WO2020248848A1 - 智能化异常细胞判断方法、装置及计算机可读存储介质 - Google Patents

智能化异常细胞判断方法、装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2020248848A1
WO2020248848A1 PCT/CN2020/093546 CN2020093546W WO2020248848A1 WO 2020248848 A1 WO2020248848 A1 WO 2020248848A1 CN 2020093546 W CN2020093546 W CN 2020093546W WO 2020248848 A1 WO2020248848 A1 WO 2020248848A1
Authority
WO
WIPO (PCT)
Prior art keywords
cell
gray
cell set
value
abnormal
Prior art date
Application number
PCT/CN2020/093546
Other languages
English (en)
French (fr)
Inventor
郭冰雪
王季勇
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020248848A1 publication Critical patent/WO2020248848A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to an intelligent abnormal cell judgment method, device, and computer-readable storage medium.
  • Abnormal cells such as cancer cells
  • cancer cells are often the fuse for humans to produce major illnesses. According to surveys, there are 500,000 new cases and 274,000 deaths worldwide each year, of which 85% of the new cases are due to early identification of abnormal cells The reason for the low recognition rate.
  • cervical cancer is currently the only cancer that can be detected and cured early, so early recognition is crucial for the treatment of the disease.
  • the inventor found that the cell fluid inspection method is currently the most commonly used method for identifying abnormal cells.
  • This application provides an intelligent method and device for judging abnormal cells, and a computer-readable storage medium. Its main purpose is to simplify, quickly and accurately determine whether the cell picture or video contains abnormalities when the user inputs a cell picture or video. Cell and output the judgment result.
  • an intelligent method for judging abnormal cells includes:
  • the obtaining a cell set includes:
  • the mucosa and secretions of the cells after the staining process are photographed to obtain the cell set.
  • the noise reduction adopts the following adaptive image noise reduction filtering method:
  • (x, y) represents the coordinates of the image pixels in the cell set
  • f(x, y) is the output data of the cell set after noise reduction processing based on the adaptive image noise reduction filtering method
  • ⁇ (x,y) is noise
  • g(x,y) is the set of cells
  • Is the total variance of the noise of the cell set Is the average gray value of the pixel (x, y)
  • L represents the current pixel coordinates.
  • the image segmentation of the cell set completed by the preprocessing operation based on the multi-threshold segmentation model includes:
  • the inter-class variance set is calculated, and the inter-class variance with the largest value in the inter-class variance set is selected to reset the gray interval of the cell set to obtain more Gray scale cell set.
  • the inter-class variance set ⁇ T is:
  • T is the preset threshold interval
  • t 1 is not less than the value
  • t m is not greater than the value 255
  • the present application also provides an intelligent abnormal cell judgment device, which includes a memory and a processor, and the memory stores an intelligent abnormal cell judgment program that can run on the processor.
  • the intelligent abnormal cell judgment program is executed by the processor, the following steps are implemented:
  • the obtaining a cell set includes:
  • the mucosa and secretions of the cells after the staining process are photographed to obtain the cell set.
  • the noise reduction adopts the following adaptive image noise reduction filtering method:
  • (x, y) represents the coordinates of the image pixels in the cell set
  • f(x, y) is the output data of the cell set after noise reduction processing based on the adaptive image noise reduction filtering method
  • ⁇ (x,y) is noise
  • g(x,y) is the set of cells
  • Is the total variance of the noise of the cell set Is the average gray value of the pixel (x, y)
  • L represents the current pixel coordinates.
  • the gray-level division of the cell set completed by the preprocessing operation based on a multi-threshold segmentation model to obtain a multi-level gray-scale cell set includes:
  • the inter-class variance set is calculated, and the inter-class variance with the largest value in the inter-class variance set is selected to reset the gray interval of the cell set to obtain more Gray scale cell set.
  • the inter-class variance set ⁇ T is:
  • T is the preset threshold interval
  • t 1 is not less than the value
  • t m is not greater than the value 255
  • this application also provides a computer-readable storage medium that stores an intelligent abnormal cell judgment program, and the intelligent abnormal cell judgment program can be used by one or more The processor executes to implement the steps of the intelligent abnormal cell judgment method as described above.
  • the noise reduction process can reduce the impact of noise on cell images, and the nine-channel cell set can be calculated through the Hessian matrix, which can further improve the The feature extraction is to maximize the use of existing cell features.
  • the abnormal cell judgment model described in this application has excellent feature analysis capabilities, and can efficiently and accurately analyze whether the picture contains abnormal cells. Therefore, this application can achieve accurate Intelligent abnormal cell judgment function.
  • FIG. 1 is a schematic flowchart of an intelligent abnormal cell judgment method provided by an embodiment of the application
  • FIG. 2 is a schematic diagram of the internal structure of an intelligent abnormal cell judgment device provided by an embodiment of the application.
  • FIG. 3 is a schematic diagram of modules of an intelligent abnormal cell judgment program in an intelligent abnormal cell judgment device provided by an embodiment of the application.
  • This application provides an intelligent method for judging abnormal cells.
  • FIG. 1 it is a schematic flowchart of an intelligent abnormal cell judgment method provided by an embodiment of this application.
  • the method can be executed by a device, and the device can be implemented by software and/or hardware.
  • the intelligent method for judging abnormal cells includes:
  • the operation of obtaining the cell set includes: obtaining the mucosa and secretions of the cells, staining the mucosa and secretions, and photographing the mucosa and secretions of the cells after the staining process to obtain the cell set, And respectively mark whether each picture in the cell set contains abnormal cells to obtain the label set.
  • this application is based on the scraper rotating around the cell site to be detected to obtain the mucosa and secretions of the cells to be detected, and smear the mucosa and secretions on the plexiglass and perform a staining process, and the display
  • the micro device photographs the mucous membrane and secretion cells in the organic glass, and finally obtains the cell set.
  • the dyeing treatment first fixes the plexiglass coated with mucous membranes and secretions in an alcohol liquid, and then places a hematoxylin staining agent into the alcohol liquid, and finally achieves the reduction of the mucous membranes and secretions.
  • the purpose of staining the cells is to ensure that the cells are staind.
  • the noise reduction adopts the following adaptive image noise reduction filtering method:
  • (x, y) represents the coordinates of the image pixels in the cell set
  • f(x, y) is the output data of the cell set after noise reduction processing based on the adaptive image noise reduction filtering method
  • ⁇ (x,y) is noise
  • g(x,y) is the set of cells
  • Is the total variance of the noise of the cell set Is the average gray value of the pixel (x, y)
  • L represents the current pixel coordinates.
  • the contrast enhancement is to increase the difference between the maximum value and the minimum value of the brightness in the cell set picture, because cells with low contrast will affect the subsequent judgment of abnormal cells.
  • the contrast enhancement adopts the following method:
  • a is the linear slope and b is the intercept on the Y axis. If a>1, the output image contrast is enhanced compared to the original image. If a ⁇ 1, the output image contrast is The contrast of the original image is reduced, where D a represents the gray value of the cell set, and D b represents the gray value of the output cell set.
  • the preferred embodiment of the present application traverses the gray values of the image pixels in the cell set completed by the preprocessing operation, counts the number of times each gray value appears, and calculates each gray based on the total number of pixels in the cell set.
  • the occurrence probability of the degree value is calculated based on the preset threshold interval and the occurrence probability of each gray value to obtain the inter-class variance set, and the inter-class variance set with the largest value in the inter-class variance set is selected to reset the cell set Gray-scale interval, obtain multi-level gray-scale cell set.
  • the calculation method is:
  • t i t i+1 belongs to t 1 , t 2 ,...t m , and t 1 , t 2 ,...t m respectively represent preset thresholds, and t 1 , t 2 until t m are in increasing form, t 1 Not less than the value 0, t m is not greater than the value 255, further:
