A system and method for three-dimensional reconstruction

Abstract

The present application relates to the technical field of three-dimensional modeling, especially to a three-dimensional reconstruction system and method. The present application adopts a non-coherent multiplexing frequency selection holographic method combined with GCN deep learning to realize sub-micron three-dimensional model reconstruction; three-dimensional reconstruction of sub-micron objects is quickly realized through three-view angles, model accuracy is optimized, and refractive index distribution is obtained. The three-dimensional reconstruction system proposed in the present application can quickly and automatically generate three-dimensional modeling of small objects with N target holograms, improve model accuracy through combination with phase, improve the speed of constructing organ models of medical CT, and improve the three-dimensional mapping efficiency of the surrounding environment in autonomous driving.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and in particular to a system and method for 3D reconstruction. Background Technology

[0002] 3D reconstruction has wide applications in fields such as autonomous driving, virtual reality, and medical CT. Currently, rapid and automatic 3D model reconstruction is mostly limited, such as the need to observe objects intensively from multiple perspectives, the object's reflection function being non-reflective, the albedo being non-uniform, and the inability to quickly and efficiently achieve 3D reconstruction of objects at the micrometer level.

[0003] Furthermore, while existing 3D reconstruction based on neural network models can train and quickly predict models based on prior datasets, the gap between the accuracy of the models and the true values ​​remains large, and the models are significantly dependent on existing datasets. Summary of the Invention

[0004] To overcome the problems of low accuracy of 3D reconstruction models, unpredictable materials, and low reconstruction efficiency of submicron-sized objects in the existing technologies, this invention proposes a 3D reconstruction method that can quickly achieve 3D reconstruction of submicron-sized objects from three perspectives, optimize model accuracy, and obtain refractive index distribution.

[0005] The present invention proposes a three-dimensional reconstruction method, comprising the following steps:

[0006] S1. A three-dimensional reconstruction system is constructed. The three-dimensional reconstruction system has multiple optical paths passing through the sample position. The propagating light on each optical path is a split of the light emitted by the light source. After each split light passes through the sample at the sample position, the split light carrying the sample information is combined. The combined light is modulated and demodulated by the optical modulation module to obtain the thickness information of each position of the sample.

[0007] S2. Set a mask on the incident surface of the optical modulation and demodulation module of the 3D reconstruction system;

[0008] S3. Start the light source and obtain the sample position sample multiplexed hologram through the 3D reconstruction system, denoted as the in-situ hologram; after the beam splitting on one of the optical paths undergoes a π / 2 phase shift before the incident sample position, obtain the sample position sample multiplexed hologram through the 3D reconstruction system, denoted as the phase-shifted hologram; apply a π / 2 phase shift to each optical path respectively, and obtain N phase-shifted holograms; N is the number of optical paths in the 3D reconstruction system;

[0009] S4. After replacing the mask, execute step S3 again; obtain two sets of 2N+2 multiplexed holograms corresponding to the sample; perform spectral separation on the 2N+2 multiplexed holograms to obtain N target holograms, and the N target holograms correspond one-to-one with each optical path viewpoint;

[0010] S5. Construct a 3D reconstruction model and training samples; the training samples are the thickness data of the N target holograms corresponding to the sample and the 3D model of the sample; the input of the 3D reconstruction model is the N target holograms corresponding to the sample, and the output is the 3D model of the sample; the 3D reconstruction model performs machine learning on the training samples until convergence;

[0011] S6. When performing 3D reconstruction of the target sample, first place the target sample at the sample position of the 3D reconstruction system, then execute steps S2-S4 to obtain N target holograms of the target sample, and then input the N target holograms of the target sample into the 3D reconstruction model to obtain the 3D model output by the 3D reconstruction model.

[0012] Preferably, the method for obtaining the 3D model in the learning sample is as follows: First, the phase reconstruction algorithm is used to obtain the phase distribution of each target hologram in the N target holograms in the learning sample. Combined with the calibrated sample refractive index, the phase thickness conversion algorithm is used to obtain the phase thickness data of the sample as the thickness data of its 3D model.

