A method and device for identifying suspended matter in a urine test according to Tibetan medicine

Abstract

The application discloses a Tibetan medicine urine diagnosis suspended matter identification method and device, which is applied to the field of image processing and aims at the problem of poor interpretability of existing urine suspended matter identification results. First, a cup body detection segmentation and liquid surface extraction correction method is adopted, so that the model only focuses on the middle area of the image, sliding scanning operation on invalid areas is reduced, and the model efficiency is improved. Secondly, interval sampling is used in the liquid surface area to filter rgb color values, and a sub-area division method is used to filter texture characteristic values, so that pre-judgment of the presence or absence of suspended matter is efficiently performed, and the process of most images entering a convolutional neural network for operation is reduced. Finally, through interval sampling of RGB and definition of suspended matter disturbance evaluation parameters, the judgment process of the suspended matter is expressed, and the interpretability of the model is improved.

Description

Technical Field

[0001] This invention belongs to the field of image processing, and specifically relates to a Tibetan medicine urine diagnosis suspended matter identification technology and device. Background Technology

[0002] Urine diagnosis is one of the most distinctive traditional diagnostic methods in Tibetan medicine. Among various traditional medicines worldwide, Tibetan urine diagnosis is one of the most detailed, including observing urine color, vapor, odor, floating matter, suspended matter, and consistency, as well as observing changes in urine bubbles by agitating the urine. Urine diagnosis is a unique diagnostic system developed by Tibetans based on their understanding of the digestion and transformation of food and drink in the body and their long-term practice in the high-altitude environment. As early as the 8th century, the Tibetan medical classic *The Four Medical Tantras: Urine Diagnosis* recorded that the precipitates (goya) in urine originate from liver blood and bile and can be used to determine cold or heat-related illnesses. Clinically, the thickness or thinness of the precipitates (goya) in urine can indicate metabolic changes in liver blood and bile, and changes in the precipitates (goya) can indicate the excess or disorder of Lung, Tripa, and Pekan.

[0003] Urine diagnosis primarily employs the "Three Times and Nine Examinations" method for identifying diseases. The three times refer to observing three stages in the cooling process of urine: the hot urine stage, the warm urine stage, and the cool urine stage. The nine examinations are the specific observations for each stage: the first stage observes urine color, steam, odor, and foam; the second stage observes suspended and floating matter; and the third stage observes the timing and manner of changes in urine, as well as the resulting color. These points help determine the syndrome and disease type to assist the doctor in prescribing further treatment. The state of suspended matter is a crucial characteristic in the urine diagnosis process. Suspended matter is typically categorized as hair-like, cheese-like, horsehair-like, cloud-like, purulent, or fine sand-like.

[0004] Current research has applied machine learning algorithms to the identification of impurities in liquids. Yao et al. proposed a feature-based liquid impurity identification method (Yao Kang, Yang Ping, Ma Shiqing. A feature-based liquid impurity detection method [J]. Semiconductor Optoelectronics, 2019, 40(05): 719-725), which mainly includes bilateral filtering preprocessing of images, improved multi-scale wavelet transform detection of target edges, and selection of target features for identifying impurities in liquids.

[0005] Current technologies for liquid impurity identification rely on algorithms to obtain the location information of bubbles and impurities at multiple points in an image. However, for the task of identifying suspended matter in Tibetan medicine urine diagnosis, the above methods have the following drawbacks:

[0006] 1) Suspended matter in urine is usually concentrated in the middle of the urine image, which means that the region of interest scanning process in the background technique is redundant and inefficient;

[0007] 2) The proportion of cases with suspended matter in actual urine images is small, far lower than the probability of bubbles and impurities in liquid. This means that if each image goes through the complete calculation process of background technology, the computing power required to obtain one effective piece of information on the category of suspended matter on average is far higher than the computing power required for images that actually contain suspended matter.

[0008] 3) In the application scenario of impurity identification, only the accuracy of impurity identification is of concern; while in the scenario of urine diagnosis, not only the accuracy of the results is of concern, but also the interpretability of the model, so as to enhance the persuasiveness of the results and the pathological interpretation. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention proposes a method and device for identifying suspended matter in Tibetan medicine urine examinations. This method achieves objectification of suspended matter identification in urine examinations, reduces the average recognition time per image, and also reflects the process of suspended matter judgment to a certain extent, improving the interpretability of the model and contributing to the inheritance and interpretation of the suspended matter identification process.

