Cell miniature microscopic image acquisition system based on artificial intelligence

Abstract

The invention discloses a cell miniature microscopic image acquisition system based on artificial intelligence in the technical field of cell medical image acquisition, and the system comprises an image acquisition module which is used for acquiring a sample plane image; the sample plane image processing module is used for extracting at least two groups of measurement marks from the sample plane image, each group of measurement marks at least comprises two measurement points, reference measurement lines of the sample plane and the axis of the objective lens are constructed based on each group of measurement points, and every two reference measurement lines meet a mutual bisection relation; the perpendicularity judgment module is used for calculating the real-time distances between the measuring points on the same reference measuring line and the objective lens, and if the distances between the measuring points and the objective lens are equal, the current sample plane is judged to be perpendicular to the axis of the objective lens; otherwise, judging that the inclination deviation exists and calculating a deviation value; and the deviation feedback module is used for outputting a deviation value. According to the scheme, the perpendicularity of the objective lens axis and the sample plane can be monitored in real time, image distortion is reduced, and image acquisition quality is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cell medical image acquisition, and in particular is a cell microscopic image acquisition system based on artificial intelligence. Background Art

[0002] Microscopic image acquisition technology plays a key role in many fields, including life sciences and medical research. Traditional microscopic image acquisition systems rely primarily on precision optical equipment, including microscope objectives, stages, light sources, and image sensors. Researchers use manual or semi-automatic methods to place cell samples on the microscope stage and adjust the distance between the objective lens and the sample to clearly image the cell structure on the image sensor, thereby enabling cell observation and image acquisition.

[0003] For example, in cell morphology research, microscopic images can be used to observe important information such as cell size, shape, and organelle distribution. In pathological diagnosis, microscopic images can be used to identify pathological features of cells, such as abnormal nuclei and disordered cell arrangement, providing a basis for disease diagnosis.

[0004] Existing microscope optical systems, especially high numerical aperture objectives, are originally designed based on a vertical incident light path. However, in practical applications, if the objective optical axis cannot be kept strictly perpendicular to the sample stage plane (i.e., the plane where the sample is located), a series of serious problems will arise:

[0005] For example, when the objective lens axis is not perpendicular to the sample plane, the collected cell microscopic images will be distorted. This manifests as distorted cell structures at the edges of the image and inconsistent cell clarity at different locations, causing cells that should have been presented on the same plane to appear tilted and distorted. This not only affects researchers' intuitive observation of the cells' true morphology but also creates numerous difficulties in subsequent image analysis.

[0006] In view of these defects in the existing technology, there is an urgent need for a new cell micro-microscopic image acquisition system that can effectively improve the accuracy of image acquisition. Summary of the Invention

[0007] In order to solve the above problems, the purpose of the present invention is to provide an artificial intelligence-based cell micro-microscope image acquisition system, which can monitor the angular deviation between the objective lens axis and the sample plane in real time, ensuring that the objective lens axis and the sample plane remain highly perpendicular each time an image is acquired, thereby effectively reducing image distortion and inaccurate focusing problems and improving image quality.

[0008] In order to achieve the above object, the technical solution of the present invention is as follows:

[0009] A cell microscopic image acquisition system based on artificial intelligence, comprising:

[0010] An image acquisition module, used for acquiring a sample plane image;

[0011] A sample plane image processing module is configured to extract at least two groups of measurement marks from the sample plane image, each group of measurement marks including at least two measurement points, and to construct reference measurement lines between the sample plane and the objective lens axis based on each group of measurement points, wherein all reference measurement lines satisfy a mutually bisecting relationship.

[0012] The verticality judgment module is used to calculate the real-time distance between the measurement point and the objective lens on the same reference measurement line. If the distance between the measurement point and the objective lens is equal, it is determined that the current sample plane is perpendicular to the objective lens axis; otherwise, it is determined that there is a tilt deviation and the deviation value is calculated;

[0013] Deviation feedback module, used to output deviation value.

[0014] Furthermore, an interactive interface module is included for mapping the calculated deviation value to a corresponding coordinate area in the sample plane image and performing visual rendering.

