normal skin;

[0028] When 25% < the final average blue pixel ratio ≤ 50%, it is Level 2, indicating superficial skin barrier damage, that is, the skin barrier is damaged to the granular layer;

[0029] When 50% < the final average blue pixel ratio ≤ 75%, it is Level 3, indicating that the middle layer of the skin barrier is damaged, that is, the skin barrier is damaged to the spinous layer;

[0030] When 75% < the final average blue pixel ratio ≤ 100%, it is Level 4, indicating deep damage to the skin barrier, that is, the skin barrier is damaged to the basal layer.

[0031] The image adaptive scaling module of the present invention is used to perform scaling operations of multiple proportions on the images collected by the image acquisition module, thereby obtaining multi-scale images, thereby forming a multi-scale image set; the present invention adopts the K-means clustering algorithm in the unsupervised clustering learning method to perform cluster analysis on the multi-scale image set.

[0032] The data preprocessing module of the present invention uses the np.float32() function to convert pixel values ​​from 8-bit integers (0-255) to 32-bit floating-point values, thereby improving the accuracy of clustering calculations.

[0033] In the present invention, H represents the height of an image, that is, the number of pixels of the image in the vertical direction; W represents the width of an image, that is, the number of pixels of the image in the horizontal direction.

[0034] In the image clustering analysis module of the present invention, the K-means clustering algorithm is used for clustering, and the number of clusters K = 2; the label of each cluster is obtained through the labels matrix returned by cv2.kmeans(), and the cluster label refers to the category number assigned to each pixel, and the category number is 0 or 1; the center color of the two clusters refers to the representative color BGR value of each category, which is obtained through the centers matrix returned by cv2.kmeans() (2×3 matrix).

[0035] In the image clustering analysis module of the present invention, the clustering stop condition adopts a dual criterion:

[0036] Maximum number of iterations: 100 (to ensure real-time performance);

[0037] Center point movement threshold (preset minimum value): 0.01 (ε accuracy);

[0038] The calculation is terminated immediately when any condition is met, corresponding to OpenCV's criteria = (cv2.TERM_CRITERIA_EPS+cv2.TERM_CRITERIA_MAX_ITER, 100, 0.01).

[0039] In the blue cluster discrimination module of the present invention, the R channel values ​​of the two center colors are compared: np.argmin(centers[:,2]), and the cluster with the smaller R value is determined to be the blue cluster (because the barrier damage area in the skin image shows blue light characteristics).

[0040] The blue cluster discrimination module of the present invention determines the blue cluster based on the following criteria: in the color space, the smaller the R value, the darker the blue.

[0041] The present invention uses a blue cluster discrimination module and a mask generation module to adaptively distinguish the categories of pixels in an image, that is, the purpose is to accurately discriminate the blue pixels in the image and generate a corresponding image mask.

[0042] For each zoom scale s, s∈{7,5,3,1}, its blue pixel ratio P s The calculation method is:

[0043]

[0044] Among them, H s ×W s is the resolution of the scaled image, H s is the height of the scaled image, W s is the width of the scaled image;

[0045] H×W is the resolution of the original image, H is the height of the original image, and W is the width of the original image;

[0046] After scaling

[0047] M s is a binary mask matrix (size H s ×W s ),satisfy:

[0048]

[0049] It is an indicator function. When the condition is met, the value is 1, otherwise it is 0. The details are as follows:

[0050]

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. Accurate Quantification: Utilizing adaptive image scaling to acquire multi-scale images, the system takes into account both global and detailed features. Image clustering analysis and blue cluster discrimination are then used to accurately identify specific color (blue) pixels within the image and calculate their proportions. Based on this, the image is divided into four levels using different thresholds, enabling precise quantification of the coverage of specific color pixels, addressing the quantitative deficiencies of existing methods for detecting damaged skin barrier layers. Testing has shown that the average clustering accuracy exceeds 90% when processing images from different scenarios, with the different levels of classification highly consistent with the actual proportion of blue areas.

[0053] 2. Easy to operate: The K-means clustering method used is widely used in the field of image processing. The technical solution has clear steps and is easy to implement. It does not require complicated operating procedures and special equipment. Compared with invasive detection methods, it is more convenient for practical application.