[0010] One of the technical solutions adopted in this invention is: a Tibetan medicine method for identifying suspended matter in urine, comprising:

[0011] S1. Urine image acquisition; Urine is collected in uniform square transparent plastic cups.

[0012] S2, Cup body detection and segmentation;

[0013] S3. Perform liquid level extraction and correction on the image processed in step S2;

[0014] S4. Perform RGB color value filtering on the image processed in step S3.

[0015] S5. Extract texture features from the image processed in step S4;

[0016] S6. Filter the texture feature values ​​of the image processed in step S5.

[0017] S7. Extract contour features from the image processed in step S6.

[0018] S8. Based on the image after liquid level correction in step S3, the texture features extracted in step S5, and the contour features extracted in step S7, perform suspended object identification.

[0019] The second technical solution adopted in this invention is: a Tibetan medicine urine diagnosis suspended matter identification device, comprising: a urine image acquisition module, a cup detection and segmentation module, a liquid surface extraction and correction module, a first filtering module, a first feature extraction module, a second filtering module, a second feature extraction module, and a suspended matter identification module; the urine image acquisition module is used to acquire urine images contained in uniform square transparent plastic cups; the cup detection and segmentation module is used to remove invalid information outside the cup from the output results of the urine image acquisition module; the liquid surface extraction and correction module stretches the target area according to the output of the cup detection and segmentation module to obtain a square cup mouth image liquid surface area; the first filtering module filters out images of urine without suspended matter from the output results of the liquid surface extraction and correction module; the first feature extraction module extracts texture features from the output results of the first filtering module; the second filtering module filters the texture features extracted by the first feature extraction module; the second feature extraction module is used to perform edge detection on the output results of the second filtering module to obtain a binarized feature map; the suspended matter identification module identifies suspended matter according to the output results of the first feature module, the liquid surface correction module, and the second feature extraction module.

[0020] The beneficial effects of this invention are as follows: First, this invention employs cup detection and segmentation, as well as liquid surface extraction and correction methods, allowing the model to focus only on the central region of the image, reducing sliding scan operations on invalid regions and improving model efficiency. Second, in the liquid surface region, interval sampling is used for RGB color value filtering, and sub-region division is used for texture feature value filtering, efficiently performing pre-judgment of the presence or absence of suspended matter, reducing the process of most images entering the convolutional neural network for computation. Finally, by using RGB interval sampling and defining suspended matter perturbation evaluation parameters, the process of judging suspended matter is explicitly expressed, improving the interpretability of the model. The color value variance of RGB interval sampling explicitly represents the probability of suspended matter in the central region of the urine image, and also conforms to the characteristic judgment points of suspended matter distribution in clinical practice. The suspended matter perturbation evaluation parameters also explicitly represent the probability of suspended matter in the urine image, and also conform to the texture description of suspended matter such as horsehair and cloud-like structures in clinical practice. At the same time, texture features are extracted and retained for subsequent classification and recognition. Compared with the uninterpretable end-to-end model that directly uses neural networks, the method of this invention is more in line with clinical pathways and has interpretability. Attached Figure Description

[0021] Figure 1 This forms the overall framework of the network model.

[0022] Figure 2 This refers to the process of liquid level extraction and correction.

[0023] Among them, (a) is the urine image before cup segmentation; (b) is the target area obtained by assigning the corresponding original image part to the newly created blank image through the marker image; (c) is the target area to be stretched to obtain the liquid surface area of ​​the square cup mouth image.

[0024] Figure 3 This is a schematic diagram of the sub-region division;

[0025] Figure 4 This is a schematic diagram of the network structure. Detailed Implementation

[0026] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0027] This invention includes urine image acquisition, cup segmentation, liquid surface extraction and correction, RGB color value analysis, and texture feature analysis to filter out samples. Finally, a convolutional neural network model is used to automatically identify the suspended matter category in the unfiltered samples. The overall framework of this invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0028] 1. Urine image acquisition

[0029] Use standardized square transparent plastic cups to prevent the magnifying effect of round cups from negatively impacting the results. Use standardized handheld shooting devices to prevent significant differences in image quality under the same environment. Shooting distance requirement: The cup should be within 20-30 cm of the camera. Place the urine sample on a white A4 sheet of paper and shoot from above, then from the side, ensuring the liquid surface is centered in the image to complete image acquisition.

[0030] In addition, images captured from incorrect angles or that are blurry, images with strong reflections or contaminated cups (such as urine on the outside of the cup or other markings or writing on it), and images with urine volume less than 1 / 5 or more than 4 / 5 of the container were excluded.

