Microfluidic flow cell image deblurring enhancement method

Abstract

This invention discloses a method for deblurring and enhancing microfluidic flowing cell images, relating to the fields of image enhancement, microscopic image processing, and microfluidic cell imaging. The method acquires continuous flowing cell images and adjacent frames, determining the working image and flow direction reference. Based on inter-frame local displacement, it extracts non-overlapping candidate cell regions, determines the directional blur length by combining local flow velocity, exposure time, and pixel physical size, and constructs a single-cell local directional blur kernel along the channel direction. Then, it constructs a trailing confidence map by combining trailing similarity, displacement consistency, and edge extension degree, establishing a variational deconvolution model that includes data fidelity, second-order smoothing of directional decomposition, and grayscale or fluorescence total amount constraints. The fusion result is obtained through candidate recovery results, uncertainty-weighted fusion, and backoff constraints, and cell region total amount correction is performed, outputting a deblurred and enhanced flowing cell image.

Description

Technical Field

[0001] This invention belongs to the fields of image enhancement, microscopic image processing and microfluidic cell imaging technology, and specifically relates to a method for deblurring and enhancing microfluidic flowing cell images. Background Technology

[0002] Microfluidic microscopy imaging systems can continuously acquire images of flowing cells within microscale channels and have been widely used in cell counting, cell sorting, cell morphology analysis, fluorescence intensity detection, and high-throughput cell identification. However, since cells in microfluidic channels typically move continuously in a fixed direction, when exposure time, local flow velocity, and pixel physical size work together, cells will undergo directional displacement during exposure, leading to problems such as flow direction blurring, boundary diffusion, blunting of cell nucleus outlines, and blurring of weak fluorescence structures in the acquired images.

[0003] Existing methods for deblurring microscopic images often employ fixed blur kernels, empirical blur kernels, full-image blind deconvolution, ordinary Laplacian regularization, or learned enhancement models. While these methods can improve the visual clarity of images, they typically do not fully utilize physical information such as local flow velocity, exposure time, pixel physical size, and channel orientation in microfluidic imaging. It is difficult to construct locally oriented blur kernels for the exposure displacement differences of different cell candidate regions. At the same time, uniform full-image deconvolution is prone to misrecovering background noise, channel wall texture, and non-ghosting areas, and introduces ringing artifacts.

[0004] Furthermore, microfluidic cell images are often used for cell segmentation, counting, classification, and quantitative fluorescence analysis. If the deblurring enhancement process only pursues visual clarity without constraining the grayscale or fluorescence intensity statistics of the cell region, it may change the overall cell response intensity and affect the subsequent quantitative analysis results. Therefore, there is a need for an image deblurring enhancement method that can combine the physical information of microfluidic flow imaging, local directional blurring features, confidence judgment of the trailing region, directional decomposition regularization constraint, uncertainty weighted fusion and backoff constraint, and grayscale or fluorescence intensity correction of the cell region. Summary of the Invention

[0005] The purpose of this invention is to provide a method for deblurring and enhancing microfluidic flowing cell images, in order to solve the problems of existing microscopic image deblurring methods in microfluidic flowing cell imaging scenarios, such as difficulty in adapting to different local exposure displacements of cells, easy to cause false recovery of non-blurred areas, easy to damage cell boundary structures, and easy to change cell grayscale or fluorescence intensity statistics.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for deblurring and enhancing microfluidic flow cell images, comprising the following steps.

[0007] S1. Acquire continuous flow cell images and adjacent frames, and determine the current image to be processed, the working image, and the microfluidic channel orientation reference.

[0008] S2. In the current image to be processed, determine the candidate cell regions, estimate the local flow velocity of each candidate cell region based on the local displacement of adjacent frames, calculate the directional blur length during exposure by combining the exposure time and pixel physical size, and construct a local directional blur kernel based on the directional blur length and the microfluidic channel directional reference.

[0009] S3. Using the directional ambiguity length as the flow direction sampling scale, construct a trailing confidence map by combining the flow direction trailing similarity, inter-frame displacement consistency, and edge extension degree.

[0010] S4. Based on the local directional blur kernel and the trailing confidence map, construct a trailing confidence weighted data fidelity term, which increases the deconvolution data fidelity weight in areas with high trailing confidence and decreases the deconvolution data fidelity weight in areas with low trailing confidence. Combine this with the second-order smoothing term of directional decomposition, the vertical flow edge protection adjustment function, and the constraint term of grayscale or fluorescence intensity statistics in the cell region to construct a variational deconvolution model.

[0011] S5. Generate multiple candidate local directional fuzzy kernels around the directional fuzzy length, solve the regional-level candidate restoration results respectively, construct an uncertainty map based on the consistency between the convolution reprojection residual and the candidate restoration results, fuse the candidate restoration results by weighting according to the uncertainty map, and apply backoff constraints to high uncertainty positions.

[0012] S6. Perform grayscale or fluorescence intensity correction on the fusion result for the cell region, and output the deblurred and enhanced image of the flowing cells.

[0013] Preferably, in step S1, the image input and flow direction reference determination module is constructed, specifically including: assuming the nth frame of the flowing cell image is... Its adjacent previous frame is The current image to be processed is The image pixel domain is The pixel position is ; Obtain exposure time Image acquisition frame rate The actual physical size 's' corresponding to a single pixel and the microfluidic channel orientation reference; the microfluidic channel orientation reference is the microfluidic channel orientation angle. One of the following: the channel centerline direction, the local tangential direction, or the direction field obtained by fitting the channel structure; when using the microfluidic channel orientation angle. At that time, construct the flow direction unit vector. and vertical flow unit vector Their expressions are as follows: ; When using the channel centerline direction, local tangential direction, or direction field, substitute the tangential angle of the corresponding position or cell candidate region into the above expression to obtain the corresponding flow direction unit vector and vertical flow direction unit vector; for the current image to be processed Perform grayscale normalization or fluorescence intensity normalization to obtain the normalized image to be processed. Its expression is: The working image used in the variational deconvolution model computation is defined as follows: The working image For non-negative intensity images; when performing deblurring enhancement in the original intensity domain, let When performing deblurring enhancement in the normalized intensity domain, let ; in, and These represent the minimum and maximum intensity values ​​in the current image to be processed, respectively. It is a positive stability constant used to avoid zero denominators or numerical instability during the calculation process.