[0054] 3. Strong objectivity: By automatically identifying pixel categories and dividing levels based on fixed thresholds, the system avoids interference from subjective factors, ensures the objectivity and accuracy of image analysis results, and provides reliable data support for subsequent related research and applications. It helps doctors accurately judge the damage to the skin barrier and provides a more accurate basis for the development of skin care products.

[0055] 4. Significant economic and social benefits: In the diagnosis and assistance of medical skin diseases, the speed and efficiency of auxiliary diagnosis are greatly improved, and it can be operated without the need for professionals, facilitating precise treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a schematic diagram of the structure of the skin barrier layer detection system based on AI image recognition of the present invention;

[0057] Figure 2collecting a skin image of a person to be detected for the image acquisition module;

[0058] Figure 3 This is the corresponding image after cluster analysis and blue cluster identification;

[0059] Figure 4 is the image mask;

[0060] Figure 5 This is a diagram of the skin structure;

[0061] Figure 6 The skin image to be detected of a user collected by the image acquisition module;

[0062] Figure 7 This is a result diagram obtained by the method of the present invention;

[0063] Figure 8 The result diagram obtained by the prior art processing;

[0064] Figure 9 Graphs of software operation and results from prior art; (a) is a diagram showing the accuracy of results corrected by manual adjustment; and (b) is a diagram showing the result data values. DETAILED DESCRIPTION

[0065] The present invention is described in further detail below with reference to the embodiments.

[0066] Those skilled in the art will understand that the following examples are intended to illustrate the present invention only and should not be construed as limiting the scope of the present invention. Where specific techniques or conditions are not specified in the examples, the techniques or conditions described in the literature in the art or in the product specifications were used. Materials or equipment used without manufacturer identification are commercially available conventional products.

[0067] Example 1

[0068] like Figure 1 As shown, a skin barrier layer detection system based on AI image recognition includes:

[0069] An image acquisition module 101 is used to acquire skin images to be detected;

[0070] The image adaptive scaling module 102 is connected to the image acquisition module 101 and is used to perform scaling operations of various ratios on the image acquired by the image acquisition module 101;

[0071] The data preprocessing module 103 is connected to the image adaptive scaling module 102 and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating-point type;

[0072] The image cluster analysis module 104 is connected to the data preprocessing module 103 and is used to perform unsupervised clustering on the preprocessed floating-point pixel array to obtain the cluster label to which each pixel belongs and determine the center color value of each cluster;

[0073] The blue cluster identification module 105 is connected to the image cluster analysis module 104 and is used to identify which cluster represents blue by comparing the R component in the color value of each cluster obtained by the image cluster analysis module 104;

[0074] The mask generation module 106 is connected to the blue cluster discrimination module 105 and is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 and the values ​​corresponding to other pixels to 0, thereby generating an image mask;

[0075] The grade classification module 107 is connected to the mask generation module 106 and is used to count the blue pixel ratios of each scaled image based on the image mask generated by the mask generation module 106, and take the average of the blue pixel ratios of all scaled ratios as the final average blue pixel ratio; then, the grade is divided according to the final average blue pixel ratio to obtain the grade and detection result corresponding to the skin to be detected.

[0076] Example 2

[0077] like Figure 1 As shown, a skin barrier layer detection system based on AI image recognition includes:

[0078] An image acquisition module 101 is used to acquire skin images to be detected;

[0079] The image adaptive scaling module 102 is connected to the image acquisition module 101 and is used to perform scaling operations of various ratios on the image acquired by the image acquisition module 101;

[0080] The data preprocessing module 103 is connected to the image adaptive scaling module 102 and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating-point type;

[0081] The image cluster analysis module 104 is connected to the data preprocessing module 103 and is used to perform unsupervised clustering on the preprocessed floating-point pixel array to obtain the cluster label to which each pixel belongs and determine the center color value of each cluster;

[0082] The blue cluster identification module 105 is connected to the image cluster analysis module 104 and is used to identify which cluster represents blue by comparing the R component in the color value of each cluster obtained by the image cluster analysis module 104;

[0083] The mask generation module 106 is connected to the blue cluster discrimination module 105 and is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 and the values ​​corresponding to other pixels to 0, thereby generating an image mask;

[0084] The grade classification module 107 is connected to the mask generation module 106 and is used to count the blue pixel ratios of each scaled image based on the image mask generated by the mask generation module 106, and take the average of the blue pixel ratios of all scaled ratios as the final average blue pixel ratio; then, the grade is divided according to the final average blue pixel ratio to obtain the grade and detection result corresponding to the skin to be detected.

