Tongue image classification method for patients with systemic lupus erythematosus for different traditional Chinese medicine syndromes

Abstract

The invention relates to the technical field of image analysis, in particular to a systemic lupus erythematosus patient tongue image classification method for different traditional Chinese medicine syndromes, and the method comprises the steps: obtaining a target tongue surface image corresponding to a to-be-detected lupus erythematosus patient, and carrying out the tongue image classification through a tongue image classification network; performing interval division on the S channel value in each obtained lupus erythematosus lingual surface image; performing connected domain extraction on a region corresponding to each S channel interval; carrying out missing condition analysis processing on the holes in each multi-connected region; based on the determined comprehensive supplement index, carrying out adaptive supplement division analysis; and based on the lupus erythematosus feature obvious factors, screening out a reference image set. According to the invention, through image analysis, the obvious condition of lupus erythematosus features of the lupus erythematosus lingual surface image is quantified, and the possibility that the tongue image classification network learns inaccurate lupus erythematosus feature information is reduced to a certain extent, so that the accuracy of tongue image classification of the lupus erythematosus patient is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a tongue image classification method for patients with systemic lupus erythematosus with different traditional Chinese medicine syndromes. Background Art

[0002] With the development of science and technology, neural networks are being used more and more widely. For example, they can be used to classify tongue images of patients with systemic lupus erythematosus with different TCM syndromes. Since the selection of neural network training sets often affects the training results of neural network, the acquisition of training sets is crucial. At present, when acquiring training sets for neural networks, the method usually used is to obtain images of objects of the same category as the object to be detected to form a training set.

[0003] However, when obtaining a set of lupus erythematosus tongue images and a set of normal tongue images to form a training set for a neural network, and training a neural network for lupus erythematosus patient tongue image classification, the following technical problems often occur: Since a large number of training samples are often needed for training neural networks, and the sources of training samples are often different, and the shooting methods and resolutions of shooting equipment are often different, the lupus features in some lupus tongue images in the lupus tongue image set may not be obvious, which may cause the neural network to learn inaccurate information in the training set, affecting the generalization ability of the neural network, and further resulting in poor accuracy in the classification of tongue images of lupus patients. Summary of the invention

[0004] In order to solve the technical problem of poor accuracy in tongue image classification of systemic lupus erythematosus patients, the present invention proposes a tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes.

[0005] In a first aspect, the present invention provides a method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes, the method comprising: Acquire a target tongue surface image corresponding to a lupus erythematosus patient to be detected; According to the target tongue surface image, the target tongue surface image is classified by a pre-trained tongue image classification network, wherein the training set of the tongue image classification network is the target training set, and the target training set includes a normal tongue surface image set and a reference image set; The method for obtaining the reference image set includes: Obtain the lupus erythematosus tongue surface image set and the S channel value corresponding to each pixel point in each lupus erythematosus tongue surface image in the lupus erythematosus tongue surface image; The S channel values ​​corresponding to all the pixels in each lupus erythematosus tongue surface image are divided into intervals to obtain the S channel interval corresponding to each lupus erythematosus tongue surface image; For each S channel interval, the connected domain is extracted from the region in the lupus erythematosus tongue surface image to which it belongs, and multi-connected regions are screened out from the extracted connected domains; Analyze and process the missing conditions of all holes in each multi-connected region to obtain the target missing index corresponding to each multi-connected region; According to the target missing index corresponding to all multi-connected regions corresponding to each S channel interval and the distribution of the connected domains of the corresponding S channel intervals as adjacent S channel intervals, the comprehensive supplementary index corresponding to each S channel interval is determined; Based on the comprehensive supplementary indexes corresponding to all S channel intervals in each lupus erythematosus tongue surface image, an adaptive supplementary division analysis is performed to obtain the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue surface image; The lupus erythematosus tongue surface images whose corresponding lupus erythematosus characteristic obvious factors are greater than the preset characteristic threshold are screened out from the lupus erythematosus tongue surface image set to form a reference image set.

[0006] In combination with the first aspect above, in a possible implementation, the S channel values ​​corresponding to all pixels in each lupus erythematosus tongue surface image are divided into intervals to obtain the S channel interval corresponding to each lupus erythematosus tongue surface image, including: Determine any lupus erythematosus tongue image as a marked image, and make an S channel histogram corresponding to the marked image, recorded as a marked S channel histogram; The S channel values ​​between the S channel values ​​corresponding to every two adjacent minimum values ​​in the marked S channel histogram constitute an S channel interval.

[0007] In combination with the first aspect above, in a possible implementation, performing missing condition analysis on all holes in each multi-connected region to obtain a target missing index corresponding to each multi-connected region includes: The area ratio of each hole in the multiply connected region to which it belongs is determined as the initial missing factor corresponding to each hole; Determine the erythema missing contribution factor corresponding to each hole in each multi-connected region according to the difference between the initial missing factor corresponding to each hole in each multi-connected region and the initial missing factors corresponding to other holes; According to the erythema missing contribution factor and the initial missing factor corresponding to all the holes in each multi-connected region, the target missing index corresponding to each multi-connected region is determined, wherein the erythema missing contribution factor and the initial missing factor are positively correlated with the target missing index.

[0008] In combination with the first aspect above, in a possible implementation manner, determining the erythema missing contribution factor corresponding to each hole in each multiply connected region according to the difference between the initial missing factor corresponding to each hole in each multiply connected region and the initial missing factors corresponding to other holes includes: Determine any multiply connected region as a marked multiply connected region, determine any hole in the marked multiply connected region as a marked hole, and determine each hole in the marked multiply connected region except the marked hole as a reference hole; The accumulated value of the absolute value of the difference between the initial missing factor corresponding to the marked hole and the initial missing factor corresponding to each reference hole is determined as the target missing difference corresponding to the marked hole; According to the target loss difference corresponding to the marked hole, the erythema loss contribution factor corresponding to the marked hole is determined, wherein the target loss difference is positively correlated with the erythema loss contribution factor.