[0014] Preferably, in step S1, by acquiring continuous flowing cell images and adjacent frames, the current image to be processed, the normalized image to be processed, and the working image are determined. Combined with exposure time, frame rate, pixel physical size, and microfluidic channel orientation reference, a flow direction unit vector and a vertical flow direction unit vector are constructed. The microfluidic channel orientation reference can adopt a fixed orientation angle, channel centerline direction, local tangential direction, or orientation field obtained by fitting the channel structure, providing a unified intensity basis and orientation reference for subsequent cell candidate region extraction, orientation blur length determination, trailing confidence map construction, and variational deconvolution recovery.

[0015] Preferably, in step S2, the cell candidate region and local orientation fuzzy kernel generation module is constructed, specifically including: S21. The initial region of the cell candidate region is obtained through at least one of the following methods: grayscale thresholding, fluorescence response thresholding, edge detection, connected component analysis, or existing cell segmentation results. The initial region is then filtered or corrected based on local displacements in adjacent frames. and Local displacement, extracting candidate cell regions ,in, Indicates the first Each cell candidate region The number of candidate cell regions is represented by the local displacement, which is obtained through optical flow, block matching, or phase correlation. For overlapping candidate cell regions, the overlapping pixels are assigned to the candidate cell region with the closest distance to the region center or the smallest difference between the local displacement and the average displacement of the region, thus obtaining a set of non-overlapping candidate cell regions. S22, Let the first The average inter-frame displacement vector of each candidate cell region is Based on the image acquisition frame rate Calculate the local flow velocity based on the actual physical size s corresponding to a single pixel. Its expression is: ; in, Represents the magnitude of the average inter-frame displacement vector; S23, Based on local flow velocity Exposure time Calculate the actual physical size s corresponding to a single pixel, and the first pixel. directional blur length of individual cell candidate regions during exposure Its expression is: ; in, This indicates rounding to the nearest integer. This means that the value will be truncated to the range [1, 11]. S24, with To maximize the support window, a reference is generated along the microfluidic channel direction. Corresponding single-cell local orientation blurry nucleus ,make The non-zero response length and the first The pixel displacement of each candidate cell region during exposure corresponds to the pixel displacement. It should be noted that the directional blur length... Not only used to generate localized, fuzzy nuclei for single cells Instead, it serves as a regional-level constraint on motion degradation throughout the subsequent recovery process. On the one hand... By the The local flow rate, exposure time, and pixel physical size of each cell candidate region are jointly determined to characterize the pixel displacement generated in that cell candidate region along the microfluidic channel direction during a single exposure; on the other hand... Further as a similarity measure of flow direction to shadow The sampling scale is used as the center length of the candidate directional blur kernel set, so that the motion blur confidence judgment, candidate recovery result generation and uncertainty backoff are all constrained by the same exposure displacement estimation result, thereby avoiding the blur kernel construction, motion blur judgment and candidate fusion being independent of each other.

[0016] Preferably, in step S2, initial candidate regions of cells are obtained by at least one of grayscale thresholding, fluorescence response thresholding, edge detection, connected component analysis, or existing cell segmentation results. The initial candidate regions are then screened or corrected by combining the local displacement between the current image to be processed and adjacent frames. Overlapping regions are uniquely assigned, and the average inter-frame displacement and local flow velocity of each candidate cell region are further calculated. The directional blur length during exposure is determined based on the local flow velocity, exposure time, and pixel physical size. A single-cell local directional blur kernel is constructed along the microfluidic channel directional reference, providing regional-level motion degradation constraints for subsequent trailing confidence map construction, variational deconvolution data fidelity terms, and candidate directional blur kernel generation.

[0017] Preferably, in step S3, the module for generating the motion cell trailing confidence map specifically includes: for pixel positions ,according to Along the microfluidic channel direction in directional ambiguity length Continuous grayscale similarity within the range constructs the flow path trailing similarity quantity According to pixel position Local displacement vector at point and Differences in constructing inter-frame displacement consistency and according to Gradients in the parallel and perpendicular flow directions construct edge extension. Their expressions are as follows: ; ; ; in, , , Represents the normalized image to be processed At pixel position Image gradient at that location, according to , and Construct the first Confidence map of motion cell trailing corresponding to candidate cell regions Its expression is: ; in, For displacement count index along the flow direction, A positive flow direction grayscale difference compression parameter. For positive displacement difference scaling parameters, , , These are non-negative weight parameters. The threshold for determining the confidence level of the ghosting; when When the value is a non-integer pixel, bilinear interpolation is used. However, when the value exceeds the image pixel domain... When using boundary replication or zero-normal boundary conditions, the values ​​are taken; for The pixel position, let Confidence map of motion cell trailing Used to indicate pixel position Belongs to the The reliability of flow from a candidate cell region to a trailing region differs from that of a binary mask used only for foreground selection. As continuous weights, they directly participate in the motion blur confidence-weighted data fidelity term, enabling the restored image, after convolution with a single-cell local directional blur kernel, to more strongly constrain its relationship with the working image within the motion blur confidence region. The consistency is improved, while the deconvolution recovery intensity is reduced in non-ghosting reliable regions, thereby reducing false recovery of background regions, channel wall textures and non-ghosting regions.

[0018] Preferably, in step S3, the directional blur length of each candidate cell region is used as the flow direction sampling scale. Combined with the continuous gray-level similarity of the normalized image to be processed along the microfluidic channel direction, the consistency between the local displacement of pixels and the average displacement of the region, and the edge extension degree formed by the parallel flow direction and the vertical flow direction gradient, a motion cell trailing confidence map is constructed. This provides a pixel-level reliable basis for subsequent data fidelity weighting, candidate restoration result weighted fusion, and high uncertainty position backoff constraint.