  • n i is the number of occurrences of each gray value, i is in the range of 0 to 255, and N is the total number of occurrences of gray values.
  • the between-class variance set ⁇ T is
  • T is the preset threshold interval
  • t 1 is not less than the value
  • t m is not greater than the value 255
  • the resetting method for resetting the gray-scale interval of the cell set is:
  • t 1 ⁇ Tmax is the maximum variance between clusters
  • I(x,y) is the gray value of the cell set
  • (x,y) is the coordinate of each pixel in the cell set.
  • the method calculates the new gray value, and calculates other pixels in turn, until the final multi-level gray cell set is obtained.
  • the Hessian matrix is a matrix constructed by high-order differentiation of an image and capable of reflecting image characteristics.
  • This application preferably first calculates the scale space function I ab of the multi-level gray-scale cell set, obtains the Hessian matrix based on the inverse derivation of the scale space function I ab , and uses the Hessian matrix to solve the multi Two feature value maps corresponding to each channel in the RGB gray-scale cell set, and the feature maps are added to the original channel of the multi-level gray-scale cell set to obtain a nine-channel cell set.
  • the scale space function I ab is:
  • ab has the same meaning as the parameter in the contrast enhancement
  • a is the linear slope
  • b is the intercept on the Y axis
  • ⁇ is the scale space function parameter
  • I(x,y) is the gray scale of each channel of the cell set.
  • this application includes three channels of R, G, and B, and G(X, Y; ⁇ ) is a Gaussian function.
  • the Gaussian function is:
  • Hessian matrix H is obtained by the reverse derivation process of the scale space function I ab :
  • x n and y n represent the coordinates of different pixels in the multi-level gray-scale cell set, and the eigenvalue ⁇ of the corresponding determinant is solved based on the Hessian matrix H:
  • the eigenvalue ⁇ is generally set in three dimensions, namely ⁇ 1 , ⁇ 2 , and ⁇ 3 .
  • the Hessian matrix H is solved for the three color channels to obtain H R , H G , H B , and then the The eigenvalues of the three color channels obtain ⁇ R1 , ⁇ R2 , ⁇ R3 , ⁇ G1 , ⁇ G2 , ⁇ G3 , ⁇ B1 , ⁇ B2 , and ⁇ B3 to form a nine-channel matrix. Therefore, the nine-channel matrix is solved for each pixel of the multi-level gray-scale cell set based on the above method, and a nine-channel cell set ⁇ RGB is finally obtained.
  • the abnormal cell judgment model includes a feature extraction layer and an abnormal cell recognition layer, and is based on a convolutional neural network.
  • the feature extraction layer receives the nine-channel cell set, and performs feature extraction based on a convolution operation and a pooling operation, and the convolution operation is:
  • ⁇ ' is the output value after the convolution operation, generally in the form of a multi-dimensional matrix
  • ⁇ RGB is the nine-channel cell set
  • k is the size of the convolution kernel, usually 2*2 dimensions
  • s is the stride of the convolution operation, which can take the value 1
  • p is the data zero-filling matrix.
  • the output value after the convolution operation is subjected to the pooling operation, and the pooling operation searches for the value with the largest matrix value among the output values of the convolution operation and forms a feature set ⁇ .
  • the present application inputs the feature set and the label set to the abnormal cell identification layer, and the abnormal cell identification layer performs the convolution operation on the feature set and then inputs it to the activation function to obtain a judgment value set , Inputting the judgment value set and the label set into a loss value calculated based on a loss function, and if the loss value is less than a preset threshold, the abnormal cell judgment model exits training.
  • the activation function is:
  • n is the size of the label set
  • y t is the judgment value set
  • ⁇ t is the label set.
  • S5. Receive the user's cell picture and input it into the abnormal cell judgment model to judge whether the cell includes abnormal cells and output the judgment result.
  • the abnormal cell judgment model recognizes that the cell set A contains abnormal cells and outputs the judgment result.
  • the output mode includes screen printing or voice broadcast And so on.
  • the invention also provides an intelligent device for judging abnormal cells.
  • FIG. 2 it is a schematic diagram of the internal structure of an intelligent abnormal cell judgment device provided by an embodiment of this application. (Corresponding modification)
  • the intelligent abnormal cell judging device 1 may be a PC (Personal Computer, personal computer), or a terminal device such as a smart phone, a tablet computer, or a portable computer, or a server.
  • the intelligent abnormal cell judgment device 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.
  • the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 11 may be an internal storage unit of the intelligent abnormal cell judgment device 1 in some embodiments, for example, the hard disk of the intelligent abnormal cell judgment device 1.
  • the memory 11 may also be an external storage device of the intelligent abnormal cell judging device 1, such as a plug-in hard disk or a smart media card (SMC) equipped on the intelligent abnormal cell judging device 1. Secure Digital (SD) card, flash card (Flash Card), etc.
  • SD Secure Digital
  • flash card Flash Card
  • the memory 11 may also include both an internal storage unit of the intelligent abnormal cell judgment device 1 and an external storage device.
  • the memory 11 can be used not only to store application software and various data installed in the intelligent abnormal cell judgment device 1, such as the code of the intelligent abnormal cell judgment program 01, etc., but also to temporarily store data that has been output or will be output. .
  • the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip, and is used to run the program code or processing stored in the memory 11 Data, such as execution of intelligent abnormal cell judgment program 01, etc.
  • CPU central processing unit
  • controller microcontroller
  • microprocessor or other data processing chip
  • the communication bus 13 is used to realize the connection and communication between these components.
  • the network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.
  • the device 1 may also include a user interface.
  • the user interface may include a display (Display) and an input unit such as a keyboard (Keyboard).
  • the optional user interface may also include a standard wired interface and a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, etc.
  • the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the intelligent abnormal cell judgment device 1 and to display a visualized user interface.
  • Figure 2 only shows the intelligent abnormal cell judging device 1 with components 11-14 and the intelligent abnormal cell judging program 01. Those skilled in the art will understand that the structure shown in Figure 1 does not constitute an intelligent abnormal
  • the definition of the cell judgment device 1 may include fewer or more components than shown, or a combination of certain components, or different component arrangements.
  • the memory 11 stores the intelligent abnormal cell judgment program 01; when the processor 12 executes the intelligent abnormal cell judgment program 01 stored in the memory 11, the following steps are implemented:
  • Step 1 Obtain a cell set and a label set, and perform preprocessing operations including noise reduction and contrast enhancement on the cell set.
  • the operation of obtaining the cell set includes: obtaining the mucosa and secretions of the cells, staining the mucosa and secretions, and photographing the mucosa and secretions of the cells after the staining process to obtain the cell set, And respectively mark whether each picture in the cell set contains abnormal cells to obtain the label set.
  • this application is based on the scraper rotating one week at the cell site to be detected to obtain the mucosa and secretions of the cells to be detected, and smear the mucosa and secretions on the organic glass and perform staining treatment, and at the same time, the display
  • the micro device photographs the mucous membrane and secretion cells in the organic glass, and finally obtains the cell set.
  • the dyeing treatment first fixes the plexiglass coated with mucous membranes and secretions in an alcohol liquid, and then places a hematoxylin staining agent into the alcohol liquid, and finally achieves the reduction of the mucous membranes and secretions.
  • the purpose of staining the cells is to ensure that the cells are staind.
  • the noise reduction adopts the following adaptive image noise reduction filtering method:
  • (x, y) represents the coordinates of the image pixels in the cell set
  • f(x, y) is the output data of the cell set after noise reduction processing based on the adaptive image noise reduction filtering method