[0009] In combination with the first aspect above, in a possible implementation, the comprehensive supplementary index corresponding to each S channel interval is determined according to the target missing index corresponding to all multi-connected regions corresponding to each S channel interval and the distribution of the connected domains of the corresponding S channel intervals as adjacent S channel intervals, including: According to all connected domains corresponding to the adjacent S channel intervals of the S channel interval corresponding to each hole, a supplementary candidate connected domain is screened out from each hole; According to the target missing index corresponding to each multi-connected region, an initial supplement factor corresponding to each supplementary candidate connected domain in each hole in each multi-connected region is determined; Determine the supplementation effectiveness corresponding to each supplementary candidate connected domain in each hole in each multi-connected region according to the number of edge pixels of each hole and each supplementary candidate connected domain in each multi-connected region; According to the initial supplementation factors and supplementation effectiveness corresponding to all supplementary candidate connected domains in each hole in each multi-connected region, the adjacent supplementation index corresponding to each hole in each multi-connected region is determined, wherein the initial supplementation factor and the supplementation effectiveness are both positively correlated with the adjacent supplementation index; The average of the adjacent supplementary indices corresponding to all holes in all multi-connected regions corresponding to each S channel interval is determined as the comprehensive supplementary index corresponding to each S channel interval.

[0010] In combination with the first aspect, in a possible implementation, the selecting a supplementary candidate connected domain from each hole according to all connected domains corresponding to adjacent S channel intervals of the S channel interval corresponding to each hole includes: Determine any hole as a candidate hole, and determine the S channel interval corresponding to the multiply connected region to which the candidate hole belongs as a candidate S channel interval; Determine each S channel interval adjacent to the candidate S channel interval as an S channel interval to be supplemented, and determine each connected domain corresponding to each S channel interval to be supplemented as a calibrated connected domain; Each calibrated connected domain within the candidate hole is determined as a supplementary candidate connected domain.

[0011] In combination with the first aspect above, in a possible implementation, the formula corresponding to the initial supplementation factor corresponding to the supplementary candidate connected domain in the hole in the multi-connected region is: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The initial supplementary factor corresponding to the supplementary candidate connected domains; It is the serial number of the lupus erythematosus tongue surface image; It is The serial number of the multi-connected regions in the lupus erythematosus tongue image; It is The sequence number of the holes in the multiply connected regions; It is The serial number of the candidate connected domain to be added in each hole; is the normalization function; It is In the image of lupus erythematosus tongue The target missing index corresponding to the multi-connected regions; is the target missing index corresponding to the target connected domain; the method for obtaining the target connected domain is: The supplementary candidate connected regions are filled into the In a multiply connected region holes, so that the The area reduced by the first hole is equal to The area corresponding to the supplementary candidate connected domains is The area of ​​the multiply connected region is equal to the The area corresponding to the supplementary candidate connected domain is realized. In the image of lupus erythematosus tongue The updated multi-connected region is recorded as the target connected domain.

[0012] In combination with the first aspect above, in a possible implementation, the formula corresponding to the supplementary validity degree of the supplementary candidate connected domains in the holes in the multi-connected regions is: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The degree of supplementary effectiveness corresponding to the supplementary candidate connected domains; It is the serial number of the lupus erythematosus tongue surface image; It is The serial number of the multi-connected regions in the lupus erythematosus tongue image; It is The sequence number of the holes in the multiply connected regions; It is The serial number of the candidate connected domain to be added in each hole; is the normalization function; It is In the image of lupus erythematosus tongue In a multiply connected region The edge of the hole and the The number of pixels in the intersection between the edges of the supplementary candidate connected domains; It is In the image of lupus erythematosus tongue In a multiply connected region The number of pixels on the edge of a hole; It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The number of pixels on the edge of the supplementary candidate connected domain.

[0013] In combination with the first aspect above, in a possible implementation, the adaptive supplementary division analysis is performed based on the comprehensive supplementary indicators corresponding to all S channel intervals in each lupus erythematosus tongue image to obtain the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue image, including: According to the comprehensive supplementary index corresponding to all S channel intervals in each lupus erythematosus tongue surface image, the S channel intervals in each lupus erythematosus tongue surface image are clustered into two categories, and the continuous S channel intervals in the category with a higher comprehensive supplementary index constitute a continuous supplementary interval, and the continuous S channel intervals in the other category with a lower comprehensive supplementary index constitute a continuous separation interval; According to the area filling situation between adjacent S channel intervals in each continuous supplementary interval, the characteristic separation interval is selected from the continuous separation intervals adjacent to each continuous supplementary interval; According to all the continuous supplement intervals in each lupus erythematosus tongue surface image and its corresponding characteristic separation interval, the formula corresponding to the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue surface image is determined as follows: ;in, It is Lupus erythematosus characteristic obvious factors corresponding to lupus erythematosus tongue images; It is the serial number of the lupus erythematosus tongue surface image; is the normalization function; It is The number of consecutive supplementation intervals in the lupus erythematosus tongue surface images; It is The serial number of the continuous supplement interval in the lupus erythematosus tongue image; It is In the image of lupus erythematosus tongue The number of S channel values ​​included in a continuous supplementary interval; It is In the image of lupus erythematosus tongue The mean value of the target missing index corresponding to all multi-connected regions corresponding to all S channel intervals in the feature separation interval corresponding to the continuous supplementary interval.

[0014] In combination with the first aspect, in a possible implementation, the step of selecting a characteristic separation interval from continuous separation intervals adjacent to each continuous supplementary interval according to the area filling condition between adjacent S channel intervals in each continuous supplementary interval includes: Determine any one of the continuous supplement intervals as a marked continuous supplement interval, and sequentially determine any two adjacent S channel intervals in the marked continuous supplement interval as a marked left S channel interval and a marked right S channel interval; The total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked right S channel interval and belonging to the marked left S channel interval is determined as the left filling area between the marked left S channel interval and the marked right S channel interval; The total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked left S channel interval and belonging to the marked right S channel interval is determined as the right filling area between the marked left S channel interval and the marked right S channel interval; Determine the accumulated value of the left padding areas between all adjacent S channel intervals in the marked continuous supplement interval as the overall left padding factor corresponding to the marked continuous supplement interval; Determine the accumulated value of the right filling area between all adjacent S channel intervals in the marked continuous supplement interval as the overall right filling factor corresponding to the marked continuous supplement interval; If the overall left padding factor corresponding to the continuous mark supplement interval is greater than or equal to the overall right padding factor corresponding to it, then the continuous separation interval adjacent to the continuous mark supplement interval and on the left side of the continuous mark supplement interval is determined as the characteristic separation interval corresponding to the continuous mark supplement interval; If the overall left padding factor corresponding to the marked continuous supplement interval is smaller than its corresponding overall right padding factor, the continuous separation interval adjacent to the marked continuous supplement interval and on the right side of the marked continuous supplement interval will be determined as the characteristic separation interval corresponding to the marked continuous supplement interval.

[0015] In a second aspect, the present invention provides a tongue image classification system for patients with systemic lupus erythematosus with different TCM syndromes, the system comprising: A tongue surface image acquisition module is used to acquire a target tongue surface image corresponding to the lupus erythematosus patient to be detected; The tongue image classification module is used to classify the target tongue surface image according to the target tongue surface image through a pre-trained tongue image classification network.

