Turbidity evaluation processing method and system for blood separation plasma sample and storage medium

Abstract

The invention discloses a turbidity evaluation processing method and system for a blood separation plasma sample and a storage medium, and the method comprises the steps: firstly, processing a first image containing target cells to obtain a second image, and intercepting a third image only containing the target cells from the second image; therefore, errors caused by subsequent identification of the attribute information of the target cell due to factors such as the surrounding environment of the target cell are reduced; inputting the third image into an image recognition model trained through operations such as a convolutional layer, a pooling layer and an activation function, detecting, recognizing and outputting five-level associated feature information of the target cell, performing associated recognition on the five-level associated feature information and a preset contrast feature information table, and outputting attribute information of the target cell; by guaranteeing the recognition precision of the sample image, the turbidity of the plasma sample is evaluated and monitored.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a turbidity evaluation and processing method, system and storage medium for blood separation plasma samples. Background Art

[0002] A blood component separator (commonly called a blood separator or blood component separation device) is a medical device used to separate whole blood into different components. Its main purpose is to separate blood into multiple useful components, such as red blood cells, plasma, platelets and other blood derivatives, based on the physical and chemical properties of each component in the blood through specific separation techniques. These separated components are widely used in clinical treatment, research, and the pharmaceutical industry.

[0003] When operating a blood component separator, turbidity is an important reference indicator; turbidity refers to the degree of light scattering and absorption by tiny particulate matter suspended in the blood component separator fluid, and is usually used to measure the clarity or transparency of the blood component separator fluid. The higher the turbidity, the more turbid the blood component separator fluid looks. When using a blood component separator for blood separation, the turbidity of the plasma sample may be affected by the following reasons:

[0004] For example: fat emulsion (fat particle suspension) Reason: The plasma of patients with hyperlipidemia contains more lipid substances, or the fat particles are not completely removed during the blood collection and separation process. The suspended fat particles will increase the turbidity of the plasma, making the sample look milky white. At the same time, there may be some parasites or microorganisms in the blood sample: for example, certain bacteria (such as streptococci), spirochetes (such as Treponema pallidum) or malarial parasites. For example: Cell fragments and hemolysis Reason: During the blood separation process, if the operation is improper (such as excessive centrifugal force or too long time), it may cause red blood cells, white blood cells, etc. to rupture and release intracellular components. Cell fragments and released cellular substances make the plasma turbid, affecting the quality of the sample.

[0005] With the development of high-definition electron microscope technology, this technical means can not only apply image processing technology to identify various types of image features, but also further identify the attribute information of the image; however, further research has found that when the sample is sampled in high-definition, the influence of the fat particle suspension around the target cell will further lead to recognition errors; for example, when taking a photo to identify the target cell in an image, there will be fat particle suspension around the target cell, and even various textures on the surface of the fat particle suspension, which will lead to the recognition result that there will be textures on the fat particle suspension, and thus mistakenly think that it is the target cell (i.e., red blood cells and white blood cells), which is very different from the real recognition purpose;

[0006] Therefore, how to ensure accurate identification of the type and quantity of interfering substances (i.e., malarial parasites and suspended fat particles) in the blood component separation fluid is an important basis for the turbidity assessment of plasma samples. Summary of the invention

[0007] The object of the present invention is to provide a method, system and storage medium for evaluating the turbidity of blood separated plasma samples, which solves the above technical problems pointed out in the prior art.

[0008] A method for evaluating the turbidity of a blood separation plasma sample comprises the following steps:

[0009] acquiring a first image containing a target cell;

[0010] Performing recognition processing on the first image to obtain a second image in the first image;

[0011] The second image is a target cell region image;

[0012] The third image is obtained by performing recognition and analysis on the second image through an edge detection algorithm;

[0013] The third image is the target cell image;

[0014] Inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell;

[0015] The attribute information of the target cell includes name information of the target cell;

[0016] According to the number of target cells identified per unit area, the turbidity rating information of the sample is determined and the turbidity rating information is output.

[0017] Preferably, inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell includes:

[0018] Performing initial recognition on the third image using the image recognition model, and obtaining five-level associated feature information of the current third image by initial recognition;

[0019] The five-level associated feature information includes color feature information, shape feature information, structure feature information, local feature information, and global feature information;

[0020] Calculate and obtain the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table;

[0021] The control attribute information of the control feature information corresponding to the highest similarity is determined to be the attribute information of the target cell.

[0022] Preferably, the initial recognition of the third image by using the image recognition model to obtain five-level associated feature information of the current third image includes the following steps:

[0023] Establish an initial image recognition model;

[0024] The third image is input into a trained image recognition model obtained by training the initial image recognition model, and then the five-level associated feature information corresponding to the third image is output.

[0025] Preferably, the step of inputting the third image into a trained image recognition model obtained by training the initial image recognition model and then outputting the five-level associated feature information corresponding to the third image comprises the following steps:

[0026] Acquire a training set, wherein the training set includes a plurality of training data and annotated data corresponding to each of the training data; the training data is a training image; the annotated data is annotated attribute information corresponding to a target cell in the training data;

[0027] Input the training data into a pre-established initial image recognition model, and output an image recognition vector;

[0028] Obtaining a loss value by calculating a loss function based on the image recognition vector and the labeled data;

[0029] Back-propagating the loss value to update the parameters of the initial image recognition model, performing iterative training by presetting a maximum number of iterations and a maximum threshold of the loss value, to obtain a trained image recognition model;

[0030] The third image is input into the trained image recognition model, and the five-level associated feature information corresponding to the third image is output.

[0031] Preferably, the step of inputting the third image into the trained image recognition model and outputting the five-level associated feature information corresponding to the third image comprises the following steps:

[0032] Forward propagation of the third image through the trained image recognition model to output an image vector;

[0033] The image vector is decoded to obtain five-level associated feature information.

