A cell suspension sorting system and method based on image recognition

Through multi-scale Retinex algorithm and pseudo-color mapping technology, combined with convolutional neural network, the problem of light inhomogeneity and inaccurate color extraction in cell suspension color distinction is solved, and high accuracy and efficient cell suspension analysis is achieved .

CN118788633BActive Publication Date: 2024-11-26NANJING ADVANCED ACAD OF LIFE & HEALTH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411277892.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-12
Publication Date
2024-11-26
Estimated Expiration
2044-09-12

AI Technical Summary

Technical Problem

The existing cell suspension color distinction technology has problems such as uneven light, inaccurate color feature extraction and large errors in manual judgment, resulting in poor accuracy and consistency of analysis results.

Method used

Multi-scale Retinex algorithm is used to compensate for light inhomogeneity, combine Gaussian filtering and logarithmic operations to eliminate the influence of light, and use color transmission algorithm and pseudo-color mapping technology to extract and classify the color feature of cell suspensions using convolutional neural network. .

Benefits of technology

It significantly improves the accuracy and robustness of color distinction of cell suspensions, enhances the reliability and efficiency of analysis, and reduces the error of manual judgment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118788633B_ABST
    Figure CN118788633B_ABST
Patent Text Reader

Abstract

This invention relates to a cell suspension sorting system and method based on image recognition. The image recognition-based cell suspension sorting system includes a sorting tank with a dropper inside, a support rod inside the tank, multiple test tubes placed on the support rod, and an indicator light ring next to each test tube connected to a computer. A slide rail is installed inside the sorting tank, and a camera is slidably connected to the slide rail. The camera is connected to the computer, and after acquiring images of each test tube, it is uploaded to the computer for processing and analysis, and the results are displayed on the indicator light ring. This invention solves the problems of uneven illumination, inaccurate color feature extraction, and large errors in manual judgment existing in traditional methods, significantly improving the accuracy and reliability of cell suspension color differentiation, and has significant application value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical engineering, and specifically relates to a cell suspension sorting system and method based on image recognition. Background Art

[0002] In recent years, with the rapid development of life sciences and biotechnology, the importance of color differentiation technology of cell suspensions in cell research and pharmaceutical fields has gradually become prominent. Cell suspension refers to the suspended state of cells in a liquid medium, which is widely used in experiments such as cell culture, drug screening and gene editing. Traditional methods of color differentiation of cell suspensions usually rely on naked eye observation or simple optical microscopes for color judgment. This method is highly subjective, has large errors, and is easily affected by the experience and fatigue of the experimenters, resulting in poor accuracy and consistency of the analysis results. With the development of image recognition technology and artificial intelligence algorithms, automated color differentiation methods based on computer vision have gradually become a research hotspot. This method improves the accuracy and efficiency of color recognition through high-precision image acquisition and complex image processing algorithms. In addition, the application of multispectral imaging technology enables the use of more dimensional spectral information in the color differentiation of cell suspensions, thereby better reflecting the state and characteristics of cells.

[0003] However, the existing cell suspension color differentiation technology still has some significant shortcomings. First, the traditional image acquisition and processing methods have great limitations in capturing illumination changes, image noise and cell color features, making it difficult to obtain stable color differentiation results in complex experimental environments. Secondly, the existing image processing algorithms usually ignore the problems of illumination inhomogeneity and image standardization when extracting and analyzing color features, resulting in significant differences between images collected in different batches and under different conditions, affecting the reliability of the analysis. In response to these shortcomings, we have invented a cell suspension color differentiation method based on image recognition, which uses a multi-scale Retinex algorithm to compensate for illumination inhomogeneity, and uses Gaussian filtering and logarithmic operations to eliminate the influence of illumination, significantly improving the degree of image standardization and the accuracy of analysis. In addition, the color characteristics of the cell suspension are enhanced through the color transmission algorithm and pseudo-color mapping technology, making the color differences between different cell suspensions more obvious, further improving the sensitivity and accuracy of color differentiation.

[0004] There are two main problems with the existing cell suspension color differentiation technology based on image recognition: one is the lack of standardized processing of lighting conditions and color characteristics, and the other is the lack of accurate recognition of the colors of diverse cell suspensions. The cell suspension color differentiation method based on image recognition invented by us solves the above problems through a variety of image processing algorithms and a deep learning model of convolutional neural networks, significantly improving the accuracy and robustness of cell suspension color differentiation, and belongs to the field of image processing and biomedical engineering technology. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the following technical problems in the prior art: the selection of cell suspension is currently achieved by operators comparing with the naked eye, which is very demanding on the operator's experience, has large judgment errors and is inefficient.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a cell suspension sorting system based on image recognition, comprising:

[0008] A liquid separation box, wherein a liquid dripping member is arranged in the liquid separation box, and a lever is also arranged in the liquid separation box, and a plurality of test tubes are placed on the lever, and an indicator light ring is arranged beside the test tubes, and the indicator light ring is connected to a computer;

[0009] A slide rail is provided in the liquid separation box, and a camera is slidably connected to the slide rail. The camera is connected to a computer. After the camera obtains a photo of a test tube, it is uploaded to the computer for processing and analysis, and the result is displayed on the indicator light ring.