[0016] In a third aspect, a server is provided, comprising a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the method in the first aspect or any possible implementation of the first aspect.

[0017] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0018] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0019] The present invention has the following beneficial effects: The tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes of the present invention quantifies the obvious lupus characteristics of lupus erythematosus tongue images through image analysis, and to a certain extent reduces the possibility of the tongue image classification network learning inaccurate lupus characteristic information, and solves the technical problem of poor accuracy in the classification of tongue images of lupus patients, thereby improving the accuracy of the classification of tongue images of lupus patients. When training the tongue image classification network, compared with directly forming a training set with a lupus erythematosus tongue image set and a normal tongue image set, the present invention comprehensively considers multiple indicators related to the obvious lupus characteristics, such as target missing indicators and comprehensive supplementary indicators. In this way, the lupus characteristic factor corresponding to each lupus tongue image is quantified, and then the lupus tongue images whose corresponding lupus characteristic factor is greater than the preset characteristic threshold are screened out, so that the lupus tongue images with smaller lupus characteristic factors do not constitute the training set, which reduces the possibility of the tongue image classification network learning inaccurate lupus characteristic information to a certain extent, thereby improving the accuracy of tongue image classification of lupus patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A flow chart of the tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes of the present invention; Figure 2 A schematic diagram of the process of obtaining a reference image set of the present invention; Figure 3 A schematic diagram of the composition structure of the tongue image classification system for systemic lupus erythematosus patients with different TCM syndromes according to the present invention; Figure 4 The figure is a schematic diagram of the structure of a computer device of the present invention. DETAILED DESCRIPTION

[0022] 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 technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0023] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0024] Systemic lupus erythematosus is a complex autoimmune disease in which the patient's immune system attacks its own tissues and organs, causing inflammation and damage. The hazards of systemic lupus erythematosus mainly include damage to multiple organs, which may lead to kidney inflammation, heart and lung problems, damage to the nervous system and anemia. In addition, long-term inflammation and immune system disorders may increase the risk of infection, which may affect the patient's quality of life or even endanger his life in severe cases. Improper treatment may lead to complications, which require regular monitoring and management.

[0025] In traditional Chinese medicine, the symptoms of systemic lupus erythematosus are mainly abnormal manifestations of the tongue and pulse. Combined with possible fever, fatigue, joint pain, and erythema on the skin, a comprehensive diagnosis of systemic lupus erythematosus is made. When diagnosing systemic lupus erythematosus through tongue images, the classification of tongue images for different TCM syndromes has important clinical significance. Tongue images often reflect the patient's overall health status and visceral function. Its accurate classification can help TCM doctors better identify the cause of the disease, judge the progression of the disease, and develop personalized treatment plans. In addition, changes in tongue images can serve as an important basis for disease monitoring and efficacy evaluation, which helps to improve treatment outcomes and the quality of life of patients.

[0026] refer to Figure 1 , shows the process of some embodiments of the method for classifying tongue images of systemic lupus erythematosus patients with different TCM syndromes according to the present invention. The method for classifying tongue images of systemic lupus erythematosus patients with different TCM syndromes comprises the following steps: Step S1, obtaining a target tongue surface image corresponding to a lupus erythematosus patient to be detected.

[0027] The lupus erythematosus patient to be detected may be a patient to be classified into lupus erythematosus syndrome conditions. The classification of lupus erythematosus syndrome conditions may be achieved by tongue image classification. Lupus erythematosus syndrome conditions may be, but are not limited to, normal, rheumatic heat arthralgia syndrome, yin deficiency and internal heat syndrome, phlegm-heat stagnation in the lung syndrome, and liver depression and blood stasis syndrome. The target tongue surface image may be a tongue surface image of the lupus erythematosus patient to be detected.

[0028] As an example, an RGB (Red Green Blue, color mode) image of the tongue of a lupus erythematosus patient to be tested can be captured by a camera, and the tongue surface area can be segmented from the RGB image through threshold segmentation, and the image block where the tongue surface area segmented at this time is located is used as the target tongue surface image.

[0029] It should be noted that the tongue surface area can also be identified through a neural network. For example, it can be identified through a convolutional neural network (CNN) with a U-Net structure. The loss function used in the training process can be a cross-entropy loss function. The input of the neural network can be a captured tongue surface image, and the output can be a main tongue surface image of the tongue surface image.

[0030] Step S2: classify the target tongue surface image according to the target tongue surface image by using a pre-trained tongue image classification network.

[0031] Among them, the tongue image classification network can be ResNet (Residual Neural Network). The training set of the tongue image classification network can be a target training set. The target training set can include a normal tongue surface image set and a reference image set. The normal tongue surface images in the normal tongue surface image set can be tongue surface images of people who do not suffer from lupus erythematosus. The reference images in the reference image set can be tongue surface images with relatively obvious lupus erythematosus features. The method for obtaining the normal tongue surface image set can be: using a camera, collecting multiple RGB images of the tongue surface of people who do not suffer from lupus erythematosus, and segmenting the tongue surface areas from these RGB images through threshold segmentation, and the image blocks where these tongue surface areas are segmented at this time constitute a normal tongue surface image set.

[0032] As an example, the target tongue surface image can be input into the tongue image classification network, and the tongue image classification of the target tongue surface image can be achieved through the tongue image classification network.

[0033] refer to Figure 2 , the method for obtaining the reference image set may include the following steps: Step 201 , obtaining a lupus erythematosus tongue surface image set and an S channel value corresponding to each pixel point in each lupus erythematosus tongue surface image in the lupus erythematosus tongue surface image set.

[0034] The lupus erythematosus tongue surface image in the lupus erythematosus tongue surface image set may be a tongue surface image of a patient suffering from lupus erythematosus. The method for obtaining the S channel value corresponding to a pixel point may be: converting the lupus erythematosus tongue surface image to which the pixel point belongs into an HSV (Hue Saturation Value, color model) image to obtain the S (Saturation, saturation) channel value of the pixel point.

[0035] As an example, a camera can be used to collect RGB images of the tongue surfaces of multiple patients with lupus erythematosus, and the tongue surface areas can be segmented from these RGB images through threshold segmentation. The image blocks containing these segmented tongue surface areas constitute a lupus erythematosus tongue surface image set.

[0036] Step 202 , dividing the S channel values ​​corresponding to all the pixels in each lupus erythematosus tongue surface image into intervals to obtain the S channel interval corresponding to each lupus erythematosus tongue surface image.