[0015] Furthermore, the deviation value mapping space mapping includes the following steps:

[0016] Coordinate system establishment:

[0017] With the intersection of the objective axis and the ideal sample plane as the origin O(0,0,0), a right-handed rectangular coordinate system is established: the Z axis points along the objective axis to the sample direction, and the X / Y axis is parallel to the sample plane, forming the imaging plane reference;

[0018] Angle-Depth Conversion:

[0019] For any point P(x i ,y i ):

[0020] Calculate the actual height of the measuring point

[0021]

[0022] Where, d i : Distance from point P to the objective lens, d Ο : ideal vertical distance, θ i : local tilt angle;

[0023] Obtain the deviation depth through spatial projection relationship:

[0024]

[0025] Where, is the average height of the plane, θi is the angle between the normal vector at point P and the Z axis;

[0026] Regionalization mapping:

[0027] Divide the sample plane into m×n rectangular grid areas;

[0028] For each region R k Calculate the mean deviation depth:

[0029]

[0030] Where N k For region R k Number of internal measurement points;

[0031] ΔH k Serves as the final deviation depth value for the area.

[0032] Furthermore, the visual rendering method includes:

[0033] The regional deviation depth is represented by gradient color levels, and the color depth is positively correlated with the deviation depth;

[0034] Or the plane is divided by contour lines, and the vertical distance between adjacent contour lines is a fixed threshold;

[0035] The rendering is superimposed on the sample plane image in real time to form a fused visualization layer.

[0036] Furthermore, the measurement mark is positioned by dynamically adjusting the radius to draw multiple concentric circles, and the measurement points are taken from the cell image features on the circumference of the concentric circles.

[0037] Furthermore, the deviation value is calculated using the space vector analysis method:

[0038] The objective lens axis direction is the reference vector

[0039] Construct the sample plane normal vector based on the distance difference between the measurement points

[0040] The tilt angle deviation value is output through the vector angle formula θ:

[0041] Furthermore, a repair module is included for performing real-time image correction based on the mapping relationship between the deviation depth and image distortion.

[0042] Furthermore, the restoration module includes a distortion modeling unit, a bimodal restoration unit, and an adaptive selector;

[0043] The distortion modeling unit is used to establish the functional relationship between the deviation depth ΔH and the distortion parameter λ:

[0044]

[0045] Where k1 is the radial distortion coefficient of the optical system; k2 is the tangential distortion gradient weight; is the spatial gradient of the deviation depth;

[0046] The dual-modal repair unit includes two repair paths: a physical correction model and a deep learning repair model. The physical correction model: when ‖ΔH‖ is not greater than the preset threshold, the distorted wavefront is decomposed using Zernike polynomials to generate the point spread function (PSF) inverse kernel for deconvolution:

[0047]

[0048] Where, I corrtected is the corrected image matrix, F is the Fourier transform operator, F -1 is the inverse Fourier transform operator, OTF(λ) is the optical transfer function, I distorted is the input distorted image tensor;

[0049] Deep learning restoration model: When ‖ΔH‖ is greater than the preset threshold, a cascaded U-Net network is used: the primary network outputs a two-dimensional deformation field T ΔH ∈R H×W×2 , where T is the transformation field, R is the real space, H is the image height, W is the image width, and 2 is the number of displacement channels; the secondary network performs feature fusion: Where Encoder is the feature encoder, Decoder is the feature decoder, I distorted is the input distorted image tensor;

[0050] Adaptive selector for real-time monitoring of gradient change rate When satisfied When the change rate exceeds the preset threshold, the deep learning repair model is forcibly started.

[0051] Furthermore, the cell image features include observable structures of the cell.

[0052] Furthermore, the interactive interface module is also used to receive user operation instructions and adjust the display mode and visualization rendering parameters of the deviation value according to the instructions to meet the needs and observation habits of different users.

[0053] The above scheme has the following beneficial effects:

[0054] 1. This solution ensures that the objective axis remains highly perpendicular to the sample plane during each image acquisition by real-time monitoring of the angular deviation between the objective axis and the sample plane, effectively reducing image distortion and focus inaccuracies. Through precise perpendicularity judgment and deviation feedback, the captured cell microscopic images more clearly and accurately reflect the true morphology and structure of cells, improving the accuracy and reliability of image acquisition.

