Image reconstruction method and system based on topological structure

Abstract

The invention discloses an image reconstruction method and system based on a topological structure, and belongs to the field of image reconstruction. Comprising the following steps: establishing a digital holographic microscopic imaging model; converting the three-dimensional space topological data into a multi-layer graph structure imaged at different focal lengths through an adaptive focusing algorithm to form a hierarchical graph model; calculating an object deflection angle according to topological data of the hierarchical graph model; performing angle correction on the hierarchical graph model according to the deflection angle to obtain a corrected hierarchical graph model; and calculating optimal imaging focus positions of different angles in the corrected hierarchical graph model, and performing surface topology reconstruction according to the optimal imaging focus positions to obtain a reconstructed image after angle adjustment. In the traditional digital holographic microscopic imaging, due to the limitation of the light scattering angle, the side edge and the large-angle inclined part of an object cannot be completely recorded, so that edge distortion or information loss of a reconstructed image occurs. According to the invention, the imaging limitation during the processing of a complex curved surface object is solved.

Description

Technical Field

[0001] The present application relates to the field of image reconstruction, and more specifically, to a method and system for image reconstruction based on topological structure. Background Art

[0002] In the fields of image processing and computer vision, image angle correction technology is an important link to ensure image quality and accuracy. Traditional image angle correction technology is mainly based on geometric transformation methods, that is, angle correction is performed through manual adjustment or by using geometric features in the image (such as straight lines and edges), such as using Hough transform to detect straight lines and calculate the tilt angle. However, this type of method relies on manual intervention or specific scene features and has poor robustness. There is also correction based on template matching, which determines the reference direction of the target object through predefined templates (such as QR codes and calibration plates), but it requires a fixed template and has low adaptability to complex scenes. These methods generally have the technical defects of relying on manually set rules and having difficulty handling angle deviations in dynamic and unstructured environments.

[0003] In the field of microscopic dynamic imaging, when a vertically incident laser light source is projected onto a spherical or curved object, the principle of light reflection prevents the camera from promptly receiving information about the side edges of the curved surface. This results in the spherical or curved object losing some information during image reconstruction, leaving only the top surface information, resulting in a "cream tip" effect. To address this technical difficulty, existing spherical surface measurements often avoid vertical incidence solutions and opt for triangulation solutions, which replace the missing information by varying the angles between the light source and the camera. However, triangulation solutions have large space requirements, are highly sensitive to relative measurement angles, rely heavily on the number of pre-acquired images, and take a long time to reconstruct the resulting image. This significantly constrains and limits their application in high-precision dynamic measurement systems.

[0004] The Chinese patent application, application number CN202411476815.7, published on February 11, 2025, discloses a compensation device and method for a digital holographic microscope, which relates to the field of optical imaging technology. By using a deformable mirror group and real-time wavefront sensing technology, the compensation device can dynamically adjust the optical response according to the real-time measured wavefront distortion data. The adaptive optical element precisely controls the light wavefront, effectively expands the imaging depth and optimizes the focus control, thereby enhancing the resolution. In addition, the compensation phase algorithm calculates the required compensation by introducing a regularization term, which not only improves the algorithm's adaptability to environmental changes, but also enhances the robustness of the algorithm during the imaging process. However, this solution lacks an automatic angle detection and correction mechanism, so the reconstruction accuracy of this solution needs to be further improved. Summary of the Invention

[0005] In traditional digital holographic microscopy, when the laser is incident perpendicularly on the surface of a curved object, the side edges and parts tilted at large angles are often not fully recorded due to the limited light scattering angle, resulting in edge distortion or information loss in the reconstructed image. This application provides a topological structure-based image reconstruction method and system. By establishing a digital holographic microscopy imaging model and forming a hierarchical graph model, it can accurately extract the deflection angle of the object and perform intelligent correction on the angle. This can effectively compensate for the loss of side information caused by the perpendicular incidence of the laser, achieving high-precision reconstruction of the complete surface of spherical or curved objects, and solving the imaging limitations of existing technologies when processing complex curved objects.