[0037] As an example, this step may include the following steps: In the first step, any lupus erythematosus tongue image is determined as a labeled image, and an S channel histogram corresponding to the labeled image is made, which is recorded as a labeled S channel histogram.

[0038] For example, a histogram constructed with the S channel value corresponding to the pixel in the marked image as the horizontal coordinate and the number of pixels with the same S channel value in the marked image as the vertical coordinate is the S channel histogram corresponding to the marked image.

[0039] In the second step, the S channel values ​​between the S channel values ​​corresponding to each two adjacent minimum values ​​in the above-mentioned marked S channel histogram constitute an S channel interval, which is recorded as the S channel interval in the marked image.

[0040] For example, if the S channel values ​​in the labeled S channel histogram are 0.1, 0.2, 0.3, 0.5, 0.6, 0.7, 0.8, 0.9 and 1, and the number of pixels corresponding to these S channel values ​​are 10, 20, 100, 50, 60, 70, 40, 50 and 30, respectively, then three S channel intervals can be obtained, which are [0.1, 0.5), [0.5, 0.8) and [0.8, 1], and [0.1, 0.5) and [0.5, 0.8) are adjacent, and [0.5, 0.8) and [0.8, 1] are adjacent.

[0041] It should be noted that, since the number of pixels belonging to different S channel values ​​in the same image is often different, the amount of information contained in each S channel value in the same image is often different. In the S channel of the tongue surface image, the S channel value with a larger number of pixels occupies a larger area ratio in the tongue surface image, and contains more information. Since the saturation of the pixels on the surface of the tongue often presents a transitional change, the S channel histogram of the tongue surface image often shows a clustering phenomenon, which is mainly manifested in the formation of multiple peaks in the histogram. Each peak and the area gathered around it represent a major area in the tongue surface image. In order to completely segment several major areas in the main image of the tongue surface, it is often necessary to merge continuous S channel values. Therefore, the S channel value between the S channel values ​​corresponding to each two adjacent minimum values ​​can often be a peak in the S channel histogram and the interval where the S channel values ​​around it are gathered, which can represent a major area in the tongue surface image.

[0042] Step 203 , extracting connected domains from the region in the lupus erythematosus tongue surface image corresponding to each S channel interval, and screening out multi-connected regions from the extracted connected domains.

[0043] As an example, any S channel interval in the marked image is determined as a temporary S channel interval, and all the pixel points of the corresponding S channel in the marked image belonging to the temporary S channel interval constitute the target area corresponding to the temporary S channel interval, and the connected domain is extracted for the target area corresponding to the temporary S channel interval. At this time, all the connected domains extracted are recorded as all the connected domains corresponding to the temporary S channel interval. Generally speaking, connected domains are often divided into two categories, one is a single connected region, and the other is a multi-connected region, and multi-connected regions are often regions with holes. Holes in multi-connected regions are also called holes.

[0044] It should be noted that the color of a normal tongue surface often shows a uniform transition change, while the tongue surface of patients with systemic lupus erythematosus often has erythema, which often causes the color of some parts of the main area to change, thus appearing as holes of different colors in the image. Therefore, the holes in the multi-connected area may be erythema.

[0045] Step 204: Analyze and process the missing conditions of all holes in each multiply connected region to obtain a target missing index corresponding to each multiply connected region.

[0046] As an example, this step may include the following steps: In the first step, the area ratio of each hole in the multiply connected region to which it belongs is determined as the initial missing factor corresponding to each hole.

[0047] For example, the formula for determining the initial missing factor corresponding to the hole can be: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region The initial missing factor corresponding to the holes. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. It is The sequence number of the holes in a multiply connected region. It is In the image of lupus erythematosus tongue In a multiply connected region, The number of pixels in a hole can represent the The area of ​​the hole. It is In the lupus erythematosus tongue image, The number of pixels in a multiply connected region, which can represent the The area of ​​a multiply connected region.

[0048] It should be noted that, in actual situations, erythema often appears as holes in the normal connected domain, and holes can be understood as missing parts of the normal connected domain. The larger the In the image of lupus erythematosus tongue In a multiply connected region The larger the area of ​​the first hole, the The greater the degree of missing multiply connected regions.

[0049] In the second step, according to the difference between the initial missing factor corresponding to each hole in each multiply connected region and the initial missing factors corresponding to other holes, determining the erythema missing contribution factor corresponding to each hole in each multiply connected region may include the following sub-steps: In the first sub-step, any multi-connected region is determined as a marked multi-connected region, and any hole in the marked multi-connected region is determined as a marked hole, and each hole in the marked multi-connected region except the marked hole is determined as a reference hole.

[0050] In the second sub-step, the accumulated value of the absolute value of the difference between the initial missing factor corresponding to the marked hole and the initial missing factor corresponding to each reference hole is determined as the target missing difference corresponding to the marked hole.

[0051] The third sub-step is to determine the erythema loss contribution factor corresponding to the marked holes according to the target loss difference corresponding to the marked holes.

[0052] Among them, the target loss difference can be positively correlated with the erythema loss contribution factor.

[0053] For example, the formula for determining the erythema loss contribution factor corresponding to the hole can be: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region The contribution factor of erythema loss corresponding to each hole. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. and It is The serial numbers of different holes in a multiply connected region. is a normalization function. It is In the image of lupus erythematosus tongue The number of holes in a multiply connected region. It is the absolute value function. It is In the image of lupus erythematosus tongue In a multiply connected region The initial missing factor corresponding to the holes. It is In the image of lupus erythematosus tongue In a multiply connected region The initial missing factor corresponding to the holes. It is In the image of lupus erythematosus tongue In a multiply connected region The target missing difference corresponding to the holes.

[0054] It should be noted that in actual situations, the number of large holes in a connected domain is often smaller than the number of small holes. For large holes, there are often fewer holes with similar areas in the multi-connected regions to which they belong, so the difference between large holes and other holes in the multi-connected regions to which they belong is relatively large. For small holes, there are often more holes with similar areas in the multi-connected regions to which they belong, so the difference between small holes and other holes in the multi-connected regions to which they belong is relatively small. Therefore, when The larger the In the image of lupus erythematosus tongue In a multiply connected region The larger the area difference between the first hole and the other holes, the larger the area difference between the first hole and the other holes. The more holes there are, the more likely they are to be missing.

[0055] The third step is to determine the target missing index corresponding to each multi-connected region based on the erythema missing contribution factor and the initial missing factor corresponding to all the holes in each multi-connected region.

[0056] Among them, both the erythema loss contribution factor and the initial loss factor can be positively correlated with the target loss index.