[0019] Preferably, in step S4, the directional decomposition variational deconvolution model module is constructed, specifically including: S41, awaiting the restoration of clear cell images As variables, based on the flow direction unit vector and vertical flow unit vector The image gradient is decomposed into parallel flow gradient and vertical flow gradient, and then based on the clear cell image to be recovered. Hessian matrix The second-order transformation of the image is decomposed into a second-order transformation along the parallel flow direction and a second-order transformation along the vertical flow direction, and their expressions are as follows: ; ; ; ; S42. Based on the vertical flow gradient Constructing a vertical flow-to-edge protection regulation function Its expression is: ; in, The edge protection adjustment parameter is positive; when When the value is large, reduce the second-order smoothness intensity at the corresponding position. When the size is small, the second-order smoothing constraint at the corresponding position is enhanced. In microfluidic flow cell images, motion blur mainly extends along the channel flow direction, while structures such as cell boundaries, cell nucleus outlines, and cell protrusions usually contain obvious vertical flow direction changes. Based on this imaging characteristic, this invention decomposes the image gradient and second-order changes into parallel flow direction components and vertical flow direction components, respectively, and uses the vertical flow direction gradient to construct an edge protection adjustment function, so that the second-order smoothing constraint is enhanced in the flat region and weakened in the vertical flow direction edge region, thereby reducing the risk of cell structure being over-smoothed while suppressing deconvolution ringing. S43, blurring the local orientation of a single cell nucleus confidence plot of motion cell trailing A data fidelity term is introduced, and a variational deconvolution model is constructed by combining a second-order smoothing term of directional decomposition and a gray-level or fluorescence intensity statistics constraint term; the optimization region composed of all candidate cell regions is defined as: The energy function of the variational deconvolution model is expressed as: ; in, Indicates the first The variational deconvolution energy function corresponding to the next iteration This indicates the data fidelity item for the confidence-weighted data of the ghosting effect. This represents the second-order smooth term in the directional decomposition. This represents a constraint term for the statistical measure of grayscale or fluorescence intensity. The weight parameters for the non-negative statistic constraint term are: The trailing confidence weighted data fidelity term is expressed as: The second-order smooth term of the directional decomposition is expressed as: When the grayscale or fluorescence intensity statistics constraint term adopts a total constraint, its expression is: ; in, The weight parameters are for the non-negative smoothing terms. This indicates the iteration number, and * indicates the convolution operation, used in the first iteration. In the above variational deconvolution model, the trailing confidence weighted data fidelity term From a single cell with a locally blurred nucleus confidence plot of motion cell trailing Common constraints are used to ensure that the restored image remains consistent with the working image within the trusted area of ​​the ghosting. Degeneracy consistency; directional decomposition of second-order smooth terms Structural continuity during the process is restored using second-order constraints on flow direction and vertical flow direction; grayscale or fluorescence intensity statistics are used as constraint terms. To constrain the statistical deviation between the restored results and the working image within the candidate cell region, three sub-items jointly constrain the image of the cell to be restored to a clear state from three aspects: degradation consistency, structural constraint, and quantitative preservation. This results in the recovery process being jointly constrained by regional-level motor degeneration, directional structure, and statistical relationships of cell strength.

[0020] Preferably, in step S4, the gradient and second-order changes of the cell image to be restored are decomposed by the unit vector of the flow direction and the unit vector of the vertical flow direction, and a vertical flow direction edge protection adjustment function is constructed. The single-cell local directional blur kernel, the confidence map of the moving cell trailing shadow, the second-order smoothing term of the directional decomposition, and the constraint term of the gray-level or fluorescence intensity statistics of the cell region are introduced into the variational deconvolution model to provide optimization constraints for the subsequent solution of candidate restoration results, weighted fusion, and high uncertainty position backoff constraints.

[0021] Preferably, in step S5, the module for constructing a weighted fusion and backoff constraint of candidate recovery results specifically includes: S51, Regarding the first Each cell candidate region, around the directional fuzzy length Multiple candidate directional blur lengths are generated. The number and interval of the candidate directional blur lengths are adaptively determined based on the local displacement estimation confidence, image noise level, or occlusion degree of the cell candidate region. The local displacement estimation confidence is determined by at least one of optical flow matching residual, block matching correlation peak, or phase correlation peak. Each candidate directional blur length is limited to the range [1, 11], thereby generating multiple candidate single-cell local directional blur kernels. ,in, For candidate kernel indexes, The number of locally oriented, ambiguous nuclei in candidate single cells, and ; S52, using working image As a cell image to be restored to clarity The initial value, for each candidate single cell local orientation blur kernel The computational domain of the trailing confidence weighted data fidelity term, the directional decomposition second-order smoothing term, and the gray-level or fluorescence intensity statistics constraint term in the variational deconvolution model is restricted to the [missing information - likely a specific region or number]. Cell candidate regions Within its convolutional neighborhood, and by minimizing the corresponding local energy function, we obtain the th... Regional-level candidate recovery results corresponding to each cell candidate region The regional-level candidate recovery results Represented in the global coordinate system, where Intra-pixel participates in iterative updates. External pixels are retained as working image Alternatively, a boundary replication method can be used to participate in the convolution calculation, and the region-level candidate restoration results can be restricted to those of the working image. The same range of intensity values; S53, according to and its corresponding For pixel position Calculate the convolution reprojection residual Its expression is: ; in, Indicates the first pixel position in each cell candidate region First The convolutional reprojection residuals of each candidate restoration result; S54. Construct local inconsistency based on the consistency among regional candidate restoration results and the convolutional reprojection residuals. Its expression is: ; in, Indicates the first The first cell candidate region Local inconsistency of candidate recovery results Indicates different Candidate recovery result index, and It is a non-negative adjustment parameter; S55. Construct an uncertainty graph with a value range of [0,1] based on the local inconsistency of the candidate recovery results at each regional level. Fusion weights with candidate recovery results Their expressions are as follows: ; ; in, For positive uncertainty compression parameters, Indicates the first The first cell candidate region Each candidate restoration result at pixel location The fusion weight at the location; S56. According to the uncertainty diagram Fusion weights with candidate recovery results The regional candidate restoration results are weighted and fused, and backoff constraints are applied to high uncertainty locations to obtain a fused, deblurred, and enhanced image. Its expression is: when hour, ;when hour, ; in, This indicates that the image is enhanced by fusion, deblurring, and pixelation at the pixel location. The intensity value at the location; during the iterative solution of the regional candidate restoration results, the iteration stops when the difference between two adjacent candidate restoration results is less than a preset stopping threshold, or when the number of iterations reaches a preset maximum number of iterations. Since the local velocity estimation of the cell candidate region may be affected by inter-frame matching errors, weak fluorescence noise, local cell deformation, or occlusion by adjacent cells, using only a single directional blur length may lead to oversharpening, under-restoration, or false structure in the restoration results. Therefore, this invention focuses on the directional blur length. Generate multiple candidate directional fuzzy kernels An uncertainty graph is constructed based on the convolutional reprojection residuals of each candidate recovery result and the consistency of the candidate results. When there are large differences between candidate restoration results or large convolutional reprojection residuals, it indicates that the restoration confidence at that location is low, and the fusion result is not compatible with the working image. Back off; when candidate results are consistent and convolution reprojection residuals are small, increase the contribution of candidate recovery results to fusion.