[0006] One aspect of the present application provides an image reconstruction method based on a topological structure, comprising: establishing a digital holographic microscopy imaging model, wherein the digital holographic microscopy imaging model provides three-dimensional spatial topological data of an object by recording and reconstructing the light field information of the object; converting the three-dimensional spatial topological data into a multi-layer graph structure imaged at different focal lengths through an adaptive focusing algorithm to form a hierarchical graph model; calculating the object deflection angle according to the topological data of the hierarchical graph model; performing angle correction on the hierarchical graph model according to the deflection angle to obtain a corrected hierarchical graph model; calculating the optimal imaging focus positions at different angles in the corrected hierarchical graph model, performing surface topology reconstruction according to the optimal imaging focus position, and obtaining a reconstructed image after angle adjustment.

[0007] In particular, by establishing a digital holographic microscopy imaging model, not only the intensity information of light is recorded, but also the phase information is obtained at the same time, which can completely capture the light field distribution of the object and provide complete three-dimensional spatial topological data for the subsequent all-round reconstruction of curved objects, overcoming the limitation of traditional imaging methods that can only obtain two-dimensional information.

[0008] Furthermore, a hierarchical graph model is formed, including: using the following formula to convert the three-dimensional spatial topological data into a multi-layer graph structure imaged at different focal lengths:

[0009]

[0010] Where d represents the different focal lengths of the input; F is the Fourier transform, F filtered is the selected spectral series, E R ' is the reference wavefront, I H is the input hologram intensity information, λ is the wavelength, (x, y) and (f x ,f y) are the coordinates corresponding to the time domain and frequency domain respectively; i represents a positive integer; each pixel is taken as a node, and connections are established between adjacent pixel nodes. The edge weights are determined by phase continuity to construct a micro-layer structure; similar regions are merged through the superpixel segmentation algorithm, and the nodes are defined as local surface patches. Connections between superpixel nodes are established through curvature similarity or spatial proximity to construct a meso-layer structure; the micro-layer structure and the meso-layer structure are integrated to obtain a hierarchical graph model.

[0011] In particular, this application achieves effective fusion of multi-scale information by constructing a dual-layer structure consisting of micro and meso layers. The micro layer captures high-frequency details using pixels as the basic unit, while the meso layer acquires low-frequency structural information using local surface patches as the unit. This multi-scale representation method enables the system to simultaneously process high-frequency details (such as microtexture) and low-frequency features (such as overall shape).

[0012] Furthermore, the hierarchical graph model maintains spatial topology through edge weights, a crucial property in the angle correction process. While traditional methods can easily introduce topological discontinuities during angle transformation, this method ensures topological consistency through phase continuity and curvature similarity constraints, avoiding reconstruction artifacts such as "false edges" and "faults."

[0013] Furthermore, through Fourier transforms and angular spectrum propagation algorithms, a theoretical bridge has been established between the physical properties of light waves and digital computational models. The formula incorporates physical parameters such as the reference wavefront, holographic intensity information, and wavelength, ensuring that the reconstruction process is highly consistent with actual optical phenomena, effectively reducing errors introduced by model approximations.

[0014] Finally, through multi-layer information fusion and topological constraints at different focal lengths, the system can compensate for the lack of side information on curved surfaces in traditional methods, achieving complete surface reconstruction without the need for additional hardware. YXA solves the "cream tip" problem in vertical incidence imaging.

[0015] Furthermore, the angle correction includes: extracting the object deflection angle θ through time domain and frequency domain analysis according to the digital holographic microscopy imaging model; performing single-layer comparative analysis on the hierarchical image model according to the deflection angle θ to determine the optimal pixel position l min ; According to the optimal pixel position l min , determine the center point coordinates of the hierarchical graph model (C x ,C y ), In the example, (i, j) are the number of (x, y) pixels of the captured image; according to the coordinates of the center point Perform angle correction on each pixel of the hierarchical graph model.

[0016] In particular, the object deflection angle extracted based on time domain and frequency domain analysis, combined with the precise coordinate transformation formula, can accurately correct the curved surface parts with large angle inclinations, so that the areas that could not be fully imaged due to the limitation of the incident angle can be effectively restored, significantly improving the imaging quality of the side areas.