[0013] The formula for calculating the relative thickness τ is:

[0014] τ=Δφ / [(n a -n r )k]

[0015] Δφ=arg([U1(x,y) / U2(x,y)])

[0016] Where, n a n represents the refractive index of the medium surrounding an object. r denoted by ; k represents the wave vector, k = 2π / λ, where λ is the wavelength; arg represents the complex argument function; Δφ represents the phase difference between the wavefront U1(x,y) carrying sample information and the wavefront U2(x,y) of the original object.

[0017] Preferably, the construction of the 3D reconstruction model in step S5 specifically includes the following steps:

[0018] S51. Divide the learning samples into training samples and test samples;

[0019] S52. Select n1 training samples and combine the n1 training samples to perform parameter fitting on the 3D reconstruction model;

[0020] S53. Select n2 test samples, input N target holograms of the test samples into the 3D reconstruction model, obtain the 3D model output by the 3D reconstruction model, and obtain the object thickness data of the 3D model as the model annotation thickness data; use the thickness data of the 3D model in the learning samples as the true thickness data.

[0021] S54. Calculate the model loss of the 3D reconstruction model by combining the model labeled thickness data and the actual thickness data, and determine whether the 3D reconstruction model has converged; if not, update the 3D reconstruction model in reverse according to the model loss; if yes, fix the 3D reconstruction model and output it.

[0022] Preferably, in S4, the method for performing spectral separation on the 2N+2 multiplexed holograms to obtain N target holograms is as follows: First, perform a Fourier transform on the obtained 2N+2 multiplexed holograms, then perform an inverse Fourier transform to obtain N target holograms; Let E 10 Let E represent the light intensity distribution of the multiplexed hologram obtained when the first mask is applied and there is no phase shift in any of the optical paths. 1i Let E represent the light intensity distribution of the multiplexed hologram obtained when the i-th mask is loaded and the first optical path is phase-shifted by π / 2. 20 Let E represent the light intensity distribution of the multiplexed hologram obtained when a second mask is applied and there is no phase shift in any of the optical paths. 2i Let N represent the light intensity distribution of the multiplexed hologram when the second mask is loaded and the i-th optical path is phase-shifted by π / 2, where 1≤i≤N;

[0023] The Fourier transform process of the i-th optical path at the corresponding viewpoint is expressed by the following formula:

[0024]

[0025]

[0026]

[0027] Where FT is the Fourier transform, H is the transfer function, and H z1 H is the transfer function when loading the first phase mask. z2 For the transfer function when loading the second phase mask, H -z1 For H z1 The conjugate inverse function, H -z2 For H z2 The conjugate inverse function, H z2-z1 and H z1-z2 All are transition functions; u and v constitute the spatial coordinates in the frequency domain; z1 and z2 represent the optical path from the CCD to the sample position, and the difference between z2 and z1 is half the wavelength; m represents a set constant.

[0028] The complex amplitude u of the target hologram at the viewpoint corresponding to the i-th optical path i (x,y) is:

[0029]

[0030] α=[f mο (l i +l)-li l] / (f mο -l i l)

[0031] Where Q is the quadratic phase function, a represents the focal length of the mask, l4 is the optical path from the optical modulation module SLM to the electronic coupling device CCD, α is the transition parameter, e is a natural number, θ is the set frequency, and exp(iθ) is equal to the phase value when the i-th beam reaches the surface of the electronic coupling device CCD.

[0032] f mo Indicates the focal length of the microscope objective, l i Let l be the optical path length of the i-th beam splitter from the sample position to the microscope objective, and l be the optical path length from the microscope objective MO to the optical modulation module SLM.

[0033] The intensity H of the target hologram at the viewpoint corresponding to the i-th optical path i (x,y) is:

[0034]

[0035] in, Represents convolution operation; g(x0,y0,z0) represents the object point source, that is, the three-dimensional spatial representation of each point on the object surface; I Fi (x,y) represents the point spread function, specifically the point spread function corresponding to the optical path from the sample position to the CCD on the i-th optical path.