[0010] As a preferred technical solution for a cell suspension sorting system based on image recognition, it also includes an image preprocessing module, a feature extraction and enhancement module, and a classification analysis module. The camera takes pictures of multiple test tubes in sequence and uploads them to a computer. The image preprocessing module preprocesses the collected images to eliminate illumination unevenness and noise. The feature extraction and enhancement module enhances the color features and performs pseudo-color coding on the preprocessed images to improve the recognizability and classification accuracy of cells. The classification analysis module uses a deep learning algorithm to classify cells on the enhanced and encoded images to determine whether the cell suspension is qualified or not. After analysis, the analysis results are displayed on the corresponding indicator light ring.

[0011] As a preferred technical solution for a cell suspension sorting system based on image recognition, the camera can provide high-quality image input under different lighting conditions. The image acquisition module includes two sub-functions: autofocus and adaptive exposure control. Autofocus is to select the optimal focal length by comparing the clarity of the image at different focal lengths. Adaptive exposure control is to automatically adjust the exposure time of the camera according to changes in ambient light to ensure that the brightness of the image is uniform and moderate.

[0012] As a preferred technical solution for a cell suspension sorting system based on image recognition, the image preprocessing module also includes illumination compensation, which uses a multi-scale Retinex algorithm to eliminate illumination inhomogeneity;

[0013] The feature extraction and enhancement module includes color feature enhancement, which uses a color transmission algorithm based on histogram equalization to optimize the color features of cells, making the colors of different types of cells more obvious.

[0014] As a preferred technical solution for a cell suspension sorting system based on image recognition, the pseudo-color encoding converts the RGB color space into the HSV color space, and then applies a color mapping table to convert subtle color differences into pseudo-color displays that are easily distinguished by the human eye, and then encodes them through the pseudo-color mapping table to obtain a pseudo-color image.

[0015] As a preferred technical solution for a cell suspension sorting system based on image recognition, an illuminating lamp is provided in the liquid separation box at the position corresponding to the test tube, and a background plate is rotatably connected below the opening of the liquid separation box, and the back of the background plate is coated with a black coating.

[0016] As a preferred technical solution for a cell suspension sorting system based on image recognition, a screw and a guide rod are arranged above the opening of the liquid separation box, the dripping piece is connected to the screw and the guide rod, and the liquid separation box is obliquely provided with an elliptical groove, and the two long axis ends of the elliptical groove correspond to the screw and the slide rail respectively.

[0017] As a preferred technical solution for a cell suspension sorting system based on image recognition, the liquid separation box is also provided with a circular plate controlled by a motor, the circular plate is provided with radial grooves, the lever passes through the radial grooves, and the lever is embedded in the elliptical grooves.

[0018] The present invention also discloses a method based on the aforementioned cell suspension sorting system based on image recognition, characterized in that it comprises the following steps:

[0019] S1, start the image acquisition module, adjust the camera exposure time, capture high-resolution images of cell suspensions in multiple color test tubes, and transmit them to the image preprocessing module;

[0020] S2, the image preprocessing module uses a multi-scale Retinex algorithm to compensate for illumination non-uniformity of the received image, eliminates illumination effects through Gaussian filtering and logarithmic operations, and outputs a standardized image;

[0021] S3, normalizing the images after illumination compensation to eliminate the color and brightness differences between different images;

[0022] S4, using a color transmission algorithm to enhance the color characteristics of the cells by adjusting the brightness and contrast of pixels in the image, so that the color differences between the cell suspensions in each test tube are more obvious;

[0023] S5, converting the enhanced image from the RGB color space to the HSV color space to extract hue information;

[0024] S6, using a pseudo color mapping table to convert the hue information into a pseudo color image, so that the colors of the cell suspensions in each test tube in the image are displayed as significantly different colors;

[0025] S7. Then, the pseudo-color image is input into the analysis model constructed based on the convolutional neural network for analysis.

[0026] As a preferred technical solution for a cell suspension sorting method based on image recognition, the analysis in the analysis model includes:

[0027] The significant color displayed by each cell suspension in the pseudo-color image is compared with the set qualified standard to determine whether the cell suspension is qualified. If the significant color displayed by the cell suspension is within the qualified standard range, it is determined to be qualified, otherwise, it is determined to be unqualified.

[0028] Beneficial effects of the invention: The invention solves the problems existing in traditional methods such as uneven illumination, inaccurate color feature extraction, large errors in manual judgment, etc., greatly improves the accuracy and reliability of color distinction of cell suspensions, and has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0030] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0031] Figure 2 It is a structural schematic diagram of the liquid separation box in the present invention;

[0032] Figure 3 It is a schematic diagram of the module connection structure in the present invention.

[0033] Figure numerals: test tube 400, computer 600, indicator light ring 500, camera 102, lighting lamp 103, background plate 104, dripping part 200, guide rod 106, screw rod 105, slide rail 101, liquid separation box 100, circular plate 108, radial groove 109, lever 300, elliptical groove 107. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0037] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0038] Reference Figure 1~Figure 3 The present embodiment provides a cell suspension sorting system based on image recognition, including a liquid separation box 100, wherein a liquid dropper 200 is disposed in the liquid separation box 100, a lever 300 is further disposed in the liquid separation box 100, a plurality of test tubes 400 are placed on the lever 300, an indicator light ring 500 is disposed next to the test tubes 400, and the indicator light ring 500 is connected to a computer 600;

[0039] A slide rail 101 is provided in the liquid separation box 100, and a camera 102 is slidably connected to the slide rail 101. The camera 102 is connected to the computer 600. The camera 102 obtains a test tube photo and uploads it to the computer 600. After processing and analysis, the result is displayed on the indicator light ring 500.