[0057] For example, the formula for determining the target missing index corresponding to the multi-connected region can be: ;in, It is In the image of lupus erythematosus tongue The target missing indicators corresponding to the multi-connected regions. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. It is In the image of lupus erythematosus tongue The number of holes in a multiply connected region. It is The sequence number of the holes in a multiply connected region. It is In the image of lupus erythematosus tongue In a multiply connected region The contribution factor of erythema loss corresponding to each hole. It is In the image of lupus erythematosus tongue In a multiply connected region The initial missing factor corresponding to the holes.

[0058] It should be noted that when The larger the In the image of lupus erythematosus tongue In a multiply connected region The larger the area difference between the first hole and the other holes, the larger the area difference between the first hole and the other holes. The more holes there are, the more likely they are to be holes with more missing parts. The larger the In the image of lupus erythematosus tongue In a multiply connected region The larger the area of ​​the first hole, the The greater the degree of missing multi-connected regions, the greater the missing degree of multi-connected regions. The larger the In the image of lupus erythematosus tongue The greater the overall missingness of multiple connected regions.

[0059] Step 205 , determining a comprehensive supplementary index corresponding to each S channel interval according to target missing indexes corresponding to all multi-connected regions corresponding to each S channel interval and distribution of connected domains of the corresponding S channel intervals therein as adjacent S channel intervals.

[0060] As an example, this step may include the following steps: In the first step, based on all connected domains corresponding to the adjacent S channel intervals of the S channel interval corresponding to each hole, selecting supplementary candidate connected domains from each hole may include the following sub-steps: In the first sub-step, any hole is determined as a candidate hole, and the S channel interval corresponding to the multiply connected region to which the candidate hole belongs is determined as a candidate S channel interval.

[0061] In the second sub-step, each S channel interval adjacent to the candidate S channel interval is determined as an S channel interval to be supplemented, and each connected domain corresponding to each S channel interval to be supplemented is determined as a calibrated connected domain.

[0062] In the third sub-step, each calibrated connected domain in the candidate hole is determined as a supplementary candidate connected domain.

[0063] It should be noted that the holes caused by the normal tongue transition process can often be filled by the connected domains corresponding to the adjacent S channel intervals, while the holes caused by the erythema area are often difficult to be filled by the connected domains corresponding to the adjacent S channel intervals due to color changes. Therefore, selecting supplementary candidate connected domains from each hole can facilitate the subsequent analysis of the degree of tongue erythema.

[0064] In the second step, according to the target missing index corresponding to each multi-connected region, the initial supplement factor corresponding to each supplementary candidate connected domain in each hole in each multi-connected region is determined.

[0065] For example, the formula for determining the initial supplementary factor corresponding to the supplementary candidate connected domain in the hole in the multi-connected region can be: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The initial supplementary factor corresponding to the supplementary candidate connected domains. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. It is The sequence number of the holes in a multiply connected region. It is The sequence number of the candidate connected domain to be added in the hole. is a normalization function. It is In the image of lupus erythematosus tongue The target missing indicators corresponding to the multi-connected regions. is the target missing index corresponding to the target connected domain; the method for obtaining the target connected domain is: The supplementary candidate connected regions are filled into the In a multiply connected region holes, so that the The area reduced by the first hole is equal to The area corresponding to the supplementary candidate connected domains is The area of ​​the multiply connected region is equal to the The area corresponding to the supplementary candidate connected domain is realized. In the image of lupus erythematosus tongue The updated multi-connected region is recorded as the target connected domain.

[0066] It should be noted that and They can be expressed respectively The missing state of the multi-connected regions before and after the supplementary candidate connected regions are filled. The larger the The greater the difference in missing conditions between the multi-connected regions before and after filling the first supplementary candidate connected domain, the greater the difference in missing conditions between the first supplementary candidate connected domain. The supplementary candidate connected domains are The greater the degree of complementarity of the multiply connected regions.

[0067] The third step is to determine the supplementary effectiveness corresponding to each supplementary candidate connected domain in each hole in each multi-connected region according to the number of edge pixels of each hole in each multi-connected region and each supplementary candidate connected domain in it.

[0068] For example, the formula for determining the supplementary validity degree corresponding to the supplementary candidate connected domain in the hole in the multi-connected region can be: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The supplementary effectiveness corresponding to the supplementary candidate connected domains. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. It is The sequence number of the holes in a multiply connected region. It is The sequence number of the candidate connected domain to be added in the hole. is a normalization function. It is In the image of lupus erythematosus tongue In a multiply connected region The edge of the hole and the The number of pixels in the intersection between the edges of the supplementary candidate connected components. It is In the image of lupus erythematosus tongue In a multiply connected region The number of pixels on the edge of a hole. It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The number of pixels on the edge of the supplementary candidate connected domain.

[0069] It should be noted that when The larger the The edge of the hole and the The more overlapped parts between the edges of the supplementary candidate connected domains, the more likely it is that the The hole and The more likely the supplementary candidate connected domains are to fit closely, the more likely the The supplementary candidate connected domains are The more effectively the supplement of multiple connected regions is, the more effective it is.

[0070] The fourth step is to determine the adjacent supplementation index corresponding to each hole in each multi-connected region according to the initial supplementation factor and supplementation effectiveness corresponding to all supplementary candidate connected domains in each hole in each multi-connected region.

[0071] Among them, the initial supplementation factor and the effectiveness of supplementation can be positively correlated with adjacent supplementation indicators.

[0072] For example, the formula for determining the adjacent supplementary indexes corresponding to the holes in the multiply connected region can be: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region The adjacent supplementary indicators corresponding to the holes. It is the serial number of the lupus erythematosus tongue image. It is The serial numbers of the multiply connected regions in the lupus erythematosus tongue image. It is The sequence number of the holes in a multiply connected region. It is The sequence number of the candidate connected domain to be added in the hole. It is The number of candidate connected domains to be supplemented in a hole. It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The supplementary effectiveness corresponding to the supplementary candidate connected domains. It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The initial supplementary factor corresponding to the supplementary candidate connected domains.

[0073] It should be noted that when The larger the The greater the difference in missing conditions between the multi-connected regions before and after filling the first supplementary candidate connected domain, the greater the difference in missing conditions between the first supplementary candidate connected domain. The supplementary candidate connected domains are The greater the degree of complementarity of the multi-connected regions, the greater the The larger the The edge of the hole and the The more overlapped parts between the edges of the supplementary candidate connected domains, the more likely it is that the The hole and The more likely the supplementary candidate connected domains are to fit closely, the more likely the The supplementary candidate connected domains are The more effectively the multi-connected regions are supplemented, the more effective they are. The larger the The supplementary candidate connected domains in the holes are The better the filling effect of the first hole, the better the The supplementary candidate connected domains in the holes are The more connected regions are, the better the complement effect is.

