A method and system for cell identification in urine based on multiple models

Through the deep convolutional neural network technology combined with multiple models, different cell combinations in urine are identified, which solves the problem of poor recognition effect of single model, and improves the accuracy of urine cell recognition and data set production efficiency.

CN114495097BActive Publication Date: 2024-11-08陆建
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210108663.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-28
Publication Date
2024-11-08
Estimated Expiration
2042-01-28

AI Technical Summary

Technical Problem

The prior art has large differences in cell morphology in urine cell recognition, and the recognition effect of a single model is not ideal, resulting in low detection efficiency and imbalance in the number of cells affecting the production of data sets.

Method used

Using a multi-model-based urine cell recognition method, feature extraction, candidate regions and classification networks are constructed through deep convolutional neural networks, multiple image recognition models are trained for different cell combinations, and the images to be recognized are identified in combination with these models.

Benefits of technology

It improves the recognition accuracy of different cell combinations in urine, solves the problem of imbalance in cell count, and is conducive to the production of data sets and the improvement of identification efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114495097B_ABST
    Figure CN114495097B_ABST
Patent Text Reader

Abstract

The present invention proposes a method and system for cell recognition in urine based on multiple models, wherein the method includes: collecting, selecting and labeling urine images to form a data set; constructing an image recognition model based on a deep convolutional neural network; using the data set to train and test the image recognition model for different cell recognition combinations, and obtaining multiple image recognition models trained for different cell recognition combinations; and using the combination of the multiple image recognition models trained for different cell recognition combinations to recognize the image to be recognized. The solution proposed by the present invention uses different models to separately recognize different cell combinations in urine, which greatly improves the recognition accuracy; it solves the problem of unbalanced cell numbers and is conducive to the preparation of data sets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of urine routine image recognition, and in particular relates to a urine cell recognition method and system based on multiple models. Background Art

[0002] Identifying various cell morphologies in urine can help doctors diagnose diseases. In the early days, urine was mainly observed and identified manually under microscopes, which required a large workload, low detection efficiency, and high professional capabilities of personnel. In recent years, with the development of artificial intelligence technology, image recognition technology has also been applied to routine urine projects, which not only greatly reduced the workload but also greatly improved the detection efficiency. However, the cell morphologies present in urine are diverse and vary greatly. At present, most of the work is limited to using a single model to realize urine cell recognition. For cases where the cell morphology is very different, the recognition effect is often not very ideal. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a technical solution of a urine cell recognition method based on multiple models to solve the above technical problems.

[0004] The first aspect of the present invention discloses a method for identifying cells in urine based on multiple models; the method comprises:

[0005] Step S1, collecting, selecting and annotating urine images to form a data set;

[0006] Step S2: constructing an image recognition model based on a deep convolutional neural network;

[0007] Step S3, for different cell recognition combinations, using the data set to train and test the image recognition model, to obtain a plurality of trained image recognition models for different cell recognition combinations;

[0008] Step S4: using the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized.

[0009] According to the method of the first aspect of the present invention, in step S1, the specific method of labeling the urine image includes:

[0010] The images from the 10x microscope are annotated with small objects and epithelial cells, and the images from the 40x microscope are annotated with red blood cells and white blood cells.

[0011] According to the method of the first aspect of the present invention, in step S2, the specific structure of the image recognition model based on the deep convolutional neural network includes:

[0012] Feature extraction network, candidate region generation network and classification network;

[0013] The feature extraction network extracts features of the urine image;

[0014] The candidate region generation network processes the features of the extracted urine image and selects potential regions containing the identification object in the original image;

[0015] The classification network calculates the probability that the potential area contains the identification target through the features of the urine image.

[0016] According to the method of the first aspect of the present invention, in step S2, the specific structure of the image recognition model based on the deep convolutional neural network also includes: a position correction network;

[0017] The position correction network fine-tunes the potential region by the features of the urine image.

[0018] According to the method of the first aspect of the present invention, in step S3, the different cell recognition combinations specifically include:

[0019] Multiple recognition combinations of the small targets, epithelial cells, red blood cells and white blood cells.