[0034] Preferably, forward propagating the third image through the trained image recognition model to output an image vector comprises the following steps:

[0035] Acquire image and model parameter data; in an initial state, set p convolution kernels of size a×b; based on the convolution kernel and the image and model parameter data, perform a convolution operation on all pixels in the third image to obtain a signal value corresponding to each pixel;

[0036] The image and model parameter data include all pixels in the third image and the coordinates (x, y) corresponding to all pixels, the intensity values ​​t corresponding to the pixels i , the weight value corresponding to the trained image recognition model; wherein p is greater than or equal to 1;

[0037] The signal value corresponding to the pixel point is expressed as:

[0038]

[0039] Performing a nonlinear mapping operation on the signal value through an activation function to obtain an activation signal corresponding to the signal value;

[0040] Downsampling the activation signal through a pooling layer to obtain a target activation signal;

[0041] The target activation signal is converted through a fully connected layer to obtain an image vector.

[0042] Preferably, the calculating and obtaining the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table respectively comprises the following operation steps:

[0043] Setting header data for the five-level associated characteristic information;

[0044] The header data includes color feature header data O, shape feature header data P, structure feature header data Q, local feature header data R, and global feature header data S;

[0045] Traversing all the comparison characteristic information in the characteristic information comparison table, and obtaining the comparison header data in the comparison characteristic information;

[0046] The comparison header data includes comparison color feature header data O', comparison shape feature header data P', comparison structure feature header data Q', comparison local feature header data R', and comparison global feature header data S';

[0047] Establishing an association relationship between the comparison header data and the header data to obtain a comparison association matrix D;

[0048] The control association matrix is ​​expressed as:

[0049]

[0050] The similarity of each factor in the comparison association matrix is ​​calculated and obtained.

[0051] The present invention also provides a turbidity evaluation and processing system for blood separation plasma samples, comprising a first image acquisition module, a second image acquisition module, a third image acquisition module, and an image recognition module;

[0052] Wherein, the first image acquisition module is used to acquire a first image containing target cells;

[0053] The second image acquisition module is used to perform recognition processing on the first image to obtain a second image in the first image;

[0054] The second image is a target cell region image;

[0055] The third image acquisition module is used to identify and analyze the second image through an edge detection algorithm and then capture the third image;

[0056] The third image is the target cell image;

[0057] An image recognition module, used for inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell;

[0058] The attribute information of the target cell includes name information of the target cell.

[0059] Preferably, the image recognition module includes an initial recognition submodule, a calculation submodule, and a screening and acquisition submodule;

[0060] The initial recognition submodule is used to perform initial recognition on the third image using the image recognition model, and obtain five-level associated feature information of the current third image by initial recognition;

[0061] The five-level associated feature information includes color feature information, shape feature information, structure feature information, local feature information, and global feature information;

[0062] The calculation submodule is used to calculate and obtain the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table;

[0063] The screening and obtaining submodule is used to determine that the control attribute information of the control feature information corresponding to the highest similarity is the attribute information of the target cell.

[0064] Preferably, the initial recognition submodule is specifically used to establish an initial image recognition model;

[0065] The third image is input into a trained image recognition model obtained by training the initial image recognition model, and then the five-level associated feature information corresponding to the third image is output.

[0066] Preferably, the initial recognition submodule, in specific implementation, is used to obtain a training set, the training set including a plurality of training data and annotated data corresponding to each of the training data; the training data is a training image; the annotated data is attribute information corresponding to target cells in the training data annotated with the training data;

[0067] Input the training data into a pre-established initial image recognition model, and output an image recognition vector;

[0068] Obtaining a loss value by calculating a loss function based on the image recognition vector and the labeled data;

[0069] Back-propagating the loss value to update the parameters of the initial image recognition model, performing iterative training by presetting a maximum number of iterations and a maximum threshold of the loss value, to obtain a trained image recognition model;

[0070] The third image is input into the trained image recognition model, and the five-level associated feature information corresponding to the third image is output.

[0071] Preferably, the initial recognition submodule, in specific implementation, is further used to forward-propagate the third image through the trained image recognition model to output an image vector;

[0072] Decoding the image vector to obtain five-level associated feature information;

[0073] Preferably, the initial recognition submodule, in a specific implementation, is further used to obtain image and model parameter data; in an initial state, p convolution kernels of a×b size are set; based on the convolution kernel and the image and model parameter data, a convolution operation is performed on all pixel points in the third image to obtain a signal value corresponding to each pixel point;

[0074] The image and model parameter data include all pixels in the third image and the coordinates (x, y) corresponding to all pixels, the intensity values ​​t corresponding to the pixels i , the weight value corresponding to the trained image recognition model; wherein p is greater than or equal to 1;

[0075] The signal value corresponding to the pixel point is expressed as:

[0076]

[0077] Performing a nonlinear mapping operation on the signal value through an activation function to obtain an activation signal corresponding to the signal value;

[0078] Downsampling the activation signal through a pooling layer to obtain a target activation signal;

[0079] Converting the target activation signal through a fully connected layer to obtain an image vector;

[0080] Preferably, the calculation submodule is specifically used to set header data for the five-level association feature information;

[0081] The header data includes color feature header data O, shape feature header data P, structure feature header data Q, local feature header data R, and global feature header data S;

[0082] Traversing all the comparison characteristic information in the characteristic information comparison table, and obtaining the comparison header data in the comparison characteristic information;

[0083] The comparison header data includes comparison color feature header data O', comparison shape feature header data P', comparison structure feature header data Q', comparison local feature header data R', and comparison global feature header data S';

[0084] Establishing an association relationship between the comparison header data and the header data to obtain a comparison association matrix D;

[0085] The control association matrix is ​​expressed as:

[0086]

[0087] Calculate and obtain the similarity of each factor in the control association matrix;

[0088] The present invention also provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for evaluating the turbidity of a blood-separated plasma sample are implemented.