[0085] The scaling ratios in the image adaptive scaling module 102 include 1 / 7, 1 / 5, 1 / 3 and 1 / 1.

[0086] In the data preprocessing module 103, the specific method of extracting the pixel values ​​and converting them into a one-dimensional array is as follows: using the img.reshape((-1,3)) method of the OpenCV library to convert the H×W×3 image matrix into an N×3 two-dimensional array, where N=H×W, H is the height of the image, and W is the width of the image;

[0087] The specific method to convert the data type in the array to floating point type is: use the np.float32() function to convert the pixel value from 8-bit integer to 32-bit floating point type.

[0088] In the image cluster analysis module 104, K-means clustering algorithm is used for clustering;

[0089] The label of each cluster is obtained through the labels matrix returned by cv2.kmeans(). The cluster label refers to the category number to which each pixel is assigned; the center color value of the cluster refers to the representative color BGR value of each category, which is obtained through the centers matrix returned by cv2.kmeans().

[0090] In the blue cluster identification module 105 , the R channel value of the central color value of each cluster is compared, and the cluster with the smaller R value is determined to be a blue cluster.

[0091] In the level classification module 107, the specific method of level classification is as follows:

[0092] When the final average blue pixel ratio is ≤25%, it is Level 1, indicating normal skin;

[0093] When 25% < the final average blue pixel ratio ≤ 50%, it is Level 2, indicating superficial skin barrier damage, that is, the skin barrier is damaged to the granular layer;

[0094] When 50% < the final average blue pixel ratio ≤ 75%, it is Level 3, indicating that the middle layer of the skin barrier is damaged, that is, the skin barrier is damaged to the spinous layer;

[0095] When 75% < the final average blue pixel ratio ≤ 100%, it is Level 4, indicating deep damage to the skin barrier, that is, the skin barrier is damaged to the basal layer.

[0096] Example 3

[0097] like Figure 1 As shown, a skin barrier layer detection system based on AI image recognition includes:

[0098] An image acquisition module 101 is used to acquire skin images to be detected;

[0099] The image adaptive scaling module 102 is connected to the image acquisition module 101 and is used to perform scaling operations of various ratios on the image acquired by the image acquisition module 101;

[0100] The data preprocessing module 103 is connected to the image adaptive scaling module 102 and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating-point type;

[0101] The image cluster analysis module 104 is connected to the data preprocessing module 103 and is used to perform unsupervised clustering on the preprocessed floating-point pixel array to obtain the cluster label to which each pixel belongs and determine the center color value of each cluster;

[0102] The blue cluster identification module 105 is connected to the image cluster analysis module 104 and is used to identify which cluster represents blue by comparing the R component in the color value of each cluster obtained by the image cluster analysis module 104;

[0103] The mask generation module 106 is connected to the blue cluster discrimination module 105 and is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 and the values ​​corresponding to other pixels to 0, thereby generating an image mask;

[0104] The grade classification module 107 is connected to the mask generation module 106 and is used to count the blue pixel ratios of each scaled image based on the image mask generated by the mask generation module 106, and take the average of the blue pixel ratios of all scaled ratios as the final average blue pixel ratio; then, the grade is divided according to the final average blue pixel ratio to obtain the grade and detection result corresponding to the skin to be detected.

[0105] The scaling ratios in the image adaptive scaling module 102 include 1 / 7, 1 / 5, 1 / 3 and 1 / 1.

[0106] In the data preprocessing module 103, the specific method of extracting the pixel values ​​and converting them into a one-dimensional array is as follows: using the img.reshape((-1,3)) method of the OpenCV library to convert the H×W×3 image matrix into an N×3 two-dimensional array, where N=H×W, H is the height of the image, and W is the width of the image;

[0107] The specific method to convert the data type in the array to floating point type is: use the np.float32() function to convert the pixel value from 8-bit integer to 32-bit floating point type.