[0020] According to the method of the first aspect of the present invention, in step S3, the multiple identification combinations specifically include:

[0021] Combination of identification of small targets and epithelial cells; combination of identification of red blood cells and white blood cells; identification of small targets alone; combination of identification of epithelial cells, red blood cells and white blood cells;

[0022] The multiple image recognition models specifically include: an image recognition model for the recognition combination of small targets and epithelial cells; an image recognition model for the recognition combination of red blood cells and white blood cells; an image recognition model for the individual recognition of small targets; and an image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0023] According to the method of the first aspect of the present invention, in step S4, the specific method of applying the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized includes:

[0024] The image to be identified is identified by applying a combination of the trained image recognition model for the recognition combination of small targets and epithelial cells, the image recognition model for the recognition combination of red blood cells and white blood cells, the image recognition model for the recognition of small targets alone, and the image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0025] The second aspect of the present invention discloses a urine cell identification system based on multiple models; the system comprises:

[0026] A first processing module is configured to collect, select and annotate urine images to form a data set;

[0027] A second processing module is configured to construct an image recognition model based on a deep convolutional neural network;

[0028] A third processing module is configured to apply the data set to train and test the image recognition model for different cell recognition combinations, and obtain a plurality of trained image recognition models for different cell recognition combinations;

[0029] The fourth processing module is configured to use a combination of the plurality of image recognition models trained for different cell recognition combinations to recognize the image to be recognized.

[0030] The third aspect of the present invention discloses an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the multi-model based urine cell identification methods in the first aspect of the present disclosure are implemented.

[0031] The fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the multi-model based urine cell identification methods in the first aspect of the present disclosure.

[0032] The solution proposed by the present invention can:

[0033] 1. Different models are used to identify different cell combinations in urine, which greatly improves the accuracy of identification.

[0034] 2. It solves the problem of unbalanced cell numbers and is conducive to the preparation of data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] 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.

[0036] Figure 1 A flowchart of a method for identifying cells in urine based on multiple models according to an embodiment of the present invention;

[0037] Figure 2 A structural diagram of an image recognition model based on a deep convolutional neural network according to an embodiment of the present invention;

[0038] Figure 3 A comparison diagram of model recognition results according to an embodiment of the present invention;

[0039] Figure 4 A structural diagram of a urine cell recognition system based on multiple models according to an embodiment of the present invention;

[0040] Figure 5 The figure is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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 making creative work are within the scope of protection of the present invention.

[0042] The first aspect of the present invention discloses a method for identifying cells in urine based on multiple models. Figure 1 FIG. 4 is a flow chart of a method for identifying cells in urine based on multiple models according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0043] Step S1, collecting, selecting and annotating urine images to form a data set;

[0044] Step S2: constructing an image recognition model based on a deep convolutional neural network;

[0045] Step S3, for different cell recognition combinations, using the data set to train and test the image recognition model, to obtain a plurality of trained image recognition models for different cell recognition combinations;

[0046] Step S4: using the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized.

[0047] In step S1, urine images are collected, selected and annotated to form a data set.

[0048] In some embodiments, in step S1, the specific method of labeling the urine image includes:

[0049] The images from the 10x microscope are annotated with small objects and epithelial cells, and the images from the 40x microscope are annotated with red blood cells and white blood cells.

[0050] Specifically, real human urine images were obtained through 10x and 40x microscopes in conjunction with CCD cameras and related automated equipment. Among them, there are 20,000 images with a 10x microscope and 40,000 images with a 40x microscope. After manual selection, 13,000 and 32,000 images that meet the conditions of clarity and containing cells were selected respectively. Then professionals annotated them, annotating small targets and epithelial cells in 10x microscope images, and red blood cells and white blood cells in 40x microscope images. According to statistics, the number of small targets is 150,000, the number of epithelial cells is 40,000; the number of red blood cells is 40,000, and the number of white blood cells is 40,000.

[0051] In step S2, an image recognition model based on a deep convolutional neural network is constructed.

[0052] In some embodiments, in step S2, Figure 2 As shown in the figure, the specific structure of the image recognition model based on deep convolutional neural network includes:

[0053] Feature extraction network, candidate region generation network and classification network;

[0054] The feature extraction network extracts features of the urine image;

[0055] The candidate region generation network processes the features of the extracted urine image and selects potential regions containing the identification object in the original image;

[0056] The classification network calculates the probability that the potential area contains the identification target through the features of the urine image.