[0055] 2. This solution utilizes a dual restoration mechanism in the restoration module, which selects the appropriate correction method based on the depth of the deviation, achieving real-time correction of distorted images. The physical restoration model is suitable for smaller deviations and provides fast and effective correction. The deep learning restoration model handles larger deviations and complex distortions, providing higher-quality corrected images. An adaptive selector dynamically adjusts the restoration strategy based on the gradient change rate, ensuring intelligent and efficient image correction and improving the performance and practicality of the entire system.

[0056] 3. This solution establishes a right-handed rectangular coordinate system with the intersection of the objective axis and the ideal sample plane as the origin, providing a unified and precise three-dimensional spatial reference framework for subsequent deviation calculation and correction. By defining the Z axis pointing along the objective axis toward the sample, and the X / Y axes parallel to the sample plane as the imaging plane reference, it accurately describes the position and height information of each point on the sample plane, making deviation calculation and correction operations more scientific and accurate, laying a solid foundation for improving image acquisition quality.

[0057] For any point P on the sample plane, the actual height h of the measuring point is calculated i And combine the spatial projection relationship to obtain the deviation depth Δh i , achieving a precise conversion from angular deviation to depth deviation. This conversion can quantify the abstract angular deviation into a specific depth value, more intuitively reflecting the actual deviation between the objective axis and the sample plane, providing an accurate numerical basis for subsequent image correction, and helping to achieve more accurate image restoration.

[0058] 4. This solution provides two repair paths through a dual-modal repair unit, selecting the appropriate repair model based on the depth of the deviation, achieving a combination of flexibility and efficiency. The physical correction model is suitable for smaller deviations. It uses Zernike polynomials to decompose the distorted wavefront and perform deconvolution, which can quickly and effectively correct image distortion and save computing resources. The deep learning repair model can handle larger deviations and complex distortions. It outputs a two-dimensional deformation field through a cascaded U-Net network and performs feature fusion, fully leveraging the powerful fitting capabilities of deep learning to provide higher-quality corrected images, effectively solving the problem of poor correction effect of traditional physical models in complex distortion situations.

[0059] By combining the strengths of a physical correction model and a deep learning restoration model, the restoration module accurately corrects varying degrees of image distortion. For small deviations, the physical correction model quickly removes most distortions; for larger deviations, the deep learning restoration model further refines the correction and removes complex distortions. This dual restoration mechanism effectively improves the quality and reliability of image correction, resulting in clearer and more accurate cell microscopic images that better reflect the true morphology and structure of cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a system block diagram of the artificial intelligence-based cell micro-microscopic image acquisition system of the present invention. DETAILED DESCRIPTION

[0061] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0062] The following is further described in detail through specific implementation methods:

[0063] The embodiment is basically as shown in the attached Figure 1 Shown: A cell microscopic image acquisition system based on artificial intelligence, comprising:

[0064] The image acquisition module is used to capture images of the sample plane. This module utilizes a high-resolution CMOS image sensor and is paired with a micro-microscope optical system suitable for cell observation, including optical components such as an objective lens, condenser, and filters. In practice, the cell sample is placed on the stage. The stage position and objective lens height are adjusted to position the sample plane for optimal observation. The image acquisition module is responsible for capturing image data of the sample plane in real time or on demand, converting it into digital signals and transmitting it to the subsequent processing module.

[0065] The sample plane image processing module runs on an embedded computing system or computer and is equipped with specialized image processing software. The software uses a preset algorithm to identify and extract at least two sets of measurement markers from the captured sample plane image. Each set of measurement markers consists of at least two measurement points. Specifically, these points are located by dynamically adjusting the radius to draw multiple concentric circles. The measurement points are derived from cell image features (including observable cellular structures, such as pixels representing the nucleus or cell wall) along the circumference of the concentric circles. By continuously adjusting the radius, a series of concentric circles is generated, which is equivalent to creating a reference geometric structure on the image. Pixel points (at least two) are selected along the circumference of the concentric circles. These points represent measurable features (such as cellular structures) in the sample plane. This ensures that the positions of the measurement points are geometrically consistent (i.e., located on the same circumference), facilitating the subsequent construction of reference lines. By determining the positions of these two measurement points, a reference measurement line is constructed between the sample plane and the objective lens axis, forming a spatial reference for the sample plane and the objective lens axis. To ensure measurement accuracy, all reference measurement lines must bisect each other, meaning that the midpoints of every two reference measurement lines coincide, thus forming a stable measurement reference system.