[0036] The present invention also proposes a three-dimensional reconstruction system, including a light source module, a first beam splitter BS1, a second beam splitter BS2, a beam combiner BS3, an optical path control module, a first phase shift module, a second phase shift module, a third phase shift module, a microscope objective MO, an optical modulation module, and an electronic coupling device.

[0037] The emitted light from the light source module is split into three beams by the first beam splitter BS1 and the second beam splitter BS2. After being controlled by the optical path control module, the three beams are directed towards the sample position from three angles. The first phase shift module, the second phase shift module, and the third phase shift module are respectively set on the propagation path of the three beams. Each beam passes through the corresponding phase shift module and the sample position in sequence. After passing the sample, the optical path direction of the three beams is adjusted by the optical path control module so that the three beams carrying sample information are all directed into the beam combiner BS3 for beam combining. The output light from the beam combiner passes through the microscope objective and then enters the optical modulation module. The electronic coupling device is located in the output direction of the optical modulation module and is used to analyze the output light of the optical modulation module.

[0038] The angle between the three beams at the sample location is 120 degrees.

[0039] Preferably, the light source module includes: a light source, a bandpass filter, a polarizer, and a lens; the bandpass filter, polarizer, and lens are arranged sequentially in the direction of light propagation emitted from the light source, and the light emitted from the lens is directed toward the light-incident surface of the first beam splitter.

[0040] Preferably, the optical path control module includes: a first reflector, a second reflector, a third reflector, a fourth reflector, and a fifth reflector; the second reflector and the fifth reflector form a first optical path, the light output direction of the second beam splitter toward the sample position serves as a second optical path, and the third reflector and the fourth reflector form a third optical path; the second reflector, the first phase shift module, the sample position, and the fifth reflector are sequentially arranged on the first optical path, the second phase shift module is located on the second optical path, and the third reflector, the third phase shift module, the sample position, and the fourth reflector are sequentially arranged on the third optical path;

[0041] The light emitted from the first beam splitter in the first output direction is used as the first beam splitter. After being reflected by the second mirror, the first beam splitter propagates along the first optical path and is then reflected by the fifth mirror onto the incident surface of the beam combiner. The light emitted from the second beam splitter in the second output direction is incident onto the incident surface of the second beam splitter. The light emitted from the first beam splitter in the first output direction is used as the second beam splitter. After passing through the second phase shift module and the sample position, the second beam splitter is incident onto the beam combiner. The light emitted from the second beam splitter in the second output direction is used as the third beam splitter. After being reflected by the first mirror and the third mirror, the third beam propagates along the third optical path and is then reflected by the fourth mirror onto the incident surface of the beam combiner.

[0042] Preferably, the optical path between the first beam splitter and the second beam splitter is equal to the optical path between the second beam splitter and the first reflector, and the optical path between the first beam splitter and the second reflector is equal to the optical path between the second reflector and the third reflector.

[0043] Preferably, the optical modulation module uses a spatial light modulator (SLM), and the phase shift module uses a half-wave plate.

[0044] The advantages of this invention are:

[0045] (1) This invention employs an incoherent multiplexed frequency-selective holography method combined with GCN deep learning to achieve submicron-level 3D model reconstruction; it rapidly achieves 3D reconstruction of submicron-level objects through three-view perspectives, optimizes model accuracy, and obtains refractive index distribution. The 3D reconstruction system proposed in this invention only requires N target holograms to quickly and automatically generate 3D models of tiny objects. By combining with phase, it improves model accuracy, increases the speed of constructing organ models for medical CT, and improves the efficiency of 3D mapping of the surrounding environment in autonomous driving.