[0017] Furthermore, the object deflection angle θ is extracted using the following formula: Among them, I m I represents the imaginary part of the input hologram intensity information; H Represents the input hologram intensity information; R e Represents the real part of the input hologram intensity information.

[0018] Furthermore, according to the center point coordinates The angle correction is performed on each pixel of the hierarchical graph model using the following formula: x'=(xC x )cosθ+(yC y )sinθ+C x ;y'=-(xC x )sinθ+(yC y )cosθ+C y ;I a (x',y',l min )=I s (x,y,l min ); where (x', y') represents the corrected pixel; I a Represents the angle-corrected holographic intensity information; I s Represents the holographic intensity information without angle correction; l min represents the optimal pixel position.

[0019] Furthermore, a reconstructed image after angle adjustment is obtained, including: for the microscopic layer, taking each pixel as a single topological node; for the mesoscopic layer, taking a preset local surface patch as a node, using the principal component analysis algorithm PCA to process the foreground and background information of the local surface patch, and taking the processed foreground information as a single topological node; calculating the focus and weight of a single topological node in a hierarchical graph model; calculating the optimal imaging focus position according to the focus and weight; performing surface topology reconstruction according to the optimal imaging focus position to obtain a reconstructed image after angle adjustment.

[0020] In particular, the combination of pixel-level microstructures and superpixel-level mesostructures enhances the understanding of the overall topological structure of curved objects while maintaining the accuracy of reconstructed details. In particular, node connections established through curvature similarity and spatial proximity enable the system to better handle surface transition regions.

[0021] Furthermore, the focus and weight are calculated using the following formula: weight=a l , where a and b represent the corrected strength of the nodes at the same position in the hierarchical graph in different hierarchical layers l.

[0022] In particular, by calculating the focus and weight of a single topological node and combining it with the principal component analysis (PCA) algorithm to process foreground and background information, the optimal imaging position of different areas of complex surfaces is accurately located, solving the problem of unclear surface reconstruction caused by a single focal plane in traditional methods.

[0023] Furthermore, the best imaging focus position bestfocus is calculated based on the focus and weight using the following formula:

[0024] Furthermore, the surface topology is reconstructed according to the optimal imaging focus position to obtain the reconstructed image after angle adjustment using the following formula: In particular, surface topology reconstruction is performed based on the calculated optimal imaging focus position. Combined with the spatial relationship constraints of the topological structure, the reconstruction process not only considers single-point information, but also integrates the correlation of the surrounding area, effectively compensating for the information loss caused by uneven illumination or scattering angle limitations, and achieving complete and high-precision reconstruction of curved objects.

[0025] Another aspect of the present application also provides an image reconstruction system based on a topological structure, including: an imaging module, which establishes a digital holographic microscopic imaging model and provides three-dimensional spatial topological data of an object by recording and reconstructing the light field information of the object; a hierarchical graph module, which converts the three-dimensional spatial topological data into a multi-layer graph structure through an adaptive focusing algorithm, and the multi-layer graph structure includes a microscopic layer structure and a mesoscopic layer structure; an angle calculation module, which calculates the object deflection angle θ according to the topological data of the multi-layer graph structure; an angle correction module, which performs angle correction on the multi-layer graph structure according to the object deflection angle θ to obtain a corrected multi-layer graph structure; a focus calculation module, which calculates the optimal imaging focus position at different angles according to the corrected multi-layer graph structure; and a reconstruction module, which performs surface topology reconstruction according to the optimal imaging focus position to obtain a reconstructed image after angle adjustment.

[0026] Compared with the existing technology, the advantages of this application are:

[0027] On the one hand, existing technologies for imaging curved objects generally use single-focal-plane imaging methods and simple angle correction algorithms. However, these methods suffer from drawbacks such as an inability to process large-angle tilts, low computational efficiency and limited accuracy, and difficulty in achieving multi-scale information collaborative processing. This application overcomes the imaging limitations of surfaces with large angles (greater than 10 degrees) tilted by directly linking light wave phase and surface geometry through dynamic topological modeling, and based on the topological node-edge weight physical constraints, achieves physical-level precision angle self-adjustment.