[0074] In the fifth step, the average of the adjacent supplementary indices corresponding to all holes in all multi-connected regions corresponding to each S channel interval is determined as the comprehensive supplementary index corresponding to each S channel interval.

[0075] It should be noted that the larger the comprehensive supplementary index corresponding to the S channel interval is, the better the supplementary effect of the adjacent S channel interval on its corresponding multi-connected area is.

[0076] Step 206 , performing adaptive supplementary division analysis based on the comprehensive supplementary indicators corresponding to all S channel intervals in each lupus erythematosus tongue surface image, and obtaining lupus erythematosus characteristic obvious factors corresponding to each lupus erythematosus tongue surface image.

[0077] As an example, this step may include the following steps: In the first step, according to the comprehensive supplementary index corresponding to all S channel intervals in each lupus erythematosus tongue surface image, the Fisher optimal solution method is used to cluster the S channel intervals in each lupus erythematosus tongue surface image into two categories, and the continuous S channel intervals in the category with higher comprehensive supplementary index constitute continuous supplementary intervals, and the continuous S channel intervals in the other category with lower comprehensive supplementary index constitute continuous separation intervals.

[0078] There are at least two S channel intervals in the continuous supplementary interval and at least two S channel intervals in the continuous separation interval.

[0079] For example, if a category with a higher comprehensive supplementation index includes: [0.1, 0.3), [0.3, 0.5), [0.5, 0.6) and [0.9, 1], then a continuous supplementation interval can be constructed for this category, which can be [0.1, 0.6).

[0080] The second step, according to the area filling situation between the adjacent S channel intervals in each continuous supplementary interval, screening out the characteristic separation interval from the continuous separation intervals adjacent to each continuous supplementary interval may include the following sub-steps: In the first sub-step, any continuous supplement interval is determined as a marked continuous supplement interval, and any two adjacent S channel intervals in the marked continuous supplement interval are sequentially determined as a marked left S channel interval and a marked right S channel interval.

[0081] In the second sub-step, the total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked right S channel interval and belonging to the marked left S channel interval is determined as the left filling area between the marked left S channel interval and the marked right S channel interval.

[0082] In the third sub-step, the total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked left S channel interval and belonging to the marked right S channel interval is determined as the right filling area between the marked left S channel interval and the marked right S channel interval.

[0083] The fourth sub-step is to determine the accumulated value of the left padding areas between all adjacent S channel intervals within the above-mentioned marked continuous supplement interval as the overall left padding factor corresponding to the above-mentioned marked continuous supplement interval.

[0084] The fifth sub-step is to determine the accumulated value of the right filling area between all adjacent S channel intervals in the above-mentioned marked continuous supplement interval as the overall right filling factor corresponding to the above-mentioned marked continuous supplement interval.

[0085] The sixth sub-step, if the overall left padding factor corresponding to the above-mentioned marked continuous supplement interval is greater than or equal to its corresponding overall right padding factor, then the continuous separation interval adjacent to the above-mentioned marked continuous supplement interval and on the left of the above-mentioned marked continuous supplement interval will be determined as the characteristic separation interval corresponding to the above-mentioned marked continuous supplement interval.

[0086] The seventh sub-step, if the overall left filling factor corresponding to the above-mentioned marked continuous supplement interval is smaller than its corresponding overall right filling factor, then the continuous separation interval adjacent to the above-mentioned marked continuous supplement interval and on the right side of the above-mentioned marked continuous supplement interval will be determined as the characteristic separation interval corresponding to the above-mentioned marked continuous supplement interval.

[0087] The third step is to determine the formula corresponding to the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue surface image based on all the continuous supplementary intervals and their corresponding characteristic separation intervals in each lupus erythematosus tongue surface image: ;in, It is The lupus erythematosus characteristic obvious factors corresponding to the lupus erythematosus tongue surface images. It is the serial number of the lupus erythematosus tongue image. is a normalization function. It is The number of consecutive supplementation intervals in a lupus erythematosus tongue image. It is The serial number of the consecutive supplementary intervals in the lupus erythematosus tongue image. It is In the image of lupus erythematosus tongue The number of S channel values ​​included in a continuous supplement interval. It is In the image of lupus erythematosus tongue The mean value of the target missing index corresponding to all multi-connected regions corresponding to all S channel intervals in the feature separation interval corresponding to the continuous supplementary interval.

[0088] It should be noted that in actual situations, the longer the length of the continuous supplementary interval is, the higher the degree of supplementation of the missing hole is. If the degree of missing of the connected domain corresponding to the adjacent continuous separation interval is still high, it often indicates that there is a hole supplement fault phenomenon. There may be lupus defects in the hole, and its lupus characteristics are relatively more obvious. The larger the The longer the continuous supplement interval is, the longer the The larger the The greater the missing degree of the connected domain corresponding to the adjacent continuous separation intervals of the continuous supplementary intervals, the greater the missing degree of the connected domain corresponding to the adjacent continuous separation intervals. The larger the The more obvious the lupus features in a lupus erythematosus tongue image are, the more it should participate in the training of the subsequent tongue image classification network.

[0089] Step 207 , selecting lupus erythematosus tongue surface images whose corresponding lupus erythematosus characteristic obvious factors are greater than a preset characteristic threshold from the lupus erythematosus tongue surface image set to form a reference image set.

[0090] The preset feature threshold may be a preset threshold for determining whether the lupus erythematosus feature is obvious. For example, the preset feature threshold may be 0.3.

[0091] As an example, a lupus erythematosus tongue image whose corresponding lupus erythematosus characteristic significance factor is greater than a preset characteristic threshold value may be used as a reference image, and all reference images constitute a reference image set.

[0092] Optionally, the training process of the tongue image classification network may include the following steps: The first step is to build ResNet as the tongue image classification network before training.

[0093] In the second step, the lupus erythematosus syndrome conditions are manually labeled for each image in the target training set.

[0094] The lupus erythematosus syndrome condition may be the manifestation of lupus erythematosus in the patient. For example, the lupus erythematosus syndrome condition may be, but is not limited to, normal, rheumatic heat arthralgia syndrome, yin deficiency and internal heat syndrome, phlegm-heat stagnation in the lung syndrome, and liver depression and blood stasis syndrome.

[0095] In the third step, the target training set is used as the training set, and the lupus erythematosus symptoms annotated in each image in the target training set are used as training labels to train the constructed tongue image classification network to obtain a trained tongue image classification network.