  • ⁇ (x,y) is noise
  • g(x,y) is the set of cells
  • Is the total variance of the noise of the cell set Is the average gray value of the pixel (x, y)
  • L represents the current pixel coordinates.
  • the contrast enhancement is to increase the difference between the maximum value and the minimum value of the brightness in the cell set picture, because cells with low contrast will affect the subsequent judgment of abnormal cells.
  • the contrast enhancement adopts the following method:
  • a is the linear slope and b is the intercept on the Y axis. If a>1, the output image contrast is enhanced compared to the original image. If a ⁇ 1, the output image contrast is The contrast of the original image is reduced, where D a represents the gray value of the cell set, and D b represents the gray value of the output cell set.
  • Step 2 Perform gray-scale division on the cell set completed by the preprocessing operation based on a multi-threshold segmentation model to obtain a multi-level gray-scale cell set.
  • the preferred embodiment of the present application traverses the gray values of the image pixels in the cell set completed by the preprocessing operation, counts the number of times each gray value appears, and calculates each gray based on the total number of pixels in the cell set.
  • the occurrence probability of the degree value is calculated based on the preset threshold interval and the occurrence probability of each gray value to obtain the inter-class variance set, and the inter-class variance set with the largest value in the inter-class variance set is selected to reset the cell set Gray-scale interval, obtain multi-level gray-scale cell set.
  • the calculation method is:
  • t i t i+1 belongs to t 1 , t 2 ,...t m , and t 1 , t 2 ,...t m respectively represent preset thresholds, and t 1 , t 2 until t m are in increasing form, t 1 Not less than the value 0, t m is not greater than the value 255, further:
  • b i is the number of occurrences of each gray value, i is in the range of 0-255, and N is the total number of occurrences of gray value.
  • the between-class variance set ⁇ T is
  • T is the preset threshold interval
  • t 1 is not less than the value
  • t m is not greater than the value 255
  • the resetting method for resetting the gray-scale interval of the cell set is:
  • t 1 ⁇ Tmax is the maximum variance between clusters
  • I(x,y) is the gray value of the cell set
  • (x,y) is the coordinate of each pixel in the cell set.
  • the method calculates the new gray value, and calculates other pixels in turn, until the final multi-level gray cell set is obtained.
  • Step 3 Calculate the Hessian matrix based on the multi-level gray-scale cell set and obtain a nine-channel cell set.
  • the Hessian matrix is a matrix constructed by high-order differentiation of an image and capable of reflecting image characteristics.
  • This application preferably first calculates the scale space function I ab of the multi-level gray-scale cell set, obtains the Hessian matrix based on the inverse derivation of the scale space function I ab , and uses the Hessian matrix to solve the multi Two feature value maps corresponding to each channel in the RGB gray-scale cell set, and the feature maps are added to the original channel of the multi-level gray-scale cell set to obtain a nine-channel cell set.
  • the scale space function I ab is:
  • ab has the same meaning as the parameter in the contrast enhancement
  • a is the linear slope
  • b is the intercept on the Y axis
  • ⁇ is the scale space function parameter
  • I(x,y) is the gray scale of each channel of the cell set.
  • this application includes three channels of R, G, and B, and G(X, Y; ⁇ ) is a Gaussian function.
  • the Gaussian function is:
  • Hessian matrix H is obtained by the reverse derivation process of the scale space function I ab :
  • x n and y n represent the coordinates of different pixels in the multi-level gray-scale cell set, and the eigenvalue ⁇ of the corresponding determinant is solved based on the Hessian matrix H:
  • the eigenvalue ⁇ is generally set in three dimensions, namely ⁇ 1 , ⁇ 2 , and ⁇ 3 .
  • the Hessian matrix H is solved for the three color channels to obtain H R , H G , H B , and then the The eigenvalues of the three color channels obtain ⁇ R1 , ⁇ R2 , ⁇ R3 , ⁇ G1 , ⁇ G2 , ⁇ G3 , ⁇ B1 , ⁇ B2 , and ⁇ B3 to form a nine-channel matrix. Therefore, the nine-channel matrix is solved for each pixel of the multi-level gray-scale cell set based on the above method, and a nine-channel cell set ⁇ RGB is finally obtained.
  • Step 4 Input the nine-channel cell set and label set into the abnormal cell judgment model for training, until the abnormal cell judgment model meets the preset training exit condition and then exits the training.
  • the abnormal cell judgment model includes a feature extraction layer and an abnormal cell recognition layer, and is based on a convolutional neural network.
  • the feature extraction layer receives the nine-channel cell set, and performs feature extraction based on a convolution operation and a pooling operation, and the convolution operation is:
  • ⁇ ' is the output value after the convolution operation, generally in the form of a multi-dimensional matrix
  • ⁇ RGB is the nine-channel cell set
  • k is the size of the convolution kernel, usually 2*2 dimensions
  • s is the stride of the convolution operation, which can take the value 1
  • p is the data zero-filling matrix.
  • the output value after the convolution operation is subjected to the pooling operation, and the pooling operation searches for the value with the largest matrix value among the output values of the convolution operation and forms a feature set ⁇ .
  • the present application inputs the feature set and the label set to the abnormal cell identification layer, and the abnormal cell identification layer performs the convolution operation on the feature set and then inputs it to the activation function to obtain a judgment value set , Inputting the judgment value set and the label set into a loss value calculated based on a loss function, and if the loss value is less than a preset threshold, the abnormal cell judgment model exits training.
  • the activation function is:
  • n is the size of the label set
  • y t is the judgment value set
  • ⁇ t is the label set.
  • Step 5 Receive the user's cell picture and input it into the abnormal cell judgment model to judge whether the cell includes abnormal cells and output the judgment result.
  • the abnormal cell judgment model recognizes that the cell set A contains abnormal cells and outputs the judgment result.
  • the output mode includes screen printing or voice broadcast And so on.
  • the intelligent abnormal cell judgment program can also be divided into one or more modules, and the one or more modules are stored in the memory 11 and run by one or more processors (this embodiment For example, it is executed by the processor 12) to complete this application.
  • the module referred to in this application refers to a series of computer program instruction segments that can complete specific functions, used to describe the intelligent abnormal cell judgment program in the intelligent abnormal cell judgment device The implementation process.
  • the intelligent abnormal cell determination program can be divided into The data receiving module 10, the data processing module 20, the model training module 30, and the intelligent abnormal cell judgment output module 40 are exemplary:
  • the data receiving module 10 is configured to obtain a cell set and a label set, and perform preprocessing operations including noise reduction and contrast enhancement on the cell set.
  • the data processing module 20 is configured to: perform gray-scale division on the cell set completed by the preprocessing operation based on a multi-threshold segmentation model to obtain a multi-level gray cell set, and calculate the Hessian matrix based on the multi-level gray cell set And get the nine-channel cell set.
  • the model training module 30 is configured to input the nine-channel cell set and the label set into an abnormal cell judgment model for training, until the abnormal cell judgment model meets a preset training exit condition and then exit training.
  • the intelligent abnormal cell judgment output module 40 is configured to receive a user's cell picture and input it into the abnormal cell judgment model to determine whether the cell includes an abnormal cell and output the judgment result.
  • the above-mentioned data receiving module 10, data processing module 20, model training module 30, intelligent abnormal cell judgment output module 40 and other program modules implement functions or operation steps that are substantially the same as those in the above-mentioned embodiment, and will not be repeated here.
  • the embodiment of the present application also proposes a computer-readable storage medium.
  • the computer-readable storage medium may be non-volatile or volatile.
  • the computer-readable storage medium stores an intelligent abnormal cell judgment program, and the intelligent abnormal cell judgment program can be executed by one or more processors to achieve the following operations:

