Digital holographic reconstruction optimization model and method based on cyclic consistency adaptive optimization

Abstract

The present application relates to the technical field of holographic reconstruction, and particularly relates to a digital holographic reconstruction optimization model and method based on cyclic consistency adaptive optimization. The present application provides a digital holographic reconstruction optimization model, which comprises a first optimization network, a first reconstruction unit and a second optimization network; the first optimization network optimizes a calibration image PSF, the second optimization network optimizes a hologram Ori, the first reconstruction unit combines the optimized calibration image PSF1 output by the first optimization network and the optimized hologram Ori1 output by the second optimization network to perform reconstruction, and obtains a holographic reconstruction image R. The present application optimizes PSF through a neural network and extracts features, simultaneously extracts features of a non-coherent hologram through another optimization network, combines the features of the optimized PSF and the features of the non-coherent hologram to perform reconstruction, and thus realizes a high-quality reconstructed image.

Description

Technical Field

[0001] This invention relates to the field of holographic reconstruction technology, and in particular to a digital holographic reconstruction optimization model and method based on cyclic consistency adaptive optimization. Background Technology

[0002] The point spread function (PSF), as a core characteristic function of an optical system, plays a crucial role in digital holography. It not only comprehensively describes the system's response to a point light source, reflecting the diffraction behavior and imaging quality of the optical system, but also serves as a vital bridge connecting object space and image space. Optimizing the PSF is decisive for improving the quality of digital holographic reconstruction.

[0003] In the development of PSF applications, Coordinated Aperture Holography (COACH) is a significant milestone. In 2011, Choi et al. proposed turbid lens imaging technology, which improves the effective NA (numerical aperture) of the system by scattering high-frequency waves, achieving super-resolution reconstruction of random patterns. In 2016, Anand and Rosen extended the secondary phase aperture of Fresnel correlation holography to chaotic phase aperture, pioneering a new method combining coded phase aperture imaging and incoherent digital holography. In 2017, Rosen et al. developed interference-free COACH (I-COACH) technology, which extracts 3D object information by generating speckle intensity through a single wavefront, significantly simplifying the operation process. Subsequently, researchers have made a series of improvements to I-COACH, including improved COACH systems for 3D imaging and annular coded aperture endoscopic imaging technology.

[0004] It is evident that existing technologies utilize engineered PSFs to significantly improve imaging performance, including spatial resolution and volumetric imaging throughput. However, these methods require optimization of the PSF acquisition process and improvements to the laboratory environment. PSFs obtained under existing laboratory conditions are difficult to optimize, as they may inherently contain aberrations, distortions, and noise, affecting reconstruction quality.

[0005] Significant progress has been made in the field of digital holographic reconstruction in recent years. End-to-end deep learning-based methods have enabled imaging of scattering media in off-axis digital holography; imaging networks combining Fourier transform and CNNs have achieved breakthroughs in generalization ability and inference speed; and physically embedded untrained neural networks and self-supervised learning methods based on angular spectrum forward diffraction models have further advanced research on unsupervised learning and physical consistency constraints. However, existing methods generally rely on large amounts of sample data for training, and even self-supervised models require substantial amounts of simulated data.

[0006] However, obtaining high-quality hologram datasets remains challenging due to stringent experimental requirements and variations in optical system configurations. Furthermore, inherent characteristics of digital holography, such as wrapper phase jumps, speckle noise, and phase differences, limit the direct application of traditional neural networks. Summary of the Invention

[0007] To overcome the problem of PSF's inherent defects affecting the holographic reconstruction quality in the prior art, this invention proposes a training method for a digital holographic reconstruction optimization model based on cyclic consistency adaptive optimization. This method optimizes the PSF and extracts features through a neural network, while simultaneously extracting features from the incoherent hologram through another optimization network. The features of the optimized PSF and the incoherent hologram are then combined for reconstruction, thereby achieving a high-quality reconstructed image.

[0008] This invention proposes a training method for a digital holographic reconstruction optimization model based on cyclic consistency adaptive optimization. First, a basic model and dataset {(PSF, Pri); Ori} are constructed; PSF represents the calibration image before spatial modulation, Pri represents the test object image before spatial modulation, and Ori represents the incoherent hologram of Pri after spatial modulation.

[0009] The basic model includes: a first optimization network, a first reconstruction unit, a second optimization network, an angular spectrum propagation unit, and a second reconstruction unit. The first optimization network optimizes the calibration image PSF, the second optimization network optimizes the hologram Ori, and the first reconstruction unit combines the optimized calibration image PSF1 output by the first optimization network and the optimized hologram Ori1 output by the second optimization network to reconstruct the holographic reconstructed image R. The holographic reconstructed image R is propagated through angular spectrum to obtain a cyclic hologram Cyc. The cyclic hologram Cyc and the optimized calibration image PSF1 are reconstructed by the second reconstruction unit to obtain a cyclic reconstructed image CR.