[0089] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0090] By analyzing the above-mentioned turbidity evaluation and processing method, system and storage medium for blood separation plasma samples provided by the present invention, it can be known that in specific applications, the first image containing target cells is first processed to obtain a second image, and a third image containing only target cells is intercepted from the second image, which can reduce the error caused by the subsequent identification of target cell attribute information due to factors such as the surrounding environment of the target cells; the third image is then input into a trained image recognition model to detect and identify the feature information of the output third image, and the image recognition model can be used to convert the input image into a corresponding feature vector or feature descriptor, and these feature information can accurately represent the content and features of the image;

[0091] By comparing the preset feature information comparison table with the feature information of the third image, and calculating the similarity between the acquired feature information and all the comparison feature information in the preset feature information comparison table, the feature information most similar to the input image can be quickly found; in this way, feature matching and image recognition analysis tasks can be effectively performed, and the similarity between the feature information of the third image and the comparison feature information in the feature information comparison table is calculated, so as to determine that the comparison feature information with the highest similarity is the target feature information, and then determine that the comparison attribute information corresponding to the target feature information is the attribute information of the third image, that is, the attribute information of the target cell; by determining the comparison feature information with the highest similarity, the feature information corresponding to the target image can be accurately found, and the comparison attribute information corresponding to the comparison feature information can be used to describe the attribute characteristics of the target cell, such as color, shape, texture, etc.; finally, the target cell in the image can be quickly and effectively identified, and other factors can be prevented from causing recognition errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0093] Figure 1 A schematic diagram of the overall operation steps of a method for evaluating the turbidity of a blood-separated plasma sample provided in Example 1 of the present invention;

[0094] Figure 2 A schematic diagram of the operation steps for determining the attribute information of target cells in a method for evaluating the turbidity of a blood separation plasma sample provided in Embodiment 1 of the present invention;

[0095] Figure 3A schematic diagram of the operation steps for obtaining five-level correlation feature information of a method for evaluating the turbidity of a blood-separated plasma sample provided in the first embodiment of the present invention;

[0096] Figure 4 A more detailed schematic diagram of the operation steps for obtaining five-level correlation feature information of the turbidity evaluation and processing method for blood separation plasma samples provided in the first embodiment of the present invention;

[0097] Figure 5 A schematic diagram of the operation steps for obtaining five-level correlation feature information by further explaining the turbidity evaluation and processing method for blood separation plasma samples provided in the first embodiment of the present invention;

[0098] Figure 6 A schematic diagram of the operation steps of obtaining an image vector in a method for evaluating the turbidity of a blood separation plasma sample provided in the first embodiment of the present invention;

[0099] Figure 7 A schematic diagram of the operation steps of calculating the similarity between five-level correlation feature information and control feature information in a method for evaluating the turbidity of a blood separation plasma sample provided in Example 1 of the present invention;

[0100] Figure 8 This is a schematic diagram of the overall architecture of a turbidity evaluation and processing system for blood separation plasma samples provided in Embodiment 2 of the present invention.

[0101] Reference numerals: first image acquisition module 10 ; second image acquisition module 20 ; third image acquisition module 30 ; image recognition module 40 ; initial recognition submodule 41 ; calculation submodule 42 ; screening acquisition submodule 43 . DETAILED DESCRIPTION

[0102] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0103] The present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0104] Embodiment 1

[0105] like Figure 1 As shown, accordingly, the present invention also proposes a turbidity evaluation and processing method for blood separation plasma samples, comprising the following operating steps:

[0106] Step S10: acquiring a first image containing target cells;

[0107] The first image is an initial image, which is a plurality of initial images for identifying target cells in an image obtained by taking a camera or manually collecting;

[0108] Step S20: performing recognition processing on the first image to obtain a second image in the first image;

[0109] The second image is a target cell region image;

[0110] It should be noted that in the above-mentioned embodiment of the present application, the first image can be identified and processed through the YOLO model or the SSD model, so as to select the image of the area where the target cells are located in the first image (that is, the above-mentioned second image), and the area outside the area where the target cells are located in the first image is screened out, and only the target cell area that is of great significance for the target cell identification is retained, thereby reducing the amount of data processed by the subsequent server, thereby improving data processing efficiency.

[0111] Step S30: using an edge detection algorithm to identify and analyze the second image to obtain edge information of the target cell; and obtaining a third image based on the edge information;

[0112] The third image is the target cell image;

[0113] It should be noted that the target cell area is identified by the edge detection algorithm to obtain the edge information of the target cell, and then the target cell image can be intercepted according to the edge information in the subsequent step, which can reduce the recognition error caused by the influence of the pixel points of the image outside the target cell image when the attribute information of the target cell is subsequently identified; the edge detection algorithm includes Canny edge detection (detecting the edge in the image through Gaussian filtering, calculating gradient, non-maximum suppression and double threshold processing), Sobel edge detection (obtaining the edge image by calculating the gradient of the image in the horizontal and vertical directions, and then merging the two gradient results), Laplacian edge detection (detecting the edge in the image by calculating the Laplacian operator of the image), etc., which can realize the operation of automatic recognition and detection. The above edge detection algorithms are all prior art and will not be repeated in the embodiments of the present application;

[0114] The above-mentioned edge information refers to the outline of the outermost part of the target cell in the target cell region; based on this outline, an image containing only the target cell can be cut out from the target region image.

[0115] Step S40: inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell;

[0116] The target cell attribute information includes the name information of the target cell and the type information of the target cell; the target cell can be a red blood cell or a white blood cell, etc. However, what affects the identification of the target cell in the blood component separation solution is the interference in the collected image; for example, the interference in the blood component separation solution includes parasites such as malarial parasites or microorganisms and fat particle suspensions;

[0117] In the specific operation, the separated blood components are made into a smear and stained (such as Giemsa staining). A high-precision electron image microscope is used to check whether there are cells or parasites in the form of long strips. It should be noted that the technical solution adopted in the above-mentioned embodiment of the present application can directly cut out the target cell image from the target cell area through edge information, thereby obtaining an image containing only the target cells, ensuring that when the image recognition model is used to identify and analyze the attribute information of the target cells in the subsequent process, it will not be affected by other factors (these factors will include the remaining pixels in the target cell area that do not belong to the target cell image), thereby ensuring that the recognition accuracy is improved;

[0118] In the above-mentioned embodiment of the present application, the first image containing the target cells is processed to obtain the second image, and the third image containing only the target cells is cut out from the second image, which can reduce the error caused by the subsequent identification of the target cell attribute information due to factors such as the surrounding environment of the target cells; the third image is then input into the trained image recognition model to detect and identify the feature information of the output third image. Through the image recognition model, the input image can be converted into a corresponding feature vector or feature descriptor. This feature information can accurately represent the content and features of the image; the feature information of the third image is compared with the feature information of the preset feature information comparison table, and the similarity between the acquired feature information and all the comparison feature information in the preset feature information comparison table is calculated, so that the image can be quickly found. to the feature information that is most similar to the input image; in this way, feature matching and image recognition analysis tasks can be effectively performed, and the similarity between the feature information of the third image and the control feature information in the feature information comparison table is calculated, so as to determine that the control feature information with the highest similarity is the target feature information, and then determine that the control attribute information corresponding to the target feature information is the attribute information of the third image, that is, the attribute information of the target cell. By determining the control feature information with the highest similarity, the feature information corresponding to the target image can be accurately found, and the control attribute information corresponding to the control feature information can be used to describe the attribute characteristics of the target cell, such as color, shape, texture, etc.; finally, the target cell in the image can be quickly and effectively identified, and other factors can be prevented from causing recognition errors.