[0046] (2) Traditional multi-angle holograms require the construction of a holographic system using multiple Fresnel incoherent digital holographic optical paths, resulting in complex equipment, high costs due to the use of multiple LEDs, CCDs, and other devices, and the acquisition of nine holograms for phase extraction. In this system, the system is constructed using a single LED light source and CCD. By employing a multiplexing frequency selection algorithm, only eight multiplexed holograms are needed to complete phase extraction, simplifying the experimental setup, optimizing the efficiency of phase extraction from holograms under multiple viewing angles, and expanding the field of view. At the same time, the multiplexing method ensures the real-time performance of the system.

[0047] (3) By using multiplexed holograms as perceptual features extracted from the input image, the network generates the correct geometric shape by progressively deforming an ellipsoid. To stabilize the entire deformation process, a coarse-to-fine strategy is adopted to produce a visually appealing and physically accurate 3D model. The relative object thickness obtained from phase conversion under multiple viewpoints is used; the relative object thickness characterizes the distance between model pixels in the input image viewpoint. Based on the input viewpoint in the neural network, the distance between corresponding pixels in the model is adjusted, improving modeling accuracy. Attached Figure Description

[0048] Figure 1 It is a 3D reconstruction system that includes 3 optical paths;

[0049] Figure 2 For: the phase mask of the double-reflected spherical wave loaded by the spatial light modulator;

[0050] Figure 3 The images are: the input image and output model of the neural network after the background removal of the paramecium hologram; where (a) is the target hologram under one of the views input to the 3D reconstruction model, and (b) is the 3D model output by the 3D reconstruction model.

[0051] Figure 4(a) shows the phase diagram corresponding to the target hologram from one of the perspectives of the paramecium.

[0052] Figure 4(b) shows the phase distribution of Paramecium at different positions obtained after the phase thickness transformation algorithm in Figure 4(a).

[0053] Figure 4(c) shows the thickness of Paramecium at different positions on the fixed plane of the y-axis.

[0054] Figure 5 The image shows a hologram of a paramecium and its background from one of the viewpoints; where (a) is the target hologram of the paramecium and (b) is the background image of the target hologram.

[0055] Figure 6 This is a flowchart of the 3D reconstruction method. Detailed Implementation

[0056] The technical solutions of 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] like Figure 1 As shown, this embodiment proposes a three-dimensional reconstruction system with three optical paths, including: a light source LED, a bandpass filter BF, a polarizer P, a lens L, a first beam splitter BS1, a second beam splitter BS2, a beam combiner BS3, a first reflector M1, a second reflector M2, a third reflector M3, a fourth reflector M4 and a fifth reflector M5, a first phase shifting module HWP1, a second phase shifting module HWP2, a third phase shifting module HWP3, a microscope objective MO, an optical modulation module and an electronic coupling device CCD.

[0058] In this embodiment, lens L is a focusing lens, the first phase shift module HWP1, the second phase shift module HWP2 and the third phase shift module HWP3 are all half-wave plates, and the optical modulation module is a spatial light modulator (SLM).

[0059] In this 3D reconstruction system, the distance between the first beam splitter BS1 and the second beam splitter BS2 should be equal to the distance between the second beam splitter BS2 and the first reflector M1, the distance between the first beam splitter BS1 and the second reflector M2 should be equal to the distance between the second reflector M2 and the third reflector M3, and the included angle between the three optical paths at the sample position S should be equal to 120 degrees.

[0060] Relative thickness describes the optical path difference that exists when light passes through an object or reflects off its surface, and is not constrained by the system's numerical aperture. If the object is a uniformly distributed transparent medium, even if the object has a certain thickness, there is no phase difference because light does not generate an optical path difference when passing through the object's interior. Therefore, the phase difference only reflects the difference in axial thickness distribution and is independent of the system's axial resolution. In this embodiment, standard objects are used to construct learning samples. Standard objects refer to: uniform objects with known average refractive index and geometric dimensions, or objects with known refractive index distribution and geometric dimensions.

[0061] In this embodiment, two phase masks for reflected spherical waves are used before and after the spatial modulation module (SLM), with an optical path difference of half a wavelength between the two phase masks. The two phase masks are respectively as follows: Figure 2 As shown in Figures (a) and (b).