[0040] Specifically, the computer 600 also controls the movement of the camera 102 on the slide rail 101. Specifically, the camera 102 moves on the slide rail 101 and takes pictures of each test tube 400 in sequence, which are uploaded and identified and analyzed by the computer 600 to distinguish qualified and unqualified subpackaged suspensions, and the results are displayed on the indicator light ring 500 next to the corresponding test tube 400. Specifically, the indicator light ring 500 is a red and blue indicator light, red represents unqualified, and blue represents qualified.

[0041] It also includes an image preprocessing module, a feature extraction and enhancement module, and a classification analysis module. The camera 102 takes pictures of multiple test tubes 400 in sequence and uploads them to the computer 600. The image preprocessing module preprocesses the collected images to eliminate lighting unevenness and noise. The feature extraction and enhancement module enhances the color features and performs pseudo-color coding on the preprocessed images to improve the recognizability and classification accuracy of cells. The classification analysis module uses a deep learning algorithm to classify cells on the enhanced and encoded images to determine whether the cell suspension is qualified or not. After analysis, the analysis results are displayed on the corresponding indicator light ring 500.

[0042] An illuminating lamp 103 is arranged in the liquid separation box 100 at a position corresponding to the test tube 400. A background plate 104 is rotatably connected below the opening of the liquid separation box 100. The back of the background plate 104 is coated with a black coating.

[0043] A screw rod 105 and a guide rod 106 are arranged above the opening of the liquid separation box 100, and the dripping member 200 is connected to the screw rod 105 and the guide rod 106. The liquid separation box 100 is inclinedly provided with an elliptical groove 107, and the two long axis ends of the elliptical groove 107 correspond to the screw rod 105 and the slide rail 101 respectively.

[0044] The liquid separation box 100 is further provided with a circular plate 108 controlled by a motor. The circular plate 108 is provided with a radial groove 109 . The lever 300 is provided through the radial groove 109 , and the lever 300 is embedded in the elliptical groove 107 .

[0045] It should be noted that the test tube is placed on the lever 300 through a snap-fit ​​structure, and a counterweight is provided on the lever 300 to keep the test tube placed vertically.

[0046] It should be noted that, during the rotation of the circular plate 108 , the lever 300 will be driven to move in the elliptical groove 107 , and two levers 300 are provided, which are centrally symmetrically arranged.

[0047] In the present invention, the dripping process is realized at the top of the opening of the liquid separation box 100. The lever 300 with the empty test tubes is placed corresponding to the bottom of the dripping part 200, and the dripping part performs dripping. At the same time, another lever 300 is placed with the test tubes filled with cell suspension, corresponding to the slide rail 101. The camera 102 moves along the slide rail 101, the background plate 104 is erected, and each test tube is photographed. After computer processing and analysis, the detection result is displayed on the indicator light ring 500, and then the background plate 104 is put down, the circular plate 108 is rotated 180°, and the background plate 104 is erected again. The test tubes just filled with cell suspension are turned to the camera 102 for photo identification. At the same time, the test tubes with results are removed and classified according to the results displayed on the indicator light ring 500. Then the empty test tubes are placed on the lever 300 to continue dripping.

[0048] Furthermore, the system uses the camera 102 to take pictures to analyze whether the test tube containing the cell suspension is qualified. The function is mainly composed of the following modules: image acquisition module (camera 102), image preprocessing module, feature extraction and enhancement module, classification analysis module and result display module (indicator ring 500). The functional application and working principle of each module are described as follows:

[0049] Image acquisition module: The main function of the image acquisition module is to acquire high-resolution images of cell suspensions in real time. The module includes a high-speed camera that can provide high-quality image input under different lighting conditions. The image acquisition module includes two sub-functions: autofocus and adaptive exposure control: Autofocus: By comparing the clarity of the image at different focal lengths, the best focal length is selected. This can be achieved by maximizing the energy of high-frequency components, and adaptive exposure control: Automatically adjust the exposure time of the camera according to changes in ambient light to ensure that the brightness of the image is uniform and moderate. Adaptive exposure can be achieved by calculating the difference between the average brightness of the image and the target brightness to adjust the exposure time.

[0050] Image preprocessing module: The main function of the image preprocessing module is to perform preliminary processing on the collected images, eliminate illumination unevenness and noise, and provide standardized input images for subsequent color feature enhancement and pseudo-color coding. Illumination compensation: The multi-scale Retinex (MSR) algorithm is used. MSR can effectively eliminate illumination unevenness. (The image preprocessing module standardizes the image so that the subsequent feature extraction and enhancement modules can perform color enhancement and pseudo-color coding based on uniform illumination conditions.)