[0010] The second optimized network has the same structure as the first optimized network, and the parameters of the second optimized network are randomly initialized and fixed; the base model is trained on the dataset, and the first optimized network is updated with the goal of minimizing the loss function during the training process;

[0011] After the basic model is trained, the connection structure of the first optimization network, the first reconstruction unit, and the second optimization network is extracted to form a holographic reconstruction optimization model.

[0012] Preferably, the loss function is L = λ1L f +λ2L b +λ3L c Or, L = λ1L f +λ2L b L f L represents the mean squared error loss between the reconstructed image R and the test image Pri;b L represents the mean squared error loss of the hologram Ori and the cyclic hologram Cyc. c λ represents the mean squared error loss between the cyclic reconstruction image CR and the test object image Pri; λ1, λ2, and λ3 all represent weighting coefficients that take values ​​in the interval (0,1).

[0013] Preferably, the first and second optimized networks are fully connected networks.

[0014] Preferably, the first reconstruction unit and the second reconstruction unit employ convolutional networks.

[0015] Preferably, the construction of the dataset includes the following steps:

[0016] A holographic acquisition system is constructed, which modulates the light wave carrying the test object information through a second spatial modulator. The phase mask loaded on the second spatial modulator adopts a phase-constrained amplitude with a concentric ring structure, so that the output light of the second spatial modulator forms multiple focal points on the optical axis. The electronic coupler moves on the output optical axis of the second spatial modulator and outputs holograms Ori at different focal points. These are combined with the calibration image PSF used to modulate the holographic acquisition system, the test object image Pri, and the holograms Ori to form a sample.

[0017] Preferably, the holographic acquisition system includes: a white light source, a first polarizer, a first beam splitter, a first spatial modulator, a second polarizer, a collimating lens, a second beam splitter, a second spatial modulator, and an electronic coupler;

[0018] The white light emitted from the light source is polarized by the first polarizer and then redirected by the first beam splitter before being projected onto the first spatial modulator. The first spatial modulator loads the test object image Pri. The reflected light from the first spatial modulator carries the test object information. This reflected light is polarized by the second polarizer, calibrated by the collimating lens, and then redirected by the beam splitter before being projected onto the second spatial modulator. The second spatial modulator reflects the light onto the electronic coupler, which outputs a hologram.

[0019] The present invention proposes a digital holographic reconstruction optimization method, characterized in that: firstly, a holographic reconstruction optimization model, a holographic acquisition system, and a calibration image PSF are acquired; after calibrating the holographic acquisition system using the calibration image, a test object image Pri is input into the holographic acquisition system; the holographic acquisition system spatially modulates the test object image Pri and outputs a hologram Ori through an electronic coupler CCD; the PSF, Pri, and Ori are input into the holographic reconstruction optimization model to obtain the holographic reconstruction image R output by the holographic reconstruction optimization model.

[0020] Preferably, holograms Ori at different focal points are acquired using a holographic system, and each hologram Ori is reconstructed using a holographic reconstruction optimization model. Then, a three-dimensional hologram of the test object is constructed by combining the holographic reconstruction image R corresponding to each focal point.

[0021] The present invention proposes a digital holographic reconstruction optimization system, which includes a memory and a processor. The memory stores a computer program, and the processor is connected to the memory. The processor is used to execute the computer program to realize the digital holographic reconstruction optimization method.

[0022] The present invention proposes a storage medium storing a computer program, which, when executed, is used to implement the digital holographic reconstruction optimization method.

[0023] The advantages of this invention are:

[0024] (1) The present invention proposes a digital holographic reconstruction optimization method, which constructs a PSF adaptive optimization network and achieves dynamic optimization of PSF through deep learning methods, providing a new technical path for achieving high-quality two-dimensional (axial slice) super-resolution reconstruction in multi-depth digital holography. This breakthrough not only solves the limitations of traditional methods, but also opens up new directions for the further development of digital holographic technology.

[0025] (2) In this invention, the incoherent hologram Ori is processed by a second optimization network with fixed and randomized weights to optimize the angle of feature extraction and transformation. This allows for specific feature extraction and transformation of the input hologram, enabling better learning of the relationships between features and facilitating the subsequent reconstruction process. Furthermore, the second optimization network, as a stable nonlinear transformer, helps reduce data redundancy, provides a consistent feature representation, and achieves high-quality data preprocessing.