[0119] Then, based on the number of target cells identified per unit area, turbidity rating information of the sample is determined and output.

[0120] In cell analysis, the number of cells and their distribution can reflect the turbidity level of the sample. The rating standards can be set according to the specific application scenario and are generally divided into the following levels: Low Turbidity: The number of cells per unit area is small, indicating that the sample is relatively clear. Moderate Turbidity: The number of cells per unit area is moderate, indicating that the sample has a certain degree of turbidity. High Turbidity: The number of cells per unit area is large, indicating that the sample is relatively turbid.

[0121] For example, calculate the number of target cells per unit area

[0122] To determine the analysis area, first determine the correspondence between the image size and the actual area: First, you need to know the actual physical area (such as square microns, square millimeters, etc.) corresponding to each pixel in the image. This requires pre-calibration of the magnification and resolution of the microscope or imaging device. For example: If the magnification of the microscope is 1000 times, the actual area corresponding to each pixel is 0.1 square microns. The image size is 1024x1024 pixels, so the actual analysis area is 1024*0.1=102.4 microns×102.4 microns.

[0123] Cell counting, identification and marking of target cells: Through the pre-trained image recognition model, the system has identified and marked all target cells in the second image (target cell area image).

[0124] Calculate the cell density per unit area. Use the following formula to calculate the target cell density per unit area: cell density = total number of target cells divided by the area of ​​the analysis region; for example: total number of target cells: 500, analysis region: 100 square microns; cell density = 500 / 100 = 5 cells / square micron;

[0125] Determine turbidity rating: Based on the calculated cell density, map it to a predefined turbidity rating standard. The following is an exemplary rating mechanism: Cell density range (cells / square micron) Turbidity level 0-1 Very low turbidity; 1-5 Low turbidity; 5-10 Medium turbidity; 10-20 High turbidity; >20 Very high turbidity.

[0126] Specifically, Figure 2 As shown, in step S40, the third image is input into a pre-trained image recognition model to determine the attribute information of the target cell, including:

[0127] Step S41: using the image recognition model to perform initial recognition on the third image, and obtaining five-level associated feature information of the current third image by initial recognition;

[0128] The five-level associated feature information includes color feature information, shape feature information, structure feature information, local feature information, and global feature information;

[0129] It should be noted that after the color feature information, shape feature information, structure feature information, local feature information, and global feature information of the output third image are identified by the image recognition model, the attribute information corresponding to the target cells in the third image can be identified by the preset feature information comparison table, that is, the attribute information of the target cells, that is, the type information and name information of the target cells;

[0130] For example: the color feature information of the third image output by the image recognition model is C.01-red; the shape feature information is S.314-biconcave disc, the structure feature information is St.152-no nucleus and St.02-with cell membrane body; the local feature information is L.166-no edges; the global feature information is G.152-red blood cells in human blood; therefore, the object in the current third image can be identified as {C.01, S.314, St.152, St.02, L.166, G.152}-red blood cells according to the feature information comparison table.

[0131] Step S42: Calculate and obtain the similarity between the five-level associated feature information and all the reference feature information in the preset feature information comparison table;

[0132] Step S43: determining the control feature information corresponding to the highest similarity as the target control feature information; and determining the control attribute information corresponding to the target feature information as the attribute information of the target cell.

[0133] It should be noted that the technical solution adopted in the above-mentioned embodiment of the present application detects and identifies the feature information of the output third image through a pre-trained image recognition model. Through the image recognition model, the input image can be converted into a corresponding feature vector or feature descriptor, and these feature information can accurately represent the content and features of the image; by comparing the feature information of the third image with the preset feature information comparison table, and by calculating the similarity between the acquired feature information and all the comparison feature information in the preset feature information comparison table, the feature information most similar to the input image can be quickly found; in this way, feature matching and image recognition analysis tasks can be effectively performed, and the similarity between the feature information of the third image and the comparison feature information in the feature information comparison table is calculated, so as to determine that the comparison feature information with the highest similarity is the target feature information, and then determine that the comparison attribute information corresponding to the target feature information is the attribute information of the third image, that is, the attribute information of the target cell. By determining the comparison feature information with the highest similarity, the feature information corresponding to the target image can be accurately found, and the comparison attribute information corresponding to the comparison feature information can be used to describe the attribute characteristics of the target cell, such as color, shape, texture, etc.

[0134] Specifically, Figure 3 As shown, in step S41, the image recognition model is used to perform initial recognition on the third image, and the initial recognition obtains five-level associated feature information of the current third image, including the following operation steps:

[0135] Step S411: establishing an initial image recognition model;

[0136] Step S412: inputting the third image into a trained image recognition model obtained by training the initial image recognition model and outputting the five-level correlation feature information corresponding to the third image;

[0137] It should be noted that, in the above-mentioned embodiment of the present application, after the initial image recognition model is established, the initial image recognition model is trained to obtain a trained image recognition model, and then the third image is input into the trained image recognition model, and the five-level associated feature information corresponding to the third image is output, that is, the third image containing only the target cells is input into the trained image recognition model, and the color feature information, shape feature information, structural feature information, local feature information, and global feature information corresponding to the target cells in the third image are output, so as to identify the attribute information of the target cells in the third image according to the five-level feature information in the third image;

[0138] In this embodiment, after the image containing only the above-mentioned red blood cells is input into the trained image recognition model, the color feature information of the output red blood cell image (i.e., the third image) is C.01-red; the shape feature information is S.314-biconcave disc-shaped, the structural feature information is St.152-no cell nucleus and St.02-with cell membrane body; the local feature information is L.166-no edges; and the global feature information is G.152-red blood cells in human blood.