[0108] In the image cluster analysis module 104, K-means clustering algorithm is used for clustering;

[0109] The label of each cluster is obtained through the labels matrix returned by cv2.kmeans(). The cluster label refers to the category number to which each pixel is assigned; the center color value of the cluster refers to the representative color BGR value of each category, which is obtained through the centers matrix returned by cv2.kmeans().

[0110] The number of clusters K = 2; the category number is 0 or 1.

[0111] The clustering stopping condition is when the change of the cluster center is less than a preset minimum value, or after multiple consecutive iterations.

[0112] The minimum value is 0.01.

[0113] In the blue cluster identification module 105 , the R channel value of the central color value of each cluster is compared, and the cluster with the smaller R value is determined to be a blue cluster.

[0114] In the level classification module 107, the specific method of level classification is as follows:

[0115] When the final average blue pixel ratio is ≤25%, it is Level 1, indicating normal skin;

[0116] When 25% < the final average blue pixel ratio ≤ 50%, it is Level 2, indicating superficial skin barrier damage, that is, the skin barrier is damaged to the granular layer;

[0117] When 50% < the final average blue pixel ratio ≤ 75%, it is Level 3, indicating that the middle layer of the skin barrier is damaged, that is, the skin barrier is damaged to the spinous layer;

[0118] When 75% < the final average blue pixel ratio ≤ 100%, it is Level 4, indicating deep damage to the skin barrier, that is, the skin barrier is damaged to the basal layer.

[0119] Example 4

[0120] A skin barrier layer detection system based on AI image recognition, including:

[0121] 1. Image acquisition module 101, used to acquire a skin image to be detected; for example, acquiring a skin image of a person to be detected is shown in Figure 2;

[0122] 2. Image adaptive scaling module 102: This module aims to obtain images with different scale features to meet the needs of subsequent multi-scale clustering analysis. Using the image scaling algorithm, starting from the original image collected by the image acquisition module 101, the image is scaled according to the scaling ratios of 1 / 7, 1 / 5, 1 / 3, and 1 / 1 (i.e., the original image size). Among them, large-scale scaling (such as 1 / 7) can effectively process the global features of the image and avoid the interference of local noise on the overall analysis; while small-scale scaling (such as 1 / 1) can retain the detailed features of the image, which is convenient for subsequent fine-grained analysis. Through this operation, a multi-scale image set containing information at different levels can be obtained, providing a rich data foundation for subsequent clustering analysis.

[0123] Among them, the image scaling algorithm is bicubic interpolation, so as to achieve image scaling. The specific implementation method is based on cv2.resize(img,None,fx=1 / size,fy=1 / size,interpolation=cv2.INTER_CUBIC) of the OpenCV library.

[0124] The scaling ratio refers to the ratio of the length of the image after scaling to the length of the original image. For example, 1 / 7 means that the length and width are both reduced to 1 / 7 of the original image.

[0125] 3. The data preprocessing module 103 is connected to the image adaptive scaling module 102 and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating-point type for subsequent numerical calculation and processing.

[0126] The specific method of extracting its pixel values ​​and converting them into a one-dimensional array is as follows: Use the img.reshape((-1,3)) method of the OpenCV library to convert the H×W×3 image matrix into an N×3 two-dimensional array, where N=H×W, H is the height of the image, and W is the width of the image;

[0127] The specific method to convert the data type in the array to floating point type is: use the np.float32() function to convert the pixel value from 8-bit integer to 32-bit floating point type.

[0128] 4. Image cluster analysis module 104, connected to data preprocessing module 103, is used to perform unsupervised clustering on the preprocessed floating-point pixel array to obtain the cluster label to which each pixel belongs and determine the center color value of each cluster;

[0129] Through this process, the preliminary classification of image pixels is achieved, laying the foundation for the subsequent determination of pixel categories.

[0130] Parameter settings in the image clustering analysis module 104: Based on the characteristics of the acquired image, define the stopping condition and the number of clusters, K, for the K-means clustering algorithm. In this invention, K is set to 2, indicating that the image pixels are expected to be divided into two clusters, blue and white (because barrier damage areas in skin images exhibit blue light characteristics, while white clusters represent normal skin areas). The stopping condition can be set to stop when the change in cluster center is less than a preset minimum value (e.g., 0.01), or when the clustering results no longer change significantly after multiple iterations.