[0022] Preferably, in step S5, multiple candidate single-cell local directional blur kernels are generated around the directional blur length of each candidate cell region, and the region-level candidate restoration results are solved based on the variational deconvolution model constructed in S4. Then, an uncertainty map and fusion weights are constructed based on the convolution reprojection residual and the consistency of the candidate restoration results, and the candidate restoration results are weighted and fused. The uncertainty map is used to constrain the high uncertainty positions to backtrack to the working image, thereby reducing the false restoration caused by the blur kernel estimation bias.

[0023] Preferably, in step S6, the cell region intensity correction and image output module is constructed, specifically including: the grayscale or fluorescence intensity correction includes at least one statistical constraint among the total cell region intensity, average region intensity, grayscale distribution, or boundary contrast; when the total cell region constraint is used, the fused deblurred and enhanced image is... Each candidate region in the cell Perform grayscale or total fluorescence correction, correction factor Represented as: ;in, and These are the preset lower and upper limits of the correction coefficients, respectively, and they satisfy the following: The final deblurred and enhanced working domain image is obtained according to the following expression. :when hour, ;when hour, ,in, Indicates the first The grayscale or total fluorescence correction coefficient for each candidate cell region. This indicates the pixel location of the final deblurred and enhanced working domain image. The intensity value at the location; when Finally, the blurred and enhanced image of the flowing cells. Represented as: ;when Finally, the blurred and enhanced image of the flowing cells. Represented as: ; in, This indicates the pixel location of the final deblurred and enhanced fluid cell image. The intensity value at the location; Limiting the intensity range to the same range as the current image to be processed, for microfluidic flow cell images, the grayscale or fluorescence intensity statistics of cell regions are often used for subsequent cell counting, classification, or fluorescence intensity analysis. If only edges or details are enhanced during the deblurring and restoration process, it may change the overall response intensity of the cell region and affect the subsequent quantitative analysis results. Therefore, this invention calculates the working image separately for each candidate cell region after fusing the candidate restoration results. With fusion results The regional intensity statistical relationship was analyzed using correction coefficients. Regional intensity correction is performed on the fusion results to ensure that the restored image maintains the stability of the overall response relationship of the cellular region while enhancing structural clarity.

[0024] Preferably, in step S6, grayscale or fluorescence intensity correction is performed on each candidate cell region in the fused deblurred and enhanced image. The correction coefficient is obtained by calculating the intensity statistical relationship between the working image and the fusion result in the corresponding cell region, and the intensity of the fusion result is adjusted at the regional level. When the working image is a normalized image, the corrected working domain image is inversely transformed to the original intensity domain to obtain the deblurred and enhanced flowing cell image, thereby constraining the intensity drift of the cell region caused by the recovery process.

[0025] Compared with the prior art, the present invention has the following beneficial effects: The present invention determines the directional blur length based on the local flow rate, exposure time and pixel physical size of the cell candidate region, so that the local directional blur kernel corresponds to the cell exposure displacement; the directional blur length constrains the generation of the trailing confidence map and the candidate blur kernel, and the trailing confidence map is introduced into the data fidelity term, which can reduce the false recovery of non-trailing regions; through the second-order smoothing term of directional decomposition, the weighted fusion of the uncertainty of the candidate recovery result and the backoff constraint, and the correction of the gray level or fluorescence intensity of the cell region, the influence of the recovery process on the cell boundary structure and quantitative intensity information is reduced while suppressing deconvolution ringing and false structure. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the overall process of a microfluidic flow cell image deblurring and enhancement method according to the present invention.

[0027] Figure 2 This is a flowchart of the image input, working image determination, and flow direction unit vector construction in step S1 of this invention.

[0028] Figure 3 This is a flowchart of the cell candidate region extraction, local flow rate calculation, and single-cell local orientation fuzzy kernel construction in step S2 of this invention.

[0029] Figure 4 This is a flowchart of the construction of the motion cell trailing confidence map in step S3 of this invention.

[0030] Figure 5 This is a flowchart of the directional decomposition second-order smoothness constraint and variational deconvolution model construction in step S4 of this invention.

[0031] Figure 6 This is a flowchart of the candidate recovery result generation, uncertainty weighted fusion, and backoff constraint in step S5 of this invention.

[0032] Figure 7 This is a flowchart of the cell region grayscale or fluorescence intensity correction and final image output in step S6 of this invention. Detailed Implementation

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0034] This invention provides a method for deblurring and enhancing microfluidic flowing cell images, which includes image input and orientation reference construction, cell candidate region extraction, orientation blur length calculation and local orientation blur kernel construction, moving cell trailing confidence map construction, orientation decomposition variational deconvolution model construction, candidate restoration result weighted fusion and high uncertainty position backoff constraint, and cell region grayscale or fluorescence intensity correction.

[0035] Please see Figure 1 , Figure 1 The overall flow of a microfluidic flow cell image deblurring and enhancement method according to an embodiment of the present invention is shown.

[0036] S1. Acquire continuous flow cell images and adjacent frames, and determine the current image to be processed, the working image, and the microfluidic channel orientation reference.

[0037] Furthermore, such as Figure 2As shown, the image input and flow direction reference determination module is constructed. The specific steps are as follows: Let the nth frame of the flowing cell image be... Its adjacent previous frame is The current image to be processed is The image pixel domain is The pixel position is ; Obtain exposure time Image acquisition frame rate The actual physical size 's' corresponding to a single pixel and the microfluidic channel orientation reference; the microfluidic channel orientation reference is the microfluidic channel orientation angle. One of the following: the channel centerline direction, the local tangential direction, or the direction field obtained by fitting the channel structure; when using the microfluidic channel orientation angle. At that time, construct the flow direction unit vector. and vertical flow unit vector Their expressions are as follows: ; When using the channel centerline direction, local tangential direction, or direction field, substitute the tangential angle of the corresponding position or cell candidate region into the above expression to obtain the corresponding flow direction unit vector and vertical flow direction unit vector; for the current image to be processed After performing grayscale normalization or fluorescence intensity normalization, the normalized image F to be processed is obtained, and its expression is: The working image used in the variational deconvolution model computation is defined as follows: The working image For non-negative intensity images; when performing deblurring enhancement in the original intensity domain, let When performing deblurring enhancement in the normalized intensity domain, let ;in, and These represent the minimum and maximum intensity values ​​in the current image to be processed, respectively. This is a positive stability constant used to avoid zero denominators or numerical instability during calculation. In this embodiment, the continuous flow cell image is... Pixel grayscale image or fluorescence image, image acquisition frame rate Pick Exposure time Pick The actual physical size s corresponding to a single pixel is taken as Microfluidic channel orientation angle Pick Positive stability constant Pick .