[0051] Feature extraction and enhancement module: The function of this module is to enhance the color features and pseudo-color coding of the pre-processed image to improve the recognizability and classification accuracy of cells. Color feature enhancement: The color transmission algorithm based on histogram equalization is used to optimize the color features of cells, making the colors of different types of cells more obvious. The specific implementation is as follows:

[0052] Working principle implementation:

[0053] Color feature enhancement: A color transmission algorithm based on histogram equalization is used to optimize the color features of cells, making the colors of different types of cells more obvious. The specific implementation is as follows:

[0054]

[0055] in, is the image after illumination compensation and standardization. and are the mean and standard deviation of the standardized image, and are the mean and standard deviation of the reference image.

[0056] Pseudo color encoding: Pseudo color encoding converts the RGB color space into the HSV color space and then applies a color mapping table to convert subtle color differences into pseudo color displays that are easily distinguished by the human eye. The formula is as follows:

[0057]

[0058] It is then encoded through a pseudo color mapping table to obtain a pseudo color image.

[0059] Classification and analysis module: The main function of the classification and analysis module is to use deep learning algorithms to classify cells in enhanced and encoded images and determine whether the cell suspension is qualified or not. Convolutional neural network (CNN) classification: CNN includes multiple layers of convolutional layers, pooling layers, and fully connected layers. Each convolutional layer extracts the spatial features of the image, the pooling layer is used for feature dimensionality reduction, and the fully connected layer is used for feature classification. The output Softmax layer is used to calculate the probability of each cell category.

[0060] Result display module: The function of the result display module is to display the classification results in real time and store them for subsequent analysis and verification. According to the output of the classification analysis module, the LED indicator and display screen display the cell classification results in real time. The green light indicates qualified, the red light indicates unqualified, and the classification data is stored in the database.

[0061] Also included is a method for implementing the sorting system, characterized in that the method comprises the following steps:

[0062] S1, start the image acquisition module, adjust the camera exposure time, capture high-resolution images of cell suspensions in multiple color test tubes, and transmit them to the image preprocessing module;

[0063] S2, the image preprocessing module uses a multi-scale Retinex algorithm to compensate for illumination non-uniformity of the received image, eliminates illumination effects through Gaussian filtering and logarithmic operations, and outputs a standardized image;

[0064]

[0065] The formula includes the following functions and parameter definitions and selections:

[0066] 1. is the pixel value of the output normalized image.

[0067] 2. is the pixel value of the input high-resolution image, which is the captured image after the exposure time is adjusted.

[0068] 3. is the average brightness of the local area of ​​the image, which can be calculated by Gaussian convolution. The formula is

[0069]

[0070] 4. is the standard deviation of the local area of ​​the image, and the formula is

[0071]

[0072] 5. To prevent small constants with zero denominators, numerical stability is ensured.

[0073] 6. is the standard deviation of the Gaussian filter, controlling the smoothness.

[0074] 7. is the illumination compensation image calculated by the multi-scale Retinex algorithm, and the calculation formula is: ,in are illumination components of different scales, and δ is a small amount to avoid the denominator being zero.

[0075] 8. It is a nonlinear adjustment parameter in the logarithmic operation, which controls the degree of image contrast enhancement.

[0076] 9. is the information filtering weight function, which is used to control the contribution of different pixel areas to the result, and is defined as , where α and β are adjustment parameters that control the intensity of information filtering.

[0077] The output image of the formula The range of depends mainly on the dynamic range of the input image and the parameter settings in the preprocessing step. After normalization and logarithmic enhancement, The value range of is usually [0,1], which represents the standardized brightness value of the image. Among them, lower values ​​(close to 0) represent dark areas or low brightness areas in the image, and higher values ​​(close to 1) represent bright areas or high brightness areas. Through the processing of this formula, the uneven illumination of the image is compensated, and the brightness and color characteristics are more uniform, which is helpful for subsequent image analysis and processing.

[0078] S3, normalizing the images after illumination compensation to eliminate the color and brightness differences between different images;

[0079] The normalization formula is as follows:

[0080]

[0081] The formula contains the following functions and parameters defined and selected:

[0082] 1. Represents the pixel value of the normalized image. After normalization, the pixel value of the output image will be in the range of [0,1].

[0083] 2. is the pixel value of the input image after illumination compensation, which is the image processed by the previous multi-scale Retinex algorithm and illumination compensation.

[0084] 3. Indicates the minimum pixel value in the image after illumination compensation, used as the lower limit for normalization.

[0085] 4. Indicates the maximum pixel value in the image after illumination compensation, which is used as the upper limit for normalization.

[0086] 5. It is the Gaussian weighted integral of the image after illumination compensation in the neighborhood A, which is used to smooth the image and enhance the contrast in the local area. σ represents the standard deviation of the Gaussian function.

[0087] 6. is the normalization constant of the Gaussian kernel, which is used to ensure that the pixel values ​​of the image after Gaussian smoothing are in the correct range.

[0088] Range explanation:

[0089] Normalized image The range of is [0,1]. Normalization makes the minimum and maximum values ​​of the image mapped to 0 and 1 respectively, and the relative positions of other pixel values ​​remain unchanged. The result of normalization ensures that different images have the same brightness and color range when compared, thereby eliminating the differences caused by different imaging conditions or equipment, making the results of image analysis more reliable and consistent.