[0131] The labels matrix returned by cv2.kmeans() is used to obtain the labels of each cluster. The cluster labels are the class numbers assigned to each pixel. The center color values ​​of the clusters are the BGR values ​​representing the representative colors of each class, which are obtained by using the centers matrix returned by cv2.kmeans(). Preferably, the class numbers are 0 or 1.

[0132] 5. The blue cluster identification module 105 is connected to the image cluster analysis module 104 and is used to determine which cluster represents blue by comparing the R component in the color value based on the central color value of each cluster obtained by the image cluster analysis module 104. That is, the R channel value of the central color value of each cluster is compared and the cluster with the smaller R value is determined as the blue cluster. In the color space, the smaller the R value, the darker the blue. Based on this, the blue cluster is determined. The corresponding image after the identification is as follows: Figure 3 As shown;

[0133] 6. The mask generation module 106 is connected to the blue cluster discrimination module 105 and is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 (indicating white, which is often used to represent the target object in a binary image) and the values ​​corresponding to other pixels to 0 (indicating black, which is often used to represent the background in a binary image), thereby generating an image mask. For example, the generated image mask is shown in FIG. Figure 4 As shown; in this way, the blue pixels in the image are distinguished from other pixels, which facilitates the subsequent statistics and analysis of the blue pixels.

[0134] That is, after the blue cluster is determined by the blue cluster identification module 105 , the mask generation module 106 generates an image mask having the same size as the scaled image.

[0135] 7. A grading module 107 is connected to the mask generation module 106 and is configured to calculate the blue pixel ratio of each scaled image based on the image mask generated by the mask generation module 106, and take the average of the blue pixel ratios at all scaled ratios as the final average blue pixel ratio. The grading module 107 then divides the skin into grades based on the final average blue pixel ratio to obtain the grade corresponding to the skin to be detected and the specific method for grading the detection results is as follows:

[0136] When the final average blue pixel ratio is ≤25%, it is Level 1, indicating normal skin;

[0137] When 25% < the final average blue pixel ratio ≤ 50%, it is Level 2, indicating superficial skin barrier damage, that is, the skin barrier is damaged to the granular layer;

[0138] When 50% < the final average blue pixel ratio ≤ 75%, it is Level 3, indicating that the middle layer of the skin barrier is damaged, that is, the skin barrier is damaged to the spinous layer;

[0139] When 75% < the final average blue pixel ratio ≤ 100%, it is Level 4, indicating deep damage to the skin barrier, that is, the skin barrier is damaged to the basal layer.

[0140] The corresponding skin structure diagram is as follows Figure 5 shown.

[0141] Through the above series of operations, the present invention can achieve accurate quantification and grading of the blue pixel coverage in the image, and has important application value in related fields such as image analysis and quantitative evaluation.

[0142] Application Examples

[0143] The image acquisition module collects a user's skin image to be tested - the skin keratinocyte shedding staining image, the original microscope image of which is as follows Figure 6 As shown, using the system of the present invention, the final average blue pixel ratio is detected to be 44.98%, that is, the superficial layer of the skin barrier is damaged. The detection result is shown in the figure Figure 7 As shown.

[0144] The result obtained by using existing technology is that the proportion of exfoliated cells is 0.9157%, which is normal skin. The test results are shown in the figure below. Figure 8 As shown. Among them, the existing technology is a binary method. The principle is: based on the grayscale or color value of the pixel, a threshold is set (manually or adaptively), and the image pixels are simply divided into two categories of color, blue area and non-blue area; for example: if the pixel value ≥ the threshold → it is set to blue, otherwise it is set to black, non-blue. Software operation and results of the existing technology Figure 9 shown.

[0145] Afterwards, through the evaluation of clinical symptoms and signs, VISIA skin tester, and West China Sensitive Skin Questionnaire survey results, it was analyzed that the user's skin barrier was indeed damaged and not normal skin.

[0146] because Figure 5 The dyeing color is relatively light, which makes the original technology unable to identify the lightly dyed blue pixels, and the human eye cannot make normal judgments; however, the system of the present invention can solve the above technical problems and obtain accurate detection results.