[0062] When the SLM loads the first mask, the intensity distribution of the multiplexed hologram acquired by the CCD is E. 10Applying a phase shift only to the first optical path, the intensity distribution of the hologram acquired by the CCD is E. 11 Applying a phase shift only to the second optical path, the intensity distribution of the hologram acquired by the CCD is E. 12 Applying a phase shift only to the third optical path, the intensity distribution of the hologram acquired by the CCD is E. 13 .

[0063] When the spatial light modulator loads the second mask, the light intensity distribution of the multiplexed hologram acquired by the CCD is E. 20 Applying a phase shift only to the first optical path, the intensity distribution of the hologram acquired by the CCD is E. 21 Applying a phase shift only to the second optical path, the intensity distribution of the hologram acquired by the CCD is E. 22 Applying a phase shift only to the third optical path, the intensity distribution of the hologram acquired by the CCD is E. 23 .

[0064] By separating and extracting information from samples at different angles in a multiplexed hologram, complex amplitude and phase information from multiple viewpoints can be obtained. Multiplying the obtained object image information by the corresponding transfer function and performing an inverse Fourier transform, the object image information from each viewpoint can be reconstructed, with each viewpoint corresponding to a specific optical path.

[0065] Fourier frequency domain transform was performed on eight multiplexed holograms with light intensity distributions of E10, E11, E12, E13, E20, E21, E22, and E23, i.e., Fourier transform was performed first and then inverse Fourier transform, to obtain three target holograms.

[0066] Let FT{O} denote the Fourier transform process at the corresponding viewpoint of the first optical path:

[0067]

[0068] Where FT is the Fourier transform, H is the transfer function, and H z1 For the transfer function when loading the first phase mask, H z2 For the transfer function when loading the second phase mask, H -z1 For H z1 The conjugate inverse function, H -z2 For H z2 The conjugate inverse function, H z2-z1 and H z1-z2 Both are transition functions; z1 and z2 represent the optical path from the CCD to the sample position, and the difference between z2 and z1 is half the wavelength.

[0069]

[0070]

[0071] u and v constitute the spatial coordinates in the frequency domain; exp is an exponential function, and m represents a set constant; let FT{S} represent the Fourier transform process under the corresponding viewpoint of the second optical path:

[0072]

[0073] Let FT{M} denote the Fourier transform process at the corresponding viewpoint of the third optical path:

[0074]

[0075] Let u i (x,y) represents the complex amplitude of the target hologram from one viewpoint, that is, the complex amplitude of the hologram formed by the interference of the beam carrying sample information through the optical path to the CCD surface:

[0076]

[0077] α=[f mο (l i +l)-l i l] / (f mο -l i l)

[0078] Among them, l i l is the optical path length of the i-th beam splitter from the sample position to the microscope objective; l is the optical path length from the microscope objective MO to the optical modulation module SLM; l4 is the optical path length from the optical modulation module SLM to the electronic coupling device CCD;

[0079] e is a natural number, θ is a set frequency, and exp(iθ) is equal to the phase value when the i-th beam reaches the CCD surface of the electron coupler; f mo 'a' represents the focal length of the microscope objective; 'a' represents the focal length of the mask. It is worth noting that the propagation distance of the two masks is the same.

[0080] Let H i (x,y) represents the light intensity distribution of the target hologram under the viewpoint corresponding to the i-th optical path;

[0081]

[0082] Represents convolution operation; g(x0,y0,z0) represents the object point source, that is, the three-dimensional spatial representation of each point on the object surface; I Fi (x,y) represents the point spread function, specifically the point spread function corresponding to the optical path from the sample position to the CCD on the i-th optical path.

[0083] Phase maps from three perspectives are obtained by performing phase reconstruction on the separated and reconstructed target hologram. Specifically, phase reconstruction can employ the angular spectrum method or any other existing phase reconstruction algorithm.