[0026] (3) Optimize the network by using a fully connected layer network, which can better learn the nonlinear relationship between features.

[0027] (4) The PSF (Point Spread Function), as the core point response function of a holographic system, determines the diffraction behavior and imaging sharpness of the hologram. The PSF is a key feature of the imaging system, and its quality directly determines the quality of the reconstructed image. This invention focuses on optimizing the PSF to address the root cause of holographic reconstruction problems. The PSF contains optical characteristic information of the imaging system; optimizing the PSF is equivalent to optimizing the entire imaging system. Furthermore, the PSF can be used for different hologram reconstructions, making this method more consistent with the essence of physical systems. This invention is applicable to holographic reconstruction in different scenarios, facilitating new breakthroughs in the field of holographic reconstruction.

[0028] (5) The present invention proposes a training method for a digital holographic reconstruction optimization model based on cyclic consistency adaptive optimization. First, a holographic acquisition system is used to generate datasets according to different imaging requirements, and the axial response characteristics at different imaging depths are analyzed. Specifically, in 3D holographic imaging, corresponding focusing PSFs are generated for target points at different axial positions to ensure accurate recovery of depth information. During the data generation stage, diverse PSFs are generated for different types of 3D targets and different environmental interference factors to ensure the diversity and adaptability of subsequent training data, laying a data foundation for the training of the deep learning model.

[0029] (6) This invention employs a triple loss function to comprehensively optimize the objective, covering the entire image reconstruction process. It also considers the reconstruction quality in both forward and reverse directions, increasing the constraints on optimization. In particular, the application of the cycle consistency loss function ensures that the PSF remains consistent during bidirectional conversion, reducing information loss and improving the reliability of reconstruction.

[0030] (7) This invention trains only one optimized network, which reduces computational complexity, training difficulty and memory requirements. It can concentrate computational resources on the most critical parts, making training convergence faster and more stable.

[0031] (8) In terms of network architecture design, this invention innovatively adopts a single-input fully connected network (FCN) structure to achieve adaptive optimization of feature constraints, and proposes a novel cyclic consistency constraint loss function to effectively improve reconstruction quality. Attached Figure Description

[0032] Figure 1 Optimize the model for digital holographic reconstruction;

[0033] Figure 2 The basic model;

[0034] Figure 3 This is a schematic diagram of the loss function;

[0035] Figure 4 Optimization methods for digital holographic reconstruction;

[0036] Figure 5 It is a holographic acquisition system;

[0037] Figure 6(a) shows the phase-constrained amplitude;

[0038] Figure 6(b) shows the phase mask;

[0039] Figure 7 A physical image of the holographic acquisition system;

[0040] Figure 8 The examples show the changing trends of marginal correlation coefficients using different loss functions.

[0041] Figure 9 The examples illustrate the changing trends of structural similarity using different loss functions.

[0042] Figure 10 The experimental results are shown in (a) for incoherent holograms, (b) for PSF, (c) for reconstruction of the contrast model, (d)-(f) for reconstruction of the model of the present invention after different training rounds, and (g) for intensity distribution of the dashed lines in (c) and (f).

[0043] Figure 11(a) is an incoherent hologram in the embodiment;

[0044] Figure 11(b) is a reconstructed hologram of Figure 11(a) output by the comparison model;

[0045] Figure 11(c) is the reconstructed hologram of Figure 11(a) output by the reconstructed holographic optimization model proposed in this invention. Detailed Implementation

[0046] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0047] Reference Figure 1 This embodiment provides a digital holographic reconstruction optimization model, including: a first optimization network, a first reconstruction unit, and a second optimization network; the first optimization network optimizes the calibration image before spatial modulation, i.e., the point spread function (PSF), and the second optimization network optimizes the incoherent hologram Ori of the test object image after spatial modulation; the first reconstruction unit combines the optimized calibration image PSF1 output by the first optimization network and the optimized hologram Ori1 output by the second optimization network to reconstruct the holographic reconstruction image.

[0048] Reference Figure 2 , Figure 3 This embodiment presents a training method for a digital holographic reconstruction optimization model based on cycle consistency adaptive optimization, comprising the following steps:

[0049] St1. Construct the basic model and dataset {(PSF, Pri); Ori}; PSF represents the calibration image before spatial modulation, i.e., the point spread function, which can be a binarized pinhole image; Pri represents the test object image before spatial modulation; Ori represents the hologram of Pri after spatial modulation.