[0090] S4, using a color transmission algorithm to enhance the color characteristics of the cells by adjusting the brightness and contrast of pixels in the image, so that the color differences between the cell suspensions in each test tube are more obvious;

[0091] We define the color transfer and enhancement formula of the image as follows

[0092]

[0093]

[0094] The formula contains the definition and selection of the following functions and parameters

[0095] 1. is the pixel value of the enhanced image.

[0096] 2. is the pixel value of the input image, which is the original image after preliminary processing.

[0097] 3. Represents the mean brightness of the local area of ​​the image, obtained by Gaussian convolution, the formula is

[0098] in, Gaussian filter standard deviation calculated for local brightness.

[0099] 4. is the brightness standard deviation of the local area of ​​the image, defined as

[0100] =

[0101] 5. Small constant to prevent division by zero.

[0102] 6. The standard deviation of the Gaussian filter for color enhancement, controlling the smoothness and range of local color difference enhancement.

[0103] 7. It is the weight coefficient of different feature channels in the color enhancement process, which is used to balance the contribution of different channels.

[0104] 8. and To adjust the parameters, control the color similarity calculation and enhancement strength between different feature channels.

[0105] 9. is the pixel value of the reference image or the target color feature image, and usually an image with good color contrast and target features is selected.

[0106] 10. Indicates the color difference between the current pixel value and the reference pixel value.

[0107] 11. It is the local area of ​​the image considered in the enhancement process and used to calculate the local color similarity.

[0108] Range explanation:

[0109] Enhanced image The value range of is usually kept between [0,1] or [0,255], depending on the bit depth and normalization of the input image. The Gaussian smoothing and exponential weighting operations in the formula significantly enhance the color features in the image. Higher values ​​(close to 1 or 255) represent areas of high brightness or significant color in the enhanced image, while lower values ​​(close to 0) represent dim or insignificant areas. After processing with this formula, the contrast and color features of the image are more prominent, making the color differences between cell suspensions in different test tubes more obvious, thereby improving the accuracy of image analysis and cell classification.

[0110] S5, converting the enhanced image from the RGB color space to the HSV color space, and extracting the hue information;

[0111]

[0112] The formula contains the definition and selection of the following functions and parameters:

[0113] 1. It is the output hue information, that is, the hue component of the enhanced image in the HSV color space.

[0114] 2. Respectively represent the red, green and blue components of the enhanced image in the RGB color space

[0115] 3. δ is a small constant to prevent division by zero and is used to enhance the numerical stability of the hue calculation.

[0116] 4. To avoid small constants with zero denominators, ensure computational stability.

[0117] 5. To control the nonlinear enhancement parameters in the process of hue information extraction, the relative contributions of different color components are affected.

[0118] 6. The basic calculation formula for hue is to convert the RGB color space into the angle value of the HSV color space.

[0119] 7. It is a complex information filtering function that is used to control the influence of different color components on hue calculation, and to enhance or weaken the weight of certain color components through exponential functions.

[0120] Range explanation:

[0121] Hue Information The range of is [0,2π] (radians), or is conventionally defined as 0,360] degrees, depending on the angle unit selected during calculation. In the HSV color space, the hue value represents the basic type of color. For example, 0 degrees represents red, 120 degrees represents green, and 240 degrees represents blue. After being processed by this formula, the color information of the image is enhanced, making cell suspensions of different colors more visually recognizable, which is convenient for subsequent image analysis and classification operations. The combination of integral and exponential operations in the formula further enhances the stability and robustness of the hue information, ensuring that the extraction of hue information remains accurate and reliable under different lighting and color distribution conditions.

[0122] S6, using a pseudo color mapping table to convert the hue information into a pseudo color image, so that the colors of the cell suspensions in each test tube in the image are displayed as significantly different colors;

[0123] Define pseudo color mapping function , which maps the hue information to a new color space. This mapping function uses a sine function combined with an exponential decay function to achieve nonlinear mapping of colors, ensuring that the color distribution is richer and the differences are more obvious.

[0124] The mapping function is used to perform pseudo color mapping on the smoothed and enhanced hue information to generate an intermediate pseudo color image.

[0125] The pseudo color mapping formula is as follows:

[0126]

[0127] The formula contains the definition and selection of the following functions and parameters:

[0128] 1. is the pixel value of the output pseudo color image, which is the image after pseudo color mapping processing.

[0129] 2. The hue information (Hue) of the input represents the hue component extracted after converting from RGB to HSV color space.

[0130] 3. α is the frequency parameter that controls the sine function, which is used to change the hue frequency in the pseudo-color mapping to make the color transition richer.

[0131] 4. is the phase offset, which is used to adjust the phase of the sine function, thereby affecting the hue change of the pseudo color.

[0132] 5.β is an exponential decay parameter that controls the influence range of different tones in the pseudo-color mapping process, making the color change more significant.

[0133] 6. is the mean of the hue information, which is used to calculate the center position of the hue distribution.

[0134] 7. and are the gradients of the hue information in the x and y directions, respectively, indicating the rate of change of the hue information.

[0135] 8. is the pseudo color mapping function, defined as:

[0136]

[0137] in, is the standard deviation of the mapping function, which controls the smoothness of the pseudo color mapping and the range of color changes.