[0084] During phase extraction, phase fluctuations caused by aberrations are non-negligible. Therefore, this implementation uses a double exposure method to display the phase distribution of the object wavefront. The double exposure algorithm reflects the degree of change in the object's morphology by calculating the phase difference Δφ between the sample-information-carrying object wavefront U1(x,y) and the original object wavefront U2(x,y). The sample-information-carrying object wave is the light emitted after the sample is reflected by the light source; the original object wave is the light emitted by the light source; and the wavefront refers to the set of points with the same phase along the direction of wave propagation in a wave propagation phenomenon.

[0085] A phase-thickness conversion algorithm is used to convert phase differences on a phase map into relative thicknesses. Specifically, the formula for phase-to-relative-thickness conversion is as follows:

[0086] τ=Δφ / [(n a -n r )k]

[0087] Δφ=arg([U1(x,y) / U2(x,y)])

[0088] Where τ is the relative thickness, n a n represents the refractive index of the medium surrounding an object. r The refractive index of the object is represented. k represents the wave vector, k = 2π / λ, where λ is the wavelength; arg represents the complex argument function.

[0089] Example

[0090] In this embodiment, taking the 3D modeling of paramecium as an example, the first step is to build... Figure 1 The system for 3D reconstruction is shown, and then learning samples are constructed using the system for 3D reconstruction to train the 3D reconstruction model.

[0091] In this embodiment, using Figure 6 The method shown is used to perform three-dimensional reconstruction of paramecium samples.

[0092] When modeling in 3D, the thickness value needs to be calibrated according to different locations. Since the refractive index varies at different locations in a biological cell (such as the cell nucleus), it is necessary to calculate the thickness value separately for each location. However, for objects with almost uniform refractive indices, independent calibration is not required.

[0093] In this embodiment, through Figure 1The system shown in the diagram reconstructs a 3D image. By changing the mask twice, eight multiplexed holograms of the paramecium sample at the sample location are obtained. Then, spectral separation is used to obtain three target holograms, one of which is shown below. Figure 3 As shown in Figure 4(a), Figure 4(a) is... Figure 3 The phase diagram in Figure 4(a) and Figure 4(b) are shown. Figure 3 The phase distribution in Figure (a) and the thickness at different locations on a cross section of the object in Figure 4(c); Figure 3 The middle (b) figure shows a 3D model of a paramecium.

[0094] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0095] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0096] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. A method for three-dimensional reconstruction, characterized in that, Includes the following steps: S1. A three-dimensional reconstruction system is constructed. The three-dimensional reconstruction system has multiple optical paths passing through the sample position. The propagating light on each optical path is a split of the light emitted by the light source. After each split light passes through the sample at the sample position, the split light carrying the sample information is combined. The combined light is modulated and demodulated by the optical modulation module to obtain the thickness information of each position of the sample. S2. Set a mask on the incident surface of the optical modulation and demodulation module of the 3D reconstruction system; S3. Start the light source and obtain the sample position sample multiplexed hologram through the 3D reconstruction system, denoted as the in-situ hologram; after the beam splitting on one of the optical paths undergoes a π / 2 phase shift before the incident sample position, obtain the sample position sample multiplexed hologram through the 3D reconstruction system, denoted as the phase-shifted hologram; apply a π / 2 phase shift to each optical path respectively, and obtain N phase-shifted holograms; N is the number of optical paths in the 3D reconstruction system; S4. After replacing the mask, execute step S3 again; obtain two sets of 2N+2 multiplexed holograms corresponding to the sample; perform spectral separation on the 2N+2 multiplexed holograms to obtain N target holograms, and the N target holograms correspond one-to-one with each optical path viewpoint; S5. Construct a 3D reconstruction model and training samples; the training samples are the thickness data of the N target holograms corresponding to the sample and the 3D model of the sample; the input of the 3D reconstruction model is the N target holograms corresponding to the sample, and the output is the 3D model of the sample; the 3D reconstruction model performs machine learning on the training samples until convergence; S6. When performing 3D reconstruction of the target sample, first place the target sample at the sample position of the 3D reconstruction system, then execute steps S2-S4 to obtain N target holograms of the target sample, and then input the N target holograms of the target sample into the 3D reconstruction model to obtain the 3D model output by the 3D reconstruction model.