[0057] The position correction network fine-tunes the potential region by the features of the urine image.

[0058] Specifically, the feature extraction network extracts features of urine images and is composed of five extraction units:

[0059] The first feature extraction unit consists of two convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 64 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0060] The second feature extraction unit consists of two convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 128 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0061] The third feature extraction unit consists of three convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 256 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0062] The fourth feature extraction unit consists of three convolutional layers and one maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 512 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0063] The fifth feature extraction unit consists of three convolutional layers, with a convolution kernel of 3×3 size, a depth of 512, and a stride of 1;

[0064] The candidate region generation network processes the features of the extracted urine image, selects potential regions containing the identified object in the original image, and reduces the amount of calculation in the subsequent recognition process. It consists of two convolutional layers, a softmax layer, and an SPP layer for adjusting the network size:

[0065] The convolution kernel of the first convolutional layer is 3×3 in size, 512 in depth, and 1 in stride;

[0066] The convolution kernel of the second convolution layer is 1×1 in size, 18 in depth, and 1 in stride;

[0067] The classification network calculates the probability that the potential area contains the identification target through the features of the urine image, and is composed of three fully connected layers and one softmax layer;

[0068] The position correction network fine-tunes the potential area through the features of the urine image to make the recognition range more accurate, and is composed of two fully connected layers.

[0069] In step S3, for different cell recognition combinations, the image recognition model is trained and tested using the data set to obtain a plurality of trained image recognition models for different cell recognition combinations.

[0070] In some embodiments, in step S3, the different cell recognition combinations specifically include:

[0071] Multiple recognition combinations of the small targets, epithelial cells, red blood cells and white blood cells;

[0072] The multiple identification combinations specifically include:

[0073] Combination of identification of small targets and epithelial cells; combination of identification of red blood cells and white blood cells; identification of small targets alone; combination of identification of epithelial cells, red blood cells and white blood cells;

[0074] The multiple image recognition models specifically include: an image recognition model for the recognition combination of small targets and epithelial cells; an image recognition model for the recognition combination of red blood cells and white blood cells; an image recognition model for the individual recognition of small targets; and an image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0075] Specifically, target recognition is performed on small targets, epithelial cells, red blood cells and white blood cells in human urine. Among them, small targets are a general term for several types of cells with smaller sizes observed under a low-power microscope. Their sizes are relatively small compared to epithelial cells, red blood cells and white blood cells, while the sizes of epithelial cells, red blood cells and white blood cells are similar. The following five groups of methods were experimented with the different and combined recognition targets of these four recognition objects as the recognition targets of the model:

[0076] A. Using the above model and dataset, small targets, epithelial cells, red blood cells and white blood cells are identified simultaneously. Parameters such as learning rate, loss function, termination error, and maximum number of iterations are set, and the dataset is divided into 80% for training and 20% for testing. The model is trained using the steepest descent method. The model is then evaluated using the test set.

[0077] B. Use the above model and dataset to identify small targets and epithelial cells; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0078] C. Use the above model and dataset to identify red blood cells and white blood cells simultaneously; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0079] D. Use the above model and dataset to identify small targets; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0080] E. Using the above model and dataset, epithelial cells, red blood cells, and white blood cells are identified simultaneously; parameters such as learning rate, loss function, termination error, and maximum number of iterations are set, and the dataset is divided into 80% for training and 20% for testing. The model is trained using the steepest descent method; and the model is then evaluated using the test set.

[0081] Apply the above five methods to generate five models, Model A, Model B, Model C, Model and Model E;

[0082] Model A: Target recognition of small targets, epithelial cells, red blood cells and white blood cells in human urine;

[0083] Model B: Image recognition model for the combination of small target and epithelial cell recognition;

[0084] Model C: Image recognition model for the combination of red blood cells and white blood cells;

[0085] Model D: Image recognition model for small objects.

[0086] Model E: Image recognition model for the identification of epithelial cells, red blood cells, and white blood cells.

[0087] The evaluation results of the five methods are as follows: Figure 3 As shown above, it can be seen that, by comparison, the overall effect of using a single model to recognize small targets, epithelial cells, red blood cells, and white blood cells is very poor; if model D is used to recognize small targets, and model E is used to recognize epithelial cells, red blood cells, and white blood cells, the two models can be combined to achieve the same purpose as model A, and the recognition effect will be greatly improved.