[0031] 2. Cup body detection and segmentation

[0032] In the captured images, the urine is often in the center and only occupies a portion of the image; the area around the rim of the urine cup is irrelevant information. Figure 2 As shown in (a), to obtain the liquid surface area, the cup in the image is first detected and segmented. A template-based contour extraction method is used to detect and segment the cup. The template image refers to a sample image with absolutely standard shooting angles and specifications, as detailed below:

[0033] 1) Apply Gaussian filtering to both the template image and the sample image to eliminate image noise. Gaussian filtering uses a 3×3 Gaussian filter to filter the target image. The values ​​at each coordinate position of the filter are derived from the following formula:

[0034]

[0035] Where (n,m) are the point coordinates, which can be considered integers in image processing, and σ is the standard deviation.

[0036] 2) Perform Unsharp Mask (USM) sharpening on the template and sample images. By enhancing the high-frequency components of the images, the visual effect is improved, the image clarity is enhanced, and the outlines and edges of the images are strengthened, as shown below:

[0037] y(n,m)=x(n,m)+λz(n,m)

[0038] Where x(n,m) is the input image, y(n,m) is the output image, λ is a scaling factor used to control the enhancement effect, and z(n,m) is the correction signal, which is generally obtained through high-pass filtering.

[0039] z(n,m)=4x(n,m)-x(n-1,m)-x(n+1,m)-x(n,m-1)-x(n,m+1)

[0040] 3) Convert the template image and sample image to grayscale images, using the following formula to convert the corresponding pixels:

[0041] Gray=0.1140*Blue+0.5870*Green+0.2989*Red

[0042] Blue, Green, and Red correspond to the pixel values ​​of different channels in the original RGB image.

[0043] 4) Use the Laplacian second-order differential operator to sharpen the template and sample images after they have been converted to grayscale, emphasizing the edges and details of the image and improving the contrast. The operator is expressed by the following formula:

[0044]

[0045] 5) Binarize the sharpened template image and sample image from the previous step, i.e., if the pixel value is greater than the set threshold, it is 1, and if it is less than 0, then the image only has outline lines, which are used for subsequent outline matching.

[0046] 6) Perform contour matching using the binarized images of the template image (A) and the sample image (B), and find the contour with the smallest matching value. The contour matching value is defined as follows:

[0047]

[0048] Where H represents the Hu invariant moments generated from the image, which includes 7 invariant moments constructed using second- and third-order normalized central moments, i is used to indicate the index of the Hu invariant moments, i = 0, 1, 2, 3, 4, 5, 6, and || represents the absolute value operation.

[0049] Select the contour with the smallest matching value, such as Figure 2 As shown in (b), the contour with the smallest matching value is filled with a marker, and the corresponding original sample image is assigned to the newly created blank image through the marker image to segment the target region.

[0050] 3. Liquid level extraction correction

[0051] To eliminate the distortion caused by the shooting, the target area is stretched to obtain the liquid surface area X of the square cup rim image, as shown. Figure 2 As shown in (c).

[0052] 4. RGB color value filtering

[0053] Based on the location and color characteristics of suspended matter in clinical urinalysis and the clinical urinalysis diagnostic procedure, suspended matter usually appears in the central region of urine and is mostly white and opaque or black impurities. That is, the color of the suspended matter differs significantly from the usual color range of urine and is discretely distributed in the central region. Therefore, the following steps are used for identification:

[0054] 1) Within the X domain, average samples are taken from the Red, Green, and Blue channels of the image at intervals specified by the parameters sample_interval_x and sample_interval_y to obtain the average RGB color value. sample_interval_x and sample_interval_y are hyperparameters representing the sampling intervals in the horizontal and vertical axes, respectively; they are typically set to 10.

[0055] 2) In order to filter out the influence of light, the RGB color space is mapped to the HSV color space, the lightness component in the HSV color space is adjusted to the highest level, and then converted back to the RGB color space.

[0056] 3) Calculate the variance σ based on the adjusted RGB color values ​​of each sampling point, and set the variance threshold θ. If there are no suspended particles in the urine that disturb the image color, then σ < θ, meaning the urine is considered to be free of suspended particles, and the image is filtered out. Conversely, if σ > θ, then the urine is considered to contain suspended particles, and step 5 continues. θ is generally set to 30.