[0138] Range Explanation

[0139] Pseudo-color images The value range is [0,1] or [0,255] (depending on the bit depth and normalization of the image). After pseudo-color mapping, the image's pixel values ​​are mapped to a new color space, making the colors of the original hue information more prominent and distinguishable. Higher values ​​represent areas with higher color intensity in the mapped image, and lower values ​​represent areas with lower color intensity. Through the combination of sine and exponential functions, this formula can generate pseudo-color images based on hue information, making the color differences between cell suspensions of different colors more obvious, which is helpful for subsequent image analysis and cell classification. The complex calculations and integral operations in the formula enhance the visual effect of the hue information, ensuring that the results of pseudo-color mapping are still clear and highly recognizable under different lighting and color distribution conditions.

[0140] S7, then input the pseudo-color image into an analysis model built based on a convolutional neural network for analysis, wherein:

[0141] The significant color displayed by each cell suspension in the pseudo-color image is compared with the set qualified standard to determine whether the cell suspension is qualified. If the significant color displayed by the cell suspension is within the qualified standard range (i.e., the set qualified color interval), it is determined to be qualified (marked as 1), otherwise, it is determined to be unqualified (marked as 0).

[0142] The following is a piecewise formula for inputting a pseudo-color image into an analysis model built on a convolutional neural network for analysis.

[0143] 1. Convolution operation

[0144] First, a convolution operation is performed to extract the features of the pseudo-color image:

[0145]

[0146] in, is the convolution feature map, is the weight of the convolution kernel, is the pixel value of the pseudo-color image.

[0147] 2. Feature normalization and nonlinear mapping

[0148] The convolution features are normalized and the nonlinear mapping capability is increased through the hyperbolic tangent function tanh:

[0149]

[0150] in, is the normalized eigenvalue, is the mean value of the color feature in the pseudo-color image, is the standard deviation of the color feature, A small constant to prevent the denominator from being zero.

[0151] 3. Gradient energy calculation

[0152] Calculate the gradient energy of the pseudo-color image to measure the edge strength and change of the image:

[0153]

[0154] in, is the gradient energy, which indicates the edge strength of the image. To avoid small quantities with zero denominator.

[0155] 4. Comprehensive feature fusion

[0156] Fuse convolution features, normalized features and gradient energy to calculate the final feature output:

[0157]

[0158] in, is the activation function output value of the convolutional neural network, and σ(z) is the Sigmoid function, which is used to limit the output value between 0 and 1.

[0159] 5. Classification Decision

[0160] According to the output value F(x) and the classification threshold , determine whether the cell suspension is qualified:

[0161]

[0162] in, is the final classification result, 1 means qualified, 0 means unqualified, is the classification decision value.

[0163] The value range of the formula output is [0,1], which represents the classification confidence value of the convolutional neural network for the input pseudo-color image. The output of the Sigmoid function is used to generate probability values. A higher value (close to 1) indicates that the color of the cell suspension in the image meets the qualified standard, while a lower value (close to 0) indicates that the color does not meet the qualified standard. By setting the threshold τ, the qualified or unqualified judgment criteria can be adjusted according to specific requirements, so that the model can flexibly respond to different application scenarios and analysis needs.

[0164] In order to more clearly understand the range of the above formula and its practical significance, the range of each part of the formula and its role in the entire image analysis process will be explained in detail below, and examples will be given to illustrate how to use these ranges to achieve classification and judgment of cell suspensions.

[0165] 1. Convolution operation Range

[0166] Range explanation:

[0167] Convolution Operation The range of depends on the input pseudo-color image The pixel value range and convolution kernel weight Assume that the pixel value range of the input image is [0,255], and the weight of the convolution kernel is determined by training, and its value may be between [-1,1]. Since convolution is a summation operation, the range of the convolution feature map can be calculated as follows:

[0168]

[0169] For example:

[0170] Assume that a 3x3 convolution kernel is used (ie, N=3, M=3), and the maximum pixel value of the image is 255, then The value range is [-2295, 2295]. The convolution operation enhances the edges and features in the image, so higher positive and negative values ​​represent strong edges or significant features in the image.

[0171] 2. Feature normalization and nonlinear mapping Range

[0172] Range explanation:

[0173] Normalized features The range of depends on the range of the normalization function and the hyperbolic tangent function tanh. The normalization function converts the color features of the image into a standard normal distribution (mean 0, standard deviation 1), while the range of the tanh function is (-1,1).

[0174]

[0175] For example:

[0176] In practical applications, if the color mean of the input pseudo-color image is is 128, standard deviation is 64, then for an input pixel value of 200, the normalized value is approximately:

[0177]

[0178] After the hyperbolic tangent function transformation, This indicates that the color feature of the pixel is close to the positive side of the center value of the target color feature and is a significant feature.

[0179] 3. Gradient Energy Range

[0180] Range explanation:

[0181] Gradient Energy The range of is non-negative because it represents the intensity of changes in the image in all directions (edge ​​strength). The square term ensures that the range is always positive, and a small amount δ is added to prevent the value from being zero. Its typical range is usually [0,∞), but in practical applications of image analysis, the range is usually standardized to between [0,1] or [0,255].