[0088] In step S4, the image to be identified is identified by using a combination of the trained multiple image recognition models for different cell recognition combinations.

[0089] In some embodiments, in step S4, the specific method of using the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized includes:

[0090] The image to be identified is identified by applying a combination of the trained image recognition model for the recognition combination of small targets and epithelial cells, the image recognition model for the recognition combination of red blood cells and white blood cells, the image recognition model for the recognition of small targets alone, and the image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0091] Specifically, any image to be identified is passed through model D to realize the identification of small targets, and model E is used to realize the identification of epithelial cells, red blood cells, and white blood cells. The identification results of models D and E are combined to realize the identification of small targets, epithelial cells, red blood cells, and white blood cells.

[0092] The model D is an image recognition model for individually identifying small targets; the model E is an image recognition model for identifying epithelial cells, red blood cells and white blood cells.

[0093] In summary, the solution proposed by the present invention can:

[0094] 1. Different models are used to identify different cell combinations in urine, which greatly improves the accuracy of identification.

[0095] 2. It solves the problem of unbalanced cell numbers and is conducive to the preparation of data sets.

[0096] The second aspect of the present invention discloses a urine cell recognition system based on multiple models. Figure 4 FIG. 4 is a structural diagram of a urine cell recognition system based on multiple models according to an embodiment of the present invention; Figure 4 As shown, the system 100 includes:

[0097] The first processing module 101 is configured to collect, select and annotate urine images to form a data set;

[0098] The second processing module 102 is configured to construct an image recognition model based on a deep convolutional neural network;

[0099] The third processing module 103 is configured to apply the data set to train and test the image recognition model for different cell recognition combinations, and obtain a plurality of trained image recognition models for different cell recognition combinations;

[0100] The fourth processing module 104 is configured to use the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized.

[0101] According to the system of the second aspect of the present invention, the first processing module 101 is specifically configured to annotate images of a 10x microscope as small objects and epithelial cells, and annotate images of a 40x microscope as red blood cells and white blood cells.

[0102] Specifically, real human urine images were obtained through 10x and 40x microscopes in conjunction with CCD cameras and related automated equipment. Among them, there are 20,000 images with a 10x microscope and 40,000 images with a 40x microscope. After manual selection, 13,000 and 32,000 images that meet the conditions of clarity and containing cells were selected respectively. Then professionals annotated them, annotating small targets and epithelial cells in 10x microscope images, and red blood cells and white blood cells in 40x microscope images. According to statistics, the number of small targets is 150,000, the number of epithelial cells is 40,000; the number of red blood cells is 40,000, and the number of white blood cells is 40,000.

[0103] According to the system of the second aspect of the present invention, the second processing module 102 is specifically configured as follows: Figure 2 As shown in the figure, the specific structure of the image recognition model based on deep convolutional neural network includes:

[0104] Feature extraction network, candidate region generation network and classification network;

[0105] The feature extraction network extracts features of the urine image;

[0106] The candidate region generation network processes the features of the extracted urine image and selects potential regions containing the identification object in the original image;

[0107] The classification network calculates the probability that the potential area contains the identification target through the features of the urine image.

[0108] The position correction network fine-tunes the potential region by the features of the urine image.

[0109] Specifically, the feature extraction network extracts features of urine images and is composed of five extraction units:

[0110] The first feature extraction unit consists of two convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 64 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0111] The second feature extraction unit consists of two convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 128 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0112] The third feature extraction unit consists of three convolutional layers and a maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 256 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0113] The fourth feature extraction unit consists of three convolutional layers and one maximum pooling layer. The convolution kernel of the convolution layer is 3×3 in size, 512 in depth, and 1 in stride. The filter size of the pooling layer is 2×2 with a stride of 2.