[0057] 5. Texture Feature Extraction

[0058] Combining clinical manifestations and diagnostic procedures in urinalysis, suspended matter can manifest as hair-like, cheese-like, horsehair-like, cloud-like, purulent, or fine sand-like forms, exhibiting rich textural features. Accurate extraction of these texture features not only aids in differentiating between different types of suspended matter but also forms the basis for determining the presence or absence of suspended matter in the central region. The following steps are used for texture feature extraction:

[0059] 1) First, divide the liquid surface region X into 25 sub-regions of 5 rows and 5 columns, such as... Figure 3 As shown, each sub-region is represented by a horizontal and a vertical axis.

[0060] 2) The LBP algorithm is used to transform the sub-regions (1,1), (1,2), (1,3), (2,1), (2,2), (2,3), (3,1), (3,2), and (3,3) located in the central region, resulting in the following transformed feature map:

[0061] LBP (1,1) LBP (1,2) LBP (1,3) LBP (2,1) LBP (2,2) LBP (2,3) LBP (3,1) LBP (3,2) LBP (3,3)

[0062] Considering that suspended matter generally appears in the central area, this step uses the LBP algorithm to transform the nine sub-regions located in the central area, thus avoiding interference from factors such as the cup rim and foam in the edge areas.

[0063] 3) LBP feature map (i,j) Divided into a 16×16 grid;

[0064] 4) For a pixel in each grid, compare its grayscale value with the grayscale values ​​of its 8 neighboring pixels. If the grayscale value of the surrounding pixels is greater than that of the center pixel, the pixel is marked as 1; otherwise, it is marked as 0. In this way, the comparison of 8 points in a 3*3 neighborhood can generate an 8-bit binary number.

[0065] 5) Calculate the histogram for each grid cell, which is the frequency of each number; then normalize the histogram.

[0066] 6) Connect the statistical histograms of each grid to form a single feature vector, resulting in the following feature vector group composed of the feature vectors of the nine central sub-regions:

[0067] f (1,1) ,f (1,2) ,f (1,3) ,f (2,1),f (2,2) ,f (2,3) ,f (3,1) ,f (3,2) ,f (3,3)

[0068] 6. Texture Feature Value Filtering

[0069] Combining clinical manifestations and diagnostic procedures in urinalysis, suspended matter can be categorized into various textured types. Since suspended matter is generally located in the central region of the image, a floating matter perturbation evaluation parameter τ is defined to determine whether there are textured and edge-rich sub-blocks in the central region of the image. This allows for an explicit determination of the presence or absence of suspended matter. The filtering operation is then performed as follows:

[0070] 1) Define the suspended object disturbance evaluation parameter τ based on texture features as follows:

[0071]

[0072] Where d(·,·) represents the Euclidean geometric distance:

[0073] 2) Set a threshold ω. If there are no suspended particles in the urine that disturb the image color, then τ < ω, which means that the image is filtered out as having no suspended particles in the urine. Otherwise, if τ > ω, then the image is considered to have suspended particles in the urine, and continue to step 7. ω is generally set to 0.01.

[0074] 7. Contour Feature Extraction

[0075] The Canny operator is used to perform edge detection on the urine image to obtain a binarized feature map X. canny .

[0076] 8. Construction of Suspended Object Recognition Model

[0077] The RGB image of the liquid surface region X obtained in step 3 of the urine image to be identified contains different color channel images X. r X g X b and the transformed sub-region feature map LBP in step 5 (i,j) (i = 1, 2, 3, j = 1, 2, 3) and the binarized feature map X in step 7 canny A total of 13 feature maps were used as input to the convolutional neural network for suspended object recognition;

[0078] During convolutional neural network training, different color channels of the RGB image of the liquid surface region X obtained in "Step 3" are used, with known labels (provided by experienced clinicians for the correct Tibetan medicine category of suspended matter, including labels such as hairy, cheese-like, horsehair-like, cloud-like, pus-like, and fine sand-like). r X g X band the transformed sub-region feature map LBP in step 5 (i,j) (i = 1, 2, 3, j = 1, 2, 3) and the binarized feature map X in step 7 canny The network model was trained using image data consisting of 13 feature maps. The model includes a first convolutional unit, a second convolutional unit, a third convolutional unit, and a fully connected layer. The first and second convolutional units each contain a convolutional layer, a local response normalization layer, and a max-pooling layer. The third convolutional unit contains three convolutional layers and a max-pooling layer. Parameter settings are as follows: Figure 4 As shown:

[0079] The first convolutional unit has a convolutional layer size of 11*11, a stride of 4, 96 channels, and uses ReLU activation function; its max pooling layer has a size of 3*3 and a stride of 2.