[0028] On the other hand, existing technologies generally use simplified models or reduced resolution to increase speed in real-time imaging processing, but this has the drawback of being difficult to balance between accuracy and speed. This application significantly improves the algorithm speed and efficiency through optimization based on GPU parallel computing. Combined with a fast tilt compensation method based on Zernike polynomials and an electronic device patent application, it overcomes the constraints and influence of angles on imaging effects in the field of high-precision imaging, achieving fast and high-precision real-time calibration imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the digital reconstruction modeling process of this application;

[0030] Figure 2 This is an exemplary flow chart of a topology-based image reconstruction method of the present application;

[0031] Figure 3 This is the result of 3D holographic reconstruction of the wafer solder ball surface using the traditional algorithm;

[0032] Figure 4 This is the result of the 3D holographic reconstruction of the wafer solder ball surface using this algorithm. DETAILED DESCRIPTION

[0033] The present application is described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] like Figure 1 and Figure 2 As shown, a digital holographic microscopy model is established. This model provides three-dimensional spatial topological data of an object by recording and reconstructing its light field information. Digital holographic microscopy is based on the wave nature of light. Unlike traditional imaging, which only records light intensity, holographic imaging simultaneously records both the amplitude and phase of the light wave. When a laser illuminates a curved object, the object light wave scattered by the object and the reference light wave form an interference pattern on the interference surface, which contains complete three-dimensional spatial topological information.

[0035] The adaptive focus algorithm transforms 3D spatial topology data into a multi-layered graph structure imaged at different focal lengths, forming a hierarchical graph model. For curved objects, the optimal imaging plane varies for different parts. The adaptive focus algorithm calculates the intensity variance at different d values ​​to find the optimal focus position for each region, forming a multi-layered graph structure containing information on different focal lengths.

[0036] The object's deflection angle is calculated based on the topological data of the hierarchical graph model; the hierarchical graph model is angle-corrected based on the deflection angle to obtain a corrected hierarchical graph model; the optimal imaging focus positions at different angles in the corrected hierarchical graph model are calculated, and the surface topology is reconstructed based on the optimal imaging focus positions to obtain a reconstructed image with adjusted angles. This multi-level fusion method achieves complete reconstruction of curved surface objects under vertically incident laser light, avoiding the "cream tip" phenomenon seen in traditional methods while maintaining the system's simplicity and efficiency.

[0037] Step 1: Build the model.

[0038] First, a digital holographic microscopy (DMH) model with image clarity evaluation indicators is established. This model can evaluate the quality of imaging based on image clarity. At the same time, it uses the hologram structure to describe the spatial relationship of objects. It can provide high-resolution 3D spatial topology data by recording and reconstructing the complete light field information (including amplitude and phase) of the object. The specific modeling process can be seen in Figure 1 In detail, the core physical principles of the digital holographic microscopy (DHM) model are based on the wave nature of light and the phenomenon of interference. When coherent light strikes the surface of an object, the reflected light wave (object light wave O) meets the reference light wave R, which has not been modulated by the object, forming an interference pattern. This pattern is recorded by a digital sensor as a hologram I. The hologram records not only the intensity information of the light, but also the complete phase information.

[0039] Step 2: Hierarchical graph modeling. Hierarchical graph modeling converts holographic information into a multi-layer structure. Its physical essence is to numerically reconstruct the hologram at different reconstruction depths through the angular spectrum propagation algorithm.

[0040] In detail, through the adaptive focus algorithm, in the focus data range d∈(0,f max ], with a set step (usually set step = 1), the 3D information reconstructed by the hologram is converted into a multi-layer image structure with different focal lengths. The specific formula for the intensity information of each layer is as follows

[0041]

[0042] in is the Fourier transform, is the selected spectral order, E′ R is the reference wavefront, I H is the input hologram intensity information, λ is the wavelength, (x, y) and (f x′ f y ) are the corresponding coordinates in the time domain and frequency domain respectively.