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种智能化异常细胞判断方法、一种智能化异常细胞判断装置以及一种计算机可读存储介质，该方法包括：获取细胞集及标签集，对所述细胞集进行预处理操作（S1），基于多阈值分割模型对所述预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集（S2），基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集（S3），将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练（S4），接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果（S5），完成智能化异常细胞判断。该方法可以实现精准的智能化异常细胞判断功能。

Description

智能化异常细胞判断方法、装置及计算机可读存储介质
本申请要求于2019年06月14日提交中国专利局、申请号为201910520871.9、发明名称为“智能化异常细胞判断方法、装置及计算机可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种智能化异常细胞判断方法、装置及计算机可读存储介质。
背景技术
异常细胞，如癌细胞，往往是人类产生重大病情的导火索，据调查显示，全世界每年有50万新发病例和27.4万死亡病例，其中85％的新发病例是由于前期识别异常细胞的识别率低下的原因。特别地，宫颈癌是目前唯一可以早发现并治愈的癌症，因此早期识别对于病情的治疗相当关键。发明人发现细胞液检查方法是目前最常用的异常细胞识别方法，但在中国由于缺乏病理医生和细胞检测设备，对异常的识别率很低；另外有各种人工辅助识别的设备系统，但多数辅助识别系统一般基于传统方法，所述传统方法依靠细胞质或者细胞核的精确分割和传统图像处理算法进行特征提取与选择，因此识别的前期处理繁琐，且识别率并不高。
发明内容
本申请提供一种智能化异常细胞判断方法、装置及计算机可读存储介质，其主要目的在于当用户输入细胞图片或视频时，可精简快速并准确的判断所述细胞图片或视频中是否包括异常细胞并输出判断结果。
为实现上述目的，本申请提供的一种智能化异常细胞判断方法，包括：
获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集；
基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
可选地，所述获取细胞集包括：
获取细胞的粘膜及分泌物；
对所述粘膜及分泌物进行染色处理；
拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集。
可选地，所述降噪采用如下自适应图像降噪滤波法：
g(x,y)＝η(x,y)+f(x,y)
Figure PCTCN2020093546-appb-000001
其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
Figure PCTCN2020093546-appb-000002
为所述细胞集的噪声总方差，
Figure PCTCN2020093546-appb-000003
为所述(x,y)的像素灰度均值，
Figure PCTCN2020093546-appb-000004
为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
可选地，所述基于多阈值分割模型对所述预处理操作完成的细胞集进行图像分割，包括：
遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于总像素数量计算每个灰度值的出现概率；
基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
可选地，所述类间方差集σ T为：
Figure PCTCN2020093546-appb-000005
其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
Figure PCTCN2020093546-appb-000006
表示不同预设阈值区间内所述每个灰度值的出现概率，
Figure PCTCN2020093546-appb-000007
表示不同预设阈值区间的灰度均值。
此外，为实现上述目的，本申请还提供一种智能化异常细胞判断装置，该装置包括存储器和处理器，所述存储器中存储有可在所述处理器上运行的智能化异常细胞判断程序，所述智能化异常细胞判断程序被所述处理器执行时实现如下步骤：
获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集；
基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
可选地，所述获取细胞集包括：
获取细胞的粘膜及分泌物；
对所述粘膜及分泌物进行染色处理；
拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集。
可选地，所述降噪采用如下自适应图像降噪滤波法：
g(x,y)＝η(x,y)+f(x,y)
Figure PCTCN2020093546-appb-000008
其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
Figure PCTCN2020093546-appb-000009
为所述细胞集的噪声总方差，
Figure PCTCN2020093546-appb-000010
为所述(x,y)的像素灰度均值，
Figure PCTCN2020093546-appb-000011
为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
可选地，所述基于多阈值分割模型对所述预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集，包括：
遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于总像素数量计算每个灰度值的出现概率；
基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
可选地，所述类间方差集σ T为：
Figure PCTCN2020093546-appb-000012
其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
Figure PCTCN2020093546-appb-000013
表示不同预设阈值 区间内所述每个灰度值的出现概率，
Figure PCTCN2020093546-appb-000014
表示不同预设阈值区间的灰度均值。
此外，为实现上述目的，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有智能化异常细胞判断程序，所述智能化异常细胞判断程序可被一个或者多个处理器执行，以实现如上所述的智能化异常细胞判断方法的步骤。
本申请提出的智能化异常细胞判断方法、装置及计算机可读存储介质中，所述降噪处理可减少噪声对细胞图像的影响，通过海森矩阵计算得到九通道细胞集，可进一步提高对细胞的特征提取，最大化利用已有的细胞特征，同时本申请所述的异常细胞判断模型具有优异的特征分析能力，可高效准确的分析出图片中是否包含异常细胞，因此本申请可以实现精准的智能化异常细胞判断功能。
附图说明
图1为本申请一实施例提供的智能化异常细胞判断方法的流程示意图；
图2为本申请一实施例提供的智能化异常细胞判断装置的内部结构示意图；
图3为本申请一实施例提供的智能化异常细胞判断装置中智能化异常细胞判断程序的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供一种智能化异常细胞判断方法。参照图1所示，为本申请一实施例提供的智能化异常细胞判断方法的流程示意图。该方法可以由一个装置执行，该装置可以由软件和/或硬件实现。
在本实施例中，所述智能化异常细胞判断方法包括：
S1、获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作。
其中，所述获取细胞集的操作包括：获取细胞的粘膜及分泌物，对所述粘膜及分泌物进行染色处理，拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集，并分别标注所述细胞集中每张图片中是否包含有异常细胞，得到所述标签集。
较佳地，本申请基于刮片在待检测细胞部位旋转一周获取所述待检测细胞的粘膜及分泌物，并将所述粘膜及分泌物涂抹至有机玻璃上并进行染色处理，同时调取显微设备拍摄所述有机玻璃内的粘膜及分泌物细胞，并最终得到所述细胞集。
优选地，所述染色处理先将所述涂抹有粘膜及分泌物的有机玻璃固定于酒精液体中，然后放置苏木素染色剂至所述酒精液体中，并最终达到将所述粘膜及分泌物内的细胞进行染色的目的。
本申请较佳实施例，所述降噪采用如下自适应图像降噪滤波法：
g(x,y)＝η(x,y)+f(x,y)
Figure PCTCN2020093546-appb-000015
其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
Figure PCTCN2020093546-appb-000016
为所述细胞集的噪声总方差，
Figure PCTCN2020093546-appb-000017
为所述(x,y)的像素灰度均值，
Figure PCTCN2020093546-appb-000018
为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
进一步地，所述对比度增强的目的是增大所述细胞集图片中亮度最大值与最小值之间的差值，因为对比度低的细胞会影响后续异常细胞的判断。较佳地，所述对比度增强采用如下方法：
D b＝f(D a)＝a*D a+b
其中a为线性斜率，b为在Y轴上的截距，若a>1，此时输出的图像对比度相比原图像的对比度是增强的，若a<1，此时输出的图像对比度相比原图像的对比度是减小的，其中D a代表所述细胞集的灰度值，D b代表输出的细胞集灰度值。
S2、基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集。
优选地，本申请较佳实施例遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于所述细胞集总像素数量计算每个灰度值的出现概率，基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