[0038] S2. In the current image to be processed, determine the candidate cell regions, estimate the local flow velocity of each candidate cell region based on the local displacement of adjacent frames, calculate the directional blur length during exposure by combining the exposure time and pixel physical size, and construct a local directional blur kernel based on the directional blur length and the microfluidic channel directional reference.

[0039] Furthermore, such as Figure 3 As shown, the module for constructing cell candidate regions and locally oriented fuzzy kernels includes the following sub-steps: S21. The initial region of the cell candidate region is obtained through at least one of the following methods: grayscale thresholding, fluorescence response thresholding, edge detection, connected component analysis, or existing cell segmentation results. The initial region is then filtered or corrected based on local displacements in adjacent frames. and Local displacement, extracting candidate cell regions ,in, Indicates the first Each cell candidate region The number of candidate cell regions is represented by the local displacement, which is obtained through optical flow, block matching, or phase correlation. For overlapping candidate cell regions, the overlapping pixels are assigned to the candidate cell region with the closest distance to the region center or the smallest difference between the local displacement and the average displacement of the region, thus obtaining a set of non-overlapping candidate cell regions. S22, Let the first The average inter-frame displacement vector of each candidate cell region is Based on the image acquisition frame rate Calculate the local flow velocity based on the actual physical size s corresponding to a single pixel. Its expression is: ;in, Represents the magnitude of the average inter-frame displacement vector; in this embodiment, the first... The average inter-frame displacement vector magnitude of each cell candidate region Pick Then its local flow velocity for: ; S23, Based on local flow velocity Exposure time Calculate the actual physical size s corresponding to a single pixel, and the first pixel. directional blur length of individual cell candidate regions during exposure Its expression is: ;in, This indicates rounding to the nearest integer. This means that the value will be truncated to the range [1, 11]. S24, with To maximize the support window, a reference is generated along the microfluidic channel direction. Corresponding single-cell local orientation blurry nucleus ,make The non-zero response length and the first The pixel displacement of each cell candidate region during exposure corresponds to the pixel displacement. In this embodiment, according to... , and Calculate the first The directional fuzzy length of each cell candidate region for: Therefore, a single-cell local orientation fuzzy kernel with a non-zero response length of 3 is generated. It should be noted that the directional ambiguity length Not only used to generate localized, fuzzy nuclei for single cells Instead, it serves as a regional-level constraint on motion degradation throughout the subsequent recovery process. On the one hand... By the The local flow rate, exposure time, and pixel physical size of each cell candidate region are jointly determined to characterize the pixel displacement generated in that cell candidate region along the microfluidic channel direction during a single exposure; on the other hand... Further as a similarity measure of flow direction to shadow The sampling scale is used as the center length of the candidate directional blur kernel set, so that the motion blur confidence judgment, candidate recovery result generation and uncertainty backoff are all constrained by the same exposure displacement estimation result, thereby avoiding the blur kernel construction, motion blur judgment and candidate fusion being independent of each other.

[0040] S3. Using the directional ambiguity length as the flow direction sampling scale, construct a trailing confidence map by combining the flow direction trailing similarity, inter-frame displacement consistency, and edge extension degree.

[0041] Furthermore, such as Figure 4 As shown, the module for generating confidence maps of moving cell trails is constructed. The specific steps are as follows: for pixel positions... ,according to Along the microfluidic channel direction in directional ambiguity length Continuous grayscale similarity within the range constructs the flow path trailing similarity quantity According to pixel position Local displacement vector at point and Differences in constructing inter-frame displacement consistency And construct the edge extension amount based on the gradient of F in the parallel flow direction and the perpendicular flow direction. Their expressions are as follows: ; ; ; in, , , This indicates the pixel position of the normalized image F to be processed. Image gradient at that location, according to , and Construct the first Confidence map of motion cell trailing corresponding to candidate cell regions Its expression is: ; in, For displacement count index along the flow direction, A positive flow direction grayscale difference compression parameter. For positive displacement difference scaling parameters, , , These are non-negative weight parameters. The threshold for determining the confidence level of the ghosting; when When the value is a non-integer pixel, bilinear interpolation is used. However, when the value exceeds the image pixel domain... When using boundary replication or zero-normal boundary conditions, the values ​​are taken; for The pixel position, let In this embodiment, the flow direction grayscale difference compression parameter Set to 0.05, displacement difference scaling parameter Use 4.0, a non-negative weight parameter. , , The trailing image confidence thresholds were set to 0.40, 0.35, and 0.25 respectively. 0.50, confidence plot of motor cell trailing Used to indicate pixel position Belongs to the The reliability of flow from a candidate cell region to a trailing region differs from that of a binary mask used only for foreground selection. As continuous weights, they directly participate in the motion blur confidence-weighted data fidelity term, enabling the restored image, after convolution with a single-cell local directional blur kernel, to more strongly constrain its relationship with the working image within the motion blur confidence region. The consistency is improved, while the deconvolution recovery intensity is reduced in non-ghosting reliable regions, thereby reducing false recovery of background regions, channel wall textures and non-ghosting regions.

[0042] S4. Based on the local directional blur kernel and the trailing confidence map, construct a trailing confidence weighted data fidelity term, which increases the deconvolution data fidelity weight in areas with high trailing confidence and decreases the deconvolution data fidelity weight in areas with low trailing confidence. Combine this with the second-order smoothing term of directional decomposition, the vertical flow edge protection adjustment function, and the constraint term of grayscale or fluorescence intensity statistics in the cell region to construct a variational deconvolution model.

[0043] Furthermore, such as Figure 5 As shown, the module for constructing the directional decomposition variational deconvolution model includes the following sub-steps.