[0050] The basic model includes: a first optimization network, a first reconstruction unit, a second optimization network, an angular spectrum propagation unit, and a second reconstruction unit;

[0051] The first optimization network optimizes the calibration image PSF, and the second optimization network optimizes the spatially modulated hologram Ori of the test object image. The first reconstruction unit combines the optimized calibration image PSF1 output by the first optimization network and the optimized hologram Ori1 output by the second optimization network to reconstruct the holographic reconstructed image R. The holographic reconstructed image R is propagated through angular spectrum to obtain a cyclic hologram Cyc. The cyclic hologram Cyc and the optimized calibration image PSF1 are reconstructed by the second reconstruction unit to obtain a cyclic reconstructed image CR.

[0052] Step 2: Train the base model on the dataset {(PSF, Pri); Ori}. During training, update the first optimized network with the goal of minimizing the loss function. The formula for calculating the loss function is:

[0053] L=λ1L f +λ2L b +λ3L c

[0054] Among them, L f L represents the mean squared error loss between the reconstructed image R and the test image Pri; b L represents the mean squared error loss of the hologram Ori and the cyclic hologram Cyc. c λ1 represents the mean squared error loss between the cyclic reconstruction image CR and the test object image Pri; λ1, λ2, and λ3 all represent weighting coefficients that take values ​​in the interval (0,1). In this embodiment, λ1 = λ2 = 0.3 and λ3 = 0.2 can be specifically set.

[0055] In specific implementation, during the construction of the dataset {(PSF, Pri); Ori}, a multi-focal-length mask can be loaded onto the spatial modulator of the test object image Pri in the holographic acquisition system that generates Ori, thereby obtaining holograms Ori corresponding to different focal lengths, enriching the dataset, and enabling the first optimized network to adapt to the PSF requirements of different focal length holographic acquisition systems.

[0056] Reference Figure 5 Figure 6(a) shows a holographic acquisition system proposed in this invention, which includes: a white light source, a first polarizer P1, a first beam splitter BS1, a first spatial modulator SLM1, a second polarizer P2, a collimating lens CL, a second beam splitter BS2, a second spatial modulator SLM2, and an electronic coupler CCD.

[0057] The white light source is a xenon lamp. The light emitted by the xenon lamp is polarized by the first polarizer P1 and then redirected by the first beam splitter BS1 before being projected onto the first spatial modulator SLM1. The first spatial modulator SLM1 loads the test object image Pri. The reflected light from the first spatial modulator SLM1 carries the test object information. This reflected light is polarized by the second polarizer P2, calibrated by the collimating lens CL, and then redirected by the beam splitter BS2 before being projected onto the second spatial modulator SLM2. The second spatial modulator SLM2 is loaded with a phase mask. SLM2 reflects the light onto the electronic coupler CCD, and the electronic coupler CCD outputs a hologram.

[0058] The spatial modulator (SLM) can only modulate polarized light. In this optical path, P1 works in conjunction with SLM1, and P2 works in conjunction with SLM2 to ensure the smooth operation of spatial modulation.

[0059] In this system, the phase mask loaded on the second spatial modulator SLM2 adopts a phase-constrained amplitude structure with a concentric ring structure so that the reflected light from the second spatial modulator SLM2 forms multiple focal points on the optical axis; the moving electronic coupler CCD can then collect holograms Ori at each focal point, thereby constructing samples {(PSF, Pri); Ori} corresponding to different focal points.

[0060] Reference Figure 4 The digital holographic reconstruction optimization method proposed in this embodiment includes the following steps:

[0061] S1. Acquire the holographic acquisition system and calibration image PSF. After calibrating the holographic acquisition system with the calibration image, input the test object image Pri into the holographic acquisition system. After spatially modulating the test object image Pri, the holographic acquisition system outputs the hologram Ori through the electronic coupler CCD.

[0062] S2. Input PSF, Pri and Ori into the holographic reconstruction optimization model to obtain the holographic reconstruction image R output by the holographic reconstruction optimization model.

[0063] When the second spatial modulator SLM2 in the holographic acquisition system adopts the phase mask with phase-constrained amplitude as shown in Figure 6(a), the electronic coupler CCD needs to repeat step S1 every time it switches the focus in order to obtain the hologram Ori corresponding to different focuses.

[0064] In this embodiment, the moving CCD acquires holograms Ori at each focal point, and the Ori at each focal point is optimized using a holographic reconstruction optimization model to obtain the corresponding holographic reconstruction image R. By combining the holographic reconstruction images R corresponding to each focal point, a three-dimensional holographic image of the test object can be constructed.

[0065] The above-mentioned holographic reconstruction optimization model is verified in conjunction with specific embodiments below.