[0043] Number of layers in hierarchical graph modeling At the microscopic level of image data, each pixel is a node, carrying attributes such as phase, gradient, and curvature. Adjacent pixel nodes are connected, and the edge weight is determined by phase continuity. At the mesoscopic level, similar regions are merged through superpixel segmentation, and the nodes represent local surface patches. Superpixel nodes are connected by curvature similarity or spatial proximity. The two-layer structure is integrated to form a hierarchical graph model, which enables the system to retain high-precision details at the pixel level while understanding the overall structural characteristics of the object at the mesoscopic level, providing a solid mathematical foundation for subsequent angle detection and correction. This fusion method based on physical optics and computational topology enables the system to effectively deal with the problem of missing side information of curved objects under vertical incidence.

[0044] Step 3: Holographic topology data is used for angle calculation. In dynamic measurements, high-speed digital holography is used to record transient phase changes, and the deflection angle is extracted through correlation analysis in the time and frequency domains.

[0045]

[0046] Among them, I m I represents the imaginary part of the input hologram intensity information; H Represents the input hologram intensity information; R e Represents the real part of the input hologram intensity information.

[0047] In particular, digital holography records the complex amplitude information of light waves, including real and imaginary parts; when the surface of an object is tilted, the phase distribution of the reflected light will change; this phase change will be reflected in the complex representation of the hologram, and the tilt angle can be calculated by the ratio of the real and imaginary parts. When the surface of an object is tilted, the wavefront of the reflected light wave will also tilt accordingly. This tilt will cause a change in spatial frequency, which will then appear as a change in phase distribution in the hologram. By analyzing the complex representation of the hologram, this tilt angle can be extracted. Compared with traditional angle detection methods, this application uses phase information rather than just intensity information to improve the accuracy of angle detection. In addition, it is also effective for unstructured surfaces and does not require feature points or reference marks; finally, it can handle small angle changes (theoretical accuracy can reach 0.01 degrees), which is suitable for high-precision microscopic imaging.

[0048] Step 4: Angle correction calibration layer diagram.

[0049] Compare the single layers of the hierarchical image, calibrate the optimal pixel position for angle correction, and obtain the corrected hierarchical image.

[0050] According to the same pixel point I in different levels of images s (x, y, l), calculate the average value D of its gradient and the four pixels above, below, left and right l (Excluding edge points).

[0051]

[0052] Among them D l The number of layers corresponding to the minimum position of min The optimal pixel position layer for that pixel. Specifically, at the optimal focus layer on the object surface, the intensity gradient between adjacent pixels should be minimal. This is because the image is clearest and the intensity variation between pixels is smoothest at the optimal focus plane.

[0053] Specifically, the coordinate origin is first moved to the image center (cx, cy); a rotation transformation is then performed by angle θ; and finally, the coordinate origin is returned to its original position. The corrected pixel value I'(x, y) is equal to the original pixel value I(x', y') at the corresponding position after the rotation transformation. This process is equivalent to a geometric transformation of the light field information recorded in the hologram to compensate for the phase distribution distortion caused by the tilt of the object's surface. This transformation effectively restores the side areas of the curved surface that were originally not fully imaged due to the limited incidence angle.

[0054] According to the above logic, traverse all pixels and find the optimal pixel position I for each pixel. s (x, y, l min ), where the single layer correction strength I a (x′, y′, l) The specific steps are as follows:

[0055] First determine the coordinates of the center point of the hierarchical graph Where i and j are the number of x and y pixels of the captured image respectively. For the pixel (x′, y′) of the corrected image, its corresponding coordinate changes are

[0056] x'=(xC x )cosθ+(yC y )sinθ+C x (4)

[0057] y'=-(xC x )sinθ+(yC y )cosθ+C y (5)

[0058] I a (x', y', Imin )=I s (x, y, I min ) (6)

[0059] Specifically, this application uses a matrix transformation with an angle θ with the image center as the rotation origin, which is equivalent to redirecting the local tangent plane of the surface so that it is parallel to the imaging plane, thereby maximizing the imaging quality. This application ensures that no artifacts or distortions are introduced during the transformation process by retaining the topological relationship constraints between nodes in the hierarchical graph model, and at the same time uses a bilinear interpolation algorithm to handle non-integer coordinate mapping problems, thereby ensuring the smoothness of the reconstructed image. This angle correction method combined with topological constraints effectively solves the problem of missing side information of curved objects under vertically incident laser light, without the need to add additional light sources or cameras, and significantly improves the imaging quality while maintaining the simplicity of the system.