进一步地，所述每个灰度值的出现概率
Figure PCTCN2020093546-appb-000019
的计算方法为：
Figure PCTCN2020093546-appb-000020
其中，t it i+1属于t 1,t 2,…t m，而t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，进一步地：
Figure PCTCN2020093546-appb-000021
其中，n i为所述每个灰度值的出现次数，i在0～255范围内，N为灰度值的总出现次数。
较佳地，所述类间方差集σ T为
Figure PCTCN2020093546-appb-000022
其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
Figure PCTCN2020093546-appb-000023
表示不同预设阈值区间内所述每个灰度值的出现概率，
Figure PCTCN2020093546-appb-000024
表示不同预设阈值区间的灰度均值。
进一步地，所述重置所述细胞集的灰度区间的重置方法为：
Figure PCTCN2020093546-appb-000025
其中，t 1σ Tmax为所述最大的类间方差，I(x,y)为所述细胞集的灰度值，(x,y)为所述细胞集中各像素坐标。例如当像素坐标为(2,3)的灰度值I(2,3)＝4时，判断4的灰度值在区间t 1<I(2,3)≤t 2，则按照上述重置方法计算出新的灰度值，依次计算其他像素，直至最终得到多级灰度细胞集。
S3、基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集。
所述海森矩阵是由图像的高阶微分构造而成并能够反映图像特征的矩阵。本申请较佳地先计算所述多级灰度细胞集的尺度空间函数I ab，基于所述尺度空间函数I ab反向求导得到所述海森矩阵，利用海森矩阵求解出所述多级灰度细胞集RGB中每个通道对应的两种特征值图，并将所述特征图增加到所述多级灰度细胞集的原通道中，得到九通道细胞集。
优选地，所述尺度空间函数I ab为：
Figure PCTCN2020093546-appb-000026
其中，ab与所述对比度增强中参数含义相同，a为线性斜率，b为在Y轴上的截距，σ为尺度空间函数参数，I(x,y)为所述细胞集各通道的灰度值，本申请包括R、G、B三个通道，G(X,Y；σ)为高斯函数。
较佳地，所述高斯函数为：
Figure PCTCN2020093546-appb-000027
其中，e为无限不循环小数。进一步地，所述尺度空间函数I ab反向求导过程得到所述 海森矩阵H：
Figure PCTCN2020093546-appb-000028
其中，x n,y n表示所述多级灰度细胞集内不同像素的坐标，基于所述海森矩阵H求解对应行列式的特征值λ：
λH＝0
较佳地，所述特征值λ一般设置为3维，即λ 1，λ 2，λ 3。
进一步地，因为所述多级灰度细胞集共有R、G、B三个颜色通道，分别对所述三个颜色通道求解海森矩阵H得到H R、H G、H B，继而求解所述三个颜色通道的特征值得到λ R1，λ R2，λ R3、λ G1，λ G2，λ G3、λ B1，λ B2，λ B3共九个值组成九通道矩阵。因此基于上述方法对所述多级灰度细胞集的每个像素求解所述九通道矩阵并最终得到九通道细胞集λ RGB。
S4、将所述九通道细胞集及标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练。
较佳地，所述异常细胞判断模型包括特征提取层和异常细胞识别层，并以卷积神经网络为基础。其中，所述特征提取层接收所述九通道细胞集，并基于卷积操作和池化操作进行特征提取，所述卷积操作为：
Figure PCTCN2020093546-appb-000029
其中ω’为所述卷积操作后的输出值，一般为多维矩阵形式，λ RGB为所述九通道细胞集，k为卷积核的大小，通常为2*2维度，每个维度值为1的矩阵，s为所述卷积操作的步幅，可取值为1，p为数据补零矩阵。
本申请较佳实施例将所述卷积操作后的输出值进行所述池化操作，所述池化操作寻找所述卷积操作的输出值中矩阵数值最大的值并组成特征集ω。
进一步地，本申请将所述特征集及所述标签集输入至所述异常细胞识别层，所述异常细胞识别层对所述特征集进行所述卷积操作后输入至激活函数得到判断值集，将所述判断值集和所述标签集输入基于损失函数中计算得到损失值，若所述损失值小于预设阈值时，所述异常细胞判断模型退出训练。
较佳地，所述激活函数为：
Figure PCTCN2020093546-appb-000030
其中y为所述判断值集，e为无限不循环小数，ω为所述特征集。
本申请较佳实施例所述损失值T为：
Figure PCTCN2020093546-appb-000031
其中，n为所述标签集的大小，y t为所述判断值集，μ t为所述标签集。
S5、接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
较佳地，若用户输入的细胞集A图片中包括异常细胞，则所述异常细胞判断模型识别出所述细胞集A含有异常细胞，并输出判断结果，所述输出方式包括屏幕打印或语音播报等方式。
发明还提供一种智能化异常细胞判断装置。参照图2所示，为本申请一实施例提供的智能化异常细胞判断装置的内部结构示意图。(对应修改)
在本实施例中，所述智能化异常细胞判断装置1可以是PC(Personal Computer，个人电脑)，或者是智能手机、平板电脑、便携计算机等终端设备，也可以是一种服务器等。该智能化异常细胞判断装置1至少包括存储器11、处理器12，通信总线13，以及网络接口14。
其中，存储器11至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是智能化异常细胞判断装置1的内部存储单元，例如该智能化异常细胞判断装置1的硬盘。存储器11在另一些实施例中也可以是智能化异常细胞判断装置1的外部存储设备，例如智能化异常细胞判断装置1上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital，SD)卡，闪存卡(Flash Card)等。进一步地，存储器11还可以既包括智能化异常细胞判断装置1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于智能化异常细胞判断装置1的应用软件及各类数据，例如智能化异常细胞判断程序01的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器或其他数据处理芯片，用于运行存储器11中存储的程序代码或处理数据，例如执行智能化异常细胞判断程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口)，通常用于在该装置1与其他电子设备之间建立通信连接。
可选地，该装置1还可以包括用户接口，用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard)，可选的用户接口还可以包括标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在智能化异常细胞判断装置1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及智能化异常细胞判断程序01的智能化异常细胞判断装置1，本领域技术人员可以理解的是，图1示出的结构并不构成对智能化异常细胞判断装置1的限定，可以包括比图示更少或者更多的部件，或者组合某些部件，或者不同的部件布置。
在图2所示的装置1实施例中，存储器11中存储有智能化异常细胞判断程序01；处理器12执行存储器11中存储的智能化异常细胞判断程序01时实现如下步骤：
步骤一、获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作。
其中，所述获取细胞集的操作包括：获取细胞的粘膜及分泌物，对所述粘膜及分泌物进行染色处理，拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集，并分别标注所述细胞集中每张图片中是否包含有异常细胞，得到所述标签集。
较佳地，本申请基于刮片在待检测细胞部位旋转一周获取所述待检测细胞的粘膜及分泌物，并将所述粘膜及分泌物涂抹至有机玻璃上并进行染色处理，同时调取显微设备拍摄所述有机玻璃内的粘膜及分泌物细胞，并最终得到所述细胞集。
优选地，所述染色处理先将所述涂抹有粘膜及分泌物的有机玻璃固定于酒精液体中，然后放置苏木素染色剂至所述酒精液体中，并最终达到将所述粘膜及分泌物内的细胞进行染色的目的。
本申请较佳实施例，所述降噪采用如下自适应图像降噪滤波法：
g(x,y)＝η(x,y)+f(x,y)
Figure PCTCN2020093546-appb-000032
其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
Figure PCTCN2020093546-appb-000033
为所述细胞集的噪声总方差，
Figure PCTCN2020093546-appb-000034
为所述(x,y)的像素灰度均值，
Figure PCTCN2020093546-appb-000035
为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
进一步地，所述对比度增强的目的是增大所述细胞集图片中亮度最大值与最小值之间的差值，因为对比度低的细胞会影响后续异常细胞的判断。较佳地，所述对比度增强采用如下方法：
D b＝f(D a)＝a*D a+b
其中a为线性斜率，b为在Y轴上的截距，若a>1，此时输出的图像对比度相比原图像的对比度是增强的，若a<1，此时输出的图像对比度相比原图像的对比度是减小的，其中D a代表所述细胞集的灰度值，D b代表输出的细胞集灰度值。
步骤二、基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集。
优选地，本申请较佳实施例遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于所述细胞集总像素数量计算每个灰度值的出现概率，基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
进一步地，所述每个灰度值的出现概率
Figure PCTCN2020093546-appb-000036
的计算方法为：
Figure PCTCN2020093546-appb-000037