[0066] The verticality determination module is also implemented based on an embedded computing system or computer. It receives measurement point position data from the sample plane image processing module and calculates the real-time distance between each measurement point on the same reference measurement line and the objective lens. The specific calculation method can use a CMOS image sensor to obtain the distance from all measurement points on the same reference measurement line to the objective lens through time-of-flight (ToF) ranging or parallax (binocular) ranging. If the distance from all measurement points on the same reference measurement line to the objective lens is equal, it indicates that the current sample plane and the objective lens axis are perpendicular; conversely, if there is a difference in distance, it is determined that there is a tilt deviation between the sample plane and the objective lens axis.

[0067] Specifically, the deviation value in this embodiment is calculated using the space vector analysis method:

[0068] The objective lens axis direction is the reference vector

[0069] Construct the sample plane normal vector based on the distance difference between the measurement points

[0070] The tilt angle deviation value is output through the vector angle formula θ:

[0071] The deviation feedback module is used to output the deviation value. After the verticality determination module detects the tilt deviation and calculates the deviation value, the deviation feedback module outputs this deviation value as a digital signal. The output deviation value can be a specific numerical value or a value scaled according to a preset ratio for subsequent processing or display. This module ensures that the deviation information is promptly and accurately transmitted to the subsequent interactive interface module (to facilitate user adjustment of the objective lens) or other possible adjustment control modules (such as the automatic objective lens adjustment mechanism provided by the microscope system).

[0072] The invention also includes an interactive interface module for mapping the calculated deviation value to a corresponding coordinate area in the sample plane image and performing visual rendering.

[0073] Specifically, the interactive interface module: This module is typically a graphical user interface (GUI) running on a computer or dedicated control terminal. It receives deviation value data from the deviation feedback module and maps it to the corresponding coordinate area in the sample plane image. The specific mapping method is to determine the sample plane area corresponding to the deviation value based on the positional relationship between the sample plane image coordinate system and the measurement mark. In terms of visual rendering, the interactive interface module uses gradient color gradations or contour lines to intuitively display the deviation depth. For example, a gradient color gradation from blue to red is used to represent the change from shallow to deep deviation depth, with color depth positively correlated with deviation depth. Alternatively, contour lines are used to connect areas of the same deviation depth, with the vertical distance between adjacent contour lines set to a fixed threshold, allowing users to clearly observe the tilt of the sample plane. The rendered image is superimposed on the original sample plane image in real time to form a fused visualization layer. The user can observe this fused layer to intuitively understand the perpendicularity deviation between the objective axis and the sample plane. At the same time, the interactive interface module also supports users to operate through input devices such as mouse, keyboard or touch screen. Users can adjust the display mode of deviation values ​​and visualization rendering parameters according to their needs, such as changing the color mapping scheme, adjusting the contour line interval, zooming in and out of the image, etc., to meet different observation habits and analysis needs.

[0074] Preferably, the deviation value mapping space mapping comprises the following steps:

[0075] Coordinate system establishment:

[0076] With the intersection of the objective axis and the ideal sample plane as the origin O(0,0,0), a right-handed rectangular coordinate system is established: the Z axis points along the objective axis to the sample direction, and the X / Y axis is parallel to the sample plane, forming the imaging plane reference;

[0077] Angle-Depth Conversion:

[0078] For any point P(x i ,yi ):

[0079] Calculate the actual height of the measuring point

[0080]

[0081] Where, d i : Distance from point P to the objective lens, d Ο : ideal vertical distance, θ i : local tilt angle;

[0082] Obtain the deviation depth through spatial projection relationship:

[0083]

[0084] Where, is the average height of the plane, θ i is the angle between the normal vector at point P and the Z axis; Directly associate angles with geometric deformations to avoid black box models.

[0085] Regionalization mapping:

[0086] Divide the sample plane into m×n rectangular grid areas;

[0087] For each region R k Calculate the mean deviation depth:

[0088]

[0089] Where N k For region R k Number of internal measurement points;

[0090] ΔH k Serves as the final deviation depth value for the area.