[0060] Step 5: Calculate the optimal imaging focus position at different angles.

[0061] For each pixel (x', y') node at the micro level, calculate the focus and weight of a single topological node in the hierarchical graph

[0062]

[0063] weight=a l (8)

[0064] Where a and b represent the corrected strength I of the nodes at the same position in the hierarchical graph in different hierarchical layers l. a The quadratic term coefficient and linear term coefficient of the fitted quadratic curve.

[0065] Calculate the optimal imaging focus position of the node based on the calculated focus and weight

[0066]

[0067] At the mesoscopic level, using local surface patches as nodes, the principal component analysis (PCA) algorithm preprocesses the foreground and background information of these patches, distinguishing between foreground information (representing valid signals) and background information (representing noise). After retaining the projection data corresponding to the main singular values, the system applies the same quadratic fitting method (Formulas 7-9) as the microscopic level to the reduced dimensionality data, but in a physical sense, it focuses more on the optical response characteristics of the surface patch as a whole, rather than the local changes of individual pixels.

[0068] Step 6: Surface topology reconstruction.

[0069] The calculated best imaging focus position is brought back to formula (10) to perform surface topology reconstruction, and the reconstructed image I after large-angle self-adjustment is obtained. s.

[0070]

[0071] Specifically, from the perspective of light wave propagation theory, this reconstruction method is equivalent to extracting the amplitude and phase information of the wave field at different depths, and then reconstructing the complete three-dimensional structure based on optimal focusing conditions. This process solves the "cream tip" problem in traditional perpendicular incidence methods. Specifically, due to the law of light reflection, perpendicularly incident light waves cannot effectively capture the side edges of curved surfaces. This method achieves full reconstruction of the surface through angle correction and multi-level topological analysis. Physical constraints on node-edge weights in a hierarchical graph model ensure topological continuity during the reconstruction process.

[0072] In a specific embodiment of the present application, a spherical or curved object loses some information during the image reconstruction process, and only the top surface information exists, showing a cream tip. A wafer solder ball is selected as a measurement sample. Without using the algorithm of the present application, the result of simply performing surface morphology reconstruction is as follows: Figure 3 As shown in the figure, it can be clearly seen from the color of the legend that most of the solder ball top surface reconstructions have abruptly colored high-pointed cream tips. Figure 4 After the angle self-adjustment is performed using this application, the surface reconstruction of the wafer solder ball is found to be significantly improved, and the ball is clearly displayed. Figure 3 and Figure 4 The algorithm’s effect is intuitively demonstrated by comparison: with the traditional method, a “cream tip” appears on the top of the solder ball, while after applying this algorithm, the sphere structure is complete and clear, verifying the effectiveness of this method in practical applications.

[0073] The above schematically describes the invention of the present application and its implementation methods. This description is not restrictive. Without departing from the spirit or basic features of the present application, the present application can be implemented in other specific forms. What is shown in the drawings is only one of the implementation methods of the invention of the present application. The actual structure is not limited to this. Any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the present invention, a structural method and embodiment similar to the technical solution without creativity should fall within the scope of protection of this patent. In addition, the word "including" does not exclude other elements or steps, and the word "one" before an element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. Words such as first and second are used to indicate names and do not indicate any specific order.

Claims

1. A topology-based image reconstruction method, characterized in that: include: Establishing a digital holographic microscopy model, wherein the digital holographic microscopy model provides three-dimensional spatial topological data of an object by recording and reconstructing light field information of the object; The three-dimensional spatial topology data is converted into a multi-layer graph structure with different focal lengths through an adaptive focusing algorithm to form a hierarchical graph model; Calculate the object's deflection angle based on the topological data of the hierarchical graph model; Performing angle correction on the hierarchical graph model according to the deflection angle to obtain a corrected hierarchical graph model; The optimal imaging focus positions at different angles in the corrected hierarchical graph model are calculated, and the surface topology is reconstructed according to the optimal imaging focus positions to obtain a reconstructed image after angle adjustment.