[0139] St.152-No nucleus: Mature red blood cells lack a nucleus, a feature that distinguishes them from other types of blood cells (such as white blood cells). St.02-With cell membrane: Red blood cells have an elastic and tough cell membrane that allows them to pass through narrow capillaries during blood circulation. Local features (L.166-No edges): The surface of red blood cells is smooth and has no obvious edges, which helps them to deform flexibly.

[0140] When blood smears are processed with Giemsa stain and viewed under an optical microscope, the colors of red blood cells and white blood cells appear as follows: Red blood cell (erythrocyte) color: Pink to red. This is because the dye in the Giemsa stain binds to the hemoglobin in the red blood cells, giving them a typical pink to red hue.

[0141] Similarly, if it is (i.e. the third image), the color feature information is C.04-green; the shape feature information is S.316-arc-shaped, the structure feature information is St.196-has a nucleus and St.02-has a cell membrane body; the local feature information is L.199-has edges; and the global feature information is G.187-Plasmodium gametocytes in human blood.

[0142] Blood smears were processed with Giemsa stain and observed under an optical microscope. Plasmodium gametocytes are bow-shaped or curved crescent-shaped, which is a significant morphological feature. In immunofluorescence staining, Plasmodium gametocytes are usually labeled green. This is because one of the commonly used fluorescent markers is fluorescein isothiocyanate (FITC), which emits green fluorescence. When FITC is used to link specific antibodies to label pathogens, the gametocytes of P.falciparum will show green fluorescence under a fluorescence microscope, thereby improving the sensitivity and specificity of the detection.

[0143] Specifically, Figure 4 As shown, in step S412, the third image is input into the trained image recognition model obtained by training the initial image recognition model and then output to obtain the five-level association feature information corresponding to the third image, including the following operation steps:

[0144] Step S4121: obtaining a training set, wherein the training set includes a plurality of training data and annotated data corresponding to each of the training data; the training data is obtained from training images by crawler technology; the annotated data is annotated with the training data for the attribute information corresponding to the target cells in the training data;

[0145] Step S4122: input the training data into a pre-established initial image recognition model, and output an image recognition vector;

[0146] Step S4123: Obtaining a loss value by calculating a loss function based on the image recognition vector and the labeled data;

[0147] Step S4124: back-propagating the loss value to update the parameters of the initial image recognition model, completing the first iteration, and setting an iteration counter, increasing the number of iterations of the iteration counter by one; and repeating the above operation, recording the loss value and the number of iterations;

[0148] The iteration counter is initially set to 0 and is set to 1 when the first iteration is completed;

[0149] Step S4125: Preset a maximum number of iterations and a maximum threshold of the loss value; stop training when the loss value is less than or equal to the maximum threshold of the loss value or when the number of iterations reaches the maximum number of iterations, and obtain a trained image recognition model;

[0150] Step S4126: input the third image into the trained image recognition model, and output the five-level association feature information corresponding to the third image.

[0151] It should be noted that the above-mentioned embodiment of the present application trains the image recognition model by crawling a large amount of training data and annotated data in the training set obtained, and crawls multiple images containing target cells from the plasma sample database as training data through crawler technology, and annotates these data, that is, adds attribute information to the target cells in each image, and provides labeled training data for subsequent model training; further, the image recognition vector is obtained through the forward calculation of the model, and the image data is converted into a vector representation that can be used to calculate the loss value through the model; then the loss value is calculated through the defined loss function. The loss value is a metric that measures the gap between the model prediction result and the true label, evaluates the performance of the current model on the training set, and provides feedback signals for the next parameter update; and through repeated parameter updates and iterations, the model is gradually optimized to improve the accuracy of image recognition, and finally the termination condition of the training is set to avoid overfitting or unlimited training; through cyclic iterations, the model can be gradually optimized so that it can accurately recognize the image, and finally a trained image recognition model is obtained;

[0152] The above-mentioned embodiment of the present application annotates a large number of training images (i.e., the above-mentioned training set, which can also be considered as an image containing only target cells) with the five-level associated feature information (color feature information, shape feature information, structural feature information, and local feature information) of the target cells in the training images, and then inputs the training images into the initial image recognition model to output predicted data (i.e., the above-mentioned image recognition vector), and then calculates the loss value based on the predicted data and the annotated data (i.e., the five-level associated feature information); further, the final trained image recognition model is obtained based on the preset loss value threshold and the number of iterations threshold; and finally, the current third image is input into the trained image recognition model to output the five-level associated feature information corresponding to the current third image.

[0153] Better, such as Figure 5 As shown, in step S4126, the third image is input into the trained image recognition model, and the five-level association feature information corresponding to the third image is output, including the following operation steps:

[0154] Step S41261: forward propagating the third image through the trained image recognition model to output an image vector;

[0155] Step S41262: Decoding the image vector to obtain five-level correlation feature information;

[0156] It should be noted that the above decoding operation can obtain the five-level correlation feature information corresponding to the third image through operations such as feature visualization and deconvolution, so that the attribute information of the target cells in the image can be obtained by analyzing the five-level correlation feature information and the control feature information.

[0157] It should be noted that when the embodiment of the present application obtains the five-level associated feature information in the third image (i.e., the color feature information, shape feature information, structural feature information, and local feature information of the red blood cells mentioned above), in the trained image recognition model, forward propagation is first performed to output the image vector, and then the image vector is decoded according to a preset decoder to obtain the color feature information, shape feature information, structural feature information, and local feature information.

[0158] Specifically, Figure 6 As shown, in step S41261, the third image is forward propagated through the trained image recognition model to output an image vector, including the following steps:

[0159] Step S412611: Acquire image and model parameter data; in an initial state, set p convolution kernels of a×b size; perform a convolution operation on all pixels in the third image based on the convolution kernel and the image and model parameter data to obtain a signal value corresponding to each pixel;

[0160] The image and model parameter data include all pixels in the third image and the coordinates (x, y) corresponding to all pixels, the intensity values ​​t corresponding to the pixels i , the weight value corresponding to the trained image recognition model; wherein p is greater than or equal to 1;

[0161] The signal value corresponding to the pixel point is expressed as (that is, the signal value corresponding to the current pixel point i):

[0162]