[0114] The fifth feature extraction unit consists of three convolutional layers, with a convolution kernel of 3×3 size, a depth of 512, and a stride of 1;

[0115] The candidate region generation network processes the features of the extracted urine image, selects potential regions containing the identified object in the original image, and reduces the amount of calculation in the subsequent recognition process. It consists of two convolutional layers, a softmax layer, and an SPP layer for adjusting the network size:

[0116] The convolution kernel of the first convolutional layer is 3×3 in size, 512 in depth, and 1 in stride;

[0117] The convolution kernel of the second convolution layer is 1×1 in size, 18 in depth, and 1 in stride;

[0118] The classification network calculates the probability that the potential area contains the identification target through the features of the urine image, and is composed of three fully connected layers and one softmax layer;

[0119] The position correction network fine-tunes the potential area through the features of the urine image to make the recognition range more accurate, and is composed of two fully connected layers.

[0120] According to the system of the second aspect of the present invention, the third processing module 103 is specifically configured as follows: the different cell recognition combinations specifically include:

[0121] Multiple recognition combinations of the small targets, epithelial cells, red blood cells and white blood cells;

[0122] The multiple identification combinations specifically include:

[0123] Combination of identification of small targets and epithelial cells; combination of identification of red blood cells and white blood cells; identification of small targets alone; combination of identification of epithelial cells, red blood cells and white blood cells;

[0124] The multiple image recognition models specifically include: an image recognition model for the recognition combination of small targets and epithelial cells; an image recognition model for the recognition combination of red blood cells and white blood cells; an image recognition model for the individual recognition of small targets; and an image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0125] Specifically, target recognition is performed on small targets, epithelial cells, red blood cells and white blood cells in human urine. Among them, small targets are a general term for several types of cells with smaller sizes observed under a low-power microscope. Their sizes are relatively small compared to epithelial cells, red blood cells and white blood cells, while the sizes of epithelial cells, red blood cells and white blood cells are similar. The following five groups of methods were experimented with the different and combined recognition targets of these four recognition objects as the recognition targets of the model:

[0126] A. Using the above model and dataset, small targets, epithelial cells, red blood cells and white blood cells are identified simultaneously. Parameters such as learning rate, loss function, termination error, and maximum number of iterations are set, and the dataset is divided into 80% for training and 20% for testing. The model is trained using the steepest descent method. The model is then evaluated using the test set.

[0127] B. Use the above model and dataset to identify small targets and epithelial cells; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0128] C. Use the above model and dataset to identify red blood cells and white blood cells simultaneously; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0129] D. Use the above model and dataset to identify small targets; set parameters such as learning rate, loss function, termination error, maximum number of iterations, etc., divide the dataset into 80% for training and 20% for testing, and train the model using the steepest descent method; then use the test set to evaluate the model.

[0130] E. Using the above model and dataset, epithelial cells, red blood cells, and white blood cells are identified simultaneously; parameters such as learning rate, loss function, termination error, and maximum number of iterations are set, and the dataset is divided into 80% for training and 20% for testing. The model is trained using the steepest descent method; and the model is then evaluated using the test set.

[0131] Apply the above five methods to generate five models, Model A, Model B, Model C, Model and Model E;

[0132] Model A: Target recognition of small targets, epithelial cells, red blood cells and white blood cells in human urine;

[0133] Model B: Image recognition model for the combination of small target and epithelial cell recognition;

[0134] Model C: Image recognition model for the combination of red blood cells and white blood cells;

[0135] Model D: Image recognition model for small objects.

[0136] Model E: Image recognition model for the identification of epithelial cells, red blood cells, and white blood cells.

[0137] According to the system of the second aspect of the present invention, the fourth processing module 104 is specifically configured to use the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized. The specific method includes:

[0138] The image to be identified is identified by applying a combination of the trained image recognition model for the recognition combination of small targets and epithelial cells, the image recognition model for the recognition combination of red blood cells and white blood cells, the image recognition model for the recognition of small targets alone, and the image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

[0139] Specifically, any image to be identified is passed through model D to realize the identification of small targets, and model E is used to realize the identification of epithelial cells, red blood cells, and white blood cells. The identification results of models D and E are combined to realize the identification of small targets, epithelial cells, red blood cells, and white blood cells.

[0140] The model D is an image recognition model for individually identifying small targets; the model E is an image recognition model for identifying epithelial cells, red blood cells and white blood cells.

[0141] The third aspect of the present invention discloses an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the multi-model urine cell identification methods disclosed in the first aspect of the present invention are implemented.