[0096] Among them, the loss function of the tongue image classification network uses the cross entropy loss function.

[0097] refer to Figure 3 Based on the same inventive concept as the above method embodiment, the present invention provides a tongue image classification system for systemic lupus erythematosus patients with different TCM syndromes, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the above computer program is executed by the processor, the steps of the tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes may specifically include: The tongue surface image acquisition module 301 is used to acquire a target tongue surface image corresponding to the lupus erythematosus patient to be detected; The tongue image classification module 302 is used to classify the target tongue surface image according to the target tongue surface image by using a pre-trained tongue image classification network.

[0098] Figure 4 is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. Figure 4 As shown, the computer device 400 includes: a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402, wherein when the processor 402 executes the computer program 403, the computer device can execute any of the tongue image classification methods for systemic lupus erythematosus patients with different TCM syndromes introduced above.

[0099] Based on the same inventive concept as the above method embodiment, the present invention provides a server, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes any of the above-mentioned tongue image classification methods for systemic lupus erythematosus patients with different TCM syndromes.

[0100] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer program product, which includes: computer program code, when the computer program code runs on a computer, enables the computer to execute any of the above-mentioned methods for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes.

[0101] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer-readable storage medium, which stores a computer program code. When the computer program code runs on a computer, the computer executes any one of the above-mentioned methods for classifying tongue images of patients with systemic lupus erythematosus for different TCM syndromes.

[0102] In summary, when training the tongue image classification network, compared with directly forming the lupus erythematosus tongue image set and the normal tongue image set into a training set, the present invention comprehensively considers multiple indicators related to the obvious characteristics of lupus erythematosus, such as target missing indicators and comprehensive supplementary indicators, etc. Thus, the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue image is quantified, and then the lupus erythematosus tongue images whose corresponding lupus erythematosus characteristic obvious factor is greater than the preset characteristic threshold are screened out, so that the lupus erythematosus tongue images with smaller lupus erythematosus characteristic obvious factors do not constitute the training set, which reduces the possibility of the tongue image classification network learning inaccurate lupus erythematosus characteristic information to a certain extent, thereby improving the accuracy of tongue image classification of lupus erythematosus patients.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A tongue image classification method for patients with systemic lupus erythematosus with different TCM syndromes, characterized in that: The following steps are involved: Acquire a target tongue surface image corresponding to a lupus erythematosus patient to be detected; According to the target tongue surface image, the target tongue surface image is classified by a pre-trained tongue image classification network, wherein the training set of the tongue image classification network is the target training set, and the target training set includes a normal tongue surface image set and a reference image set; The method for obtaining the reference image set includes: Obtain the lupus erythematosus tongue surface image set and the S channel value corresponding to each pixel point in each lupus erythematosus tongue surface image in the lupus erythematosus tongue surface image; The S channel values ​​corresponding to all the pixels in each lupus erythematosus tongue surface image are divided into intervals to obtain the S channel interval corresponding to each lupus erythematosus tongue surface image; For each S channel interval, the connected domain is extracted from the region in the lupus erythematosus tongue surface image to which it belongs, and multi-connected regions are screened out from the extracted connected domains; Analyze and process the missing conditions of all holes in each multi-connected region to obtain the target missing index corresponding to each multi-connected region; According to the target missing index corresponding to all multi-connected regions corresponding to each S channel interval and the distribution of the connected domains of the corresponding S channel intervals as adjacent S channel intervals, the comprehensive supplementary index corresponding to each S channel interval is determined; Based on the comprehensive supplementary indexes corresponding to all S channel intervals in each lupus erythematosus tongue surface image, an adaptive supplementary division analysis is performed to obtain the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue surface image; The lupus erythematosus tongue surface images whose corresponding lupus erythematosus characteristic obvious factors are greater than the preset characteristic threshold are screened out from the lupus erythematosus tongue surface image set to form a reference image set.

2. According to claim 1, a tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes is characterized by: The S channel values ​​corresponding to all the pixels in each lupus erythematosus tongue surface image are divided into intervals to obtain the S channel interval corresponding to each lupus erythematosus tongue surface image, including: Determine any lupus erythematosus tongue image as a marked image, and make an S channel histogram corresponding to the marked image, recorded as a marked S channel histogram; The S channel values ​​between the S channel values ​​corresponding to every two adjacent minimum values ​​in the marked S channel histogram constitute an S channel interval.

3. According to claim 1, a tongue image classification method for systemic lupus erythematosus patients with different TCM syndromes is characterized by: The missing condition analysis of all holes in each multi-connected region is performed to obtain the target missing index corresponding to each multi-connected region, including: The area ratio of each hole in the multiply connected region to which it belongs is determined as the initial missing factor corresponding to each hole; Determine the erythema missing contribution factor corresponding to each hole in each multi-connected region according to the difference between the initial missing factor corresponding to each hole in each multi-connected region and the initial missing factors corresponding to other holes; According to the erythema missing contribution factor and the initial missing factor corresponding to all the holes in each multi-connected region, the target missing index corresponding to each multi-connected region is determined, wherein the erythema missing contribution factor and the initial missing factor are positively correlated with the target missing index.

4. A method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 3, characterized in that: Determining the erythema missing contribution factor corresponding to each hole in each multiply connected region according to the difference between the initial missing factor corresponding to each hole in each multiply connected region and the initial missing factors corresponding to other holes includes: Determine any multiply connected region as a marked multiply connected region, determine any hole in the marked multiply connected region as a marked hole, and determine each hole in the marked multiply connected region except the marked hole as a reference hole; The accumulated value of the absolute value of the difference between the initial missing factor corresponding to the marked hole and the initial missing factor corresponding to each reference hole is determined as the target missing difference corresponding to the marked hole; According to the target loss difference corresponding to the marked hole, the erythema loss contribution factor corresponding to the marked hole is determined, wherein the target loss difference is positively correlated with the erythema loss contribution factor.

5. The method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 1, characterized in that: The comprehensive supplementary index corresponding to each S channel interval is determined according to the target missing index corresponding to all multi-connected regions corresponding to each S channel interval and the distribution of the connected domains of the corresponding S channel intervals as adjacent S channel intervals, including: According to all connected domains corresponding to the adjacent S channel intervals of the S channel interval corresponding to each hole, a supplementary candidate connected domain is screened out from each hole; According to the target missing index corresponding to each multi-connected region, an initial supplement factor corresponding to each supplementary candidate connected domain in each hole in each multi-connected region is determined; Determine the supplementation effectiveness corresponding to each supplementary candidate connected domain in each hole in each multi-connected region according to the number of edge pixels of each hole and each supplementary candidate connected domain in each multi-connected region; According to the initial supplementation factors and supplementation effectiveness corresponding to all supplementary candidate connected domains in each hole in each multi-connected region, the adjacent supplementation index corresponding to each hole in each multi-connected region is determined, wherein the initial supplementation factor and the supplementation effectiveness are both positively correlated with the adjacent supplementation index; The average of the adjacent supplementary indices corresponding to all holes in all multi-connected regions corresponding to each S channel interval is determined as the comprehensive supplementary index corresponding to each S channel interval.