2. The topology-based image reconstruction method according to claim 1, wherein: Form a hierarchical graph model, including: The following formula is used to convert three-dimensional spatial topological data into a multi-layer graph structure imaged at different focal lengths: Where d represents the different focal lengths of the input; F is the Fourier transform, F filtered is the selected spectral series, E R ' is the reference wavefront, I H is the input hologram intensity information, λ is the wavelength, (x, y) and (f x ,f y ) are the coordinates corresponding to the time domain and frequency domain respectively; i represents a positive integer; Each pixel is regarded as a node, and connections between adjacent pixel nodes are established. The edge weight is determined by phase continuity to construct a microscopic layer structure. By merging similar regions through superpixel segmentation algorithms, nodes are defined as local surface patches, and connections between superpixel nodes are established through curvature similarity or spatial proximity to construct a mesoscopic layer structure. By integrating the micro-level structure and the meso-level structure, a hierarchical graph model is obtained.

3. The topology-based image reconstruction method according to claim 1, wherein: Angle correction, including: According to the digital holographic microscopy imaging model, the object deflection angle θ is extracted through time domain and frequency domain analysis; Perform single-layer comparative analysis on the hierarchical graph model according to the deflection angle θ to determine the optimal pixel position l min ; According to the optimal pixel position l min , determine the center point coordinates of the hierarchical graph model (C x ,C y ), Among them, (i, j) are the number of (x, y) pixels of the captured image; According to the center point coordinates Perform angle correction on each pixel of the hierarchical graph model.

4. The topology-based image reconstruction method according to claim 3, characterized in that: Extract the object deflection angle θ through the following formula: Among them, I m I represents the imaginary part of the input hologram intensity information; H Represents the input hologram intensity information; R e Represents the real part of the input hologram intensity information.

5. The topology-based image reconstruction method according to claim 4, characterized in that: According to the center point coordinates The angle correction is performed on each pixel of the hierarchical graph model using the following formula: x'=(x-C x )cosθ+(y-C y )sinθ+C x y'=-(x-C x )sinθ+(y-C y )cosθ+C y I a (x',y',l min )=I s (x,y,l min ) Where (x', y') represents the corrected pixel; I a Represents the angle-corrected holographic intensity information; I s Represents the holographic intensity information without angle correction; l min represents the optimal pixel position.

6. The topology-based image reconstruction method according to any one of claims 2 to 5, characterized in that: Obtain the reconstructed image after angle adjustment, including: For the microscopic layer, each pixel is treated as a single topological node; For the mesoscopic layer, a preset local surface patch is used as a node, and the principal component analysis algorithm PCA is used to process the foreground and background information of the local surface patch, and the processed foreground information is used as a single topological node; Calculate the focus and weight of a single topological node in the hierarchical graph model; Calculate the optimal imaging focus position based on the focus and weight; According to the optimal imaging focus position, the surface topology is reconstructed to obtain a reconstructed image after angle adjustment.

7. The topology-based image reconstruction method according to claim 6, characterized in that: Calculate the focus and weight using the following formula: weight=a l , in , a, b represent the corrected strength of the nodes at the same position in the hierarchical graph in different hierarchical layers l.

8. The topology-based image reconstruction method according to claim 7, characterized in that: According to the focus and weight, the best imaging focus position bestfocus is calculated using the following formula:

9. The topology-based image reconstruction method according to claim 8, characterized in that: According to the optimal imaging focus position, the surface topology is reconstructed to obtain the reconstructed image after angle adjustment using the following formula:

10. A topology-based image reconstruction system, characterized in that: include: The imaging module establishes a digital holographic microscopy imaging model and provides the three-dimensional spatial topological data of the object by recording and reconstructing the light field information of the object; A hierarchical graph module converts three-dimensional spatial topology data into a multi-layer graph structure using an adaptive focusing algorithm. The multi-layer graph structure includes a micro-layer structure and a meso-layer structure. Angle calculation module, which calculates the object deflection angle θ based on the topological data of the multi-layer graph structure; An angle correction module, which performs angle correction on the multi-layer image structure according to the object deflection angle θ to obtain the corrected multi-layer image structure; The focus calculation module calculates the optimal imaging focus position at different angles based on the corrected multi-layer image structure; The reconstruction module reconstructs the surface topology according to the optimal imaging focus position to obtain a reconstructed image after angle adjustment.

Images