[0142] Figure 5 is a structural diagram of an electronic device according to an embodiment of the present invention, such as Figure 5 As shown, the electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, near field communication (NFC) or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the housing of the electronic device, or an external keyboard, touch pad or mouse, etc.

[0143] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the technical solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0144] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any one of the steps in the multi-model based urine cell identification method disclosed in the first aspect of the present invention are implemented.

[0145] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.

Claims

1. A method for identifying cells in urine based on multiple models, characterized in that: The method comprises: Step S1, collecting, selecting and annotating urine images to form a data set; Step S2: construct an image recognition model based on a deep convolutional neural network; the specific structure of the image recognition model based on a deep convolutional neural network includes: Feature extraction network, candidate region generation network and classification network; The feature extraction network extracts features of the urine image; The candidate region generation network processes the features of the extracted urine image and selects potential regions containing the identification object in the original image; The classification network calculates the probability that the potential area contains the identification target through the features of the urine image; The specific structure of the image recognition model based on the deep convolutional neural network also includes: a position correction network; The position correction network fine-tunes the potential region according to the features of the urine image; Step S3, for different cell recognition combinations, using the data set to train and test the image recognition model, to obtain a plurality of trained image recognition models for different cell recognition combinations; The different cell recognition combinations specifically include: Combination of identification of small targets and epithelial cells; combination of identification of red blood cells and white blood cells; identification of small targets alone; combination of identification of epithelial cells, red blood cells and white blood cells; The multiple image recognition models specifically include: an image recognition model for the recognition combination of small targets and epithelial cells; an image recognition model for the recognition combination of red blood cells and white blood cells; an image recognition model for the recognition of small targets alone; an image recognition model for the recognition of epithelial cells, red blood cells and white blood cells; Step S4, using the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized; including, The image to be identified is identified by applying a combination of the trained image recognition model for the recognition combination of small targets and epithelial cells, the image recognition model for the recognition combination of red blood cells and white blood cells, the image recognition model for the recognition of small targets alone, and the image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

2. A method for identifying cells in urine based on multiple models according to claim 1, characterized in that: In step S1, the specific method of labeling the urine image includes: The images from the 10x microscope are annotated with small objects and epithelial cells, and the images from the 40x microscope are annotated with red blood cells and white blood cells.

3. A cell identification system for urine based on multiple models, characterized in that: The system comprises: A first processing module is configured to collect, select and annotate urine images to form a data set; The second processing module is configured to construct an image recognition model based on a deep convolutional neural network; the specific structure of the image recognition model based on the deep convolutional neural network includes: Feature extraction network, candidate region generation network and classification network; The feature extraction network extracts features of the urine image; The candidate region generation network processes the features of the extracted urine image and selects potential regions containing the identification object in the original image; The classification network calculates the probability that the potential area contains the identification target through the features of the urine image; The specific structure of the image recognition model based on the deep convolutional neural network also includes: a position correction network; The position correction network fine-tunes the potential region according to the features of the urine image; A third processing module is configured to apply the data set to train and test the image recognition model for different cell recognition combinations, and obtain a plurality of trained image recognition models for different cell recognition combinations; The different cell recognition combinations specifically include: Combination of identification of small targets and epithelial cells; combination of identification of red blood cells and white blood cells; identification of small targets alone; combination of identification of epithelial cells, red blood cells and white blood cells; The multiple image recognition models specifically include: an image recognition model for the recognition combination of small targets and epithelial cells; an image recognition model for the recognition combination of red blood cells and white blood cells; an image recognition model for the recognition of small targets alone; an image recognition model for the recognition of epithelial cells, red blood cells and white blood cells; The fourth processing module is configured to apply the combination of the trained multiple image recognition models for different cell recognition combinations to recognize the image to be recognized, specifically including: The image to be identified is identified by applying a combination of the trained image recognition model for the recognition combination of small targets and epithelial cells, the image recognition model for the recognition combination of red blood cells and white blood cells, the image recognition model for the recognition of small targets alone, and the image recognition model for the recognition of epithelial cells, red blood cells and white blood cells.

4. An electronic device, characterized in that: The electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the multi-model based urine cell identification method described in any one of claims 1 to 2 are implemented.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the multi-model based urine cell identification method described in any one of claims 1 to 2 are implemented.