其中，t it i+1属于t 1,t 2,…t m，而t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，进一步地：
Figure PCTCN2020093546-appb-000038
其中，b i为所述每个灰度值的出现次数，i在0～255范围内，N为灰度值的总出现次数。
较佳地，所述类间方差集σ T为
Figure PCTCN2020093546-appb-000039
其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
Figure PCTCN2020093546-appb-000040
表示不同预设阈值区间内所述每个灰度值的出现概率，
Figure PCTCN2020093546-appb-000041
表示不同预设阈值区间的灰度均值。
进一步地，所述重置所述细胞集的灰度区间的重置方法为：
Figure PCTCN2020093546-appb-000042
其中，t 1σ Tmax为所述最大的类间方差，I(x,y)为所述细胞集的灰度值，(x,y)为所述细胞集中各像素坐标。例如当像素坐标为(2,3)的灰度值I(2,3)＝4时，判断4的灰度值在区间t 1<I(2,3)≤t 2，则按照上述重置方法计算出新的灰度值，依次计算其他像素，直至最终得到多级灰度细胞集。
步骤三、基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集。
所述海森矩阵是由图像的高阶微分构造而成并能够反映图像特征的矩阵。本申请较佳地先计算所述多级灰度细胞集的尺度空间函数I ab，基于所述尺度空间函数I ab反向求导得到所述海森矩阵，利用海森矩阵求解出所述多级灰度细胞集RGB中每个通道对应的两种 特征值图，并将所述特征图增加到所述多级灰度细胞集的原通道中，得到九通道细胞集。
优选地，所述尺度空间函数I ab为：
Figure PCTCN2020093546-appb-000043
其中，ab与所述对比度增强中参数含义相同，a为线性斜率，b为在Y轴上的截距，σ为尺度空间函数参数，I(x,y)为所述细胞集各通道的灰度值，本申请包括R、G、B三个通道，G(X,Y；σ)为高斯函数。
较佳地，所述高斯函数为：
Figure PCTCN2020093546-appb-000044
其中，e为无限不循环小数。进一步地，所述尺度空间函数I ab反向求导过程得到所述海森矩阵H：
Figure PCTCN2020093546-appb-000045
其中，x n,y n表示所述多级灰度细胞集内不同像素的坐标，基于所述海森矩阵H求解对应行列式的特征值λ：
λH＝0
较佳地，所述特征值λ一般设置为3维，即λ 1，λ 2，λ 3。
进一步地，因为所述多级灰度细胞集共有R、G、B三个颜色通道，分别对所述三个颜色通道求解海森矩阵H得到H R、H G、H B，继而求解所述三个颜色通道的特征值得到λ R1，λ R2，λ R3、λ G1，λ G2，λ G3、λ B1，λ B2，λ B3共九个值组成九通道矩阵。因此基于上述方法对所述多级灰度细胞集的每个像素求解所述九通道矩阵并最终得到九通道细胞集λ RGB。
步骤四、将所述九通道细胞集及标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练。
较佳地，所述异常细胞判断模型包括特征提取层和异常细胞识别层，并以卷积神经网络为基础。其中，所述特征提取层接收所述九通道细胞集，并基于卷积操作和池化操作进行特征提取，所述卷积操作为：
Figure PCTCN2020093546-appb-000046
其中ω’为所述卷积操作后的输出值，一般为多维矩阵形式，λ RGB为所述九通道细胞集，k为卷积核的大小，通常为2*2维度，每个维度值为1的矩阵，s为所述卷积操作的步幅，可取值为1，p为数据补零矩阵。
本申请较佳实施例将所述卷积操作后的输出值进行所述池化操作，所述池化操作寻找所述卷积操作的输出值中矩阵数值最大的值并组成特征集ω。
进一步地，本申请将所述特征集及所述标签集输入至所述异常细胞识别层，所述异常细胞识别层对所述特征集进行所述卷积操作后输入至激活函数得到判断值集，将所述判断值集和所述标签集输入基于损失函数中计算得到损失值，若所述损失值小于预设阈值时，所述异常细胞判断模型退出训练。
较佳地，所述激活函数为：
Figure PCTCN2020093546-appb-000047
其中y为所述判断值集，e为无限不循环小数，ω为所述特征集。
本申请较佳实施例所述损失值T为：
Figure PCTCN2020093546-appb-000048
其中，n为所述标签集的大小，y t为所述判断值集，μ t为所述标签集。
步骤五、接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
较佳地，若用户输入的细胞集A图片中包括异常细胞，则所述异常细胞判断模型识别出所述细胞集A含有异常细胞，并输出判断结果，所述输出方式包括屏幕打印或语音播报等方式。
可选地，在其他实施例中，智能化异常细胞判断程序还可以被分割为一个或者多个模块，一个或者多个模块被存储于存储器11中，并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请，本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段，用于描述智能化异常细胞判断程序在智能化异常细胞判断装置中的执行过程。
例如，参照图3所示，为本申请智能化异常细胞判断装置一实施例中的智能化异常细胞判断程序的程序模块示意图，该实施例中，所述智能化异常细胞判断程序可以被分割为数据接收模块10、数据处理模块20、模型训练模块30、智能化异常细胞判断输出模块40示例性地：
所述数据接收模块10用于：获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作。
所述数据处理模块20用于：基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集、基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集。
所述模型训练模块30用于：将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练。
所述智能化异常细胞判断输出模块40用于：接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
上述数据接收模块10、数据处理模块20、模型训练模块30、智能化异常细胞判断输出模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同，在此不再赘述。
此外，本申请实施例还提出一种计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性。所述计算机可读存储介质上存储有智能化异常细胞判断程序，所述智能化异常细胞判断程序可被一个或多个处理器执行，以实现如下操作：
获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集，基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
接收用户的细胞图片，并输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
需要说明的是，上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素，而且还包括没有明确列出的 其他要素，或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种智能化异常细胞判断方法，其中，所述方法包括：
    获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
    基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集；
    基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
    将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
    接收用户的细胞图片，并将所述用户的细胞图片输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
  2. 如权利要求1所述的智能化异常细胞判断方法，其中，所述获取细胞集包括：
    获取细胞的粘膜及分泌物；
    对所述粘膜及分泌物进行染色处理；
    拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集。
  3. 如权利要求1或2所述的智能化异常细胞判断方法，其中，所述降噪采用如下自适应图像降噪滤波法：
    g(x,y)＝η(x,y)+f(x,y)
    Figure PCTCN2020093546-appb-100001
    其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
    Figure PCTCN2020093546-appb-100002
    为所述细胞集的噪声总方差，
    Figure PCTCN2020093546-appb-100003
    为所述(x,y)的像素灰度均值，
    Figure PCTCN2020093546-appb-100004
    为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
  4. 如权利要求1中的智能化异常细胞判断方法，其中，所述基于多阈值分割模型对所述预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集，包括：
    遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于总像素数量计算每个灰度值的出现概率；
    基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
  5. 如权利要求4所述的智能化异常细胞判断方法，其中，所述类间方差集σ T为：
    Figure PCTCN2020093546-appb-100005
    其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
    Figure PCTCN2020093546-appb-100006
    表示不同预设阈值区间内所述每个灰度值的出现概率，