[0182]

[0183] For example:

[0184] If in a flat area of ​​an image, the pixel changes are small and the gradient is close to 0, then ≈ 0; while in the edge area, the gradient changes greatly and may reach a higher value. Assuming that the maximum gradient energy is normalized to 255, then ( = 100) represent edges of medium intensity.

[0185] 4. Comprehensive feature fusion Range

[0186] Range explanation:

[0187] Comprehensive features The range of is limited to [0,1] after applying the Sigmoid function σ(z). This is because the Sigmoid function maps any real number input to a probability value between 0 and 1, which is used to indicate whether the current pseudo-color image meets the qualification criteria.

[0188]

[0189] For example:

[0190] Assuming that the feature fusion calculation result z = 2.5, the output of the Sigmoid function is:

[0191]

[0192] This means that the image area has a 92% probability of meeting the pass standard. Conversely, if Z = -1.5, then σ(-1.5) ≈ 0.18, indicating an 18% probability of not meeting the standard.

[0193] 5. Classification Decision Range

[0194] Range explanation:

[0195] Classification Decision The value range of is {0,1}. (like ), the output is 1, indicating that the cell suspension is qualified; when When , the output is 0, indicating that the cell suspension is unqualified.

[0196]

[0197] For example:

[0198] Assuming the threshold τ=0.7, if F(x)=0.92, then =1, indicating qualified; if F(x)=0.3, then =0, indicating failure.

[0199] The beneficial effects of each step of the sorting method of the present invention are:

[0200] Step S1 (image acquisition and preprocessing preparation): By starting the image acquisition module, adjusting the camera exposure time to capture high-resolution images of multiple color test tube cell suspensions, and transmitting them to the image preprocessing module, high-quality data collection of cell suspension samples is achieved. Adjusting the camera exposure time to ensure consistent and clear image data under different lighting conditions is critical to the accuracy and consistency of subsequent image processing. The acquisition of high-resolution images improves the resolution of image details, especially in capturing subtle color differences in cell suspensions. The beneficial effect of this step is that it provides a high-quality, standardized image foundation for subsequent image preprocessing and color analysis, ensuring data accuracy and stability throughout the entire processing flow.

[0201] Step S2 (compensation and standardization of illumination non-uniformity): In this step, the image preprocessing module uses the multi-scale Retinex algorithm to compensate for illumination non-uniformity of the received image, eliminates the influence of illumination through Gaussian filtering and logarithmic operations, and outputs a standardized image. The multi-scale Retinex algorithm can effectively process images under different illumination conditions, restore the brightness of the real scene, and thus reduce the image brightness and color deviation caused by illumination changes. At the same time, Gaussian filtering and logarithmic operations are used to further eliminate the influence of illumination and ensure uniform brightness of the image. The beneficial effect of this standardization process is that it greatly improves the accuracy of subsequent color feature extraction, reduces the influence of changes in illumination conditions on color analysis, and makes the processing results between each step more consistent and reliable.

[0202] Step S3 (image normalization): The image after illumination compensation is normalized to eliminate the color and brightness differences between different images, thereby achieving the standardization of image color and brightness. The main purpose of image normalization is to eliminate the differences between images caused by different acquisition equipment and ambient lighting conditions, so that subsequent color analysis is more consistent and comparable. Through normalization, it is ensured that the color and brightness data of all images are on a unified scale, thereby reducing misjudgments or errors caused by the differences in the image data itself during the analysis stage. The beneficial effect of this step is to enhance the stability and accuracy of color feature extraction, improve the robustness of the color distinction algorithm, and ensure the reliability of the analysis results.

[0203] Step S4 (color feature enhancement): The brightness and contrast of pixels in the image are adjusted by the color transmission algorithm to enhance the color features of the cells, making the color differences between the cell suspensions in each test tube more obvious. This step improves the color contrast of each cell suspension in the image by enhancing the color features in the image, so that the subtle color differences between different cell suspensions can be more significantly presented. Its beneficial effect is to improve the accuracy of color recognition and classification, provide clearer input data for subsequent image color information extraction and pseudo-color conversion, and further enhance the effect of color analysis.

[0204] Steps S5 and S6 (color information extraction and pseudo-color conversion): Convert the enhanced image from the RGB color space to the HSV color space, extract the hue information, and then apply the pseudo-color mapping table to convert the hue information into a pseudo-color image. By converting to the HSV color space, the hue information in the image can be extracted more accurately, reducing the interference of hue changes on color judgment. The pseudo-color mapping technology converts the color of the cell suspension into a pseudo-color that is easier for the human eye to distinguish, making the color differences between different samples more significant and intuitive. The beneficial effect of this step is that it improves the visualization and differentiation ability of color information, providing a clearer and more recognizable input image for the subsequent automated analysis model based on convolutional neural networks.

[0205] Step S7 (analysis model based on convolutional neural network): The pseudo-color image is input into the analysis model constructed based on the convolutional neural network for analysis, and the significant color displayed by each cell suspension in the pseudo-color image is compared with the set qualified standard to determine whether the cell suspension is qualified. By using a convolutional neural network, complex image data can be automatically analyzed to quickly and accurately determine whether the color of each cell suspension is within the qualified standard range. The beneficial effect of this step is that it significantly improves the automation level of color differentiation, reduces the errors caused by manual operation and subjective judgment, and improves the accuracy and efficiency of cell suspension qualification determination.