[0091] When there is an inclination angle θ between the objective lens axis and the sample plane, the Z-direction deviation of the surface point P(x, y) is:

[0092]

[0093] where θ x ,θ y are the component inclination angles in the X / Z and Y / Z planes (calculated by measuring the marker group).

[0094] Preferably, a repair module is also included for performing real-time image correction based on a mapping relationship between deviation depth and image distortion.

[0095] The restoration module includes a distortion modeling unit, a dual-modal restoration unit, and an adaptive selector;

[0096] The distortion modeling unit is used to establish the functional relationship between the deviation depth ΔH and the distortion parameter λ:

[0097]

[0098] Where k1 is the radial distortion coefficient of the optical system (calibrated value), which is obtained by calibration with a standard optical calibration plate; k2 is the tangential distortion gradient weight, which dynamically fits the curvature change of the sample surface; is the spatial gradient of the deviation depth;

[0099] Example calculation: If ΔH=5μm at a certain point, (High gradient area), calibration values ​​k1 = 0.02, k2 = 0.15: λ = 0.02×52+0.15×0.8=0.5+0.12=0.62.

[0100] The dual-modal repair unit includes two repair paths: a physical correction model and a deep learning repair model. The physical correction model: when ‖ΔH‖ is not greater than a preset threshold, for example, ‖ΔH‖≤3μm (the critical value is selected based on experimental data), the distorted wavefront is decomposed using Zernike polynomials to generate the point spread function (PSF) inverse kernel for deconvolution:

[0101]

[0102] Where, I corrtected is the corrected image matrix, is the Fourier transform operator, is the inverse Fourier transform operator, OTF(λ) is the optical transfer function, I distorted is the input distorted image tensor, i.e., the original captured cell microscopic image;

[0103] Deep learning repair model: When ‖ΔH‖ is greater than the preset threshold, for example, when ‖ΔH‖>3μm, a cascaded U-Net network is used:

[0104] The primary network outputs a two-dimensional deformation field Where T is the transformation field, which represents the vector field of pixel position correction; R is the real number space, and the tensor elements are floating-point numbers (the displacement can be a fractional pixel); H is the image height, which corresponds to the vertical dimension of the microscopic image (in this embodiment, it represents the sample plane image); W is the image width, which corresponds to the horizontal dimension of the microscopic image; 2 is the number of displacement channels, channel 0: x-direction displacement (horizontal offset), channel 1: y-direction displacement (vertical offset);

[0105] Example: Assume that the deformation field value at a pixel point (x,y) = (100,200) is:

[0106] T ΔH(100,200)=[Δx,Δy]=[1.5,-0.8]

[0107] Calibration process:

[0108] 1. Original position: (x, y) = (100, 200)

[0109] 2. Position after correction:

[0110] (x′,y′)=(100+1.5,200-0.8)=(101.5,199.2)

[0111] 3. Calculate the pixel value at the new coordinates through bilinear interpolation.

[0112] The secondary network performs feature fusion: Where Encoder is the feature encoder (extracting multi-scale features (VGG16 backbone)), Decoder is the feature decoder that generates the repaired image (transposed convolution layer), I distorted is the input distorted image tensor, i.e., the original captured cell microscopic image.

[0113] Adaptive selector for real-time monitoring of gradient change rate When satisfied When the rate of change is greater than a preset threshold (e.g., 10% / ms, specifically determined through experiments: exceeding this value causes motion blur), the deep learning restoration model is forcibly started.

[0114] The above is only an embodiment of the present invention, and common knowledge such as the specific structure and / or characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. An artificial intelligence-based cell micro-microscopic image acquisition system, comprising an image acquisition module for acquiring a sample plane image; characterized in that: Also includes: A sample plane image processing module is configured to extract at least two groups of measurement marks from the sample plane image, each group of measurement marks including at least two measurement points, and to construct reference measurement lines between the sample plane and the objective lens axis based on each group of measurement points, wherein all reference measurement lines satisfy a mutually bisecting relationship. The verticality judgment module is used to calculate the real-time distance between the measurement point and the objective lens on the same reference measurement line. If the distances between the measurement point and the objective lens are equal, it is determined that the current sample plane is perpendicular to the objective lens axis. Otherwise, it is determined that there is a tilt deviation and the deviation value is calculated; Deviation feedback module, used to output deviation value.