[0163] The technical solution adopted in the above-mentioned embodiment of the present application can scan and identify the image through one or more (i.e., the above-mentioned p convolution kernels) to extract p features in the image; by using one or more convolution kernels to perform convolution operations on the input image, a series of feature maps are generated; each feature map represents the distribution of a certain local feature in the image, that is, the embodiment of the present application finally outputs five feature information (i.e., the above-mentioned five-level associated feature information), then the number of convolution kernel types in the embodiment of the present application is at least five; each pixel point in the image is convolved by at least five convolution kernels to obtain the signal value corresponding to each pixel point (the signal value can be a color signal value, a shape signal value, a structure signal value, a local signal value, a global signal value, etc.), so that the signal value of each pixel point in the current third image can be finally obtained, so that in the following operation process, the operation of the activation function, the pooling layer, etc. is finally identified and analyzed to obtain the feature vector of the third image, so that in the operation of the above-mentioned step S41262, the feature vector is decoded by the decoder to obtain the five-level associated feature information; the result of the convolution operation is a two-dimensional matrix, that is, the signal value corresponding to the above-mentioned pixel point. Explanation: The five-level correlation feature information (also known as multi-level correlation feature) refers to the parameter feature information of five important levels. However, the reason why the multi-level correlation feature chooses five levels is that if the number and types of the above parameter feature information extracted by sampling are too large (7-8 types), it will cause convolution operation pressure. However, if the number and types of parameter feature information are less (for example, 2-3 types), the recognition accuracy will decrease.

[0164] In addition, the selection of the above five levels of associated feature information is five important parameter feature information selected for a specific detail texture image.

[0165] Step S412612: performing a nonlinear mapping operation on the signal value through an activation function to obtain an activation signal corresponding to the signal value;

[0166] It should be noted that the above-mentioned activation functions include ReLU (rectified linear unit), Sigmoid and Tanh, etc. They perform nonlinear mapping on each element in the feature map (i.e., the signal value corresponding to the above-mentioned pixel point) to generate new matrix information (i.e., the above-mentioned activation signal), which can enhance the expressive ability of the image recognition model.

[0167] Step S412613: downsampling the activation signal through a pooling layer to obtain a target activation signal;

[0168] It should be noted that the above pooling layer usually uses a maximum pooling layer or an average pooling layer for downsampling operations, which can reduce the spatial size of the feature signal matrix and retain important feature information; the maximum pooling selects the maximum value of each area as the pooling result, and the average pooling calculates the average value of each area;

[0169] Take the maximum pooling layer as an example to illustrate the above downsampling operation: the input feature signal matrix of size 4x4 is: And use the 2×2 maximum pooling operation to get the maximum value of each local partition as the output value (that is, the above target activation signal): In this way, after the pooling operation, 6-8-14-16 can be obtained as the output value;

[0170] Step S412614: converting the target activation signal through a fully connected layer to obtain an image vector;

[0171] It should be noted that the above fully connected layer converts the target activation signal to obtain a feature vector of fixed length (that is, the above image vector can obtain five-level associated feature information after decoding operation);

[0172] The above-mentioned embodiment of the present application extracts image features through the convolution layer, introduces nonlinear transformation through the activation function to enhance the image expressiveness, reduces the spatial size through the pooling layer, improves the processing speed, and reduces the computational processing pressure, and the fully connected layer converts the target activation signal to obtain a feature vector of fixed length, and finally obtains a result with abstract feature representation, which provides the above-mentioned trained image recognition model with effective image processing and analysis capabilities.

[0173] Specifically, Figure 7 As shown, in step S42, calculating and obtaining the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table respectively includes the following operation steps:

[0174] Step S421: Setting header data for the five-level association characteristic information;

[0175] The header data includes color feature header data O, shape feature header data P, structure feature header data Q, local feature header data R, and global feature header data S;

[0176] Step S422: traverse all the comparison feature information in the feature information comparison table, and obtain the comparison header data in the comparison feature information;

[0177] The comparison header data includes comparison color feature header data O', comparison shape feature header data P', comparison structure feature header data Q', comparison local feature header data R', and comparison global feature header data S';

[0178] Step S423: establishing an association relationship between the comparison header data and the header data to obtain a comparison association matrix D;

[0179] The control association matrix is ​​expressed as:

[0180]

[0181] Step S424: Calculate and obtain the similarity of each factor in the comparison association matrix (and repeat the above operation until all the comparison feature information in the feature information comparison table is traversed to obtain the similarity between the five-level association feature information and all the comparison feature information in the preset feature information comparison table);

[0182] It should be noted that, in the above-mentioned embodiment of the present application, by setting the header data for the five-level associated feature information, the color feature information in the above-mentioned embodiment is obtained as O#C.01, the shape feature information is P#S.314, the structure feature information is Q#St.152 and Q#St.02-there is a cell membrane body; the local feature information is R#L.166; the global feature information is S#G.152; and then, according to the reference header data of the reference feature information in the feature information comparison table and the header data of the five-level associated feature information, an association relationship is established. For example: the header data of the reference feature information in the feature information comparison table are: red blood cells = {O'#C.01, P'#S.314, Q'#St.152, Q'#St.02, R'#L.166, S'#G.152}, fat particle suspension = {O'#C.02, P'#S.04, Q'#S t.152, Q'#St.02, R'#L.25, S'#G.33}, etc., and then establish the association relationship by corresponding the header data of the five-level association feature information to the reference header data of the reference feature information in the feature information comparison table, that is, O#C.01-O'#C.01, P#S.314-P'#S.314, Q#St.152-Q'#St.152, Q#St. 02-Q'#St.02, R#L.166-R'#L.166, S#G.152-S'#G.152 and O#C.01-O'#C.02, P#S.314-P'#S.04, Q#St.152-Q'#St.152, Q#St.02-Q'#St.02, R#L.166-R'#L.25, S#G.152-S'#G.33, which is expressed as:

[0183] Then, the similarity of each factor in each matrix is ​​calculated, and the above operation is repeated to obtain the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table.