[0147] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A skin barrier layer detection system based on AI image recognition, characterized in that: include: An image acquisition module (101) is used to acquire skin images to be detected; An image adaptive scaling module (102), connected to the image acquisition module (101), is used to perform scaling operations of various proportions on the image acquired by the image acquisition module (101); A data preprocessing module (103) is connected to the image adaptive scaling module (102) and is used to extract the pixel values ​​of each scaled image and convert them into a one-dimensional array, and then convert the data type in the array into a floating point type; An image cluster analysis module (104), connected to the data preprocessing module (103), is used to perform unsupervised clustering on the preprocessed floating-point pixel array, thereby obtaining the cluster label to which each pixel belongs and determining the center color value of each cluster; A blue cluster discrimination module (105) is connected to the image cluster analysis module (104) and is used to discriminate which cluster represents blue by comparing the R component in the color value of each cluster obtained by clustering the image cluster analysis module (104); A mask generation module (106), connected to the blue cluster discrimination module (105), is used to set the values ​​corresponding to the pixels belonging to the blue cluster to 255 and the values ​​corresponding to other pixels to 0, thereby generating an image mask; The grade classification module (107) is connected to the mask generation module (106) and is used to count the blue pixel ratios of each scaled image based on the image mask generated by the mask generation module (106), and take the average of the blue pixel ratios of all scaled ratios as the final average blue pixel ratio; then, the grade is divided according to the final average blue pixel ratio to obtain the grade corresponding to the skin to be detected and the detection result.

2. The skin barrier layer detection system based on AI image recognition according to claim 1, characterized in that: The scaling ratios in the image adaptive scaling module (102) include 1 / 7, 1 / 5, 1 / 3 and 1 / 1.

3. The skin barrier layer detection system based on AI image recognition according to claim 1, characterized in that: In the data preprocessing module (103), the specific method for extracting the pixel values ​​and converting them into a one-dimensional array is as follows: using the img.reshape((-1,3)) method of the OpenCV library to convert the H×W×3 image matrix into an N×3 two-dimensional array, where N=H×W, H is the height of the image, and W is the width of the image; The specific method to convert the data type in the array to floating point type is: use the np.float32() function to convert the pixel value from 8-bit integer to 32-bit floating point type.

4. The skin barrier layer detection system based on AI image recognition according to claim 1, characterized in that: In the image cluster analysis module (104), K-means clustering algorithm is used for clustering; The label of each cluster is obtained through the labels matrix returned by cv2.kmeans(). The cluster label refers to the category number to which each pixel is assigned; the center color value of the cluster refers to the representative color BGR value of each category, which is obtained through the centers matrix returned by cv2.kmeans().

5. The skin barrier layer detection system based on AI image recognition according to claim 4 is characterized in that: The number of clusters K = 2; the category number is 0 or 1.

6. The skin barrier layer detection system based on AI image recognition according to claim 4, characterized in that: The clustering stopping condition is when the change of the cluster center is less than a preset minimum value, or after multiple consecutive iterations.

7. The skin barrier layer detection system based on AI image recognition according to claim 6, characterized in that: The minimum value is 0.

01.

8. The skin barrier layer detection system based on AI image recognition according to claim 1, characterized in that: In the blue cluster identification module (105), the R channel value of the central color value of each cluster is compared, and the cluster with the smaller R value is determined to be a blue cluster.

9. The skin barrier layer detection system based on AI image recognition according to claim 1, characterized in that: In the level classification module (107), the specific method of level classification is as follows: When the final average blue pixel ratio is ≤25%, it is Level 1, indicating normal skin; When 25% < the final average blue pixel ratio ≤ 50%, it is Level 2, indicating superficial skin barrier damage, that is, the skin barrier is damaged to the granular layer; When 50% < the final average blue pixel proportion ≤ 75%, it is Level 3, indicating that the middle layer of the skin barrier is damaged, that is, the skin barrier is damaged to the spinous layer; When 75% < the final average blue pixel ratio ≤ 100%, it is Level 4, indicating deep damage to the skin barrier, that is, the skin barrier is damaged to the basal layer.

Images