2. The three-dimensional reconstruction method as described in claim 1, characterized in that, The method for obtaining the 3D model in the learning sample is as follows: First, the phase reconstruction algorithm is used to obtain the phase distribution of each target hologram in the N target holograms in the learning sample. Combined with the calibrated sample refractive index, the phase thickness conversion algorithm is used to obtain the phase thickness data of the sample as the thickness data of its 3D model. The formula for calculating the relative thickness τ is: τ=Δφ / [(n a -n r )k] Δφ=arg([U1(x,y) / U2(x,y)]) Where, n a n represents the refractive index of the medium surrounding an object. r denoted by ; k represents the wave vector, k = 2π / λ, where λ is the wavelength; arg represents the complex argument function; Δφ represents the phase difference between the wavefront U1(x,y) carrying sample information and the wavefront U2(x,y) of the original object.

3. The three-dimensional reconstruction method as described in claim 2, characterized in that, Step S5, the construction of the 3D reconstruction model, specifically includes the following steps: S51. Divide the learning samples into training samples and test samples; S52. Select n1 training samples and combine the n1 training samples to perform parameter fitting on the 3D reconstruction model; S53. Select n2 test samples, input N target holograms of the test samples into the 3D reconstruction model, obtain the 3D model output by the 3D reconstruction model, and obtain the object thickness data of the 3D model as the model annotation thickness data; use the thickness data of the 3D model in the learning samples as the true thickness data. S54. Calculate the model loss of the 3D reconstruction model by combining the model labeled thickness data and the actual thickness data, and determine whether the 3D reconstruction model has converged; if not, update the 3D reconstruction model in reverse according to the model loss; if yes, fix the 3D reconstruction model and output it.

4. The three-dimensional reconstruction method as described in claim 3, characterized in that, In S4, the spectral separation of the 2N+2 multiplexed holograms to obtain N target holograms is performed as follows: First, a Fourier transform is performed on the obtained 2N+2 multiplexed holograms, followed by an inverse Fourier transform, to obtain N target holograms; Let E 10 Let E represent the light intensity distribution of the multiplexed hologram obtained when the first mask is applied and there is no phase shift in any of the optical paths. 1i Let E represent the light intensity distribution of the multiplexed hologram obtained when the i-th mask is loaded and the first optical path is phase-shifted by π / 2. 20 Let E represent the intensity distribution of the multiplexed hologram obtained when a second mask is applied and all optical paths are not phase-shifted. 2i Let N represent the light intensity distribution of the multiplexed hologram when the second mask is loaded and the i-th optical path is phase-shifted by π / 2, where 1≤i≤N; The Fourier transform process of the i-th optical path at the corresponding viewpoint is expressed by the following formula: Where FT is the Fourier transform, H is the transfer function, and H z1 H is the transfer function when loading the first phase mask. z2 For the transfer function when loading the second phase mask, H -z1 For H z1 The conjugate inverse function, H -z2 For H z2 The conjugate inverse function, H z2-z1 and H z1-z2 All are transition functions; u and v constitute the spatial coordinates in the frequency domain; z1 and z2 represent the optical path from the CCD to the sample position, and the difference between z2 and z1 is half the wavelength; m represents a set constant. The complex amplitude u of the target hologram at the viewpoint corresponding to the i-th optical path i (x,y) is: α=[f mο (L i +l)-l i l] / (f mο -L i L) Where Q is a quadratic phase function, a represents the focal length of the mask, l4 is the optical path from the optical modulation module to the electronic coupling device, α is the transition parameter, e is a natural number, θ is the set frequency, and exp(iθ) is equal to the phase value when the i-th beam reaches the surface of the electronic coupling device. f mo Indicates the focal length of the microscope objective, l i Let l be the optical path length of the i-th beam splitter from the sample position to the microscope objective, and l be the optical path length from the microscope objective to the optical modulation module. The intensity H of the target hologram at the viewpoint corresponding to the i-th optical path i (x,y) is: in, Represents convolution operation; g(x0,y0,z0) represents the object point source, that is, the three-dimensional spatial representation of each point on the object surface; I Fi (x,y) represents the point spread function, specifically the point spread function corresponding to the optical path from the sample position to the electronic coupling device on the i-th optical path.