[0184] In summary, the present application provides a turbidity evaluation and processing method for blood separation plasma samples. First, a first image containing target cells is processed to obtain a second image, and a third image containing only target cells is cut out from the second image, which can reduce the error caused by the subsequent identification of target cell attribute information due to factors such as the surrounding environment of the target cells; the third image is then input into a trained image recognition model to detect and identify the feature information of the output third image. Through the image recognition model, the input image can be converted into a corresponding feature vector or feature descriptor. This feature information can accurately represent the content and features of the image; the feature information of the third image is compared with a preset feature information comparison table, and the feature information obtained by calculation is compared with all the comparison feature information in the preset feature information comparison table. By calculating the similarity of the information, the feature information most similar to the input image can be quickly found; in this way, feature matching and image recognition analysis tasks can be effectively performed, and the similarity between the feature information of the third image and the control feature information in the feature information comparison table is calculated, so as to determine that the control feature information with the highest similarity is the target feature information, and then determine that the control attribute information corresponding to the target feature information is the attribute information of the third image, that is, the attribute information of the target cell. By determining the control feature information with the highest similarity, the feature information corresponding to the target image can be accurately found, and the control attribute information corresponding to the control feature information can be used to describe the attribute characteristics of the target cell, such as color, shape, texture, etc.; finally, the target cell in the image can be quickly and effectively identified, and other factors can be prevented from causing recognition errors;

[0185] Specifically, when the feature information of the third image is output by detecting the trained image recognition model, an initial image recognition model is first established, and then the initial image recognition model is trained to obtain the trained image recognition model, and then the third image is input into the trained image recognition model, and the convolution operation is performed on the pixels in the third image by the convolution kernel in the trained image recognition model to obtain the color signal value, shape signal value, structure signal value, local signal value, global signal value, etc. of each pixel, and then through the activation function, pooling layer and other operations, by enhancing the image expression and reducing the amount of data, the effect of extracting the five-level feature information of the image is achieved, and then after the feature matching is performed through the preset feature information comparison table to obtain the comparison association matrix, the attribute information of the target cell corresponding to the five-level association feature information is identified by calculating the similarity;

[0186] Embodiment 2

[0187] In addition, based on the same concept of the above-mentioned method embodiment, the embodiment of the present application also provides a turbidity evaluation and processing system for blood separation plasma samples, which is used to implement the above-mentioned method of the present application. Since the principle of solving the problem in this method embodiment is similar to that of the system, it at least has all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, which will not be repeated here one by one.

[0188] like Figure 8 As shown, the present invention proposes a turbidity evaluation and processing system for blood separation plasma samples, comprising a first image acquisition module 10, a second image acquisition module 20, a third image acquisition module 30, and an image recognition module 40;

[0189] Wherein, the first image acquisition module 10 is used to acquire a first image containing target cells;

[0190] The first image is an initial image, which is a plurality of initial images for identifying target cells in an image obtained by taking a camera or manually collecting;

[0191] The second image acquisition module 20 is used to perform recognition processing on the first image to obtain a second image in the first image;

[0192] The second image is a target cell region image;

[0193] The third image acquisition module 30 is used to identify and analyze the second image through an edge detection algorithm to obtain edge information of the target cell; and to obtain a third image according to the edge information;

[0194] The third image is the target cell image;

[0195] An image recognition module 40, configured to input the third image into a pre-trained image recognition model to determine the attribute information of the target cell;

[0196] The attribute information of the target cell includes name information of the target cell and type information of the target cell;

[0197] Preferably, the image recognition module 40 includes an initial recognition submodule 41, a calculation submodule 42, and a screening and acquisition submodule 43;

[0198] The initial recognition submodule 41 is used to perform initial recognition on the third image using the image recognition model, and obtain five-level correlation feature information of the current third image by initial recognition;

[0199] The five-level associated feature information includes color feature information, shape feature information, structure feature information, local feature information, and global feature information;

[0200] The calculation submodule 42 is used to calculate the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table;

[0201] The screening and obtaining submodule 43 is used to determine that the control feature information corresponding to the highest similarity is the target control feature information; and to determine that the control attribute information corresponding to the target feature information is the attribute information of the target cell.

[0202] Preferably, the initial recognition submodule 41 is specifically used to establish an initial image recognition model;

[0203] The third image is input into a trained image recognition model obtained by training the initial image recognition model, and then the five-level associated feature information corresponding to the third image is output.

[0204] Preferably, the initial recognition submodule 41, in a specific implementation, is used to obtain a training set, the training set including a plurality of training data and annotated data corresponding to each of the training data; the training data is obtained from training images by crawler technology; the annotated data is annotated training data to obtain attribute information corresponding to target cells in the training data;

[0205] Input the training data into a pre-established initial image recognition model, and output an image recognition vector;

[0206] Obtaining a loss value by calculating a loss function based on the image recognition vector and the labeled data;

[0207] Back-propagating the loss value to update the parameters of the initial image recognition model, completing the first iteration, and setting an iteration counter, increasing the number of iterations of the iteration counter by one; and repeating the above operation, recording the loss value and the number of iterations;

[0208] The iteration counter is initially set to 0 and is set to 1 when the first iteration is completed;

[0209] Preset a maximum number of iterations and a maximum loss value threshold; stop training when the loss value is less than or equal to the maximum loss value threshold or when the number of iterations reaches the maximum number of iterations, and obtain a trained image recognition model;

[0210] The third image is input into the trained image recognition model, and the five-level associated feature information corresponding to the third image is output.

[0211] Preferably, the initial recognition submodule 41, in a specific implementation, is further used to forward-propagate the third image through the trained image recognition model to output an image vector;

[0212] Decoding the image vector to obtain five-level associated feature information;

[0213] Preferably, the initial recognition submodule 41 is further used to obtain image and model parameter data during specific implementation; in the initial state, p convolution kernels of a×b size are set; based on the convolution kernel and the image and model parameter data, all pixels in the third image are convolved to obtain a signal value corresponding to each pixel;

[0214] The image and model parameter data include all pixels in the third image and the coordinates (x, y) corresponding to all pixels, the intensity values ​​t corresponding to the pixels i , the weight value corresponding to the trained image recognition model; wherein p is greater than or equal to 1;

[0215] The signal value corresponding to the pixel point is expressed as:

[0216]

[0217] Performing a nonlinear mapping operation on the signal value through an activation function to obtain an activation signal corresponding to the signal value;

[0218] Downsampling the activation signal through a pooling layer to obtain a target activation signal;

[0219] Converting the target activation signal through a fully connected layer to obtain an image vector;

[0220] Preferably, the calculation submodule 42 is specifically used to set header data for the five-level association feature information;

[0221] The header data includes color feature header data O, shape feature header data P, structure feature header data Q, local feature header data R, and global feature header data S;