6. A method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 5, characterized in that: The method of selecting a supplementary candidate connected domain from each hole according to all connected domains corresponding to the adjacent S channel intervals of the S channel interval corresponding to each hole includes: Determine any hole as a candidate hole, and determine the S channel interval corresponding to the multiply connected region to which the candidate hole belongs as a candidate S channel interval; Determine each S channel interval adjacent to the candidate S channel interval as an S channel interval to be supplemented, and determine each connected domain corresponding to each S channel interval to be supplemented as a calibrated connected domain; Each calibrated connected domain within the candidate hole is determined as a supplementary candidate connected domain.

7. The method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 5, characterized in that: The formula for the initial supplementary factor corresponding to the supplementary candidate connected domains in the holes in the multi-connected regions is: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The initial supplementary factor corresponding to the supplementary candidate connected domains; It is the serial number of the lupus erythematosus tongue surface image; It is The serial number of the multi-connected regions in the lupus erythematosus tongue image; It is The sequence number of the holes in the multiply connected regions; It is The serial number of the candidate connected domain to be added in each hole; is the normalization function; It is In the image of lupus erythematosus tongue The target missing index corresponding to the multi-connected regions; is the target missing index corresponding to the target connected domain; the method for obtaining the target connected domain is: The supplementary candidate connected regions are filled into the In a multiply connected region holes, so that the The area reduced by the first hole is equal to The area corresponding to the supplementary candidate connected domains is The area of ​​the multiply connected region is equal to the The area corresponding to the supplementary candidate connected domain is realized. In the image of lupus erythematosus tongue The updated multi-connected region is recorded as the target connected domain.

8. The method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 5, characterized in that: The formula for the supplementary effectiveness of the supplementary candidate connected domains in the holes in the multi-connected regions is: ;in, It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The supplementary effectiveness corresponding to the supplementary candidate connected domains; It is the serial number of the lupus erythematosus tongue surface image; It is The serial number of the multi-connected regions in the lupus erythematosus tongue image; It is The sequence number of the holes in the multiply connected regions; It is The serial number of the candidate connected domain to be added in each hole; is the normalization function; It is In the image of lupus erythematosus tongue In a multiply connected region The edge of the hole and the The number of pixels in the intersection between the edges of the supplementary candidate connected domains; It is In the image of lupus erythematosus tongue In a multiply connected region The number of pixels on the edge of a hole; It is In the image of lupus erythematosus tongue In a multiply connected region In the hole, The number of pixels on the edge of the supplementary candidate connected domain.

9. The method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 1, characterized in that: The adaptive supplementary division analysis is performed based on the comprehensive supplementary indexes corresponding to all S channel intervals in each lupus erythematosus tongue surface image to obtain lupus erythematosus characteristic obvious factors corresponding to each lupus erythematosus tongue surface image, including: According to the comprehensive supplementary index corresponding to all S channel intervals in each lupus erythematosus tongue surface image, the S channel intervals in each lupus erythematosus tongue surface image are clustered into two categories, and the continuous S channel intervals in the category with a higher comprehensive supplementary index constitute a continuous supplementary interval, and the continuous S channel intervals in the other category with a lower comprehensive supplementary index constitute a continuous separation interval; According to the area filling situation between adjacent S channel intervals in each continuous supplementary interval, the characteristic separation interval is selected from the continuous separation intervals adjacent to each continuous supplementary interval; According to all the continuous supplement intervals and their corresponding feature separation intervals in each lupus erythematosus tongue surface image, the formula corresponding to the lupus erythematosus characteristic obvious factor corresponding to each lupus erythematosus tongue surface image is determined as follows: ;in, It is Lupus erythematosus characteristic obvious factors corresponding to lupus erythematosus tongue images; It is the serial number of the lupus erythematosus tongue surface image; is the normalization function; It is The number of consecutive supplementation intervals in the lupus erythematosus tongue surface images; It is The serial number of the continuous supplement interval in the lupus erythematosus tongue image; It is In the image of lupus erythematosus tongue The number of S channel values ​​included in a continuous supplementary interval; It is In the image of lupus erythematosus tongue The mean value of the target missing index corresponding to all multi-connected regions corresponding to all S channel intervals in the feature separation interval corresponding to the continuous supplementary interval.

10. A method for classifying tongue images of patients with systemic lupus erythematosus with different TCM syndromes according to claim 9, characterized in that: The method of selecting a characteristic separation interval from the continuous separation intervals adjacent to each continuous supplementary interval according to the area filling condition between the adjacent S channel intervals in each continuous supplementary interval includes: Determine any one of the continuous supplement intervals as a marked continuous supplement interval, and sequentially determine any two adjacent S channel intervals in the marked continuous supplement interval as a marked left S channel interval and a marked right S channel interval; The total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked right S channel interval and belonging to the marked left S channel interval is determined as the left filling area between the marked left S channel interval and the marked right S channel interval; The total area of ​​all connected domains of the corresponding S channel intervals contained in all holes corresponding to the marked left S channel interval and belonging to the marked right S channel interval is determined as the right filling area between the marked left S channel interval and the marked right S channel interval; Determine the accumulated value of the left padding areas between all adjacent S channel intervals in the marked continuous supplement interval as the overall left padding factor corresponding to the marked continuous supplement interval; Determine the accumulated value of the right filling area between all adjacent S channel intervals in the marked continuous supplement interval as the overall right filling factor corresponding to the marked continuous supplement interval; If the overall left padding factor corresponding to the continuous mark supplement interval is greater than or equal to the overall right padding factor corresponding to it, then the continuous separation interval adjacent to the continuous mark supplement interval and on the left side of the continuous mark supplement interval is determined as the characteristic separation interval corresponding to the continuous mark supplement interval; If the overall left padding factor corresponding to the marked continuous supplement interval is smaller than its corresponding overall right padding factor, the continuous separation interval adjacent to the marked continuous supplement interval and on the right side of the marked continuous supplement interval will be determined as the characteristic separation interval corresponding to the marked continuous supplement interval.

Images