[0080] The second convolutional unit has a 5x5 convolutional layer with a stride of 1 and 256 channels, and uses ReLU activation function; its max pooling layer has a 3x3 convolutional layer with a stride of 2.

[0081] The first and second convolutional layers in the third convolutional unit are both 3*3 in size, with a stride of 1 and 384 channels, and both use ReLU activation function; the third convolutional layer is 3*3 in size, with a stride of 1 and 256 channels, and uses ReLU activation function; its max pooling layer is 3*3 in size and has a stride of 2.

[0082] It consists of three fully connected layers. The first and second fully connected layers each have 4096 neurons and use ReLU activation function. The third fully connected layer has 6 neurons.

[0083] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for identifying the suspended matter in urine diagnosis of Tibetan medicine, characterized in that, Comprise: S1, urine image acquisition; S2, the collected urine image is detected and segmented, and a target area is obtained; S3, the target area is corrected for liquid surface extraction; specifically: the target area is stretched to obtain a square cup opening image liquid surface area; S4, the liquid surface area is filtered by rgb color value; step S4 specifically includes the following steps: S41, in the liquid surface area obtained in step S3, average sampling is performed on the Red, Green and Blue channel images at intervals of parameters sample_interval_x and sample_interval_y to obtain average RGB color values; S42, map the RGB color space to the HSV color space, adjust the brightness component in the HSV color space to the highest, and then convert back to the RGB color space; S43、According to the RGB color value of each sampling point after adjustment, calculate the variance , set the variance threshold , if , it is considered that there is no suspended matter in the urine, filter out the image, otherwise, it is considered that there is suspended matter in the urine; S5, texture feature extraction is performed on the image processed in step S4; step S5 specifically includes the following steps: S51, the liquid surface area obtained in step S4 is divided into 5 rows and 5 columns of 25 sub-areas, and each sub-area is represented by horizontal and vertical coordinates, with coordinates from (0, 0) to (4, 4); S52, the LBP algorithm is used to transform the sub-areas located at central coordinates (1, 1), (1, 2), (1, 3), (2, 1), (2, 2), (2, 3), (3, 1), (3, 2), (3, 3) to obtain the transformed feature map as follows: ； S53, dividing each feature map after the conversion into a grid of ; S54, for a pixel point in each grid, the gray values of the adjacent 8 pixel points are compared with it, if the gray values of the adjacent 8 pixel points are greater than the gray value of the pixel point, the position of the pixel point is marked as 1, otherwise as 0; S55, calculate the histogram of each grid; then normalize the histogram; S56, connect the obtained statistical histogram of each grid into a feature vector; the feature vector group composed of the feature vectors of the central 9 sub-areas is as follows: ； S6, texture feature value filtering is performed on the image processed in step S5; step S6 specifically includes the following steps: S61. Defining a suspended matter disturbance evaluation parameter as a function of the texture characteristic As follows: ； wherein denotes the Euclidean geometric distance; S62, setting a threshold value if is considered as no suspended matter in the urine filtered out of the image, otherwise, it is considered as having suspended matter in the urine; S7, contour feature extraction is performed on the image processed in step S6; S8, according to the image corrected by the liquid surface in step S3, the texture features extracted in step S5, and the contour features extracted in step S7, suspended matter is identified.

2. A method of identifying a urine diagnostic suspension according to claim 1, characterized in that, The collection process is: the urine sample uses a unified square plastic transparent cup, the urine sample is placed on a white A4 paper and photographed from above, the horizontal is photographed, and the liquid surface is in the center of the image, and the image collection is completed.

3. A method of identifying a urine diagnostic suspension according to claim 2, characterized in that, The template graph assisted correction contour extraction method is used to realize the detection and segmentation of the cup body, and the template graph is a sample picture with standard absolute standard and shooting angle, and the specific segmentation process includes the following steps: S21, Gaussian filtering is performed on the template graph and the sample graph to eliminate image noise; S22, the template graph and the sample graph processed in step S21 are sharpened by USM, the content of the high frequency part of the image is enhanced, the definition of the image is enhanced, and the contour and edge of the image are strengthened; S23, the template graph and the sample graph processed in step S22 are converted into a gray scale image; S24, the template image and the sample image converted into a gray image are sharpened by using a Laplace second-order differential operator; S25, the template image and the sample image after sharpening are binarized; S26, the images after binarization of the template image and the sample image are used for contour matching to find a contour with the minimum matching value; S27, the contour with the minimum matching value is filled with a mark, and the corresponding part of the original sample image is assigned to a newly created blank image through the mark image to segment the target region.