    Figure PCTCN2020093546-appb-100007
    表示不同预设阈值区间的灰度均值。
  6. 如权利要求4所述的智能化异常细胞判断方法，其中，所述每个灰度值的出现概率
    Figure PCTCN2020093546-appb-100008
    的计算方法为：
    Figure PCTCN2020093546-appb-100009
    Figure PCTCN2020093546-appb-100010
    其中，t it i+1属于t 1,t 2,…t m，而t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，n i为所述每个灰度值的出现次数，i在0～255范围内，N为灰度值的总出现次数。
  7. 如权利要求4所述的智能化异常细胞判断方法，其中，所述重置所述细胞集的灰度区间的重置方法为：
    Figure PCTCN2020093546-appb-100011
    其中，t 1σ Tmax为所述最大的类间方差，I(x,y)为所述细胞集的灰度值，(x,y)为所述细胞集中各像素坐标。例如当像素坐标为(2,3)的灰度值I(2,3)＝4时，判断4的灰度值在区间t 1<I(2,3)≤t 2，则按照上述重置方法计算出新的灰度值，依次计算其他像素，直至最终得到多级灰度细胞集。
  8. 一种智能化异常细胞判断装置，其中，所述装置包括存储器和处理器，所述存储器上存储有可在所述处理器上运行的智能化异常细胞判断程序，所述智能化异常细胞判断程序被所述处理器执行时实现如下步骤：
    获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
    基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集；
    基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
    将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
    接收用户的细胞图片，并将所述用户的细胞图片输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
  9. 如权利要求7所述的智能化异常细胞判断装置，其中，所述获取细胞集包括：
    获取细胞的粘膜及分泌物；
    对所述粘膜及分泌物进行染色处理；
    拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集。
  10. 如权利要求8或9所述的智能化异常细胞判断装置，其中，所述降噪采用如下自适应图像降噪滤波法：
    g(x,y)＝η(x,y)+f(x,y)
    Figure PCTCN2020093546-appb-100012
    其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
    Figure PCTCN2020093546-appb-100013
    为所述细胞集的噪声总方差，
    Figure PCTCN2020093546-appb-100014
    为所述(x,y)的像素灰度均值，
    Figure PCTCN2020093546-appb-100015
    为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
  11. 如权利要求10所述的智能化异常细胞判断装置，其中，所述基于多阈值分割模型对所述预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集，包括：
    遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数，并基于总像素数量计算每个灰度值的出现概率；
    基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
  12. 如权利要求11所述的智能化异常细胞判断装置，其中，所述类间方差集σ T为：
    Figure PCTCN2020093546-appb-100016
    其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
    Figure PCTCN2020093546-appb-100017
    表示不同预设阈值区间内所述每个灰度值的出现概率，
    Figure PCTCN2020093546-appb-100018
    表示不同预设阈值区间的灰度均值。
  13. 如权利要求11所述的智能化异常细胞判断装置，其中，所述每个灰度值的出现 概率
    Figure PCTCN2020093546-appb-100019
    的计算方法为：
    Figure PCTCN2020093546-appb-100020
    Figure PCTCN2020093546-appb-100021
    其中，t it i+1属于t 1,t 2,…t m，而t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，n i为所述每个灰度值的出现次数，i在0～255范围内，N为灰度值的总出现次数。
  14. 如权利要求11所述的智能化异常细胞判断装置，其中，所述重置所述细胞集的灰度区间的重置方法为：
    Figure PCTCN2020093546-appb-100022
    其中，t 1σ Tmax为所述最大的类间方差，I(x,y)为所述细胞集的灰度值，(x,y)为所述细胞集中各像素坐标。例如当像素坐标为(2,3)的灰度值I(2,3)＝4时，判断4的灰度值在区间t 1<I(2,3)≤t 2，则按照上述重置方法计算出新的灰度值，依次计算其他像素，直至最终得到多级灰度细胞集。
  15. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有智能化异常细胞判断程序，所述智能化异常细胞判断程序可被一个或者多个处理器执行，以实现如权利要求1至7中任一项所述的智能化异常细胞判断方法的步骤：
    获取细胞集及标签集，对所述细胞集进行包括降噪、对比度增强的预处理操作；
    基于多阈值分割模型对预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集；
    基于所述多级灰度细胞集计算海森矩阵并得到九通道细胞集；
    将所述九通道细胞集及所述标签集输入至异常细胞判断模型中训练，直至所述异常细胞判断模型满足预设训练退出条件后退出训练；
    接收用户的细胞图片，并将所述用户的细胞图片输入至所述异常细胞判断模型中判断所述细胞是否包括异常细胞并输出判断结果。
  16. 如权利要求15所述的计算机可读存储介质，其中，所述获取细胞集包括：
    获取细胞的粘膜及分泌物；
    对所述粘膜及分泌物进行染色处理；
    拍摄进行染色处理之后的所述细胞的粘膜及分泌物得到所述细胞集。
  17. 如权利要求15或16所述的计算机可读存储介质，其中，所述降噪采用如下自适应图像降噪滤波法：
    g(x,y)＝η(x,y)+f(x,y)
    Figure PCTCN2020093546-appb-100023
    其中，(x,y)表示所述细胞集内图像像素点坐标，f(x,y)为基于所述自适应图像降噪滤波法对所述细胞集进行降噪处理后的输出数据，η(x,y)为噪声，g(x,y)为所述细胞集，
    Figure PCTCN2020093546-appb-100024
    为所述细胞集的噪声总方差，
    Figure PCTCN2020093546-appb-100025
    为所述(x,y)的像素灰度均值，
    Figure PCTCN2020093546-appb-100026
    为所述(x,y)的像素灰度方差，L表示当前像素点坐标。
  18. 如权利要求15中的计算机可读存储介质，其中，所述基于多阈值分割模型对所述预处理操作完成的所述细胞集进行灰度划分，得到多级灰度细胞集，包括：
    遍历预处理操作完成的所述细胞集内图像像素的灰度值，统计每个灰度值出现的次数， 并基于总像素数量计算每个灰度值的出现概率；
    基于预设阈值区间和所述每个灰度值的出现概率计算得到类间方差集，并选择所述类间方差集中数值最大的类间方差重置所述细胞集的灰度区间，得到多级灰度细胞集。
  19. 如权利要求18所述的计算机可读存储介质，其中，所述类间方差集σ T为：
    Figure PCTCN2020093546-appb-100027
    其中，T为所述预设阈值区间，T＝{t 1,t 2,…t m}，t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，
    Figure PCTCN2020093546-appb-100028
    表示不同预设阈值区间内所述每个灰度值的出现概率，
    Figure PCTCN2020093546-appb-100029
    表示不同预设阈值区间的灰度均值。
  20. 如权利要求18所述的计算机可读存储介质，其中，所述每个灰度值的出现概率
    Figure PCTCN2020093546-appb-100030
    的计算方法为：
    Figure PCTCN2020093546-appb-100031
    Figure PCTCN2020093546-appb-100032
    其中，t it i+1属于t 1,t 2,…t m，而t 1,t 2,…t m分别表示预设阈值，且t 1,t 2直到t m呈递增形式，t 1不小于数值0，t m不大于数值255，n i为所述每个灰度值的出现次数，i在0～255范围内，N为灰度值的总出现次数。
PCT/CN2020/093546 2019-06-14 2020-05-29 智能化异常细胞判断方法、装置及计算机可读存储介质 WO2020248848A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910520871.9 2019-06-14
CN201910520871.9A CN110363747A (zh) 2019-06-14 2019-06-14 智能化异常细胞判断方法、装置及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2020248848A1 true WO2020248848A1 (zh) 2020-12-17