[0206] In summary, the present invention achieves accurate distinction of cell suspension colors through a series of orderly and scientific steps, from image acquisition to image processing to automated analysis, solves the problems of uneven illumination, inaccurate color feature extraction, and large errors in manual judgment that exist in traditional methods, and greatly improves the accuracy and reliability of cell suspension color distinction, which has important application value.

[0207] It will be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but will be a routine task of design, fabrication, and production for those of ordinary skill having the benefit of this disclosure without undue experimentation.

[0208] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cell suspension sorting system based on image recognition, Features: include, A liquid separation box (100), wherein a liquid dripping member (200) is arranged in the liquid separation box (100), wherein a lever (300) is further arranged in the liquid separation box (100), wherein a plurality of test tubes (400) are placed on the lever (300), wherein an indicator light ring (500) is arranged next to the test tubes (400), and wherein the indicator light ring (500) is connected to a computer (600); A slide rail (101) is provided in the liquid separation box (100), a camera (102) is slidably connected to the slide rail (101), the camera (102) is connected to the computer (600), the camera (102) acquires a test tube photo and uploads it to the computer (600), and after processing and analysis, the result is displayed on the indicator light ring (500); It also includes an image preprocessing module, a feature extraction and enhancement module, and a classification analysis module. The camera (102) takes pictures of the plurality of test tubes (400) in sequence and uploads the pictures to a computer (600). The image preprocessing module preprocesses the collected images to eliminate illumination unevenness and noise. The feature extraction and enhancement module enhances the color features and performs pseudo-color coding on the preprocessed images to improve cell recognizability and classification accuracy. The classification analysis module uses a deep learning algorithm to classify cells on the enhanced and coded images to determine whether the cell suspension is qualified or not. After analysis, the analysis result is displayed on the corresponding indicator light ring (500); The camera (102) is capable of providing high-quality image input under different lighting conditions, and the image acquisition module includes two sub-functions: automatic focus and adaptive exposure control. The automatic focus is to select the best focal length by comparing the clarity of the image under different focal lengths, and the adaptive exposure control is to automatically adjust the exposure time of the camera according to the change of ambient light to ensure that the brightness of the image is uniform and moderate. The image preprocessing module also includes illumination compensation, which uses a multi-scale Retinex algorithm to eliminate illumination non-uniformity; The feature extraction and enhancement module includes color feature enhancement, which uses a color transmission algorithm based on histogram equalization to optimize the color features of cells, making the colors of different types of cells more obvious; The pseudo color encoding converts the RGB color space into the HSV color space, and then applies a color mapping table to convert the subtle color differences into pseudo color displays that are easily distinguished by the human eye, and then encodes through the pseudo color mapping table to obtain a pseudo color image; An illuminating lamp (103) is arranged in the liquid separation box (100) at a position corresponding to the test tube (400), and a background plate (104) is rotatably connected below the opening of the liquid separation box (100), and the back of the background plate (104) is coated with a black coating; The liquid separation box (100) is provided with a screw rod (105) and a guide rod (106) near the upper part of the opening, the liquid dripping member (200) is connected to the screw rod (105) and the guide rod (106), the liquid separation box (100) is provided with an elliptical groove (107) at an angle, and the two long axis ends of the elliptical groove (107) correspond to the screw rod (105) and the slide rail (101) respectively; The liquid separation box (100) is further provided with a circular plate (108) controlled by a motor, the circular plate (108) is provided with a radial groove (109), the lever (300) is provided through the radial groove (109), and the lever (300) is embedded in the elliptical groove (107).

2. A method for sorting a cell suspension based on image recognition based on the cell suspension sorting system based on image recognition according to claim 1, Features: The following steps are involved: S1, start the image acquisition module, adjust the camera exposure time, capture high-resolution images of cell suspensions in multiple color test tubes, and transmit them to the image preprocessing module; S2, the image preprocessing module uses a multi-scale Retinex algorithm to compensate for illumination non-uniformity of the received image, eliminates illumination effects through Gaussian filtering and logarithmic operations, and outputs a standardized image; S3, normalizing the images after illumination compensation to eliminate the color and brightness differences between different images; S4, using a color transmission algorithm to enhance the color characteristics of the cells by adjusting the brightness and contrast of pixels in the image, so that the color differences between the cell suspensions in each test tube are more obvious; S5, converting the enhanced image from the RGB color space to the HSV color space to extract hue information; S6, using a pseudo color mapping table to convert the hue information into a pseudo color image, so that the colors of the cell suspensions in each test tube in the image are displayed as significantly different colors; S7. Then, the pseudo-color image is input into the analysis model constructed based on the convolutional neural network for analysis.

3. The cell suspension sorting method based on image recognition according to claim 2, Features: The analyses performed in the analytical model include: The significant color displayed by each cell suspension in the pseudo-color image is compared with the set qualified standard to determine whether the cell suspension is qualified. If the significant color displayed by the cell suspension is within the qualified standard range, it is determined to be qualified, otherwise, it is determined to be unqualified.