[0222] Traversing all the comparison characteristic information in the characteristic information comparison table, and obtaining the comparison header data in the comparison characteristic information;

[0223] The comparison header data includes comparison color feature header data O', comparison shape feature header data P', comparison structure feature header data Q', comparison local feature header data R', and comparison global feature header data S';

[0224] Establishing an association relationship between the comparison header data and the header data to obtain a comparison association matrix D;

[0225] The control association matrix is ​​expressed as:

[0226]

[0227] Calculate and obtain the similarity of each factor in the comparison association matrix (and repeat the above operation until all the comparison feature information in the feature information comparison table is traversed to obtain the similarity between the five-level association feature information and all the comparison feature information in the preset feature information comparison table);

[0228] Embodiment 3

[0229] According to another aspect of the embodiments of the present application, a storage medium having a non-volatile program code executable by a processor is also provided. The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the turbidity evaluation and processing method of a blood separation plasma sample described in any of the above embodiments are implemented.

[0230] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. A person skilled in the art may modify the technical solutions described in the above embodiments, or replace part or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the turbidity of a blood plasma sample, characterized in that: The steps are as follows: acquiring a first image containing a target cell; Performing recognition processing on the first image to obtain a second image in the first image; The second image is a target cell region image; The third image is obtained by performing recognition and analysis on the second image through an edge detection algorithm; The third image is the target cell image; Inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell; The attribute information of the target cell includes name information of the target cell; According to the number of target cells identified per unit area, the turbidity rating information of the sample is determined and the turbidity rating information is output.

2. The method for evaluating the turbidity of a blood plasma sample according to claim 1, characterized in that: Inputting the third image into a pre-trained image recognition model to determine the attribute information of the target cell includes: Performing initial recognition on the third image using the image recognition model, and obtaining five-level associated feature information of the current third image by initial recognition; The five-level associated feature information includes color feature information, shape feature information, structure feature information, local feature information, and global feature information; Calculate and obtain the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table; The control attribute information of the control feature information corresponding to the highest similarity is determined to be the attribute information of the target cell.

3. The method for evaluating the turbidity of a blood plasma sample according to claim 2, characterized in that: The method of using the image recognition model to initially recognize the third image to obtain five-level associated feature information of the current third image includes the following steps: Establish an initial image recognition model; The third image is input into a trained image recognition model obtained by training the initial image recognition model, and then the five-level associated feature information corresponding to the third image is output.

4. The method for evaluating the turbidity of a blood plasma sample according to claim 3, characterized in that: The step of inputting the third image into a trained image recognition model obtained by training the initial image recognition model and then outputting the five-level associated feature information corresponding to the third image comprises the following steps: Acquire a training set, wherein the training set includes a plurality of training data and annotated data corresponding to each of the training data; the training data is a training image; the annotated data is annotated attribute information corresponding to a target cell in the training data; Input the training data into a pre-established initial image recognition model, and output an image recognition vector; Obtaining a loss value by calculating a loss function based on the image recognition vector and the labeled data; Back-propagating the loss value to update the parameters of the initial image recognition model, performing iterative training by presetting a maximum number of iterations and a maximum threshold of the loss value, to obtain a trained image recognition model; The third image is input into the trained image recognition model, and the five-level associated feature information corresponding to the third image is output.

5. The method for evaluating the turbidity of a blood separated plasma sample according to claim 4, characterized in that: The step of inputting the third image into the trained image recognition model and outputting five-level association feature information corresponding to the third image comprises the following steps: Forward propagation of the third image through the trained image recognition model to output an image vector; The image vector is decoded to obtain five-level associated feature information.

6. The method for evaluating the turbidity of a blood separated plasma sample according to claim 5, characterized in that: The forward propagation of the third image through the trained image recognition model to output an image vector comprises the following steps: Acquire image and model parameter data; in an initial state, set p convolution kernels of size a×b; based on the convolution kernel and the image and model parameter data, perform a convolution operation on all pixels in the third image to obtain a signal value corresponding to each pixel; The image and model parameter data include all pixels in the third image and the coordinates (x, y) corresponding to all pixels, the intensity values ​​t corresponding to the pixels i , the weight value corresponding to the trained image recognition model; wherein p is greater than or equal to 1; The signal value corresponding to the pixel point is expressed as: Performing a nonlinear mapping operation on the signal value through an activation function to obtain an activation signal corresponding to the signal value; Downsampling the activation signal through a pooling layer to obtain a target activation signal; The target activation signal is converted through a fully connected layer to obtain an image vector.

7. The method for evaluating the turbidity of a blood plasma sample according to claim 6, characterized in that: The step of calculating and obtaining the similarity between the five-level associated feature information and all the comparison feature information in the preset feature information comparison table comprises the following steps: Setting header data for the five-level associated characteristic information; Traversing all the comparison characteristic information in the characteristic information comparison table, and obtaining the comparison header data in the comparison characteristic information; Establishing an association relationship between the comparison header data and the header data to obtain a comparison association matrix D; The similarity of each factor in the comparison association matrix is ​​calculated and obtained.

8. The method for evaluating the turbidity of a blood plasma sample according to claim 7, characterized in that: The header data includes color feature header data O, shape feature header data P, structure feature header data Q, local feature header data R, and global feature header data S; The comparison header data includes comparison color feature header data O', comparison shape feature header data P', comparison structure feature header data Q', comparison local feature header data R', and comparison global feature header data S'; The control association matrix is ​​expressed as:

9. A turbidity evaluation and processing system for blood separation plasma samples, characterized in that: It includes a first image acquisition module, a second image acquisition module, a third image acquisition module, and an image recognition module; Wherein, the first image acquisition module is used to acquire a first image containing target cells; The second image acquisition module is used to perform recognition processing on the first image to obtain a second image in the first image; The second image is a target cell region image; The third image acquisition module is used to identify and analyze the second image through an edge detection algorithm and then capture the third image; The third image is the target cell image; The image recognition module is used to input the third image into a pre-trained image recognition model to determine the attribute information of the target cells; determine the turbidity rating information of the sample according to the number of target cells identified per unit area and output the turbidity rating information.

10. A storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the turbidity evaluation and processing method for blood separation plasma sample according to any one of claims 1 to 8.

Images