4. A method of identifying a Tibetan urine diagnostic sediment according to claim 3, wherein, The matching value calculation formula of the contour in step S26 is: ； Wherein, A is used to represent the image after the template map binarization, B represents the image after the sample map binarization, is the i-th Hu moment generated by A, is the i-th Hu moment generated by B, i=0,1,2,3,4,5,6, | | represents the absolute value operation.

5. A method of identifying a urine diagnostic suspension according to claim 4, characterized in that, Step S8 is specifically: S81, a convolutional neural network is constructed; the convolutional neural network comprises a first convolution unit, a second convolution unit, a third convolution unit and a fully connected unit; the first and second convolution units each comprise a convolution layer, a local response normalization layer and a maximum pooling layer; the third convolution unit comprises three convolution layers and a maximum pooling layer; and the fully connected unit comprises three fully connected layers; S82, the different color channel images in the RGB image of the liquid surface region obtained in step S3, the feature images of each sub-region converted in step S5 and the binarized feature images in step S7 are taken as inputs of the convolutional neural network; and a recognition result of the suspended matter is obtained.

6. A diagnostic device for identifying urine sediment in a Tibetan urine test, characterized in that, It comprises: a urine image acquisition module, a cup body detection and segmentation module, a liquid surface extraction and correction module, a first filtering module, a first feature extraction module, a second filtering module, a second feature extraction module and a suspended matter recognition module; The urine image acquisition module is used to acquire a urine image in a uniform square plastic transparent cup; the cup body detection and segmentation module is used to remove invalid information outside the cup body in the output result of the urine image acquisition module; the liquid surface extraction and correction module performs target region stretching according to the output of the cup body detection and segmentation module to obtain a square cup mouth image liquid surface region; the first filtering module filters out images without suspended matter in the urine from the output result of the liquid surface extraction and correction module; the first feature extraction module extracts texture features from the output result of the first filtering module; and the second filtering module filters the texture features extracted by the first feature extraction module. The second feature extraction module is used to perform edge detection on the output result of the second filtering module to obtain a binarized feature image; and the suspended matter recognition module performs suspended matter recognition according to the output result of the first feature module, the output result of the liquid surface correction module and the output result of the second feature extraction module. The first filtering module filters out images without suspended matter in the urine from the output result of the liquid surface extraction and correction module, and the implementation process is as follows: S41, average sampling is performed on the Red, Green and Blue channel images in the liquid surface region at intervals of sample_interval_x and sample_interval_y to obtain average RGB color values; S42, the RGB color space is mapped to the HSV color space, the brightness component in the HSV color space is adjusted to the highest, and then the RGB color space is converted back in reverse; S43、According to the RGB color value of each sampling point after adjustment, calculate the variance , set the variance threshold , if , it is considered that there is no suspended matter in the urine, filter out the image, otherwise, it is considered that there is suspended matter in the urine; The first feature extraction module extracts texture features from the output result of the first filtering module, and the implementation process is as follows: S51, divide the liquid level region into 5 rows and 5 columns of 25 sub-regions, and express each sub-region with horizontal and vertical coordinates, the coordinates from (0, 0) to (4, 4); S52, transform the sub-regions with central coordinates (1, 1), (1, 2), (1, 3), (2, 1), (2, 2), (2, 3), (3, 1), (3, 2), (3, 3) by using the LBP algorithm, and obtain the transformed feature map as follows: ； S53, dividing each feature map after the conversion into a grid of ; S54, for a pixel point in each grid, compare the gray values of the adjacent 8 pixel points with the pixel point, if the gray values of the adjacent 8 pixel points are greater than the gray value of the pixel point, the position of the pixel point is marked as 1, otherwise as 0; S55, calculate the histogram of each grid, and then normalize the histogram; S56, connect the obtained statistical histogram of each grid to form a feature vector, and obtain the feature vector group composed of the feature vectors of the 9 sub-regions in the center as follows: ； The second filtering module filters the texture features extracted by the first feature extraction module, and the implementation process is as follows: S61. Defining a suspended matter disturbance evaluation parameter as a function of the texture characteristic As follows: ； wherein denotes the Euclidean geometric distance; S62, setting a threshold value If If the value is considered as no suspended matter in the urine, otherwise, it is considered as having suspended matter in the urine.

Images