[0044] S41, awaiting the restoration of clear cell images As variables, based on the flow direction unit vector and vertical flow unit vector The image gradient is decomposed into parallel flow gradient and vertical flow gradient, and then based on the clear cell image to be recovered. Hessian matrix The second-order transformation of the image is decomposed into a second-order transformation along the parallel flow direction and a second-order transformation along the vertical flow direction, and their expressions are as follows: ; ; ; ; S42. Based on the vertical flow gradient Constructing a vertical flow-to-edge protection regulation function Its expression is: ; in, The edge protection adjustment parameter is positive; when When the value is large, reduce the second-order smoothness intensity at the corresponding position. When the value is small, the second-order smoothness constraint at the corresponding position is enhanced; in this embodiment, the edge protection adjustment parameter is... Taking 4.0, in microfluidic flow cell images, motion blur mainly extends along the channel flow direction, while structures such as cell boundaries, cell nucleus outlines, and cell protrusions usually contain significant vertical flow direction changes. Based on this imaging characteristic, this invention decomposes the image gradient and second-order changes into parallel flow direction components and vertical flow direction components, respectively, and uses the vertical flow direction gradient to construct an edge protection adjustment function, which enhances the second-order smoothing constraint in flat regions and weakens it in vertical flow direction edge regions, thereby reducing the risk of cell structures being over-smoothed while suppressing deconvolution ringing; S43, local directional blur kernel of single cells confidence plot of motion cell trailing A data fidelity term is introduced, and a variational deconvolution model is constructed by combining a second-order smoothing term of directional decomposition and a gray-level or fluorescence intensity statistics constraint term; the optimization region composed of all candidate cell regions is defined as: The energy function of the variational deconvolution model is expressed as: ; in, Indicates the first The variational deconvolution energy function corresponding to the next iteration This indicates the data fidelity item for the confidence-weighted data of the ghosting effect. This represents the second-order smooth term in the directional decomposition. This represents a constraint term for the statistical measure of grayscale or fluorescence intensity. The weight parameters for the non-negative statistic constraint term are: The trailing confidence weighted data fidelity term is expressed as: The second-order smooth term of the directional decomposition is expressed as: When the grayscale or fluorescence intensity statistics constraint term adopts a total constraint, its expression is: ; in, The weight parameters are for the non-negative smoothing terms. This indicates the iteration number, and * indicates the convolution operation, used in the first iteration. In this embodiment, the smoothing term weight parameter Take 0.02 as the weight parameter for the conservation term. Set the value to 0.10, the maximum number of iterations to 30, and the stopping threshold to [value missing]. In the above variational deconvolution model, the trailing confidence weighted data fidelity term From a single cell with a locally blurred nucleus confidence plot of motion cell trailing Common constraints are used to ensure that the restored image remains consistent with the working image within the trusted area of ​​the ghosting. Degeneracy consistency; directional decomposition of second-order smooth terms Structural continuity during the process is restored using second-order constraints on flow direction and vertical flow direction; grayscale or fluorescence intensity statistics are used as constraint terms. To constrain the statistical deviation between the restored results and the working image within the candidate cell region, three sub-items jointly constrain the image of the cell to be restored to a clear state from three aspects: degradation consistency, structural constraint, and quantitative preservation. This results in the recovery process being jointly constrained by regional-level motor degeneration, directional structure, and statistical relationships of cell strength.

[0045] S5. Generate multiple candidate local directional fuzzy kernels around the directional fuzzy length, solve the regional-level candidate restoration results respectively, construct an uncertainty map based on the consistency between the convolution reprojection residual and the candidate restoration results, fuse the candidate restoration results by weighting according to the uncertainty map, and apply backoff constraints to high uncertainty positions.

[0046] Furthermore, such as Figure 6As shown, the module for constructing weighted fusion and backoff constraints of candidate recovery results includes the following sub-steps: S51, Regarding the first Each cell candidate region, around the directional fuzzy length Multiple candidate directional blur lengths are generated. The number and interval of the candidate directional blur lengths are adaptively determined based on the local displacement estimation confidence, image noise level, or occlusion degree of the cell candidate region. The local displacement estimation confidence is determined by at least one of optical flow matching residual, block matching correlation peak, or phase correlation peak. Each candidate directional blur length is limited to the range [1, 11], thereby generating multiple candidate single-cell local directional blur kernels. ,in, For candidate kernel indexes, The number of locally oriented, ambiguous nuclei in candidate single cells, and In this embodiment, when the confidence level of the local displacement estimation is high, the fuzzy length of the candidate direction can be taken as... , and When the confidence level of local displacement estimation is low, the image noise is strong, or the candidate cell region is occluded, the candidate length interval or the number of candidate lengths is increased, and duplicate candidate lengths are deleted. The blur length of each candidate direction is limited to the range of [1, 11]; the number of candidate direction blur kernels Choose 3, non-negative adjustment parameter , The uncertainty compression parameter is set to 0.60 and 0.40 respectively. Set to 1.0; S52, with working image As a cell image to be restored to clarity The initial value, for each candidate single cell local orientation blur kernel The computational domain of the trailing confidence weighted data fidelity term, the directional decomposition second-order smoothing term, and the gray-level or fluorescence intensity statistics constraint term in the variational deconvolution model is restricted to the [missing information - likely a specific region or number]. Cell candidate regions Within its convolutional neighborhood, and by minimizing the corresponding local energy function, we obtain the th... Regional-level candidate recovery results corresponding to each cell candidate region The regional-level candidate recovery results Represented in the global coordinate system, where Intra-pixel participates in iterative updates. External pixels are retained as working image Alternatively, a boundary replication method can be used to participate in the convolution calculation, and the region-level candidate restoration results can be restricted to those of the working image. Within the same intensity range, in this embodiment, the local energy function can be iteratively minimized using gradient descent, conjugate gradient, or alternating direction multiplier methods. After each iteration, the region-level candidate restoration results are restricted to those of the working image. The same range of intensity values; S53, according to and its corresponding For pixel position Calculate the convolution reprojection residual Its expression is: ; in, Indicates the first pixel position in each cell candidate region First S54. Convolutional reprojection residuals of each candidate restoration result; S55. Construct local inconsistency based on the consistency between regional candidate restoration results and the convolutional reprojection residuals. Its expression is: ; in, Indicates the first The first cell candidate region Local inconsistency of candidate recovery results Indicates different Candidate recovery result index, and S55. Construct an uncertainty plot with a value range of [0,1] based on the local inconsistency of the candidate recovery results at each region level. Fusion weights with candidate recovery results Their expressions are as follows: ; ; in, For positive uncertainty compression parameters, Indicates the first The first cell candidate region Each candidate restoration result at pixel location Fusion weights at the point; S56, based on the uncertainty graph Fusion weights with candidate recovery results The regional candidate restoration results are weighted and fused, and backoff constraints are applied to high uncertainty locations to obtain a fused, deblurred, and enhanced image. Its expression is: when hour, ;when hour, ;in, This indicates that the image is enhanced by fusion, deblurring, and pixelation at the pixel location. The intensity value at the location; during the iterative solution of regional candidate restoration results, the iteration stops when the difference between two adjacent candidate restoration results is less than the preset stopping threshold, or when the number of iterations reaches the preset maximum number of iterations.

[0047] S6. Perform grayscale or fluorescence intensity correction on the fusion result for the cell region, and output the deblurred and enhanced image of the flowing cells.