5. A system for three-dimensional reconstruction using the method for three-dimensional reconstruction as described in any one of claims 1-4, characterized in that, It includes a light source module, a first beam splitter, a second beam splitter, a beam combiner, an optical path control module, a first phase shift module, a second phase shift module, a third phase shift module, a microscope objective, an optical modulation module, and an electronic coupling device; The emitted light from the light source module is split into three beams by the first and second beam splitters; after being controlled by the optical path control module, the three beams are directed toward the sample position from three angles. The first phase shift module, the second phase shift module, and the third phase shift module are respectively set on the propagation path of the three beams. Each beam passes through the corresponding phase shift module and the sample position in sequence. After passing the sample, the optical path direction is adjusted by the optical path control module so that the three beams carrying sample information are all injected into the beam combiner for beam combining. The output light of the beam combiner passes through the microscope objective and is injected into the optical modulation module. The electronic coupling device is located in the output direction of the optical modulation module and is used to analyze the output light of the optical modulation module. The angle between the three beams at the sample location is 120 degrees.

6. The three-dimensional reconstruction system as described in claim 5, characterized in that, The light source module includes: a light source, a bandpass filter, a polarizer, and a lens; the bandpass filter, polarizer, and lens are arranged sequentially in the direction of light propagation emitted from the light source, and the light emitted from the lens is directed toward the incident surface of the first beam splitter.

7. The three-dimensional reconstruction system as described in claim 5, characterized in that, The optical path control module includes a first reflector, a second reflector, a third reflector, a fourth reflector, and a fifth reflector; the second and fifth reflectors form a first optical path, the light output direction of the second beam splitter toward the sample position serves as a second optical path, and the third and fourth reflectors form a third optical path; the second reflector, the first phase shift module, the sample position, and the fifth reflector are sequentially arranged on the first optical path, the second phase shift module is located on the second optical path, and the third reflector, the third phase shift module, the sample position, and the fourth reflector are sequentially arranged on the third optical path; The light emitted from the first beam splitter in the first output direction is used as the first beam splitter. After being reflected by the second mirror, the first beam splitter propagates along the first optical path and is then reflected by the fifth mirror onto the incident surface of the beam combiner. The light emitted from the second beam splitter in the second output direction is incident onto the incident surface of the second beam splitter. The light emitted from the first beam splitter in the first output direction is used as the second beam splitter. After passing through the second phase shift module and the sample position, the second beam splitter is incident onto the beam combiner. The light emitted from the second beam splitter in the second output direction is used as the third beam splitter. After being reflected by the first mirror and the third mirror, the third beam propagates along the third optical path and is then reflected by the fourth mirror onto the incident surface of the beam combiner.

8. The three-dimensional reconstruction system as described in claim 7, characterized in that, The optical path between the first beam splitter and the second beam splitter is equal to the optical path between the second beam splitter and the first reflector, and the optical path between the first beam splitter and the second reflector is equal to the optical path between the second reflector and the third reflector.

9. The three-dimensional reconstruction system as described in claim 5, characterized in that, The optical modulation module uses a spatial optical modulator, and the phase shift module uses a half-wave plate.

10. The three-dimensional reconstruction system as described in claim 5, characterized in that, It also includes a computer, which stores a computer program and a 3D reconstruction model. When the computer program is executed, it acquires three target holograms of the object at the sample location and inputs them into the 3D reconstruction model. The 3D reconstruction model outputs a 3D model of the object at the sample location.

Images