2. The artificial intelligence-based cell micro-microscopic image acquisition system according to claim 1, characterized in that: The invention also includes an interactive interface module for mapping the calculated deviation value to a corresponding coordinate area in the sample plane image and performing visual rendering.

3. The artificial intelligence-based cell microscopic image acquisition system according to claim 2, characterized in that: Deviation value mapping space mapping includes the following steps: Coordinate system establishment: With the intersection of the objective axis and the ideal sample plane as the origin O(0,0,0), a right-handed rectangular coordinate system is established: the Z axis points along the objective axis to the sample direction, and the X / Y axis is parallel to the sample plane, forming the imaging plane reference; Angle-Depth Conversion: For any point P(x i ,y i ): Calculate the actual height of the measuring point Where, d i : Distance from point P to the objective lens, d Ο : ideal vertical distance, θ i : local tilt angle; Obtain the deviation depth through spatial projection relationship: Where, is the average height of the plane, θ i is the angle between the normal vector at point P and the Z axis; Regionalization mapping: Divide the sample plane into m×n rectangular grid areas; For each region R k Calculate the mean deviation depth: Where N k For region R k Number of internal measurement points; ΔH k Serves as the final deviation depth value for the area.

4. The artificial intelligence-based cell microscopic image acquisition system according to claim 3, characterized in that: Visual rendering methods include: The regional deviation depth is represented by gradient color levels, and the color depth is positively correlated with the deviation depth; Or the plane is divided by contour lines, and the vertical distance between adjacent contour lines is a fixed threshold; The rendering is superimposed on the sample plane image in real time to form a fused visualization layer.

5. The artificial intelligence-based cell microscopic image acquisition system according to claim 4, characterized in that: The measurement marks are positioned by dynamically adjusting the radius to draw multiple concentric circles, and the measurement points are taken from the cell image features on the circumference of the concentric circles.

6. The artificial intelligence-based cell microscopic image acquisition system according to claim 5, characterized in that: The deviation value is calculated using space vector analysis: The objective lens axis direction is the reference vector Construct the sample plane normal vector based on the distance difference between the measurement points The tilt angle deviation value is output through the vector angle formula θ:

7. The artificial intelligence-based cell microscopic image acquisition system according to claim 6, characterized in that: The invention also includes a restoration module for performing real-time image correction based on the mapping relationship between the deviation depth and the image distortion.

8. The artificial intelligence-based cell microscopic image acquisition system according to claim 7, characterized in that: The restoration module includes a distortion modeling unit, a dual-modal restoration unit, and an adaptive selector; The distortion modeling unit is used to establish the functional relationship between the deviation depth ΔH and the distortion parameter λ: Where k1 is the radial distortion coefficient of the optical system; k2 is the tangential distortion gradient weight; is the spatial gradient of the deviation depth; The dual-modal repair unit includes two repair paths: a physical correction model and a deep learning repair model. The physical correction model: when ‖ΔH‖ is not greater than the preset threshold, the distorted wavefront is decomposed using Zernike polynomials to generate the point spread function (PSF) inverse kernel for deconvolution: Where, I corrtected is the corrected image matrix, is the Fourier transform operator, is the inverse Fourier transform operator, OTF(λ) is the optical transfer function, I distorted is the input distorted image tensor; Deep learning restoration model: When ‖ΔH‖ is greater than the preset threshold, a cascaded U-Net network is used: the primary network outputs a two-dimensional deformation field Where T is the transformation field, is a real number space, H is the image height, W is the image width, and 2 is the number of displacement channels; the secondary network performs feature fusion: I output =Decoder(Encoder(I distorted )⊕T ΔH ), where Encoder is the feature encoder, Decoder is the feature decoder, I distorted is the input distorted image tensor; Adaptive selector for real-time monitoring of gradient change rate When satisfied When the change rate exceeds the preset threshold, the deep learning repair model is forcibly started.

9. The artificial intelligence-based cell microscopic image acquisition system according to claim 8, characterized in that: Cell image features include observable cell structures.

10. The artificial intelligence-based cell microscopic image acquisition system according to claim 9, characterized in that: The interactive interface module is also used to receive user operation instructions and adjust the display mode of the deviation value and the visual rendering parameters according to the instructions to meet the needs and observation habits of different users.

Images