[0048] Furthermore, such as Figure 7 As shown, a cell region intensity correction and image output module is constructed. The specific steps are as follows: the grayscale or fluorescence intensity correction includes at least one statistical constraint among the total cell region intensity, average region intensity, grayscale distribution, or boundary contrast; when using the total cell region constraint, the fused, deblurred, and enhanced image is... Each candidate region in the cell Perform grayscale or total fluorescence correction, correction factor Represented as: ;in, and These are the preset lower and upper limits of the correction coefficients, respectively, and they satisfy the following: In this embodiment, the lower limit of the correction coefficient is... Take 0.85 as the upper limit of the correction factor. Take 1.15, and obtain the final deblurred and enhanced working domain image according to the following expression. :when hour, ;when hour, ; in, Indicates the first The grayscale or total fluorescence correction coefficient for each candidate cell region. This indicates the pixel location of the final deblurred and enhanced working domain image. The intensity value at the location; when Finally, the blurred and enhanced image of the flowing cells. Represented as: ;when Finally, the blurred and enhanced image of the flowing cells. Represented as: ; in, This indicates the pixel location of the final deblurred and enhanced fluid cell image. The intensity value at the location; Limit the intensity range to the same range as the current image to be processed.

[0049] For microfluidic flow cell images, the grayscale or fluorescence intensity statistics of cell regions are often used for subsequent cell counting, classification, or fluorescence intensity analysis. If only edges or details are enhanced during the deblurring and restoration process, it may change the overall response intensity of the cell region and affect the results of subsequent quantitative analysis. Therefore, this invention calculates the working image separately for each candidate cell region after fusing the candidate restoration results. With fusion results The regional intensity statistical relationship was analyzed using correction coefficients. Regional intensity correction is performed on the fusion results to ensure that the restored image maintains the stability of the overall response relationship of the cellular region while enhancing structural clarity.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various changes, modifications, substitutions and variations can be made without departing from the principles and spirit of the present invention, and all such changes, modifications, substitutions and variations should fall within the protection scope of the present invention.

Claims

1. A method for deblurring and enhancing microfluidic flow cell images, characterized in that, Includes the following steps: S1. Acquire continuous flow cell images and adjacent frames, and determine the current image to be processed, the working image, and the microfluidic channel orientation reference; S2. In the current image to be processed, determine the candidate cell regions, estimate the local flow velocity of each candidate cell region based on the local displacement of adjacent frames, calculate the directional blur length during exposure by combining the exposure time and pixel physical size, and construct a local directional blur kernel based on the directional blur length and the microfluidic channel directional reference. S3. Using the directional blur length as the flow direction sampling scale, construct a trailing confidence map by combining the flow direction trailing similarity, inter-frame displacement consistency and edge extension degree. S4. Based on the local directional blur kernel and the trailing confidence map, construct a trailing confidence weighted data fidelity term, so that the deconvolution data fidelity weight is increased in areas with high trailing confidence and decreased in areas with low trailing confidence. Combine this with the second-order smoothing term of directional decomposition, the vertical flow edge protection adjustment function, and the constraint term of gray level or fluorescence intensity statistics in the cell region to construct a variational deconvolution model. S5. Generate multiple candidate local directional fuzzy kernels around the directional fuzzy length, solve the regional-level candidate restoration results respectively, and construct an uncertainty graph based on the consistency between the convolution reprojection residual and the candidate restoration results, and perform weighted fusion and backoff constraints. S6. Perform grayscale or fluorescence intensity correction on the fusion result for the cell region, and output the deblurred and enhanced image of the flowing cells.

2. The method for deblurring and enhancing microfluidic flow cell images according to claim 1, characterized in that, In step S1, let the nth frame of the flowing cell image be... Its adjacent previous frame is The current image to be processed is The image pixel domain is The pixel position is ; Obtain exposure time Image acquisition frame rate The actual physical size 's' corresponding to a single pixel and the microfluidic channel orientation reference; the microfluidic channel orientation reference is the microfluidic channel orientation angle. One of the following: the channel centerline direction, the local tangential direction, or the direction field obtained by fitting the channel structure; when using the microfluidic channel orientation angle. At that time, construct the flow direction unit vector. and vertical flow unit vector Their expressions are as follows: ; When using the channel centerline direction, local tangential direction, or orientation field, substitute the tangential angle of the corresponding position or cell candidate region into the above expression to obtain the corresponding flow direction unit vector and vertical flow direction unit vector; for the current image to be processed Perform grayscale normalization or fluorescence intensity normalization to obtain the normalized image to be processed. Its expression is: ; Define the working image used in the variational deconvolution model computation as: The working image For non-negative intensity images; when performing deblurring enhancement in the original intensity domain, let ; When performing deblurring enhancement in the normalized intensity domain, let ;in, and These represent the minimum and maximum intensity values ​​in the current image to be processed, respectively. It is a positive stability constant used to avoid zero denominators or numerical instability during calculation.

3. The method for deblurring and enhancing microfluidic flow cell images according to claim 2, characterized in that, Step S2 includes the following steps: S21. The initial region of the cell candidate region is obtained through at least one of the following methods: grayscale thresholding, fluorescence response thresholding, edge detection, connected component analysis, or existing cell segmentation results. The initial region is then filtered or corrected based on local displacements in adjacent frames. and Local displacement, extracting candidate cell regions ,in, Indicates the first Each cell candidate region The number of candidate cell regions is represented by the local displacement, which is obtained through optical flow, block matching, or phase correlation. For overlapping candidate cell regions, the overlapping pixels are assigned to the candidate cell region with the closest distance to the region center or the smallest difference between the local displacement and the average displacement of the region, thus obtaining a set of non-overlapping candidate cell regions. S22, Let the first The average inter-frame displacement vector of each candidate cell region is Based on the image acquisition frame rate Calculate the local flow velocity based on the actual physical size s corresponding to a single pixel. Its expression is: ;in, This represents the magnitude of the average inter-frame displacement vector; S23, Based on local flow velocity Exposure time Calculate the actual physical size s corresponding to a single pixel, and the first pixel. directional blur length of individual cell candidate regions during exposure Its expression is: ; in, This indicates rounding to the nearest integer. This means that the value will be truncated to the range [1, 11]. S24, with To maximize the support window, a reference is generated along the microfluidic channel direction. Corresponding single-cell local orientation blurry nucleus ,make The non-zero response length and the first The pixel displacement of each cell candidate region corresponds to the exposure time.