Family

ID=68216064

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/093546 WO2020248848A1 (zh) 2019-06-14 2020-05-29 智能化异常细胞判断方法、装置及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110363747A (zh)
WO (1) WO2020248848A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113096077A (zh) * 2021-03-25 2021-07-09 深圳力维智联技术有限公司 异常比例检测方法、装置、设备及计算机可读存储介质
CN117876380A (zh) * 2024-03-13 2024-04-12 昆明昊拜农业科技有限公司 一种烟叶环境温湿度、微层差预测方法及系统

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363747A (zh) * 2019-06-14 2019-10-22 平安科技（深圳）有限公司 智能化异常细胞判断方法、装置及计算机可读存储介质
CN110415212A (zh) * 2019-06-18 2019-11-05 平安科技（深圳）有限公司 异常细胞检测方法、装置及计算机可读存储介质
CN111652845A (zh) * 2020-04-27 2020-09-11 平安科技（深圳）有限公司 异常细胞自动标注方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015142923A1 (en) * 2014-03-17 2015-09-24 Carnegie Mellon University Methods and systems for disease classification
CN108364032A (zh) * 2018-03-27 2018-08-03 哈尔滨理工大学 一种基于卷积神经网络的宫颈癌细胞图片识别算法
CN108376400A (zh) * 2018-02-12 2018-08-07 华南理工大学 一种骨髓细胞自动分类方法
CN109308695A (zh) * 2018-09-13 2019-02-05 镇江纳兰随思信息科技有限公司 基于改进U-net卷积神经网络模型的癌细胞识别方法
CN110363747A (zh) * 2019-06-14 2019-10-22 平安科技（深圳）有限公司 智能化异常细胞判断方法、装置及计算机可读存储介质

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108197606A (zh) * 2018-01-31 2018-06-22 浙江大学 一种基于多尺度膨胀卷积的病理切片中异常细胞的识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015142923A1 (en) * 2014-03-17 2015-09-24 Carnegie Mellon University Methods and systems for disease classification
CN108376400A (zh) * 2018-02-12 2018-08-07 华南理工大学 一种骨髓细胞自动分类方法
CN108364032A (zh) * 2018-03-27 2018-08-03 哈尔滨理工大学 一种基于卷积神经网络的宫颈癌细胞图片识别算法
CN109308695A (zh) * 2018-09-13 2019-02-05 镇江纳兰随思信息科技有限公司 基于改进U-net卷积神经网络模型的癌细胞识别方法
CN110363747A (zh) * 2019-06-14 2019-10-22 平安科技（深圳）有限公司 智能化异常细胞判断方法、装置及计算机可读存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIANG, XIANGANG ET AL.: "Extraction method of brain vessels based on multi-threshold Otsu and Hessian matrix", COMPUTER ENGINEERING AND DESIGN, vol. 35, no. 5, 31 May 2014 (2014-05-31), ISSN: 1000-7024, DOI: 20200813084945Y *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113096077A (zh) * 2021-03-25 2021-07-09 深圳力维智联技术有限公司 异常比例检测方法、装置、设备及计算机可读存储介质
CN113096077B (zh) * 2021-03-25 2024-05-03 深圳力维智联技术有限公司 异常比例检测方法、装置、设备及计算机可读存储介质
CN117876380A (zh) * 2024-03-13 2024-04-12 昆明昊拜农业科技有限公司 一种烟叶环境温湿度、微层差预测方法及系统
CN117876380B (zh) * 2024-03-13 2024-05-14 昆明昊拜农业科技有限公司 一种烟叶环境温湿度、微层差预测方法及系统

Also Published As

Publication number Publication date
CN110363747A (zh) 2019-10-22

Similar Documents

Publication Publication Date Title
WO2020248848A1 (zh) 智能化异常细胞判断方法、装置及计算机可读存储介质
WO2020253508A1 (zh) 异常细胞检测方法、装置及计算机可读存储介质
WO2022134337A1 (zh) 人脸遮挡检测方法、系统、设备及存储介质
WO2019109526A1 (zh) 人脸图像的年龄识别方法、装置及存储介质
WO2019174130A1 (zh) 票据识别方法、服务器及计算机可读存储介质
WO2021217851A1 (zh) 异常细胞自动标注方法、装置、电子设备及存储介质
US20190050641A1 (en) Methods and apparatus for capturing, processing, training, and detecting patterns using pattern recognition classifiers
WO2021189912A1 (zh) 图像中目标物的检测方法、装置、电子设备及存储介质
CN110569756B (zh) 人脸识别模型构建方法、识别方法、设备和存储介质
WO2021051611A1 (zh) 基于人脸可见性的人脸识别方法、系统、装置及存储介质
WO2020199475A1 (zh) 人脸识别方法、装置、计算机设备及存储介质
WO2021012494A1 (zh) 基于深度学习的人脸识别方法、装置及计算机可读存储介质
WO2019137038A1 (zh) 注视点确定方法、对比度调节方法及对比度调节装置、虚拟现实设备及存储介质
CN113159147B (zh) 基于神经网络的图像识别方法、装置、电子设备
WO2020253127A1 (zh) 脸部特征提取模型训练方法、脸部特征提取方法、装置、设备及存储介质
CN111461165A (zh) 图像识别方法、识别模型的训练方法及相关装置、设备
US20140334736A1 (en) Face recognition method and device
US10885321B2 (en) Hand detection method and system, image detection method and system, hand segmentation method, storage medium, and device
US11886492B2 (en) Method of matching image and apparatus thereof, device, medium and program product
WO2022127112A1 (zh) 跨模态人脸识别方法、装置、设备及存储介质
WO2022127111A1 (zh) 跨模态人脸识别方法、装置、设备及存储介质
EP4047509A1 (en) Facial parsing method and related devices
WO2021189853A1 (zh) 闪光灯光斑位置识别方法、装置、电子设备及存储介质
WO2019119396A1 (zh) 人脸表情识别方法及装置
CN110852311A (zh) 一种三维人手关键点定位方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20823417

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20823417

Country of ref document: EP

Kind code of ref document: A1