4. The method for deblurring and enhancing microfluidic flow cell images according to claim 3, characterized in that, In step S3, for pixel position ,according to Along the microfluidic channel direction in directional ambiguity length Continuous grayscale similarity within the range constructs the flow path trailing similarity quantity According to pixel position Local displacement vector at point and Differences in constructing inter-frame displacement consistency and according to Gradients in the parallel and perpendicular flow directions construct edge extension. Their expressions are as follows: ; ; ; in, , , Represents the normalized image to be processed At pixel position Image gradient at that location, according to , and Construct the first Confidence map of motion cell trailing corresponding to candidate cell regions Its expression is: ; in, For displacement count index along the flow direction, A positive flow direction grayscale difference compression parameter. For positive displacement difference scaling parameters, , , These are non-negative weight parameters. The threshold for determining the confidence level of motion blur; when When the value is a non-integer pixel, bilinear interpolation is used. However, when the value exceeds the image pixel domain... When using boundary replication or zero-normal boundary conditions, the values ​​are taken; for The pixel position, let .

5. The method for deblurring and enhancing microfluidic flow cell images according to claim 4, characterized in that, Step S4 includes the following steps: S41, awaiting the restoration of clear cell images As variables, based on the flow direction unit vector and vertical flow unit vector The image gradient is decomposed into parallel flow gradient and vertical flow gradient, and then based on the clear cell image to be recovered. Hessian matrix The second-order transformation of the image is decomposed into a second-order transformation along the parallel flow direction and a second-order transformation along the vertical flow direction, and their expressions are as follows: ； ； ； ； S42. Based on the vertical flow gradient Constructing a vertical flow-to-edge protection regulation function Its expression is: ; in, The edge protection adjustment parameter is positive; when When the value is large, reduce the second-order smoothness intensity at the corresponding position. When the value is small, enhance the second-order smoothness constraint at the corresponding position; S43, blurring the local orientation of a single cell nucleus confidence plot of motion cell trailing A data fidelity term is introduced, and a variational deconvolution model is constructed by combining a second-order smoothing term of directional decomposition and a gray-level or fluorescence intensity statistics constraint term; the optimization region composed of all candidate cell regions is defined as: The energy function of the variational deconvolution model is expressed as: ; in, Indicates the first The variational deconvolution energy function corresponding to the next iteration This indicates the data fidelity item for the confidence-weighted data of the ghosting effect. This represents the second-order smooth term in the directional decomposition. This represents a constraint term for the statistical measure of grayscale or fluorescence intensity. The weight parameters for the non-negative statistic constraint term are: The trailing confidence weighted data fidelity term is expressed as: The second-order smooth term of the directional decomposition is expressed as: When the grayscale or fluorescence intensity statistics constraint term adopts a total constraint, its expression is: ; in, The weight parameters are for the non-negative smoothing term. Indicates the iteration number. This represents the convolution operation, during the first iteration. .

6. The method for deblurring and enhancing microfluidic flow cell images according to claim 5, characterized in that, Step S5 includes the following steps: S51, Regarding the first Each cell candidate region, around the directional fuzzy length Multiple candidate directional blur lengths are generated. The number and interval of the candidate directional blur lengths are adaptively determined based on the local displacement estimation confidence, image noise level, or occlusion degree of the cell candidate region. The local displacement estimation confidence is determined by at least one of optical flow matching residual, block matching correlation peak, or phase correlation peak. Each candidate directional blur length is limited to the range [1, 11], thereby generating multiple candidate single-cell local directional blur kernels. ,in, For candidate kernel indexes, The number of locally oriented, ambiguous nuclei in candidate single cells, and ; S52, using working image As a cell image to be restored to clarity The initial value, for each candidate single cell local orientation blur kernel The computational domain of the trailing confidence weighted data fidelity term, the directional decomposition second-order smoothing term, and the gray-level or fluorescence intensity statistics constraint term in the variational deconvolution model is restricted to the [missing information - likely a specific region or number]. Cell candidate regions Within its convolutional neighborhood, and by minimizing the corresponding local energy function, we obtain the th... Regional-level candidate recovery results corresponding to each cell candidate region The regional-level candidate recovery results Represented in the global coordinate system, in Intra-pixel participates in iterative updates. External pixels are retained as working image Alternatively, a boundary replication method can be used to participate in the convolution calculation, and the region-level candidate restoration results can be restricted to those of the working image. The same range of intensity values; S53, according to and its corresponding For pixel position Calculate the convolution reprojection residual Its expression is: ; in, Indicates the first pixel position in each cell candidate region First The convolutional reprojection residuals of each candidate restoration result; S54. Construct local inconsistency based on the consistency among regional candidate restoration results and the convolutional reprojection residuals. Its expression is: ; in, Indicates the first The first cell candidate region Local inconsistency of candidate recovery results Indicates different Candidate recovery result index, and It is a non-negative adjustment parameter; S55. Construct an uncertainty graph with a value range of [0,1] based on the local inconsistency of the candidate recovery results at each regional level. Fusion weights with candidate recovery results Their expressions are as follows: ; ; in, For positive uncertainty compression parameters, Indicates the first The first cell candidate region Each candidate restoration result at pixel location The fusion weight at the location; S56. According to the uncertainty diagram Fusion weights with candidate recovery results The regional candidate restoration results are weighted and fused, and backoff constraints are applied to high uncertainty locations to obtain a fused, deblurred, and enhanced image. Its expression is: when hour, ;when hour, ; in, This indicates that the image is enhanced by fusion, deblurring, and pixelation at the pixel location. The intensity value at the location; during the iterative solution of regional candidate restoration results, the iteration stops when the difference between two adjacent candidate restoration results is less than the preset stopping threshold, or when the number of iterations reaches the preset maximum number of iterations.

7. The method for deblurring and enhancing microfluidic flow cell images according to claim 6, characterized in that, In step S6, the grayscale or fluorescence intensity correction includes at least one statistical constraint among total cell area, average regional intensity, grayscale distribution, or boundary contrast; when using the total cell area constraint, the fused deblurred and enhanced image is... Each candidate region in the cell Perform grayscale or total fluorescence correction, correction factor Represented as: ; in, and These are the preset lower and upper limits of the correction coefficients, respectively, and they satisfy the following: The final deblurred and enhanced working domain image is obtained according to the following expression. : when hour, ;when hour, ; in, Indicates the first The grayscale or total fluorescence correction coefficient for each candidate cell region. This indicates the pixel location of the final deblurred and enhanced working domain image. The intensity value at the location; when Finally, the blurred and enhanced image of the flowing cells. Represented as: ;when Finally, the blurred and enhanced image of the flowing cells. Represented as: ;in, This indicates the pixel location of the final deblurred and enhanced fluid cell image. The intensity value at the location; Limit the intensity range to the same range as the current image to be processed.

Images