[0066] In this embodiment, the following is adopted: Figure 5 The holographic acquisition system shown generates a dataset, and the physical image is as follows. Figure 7 As shown in Figure 6(a), the mask loaded by the second spatial modulator SLM2 satisfies the constraints shown in Figure 6(a), and the mask parameters can be iterated through the GSA algorithm.

[0067] In this embodiment, the mask shown in Figure 6(b) is obtained by iterating the GSA algorithm 100 times under the constraints shown in Figure 6(a). Its basic parameters are as follows:

[0068] Size: 512×512 pixels;

[0069] Pixel size: 1 micrometer / pixel;

[0070] Circular ring parameters:

[0071] Quantity: 10 concentric rings;

[0072] Radius setting: [20,40,60,80,100,120,140,160,180,200] pixels;

[0073] Ring thickness: uniformly set to 8 pixels;

[0074] Spacing between adjacent rings: 12 pixels;

[0075] Scattering parameters: gradually increase from the inner ring to the outer ring, respectively [0.05, 0.06, 0.07, 0.08, 0.09, 0.1, 0.11, 0.12, 0.13, 0.14];

[0076] Initial phase: randomly distributed in the interval (0-2π).

[0077] In the mask, each ring has its own scattering degree, so that when the incident light is reflected by the mask and focused, a small portion of the light will be scattered, thus constructing a self-interference digital hologram.

[0078] In this implementation, the dataset is divided into a training set and a test set. The model continues to be trained on the training set, and after training is completed, it is validated on the test set.

[0079] In this embodiment, a resolution board is used as the test object, and both the first reconstruction unit and the second reconstruction unit adopt convolutional neural networks.

[0080] In this embodiment, the model training process is first demonstrated.

[0081] In this embodiment, for the training method of the above-mentioned digital holographic reconstruction optimization model based on cycle consistency adaptive optimization, four loss functions are used for model training. The four loss functions are as follows:

[0082] L1 = 0.3L f +0.3L b +0.2L c

[0083] L2 = 0.3L b +0.2L c

[0084] L3 = 0.3L f +0.2L c

[0085] L4 = 0.3L f +0.3L b

[0086] The model training process involving the above four loss functions is as follows: Figure 8 , Figure 9 As shown, in terms of convergence, the marginal correlation coefficient (ECC) tends to stabilize after 1000 iterations, while the structural similarity (SSIM) continues to improve up to 2000 iterations, reflecting the model's optimization characteristics for different frequency components. For both the marginal correlation coefficient (ECC) and structural similarity (SSIM), a comprehensive forward reconstruction loss (L...) was adopted. f ), reverse consistency loss (L b ) and cycle consistency constraints (L c The training methods using the L1 loss function all achieved optimal performance, combined with the forward reconstruction loss (L... f ) and reverse consistency loss (L b The training effect of loss function L4 is slightly worse than that of loss function L1, and the training effect of loss functions L2 and L3 is completely incomparable to that of loss functions L1 and L4.

[0087] As can be seen, in this invention, the key to achieving excellent model performance lies in comprehensively considering the synergistic effect of forward reconstruction loss (Lf), backward consistency loss (Lb), and cyclic consistency constraint (Lc). The model structure presented in this invention, using the given loss function, achieves high performance without relying on a large-scale training set, opening up a new avenue for digital holographic reconstruction under data-scarce conditions.

[0088] In this embodiment, the first optimization network, the first reconstruction unit, and the second optimization network are extracted from the base model trained using the L1 loss function to constitute the network provided in this invention. Figure 1 The holographic reconstruction optimization model is shown.

[0089] In this embodiment, the first optimization network and the first reconstruction unit form a comparison model. In the comparison model, the first optimization network optimizes the PSF and outputs PSF1, and the first reconstruction unit combines PSF1 and the test object image Ori to perform image reconstruction to output a reconstructed hologram R'.

[0090] Figure 10 The results are shown using a resolution board as the test object and a binarized pinhole image as the PSF. Among them:

[0091] Sub-image (a) is the test object image Ori, specifically an incoherent hologram, and sub-image (b) is a PSF;

[0092] Subgraph (c) is the reconstructed hologram R' of the contrast model combining the outputs of Ori and PSF; and the first optimized network in this contrast model is extracted from the base model trained for 2300 rounds.

[0093] Subgraph (d) shows the reconstructed hologram R output by combining the Ori and PSF of the holographic reconstruction optimization model trained for 200 epochs; Subgraph (e) shows the reconstructed hologram R output by combining the Ori and PSF of the holographic reconstruction optimization model trained for 500 epochs; Subgraph (f) shows the reconstructed hologram R output by combining the Ori and PSF of the holographic reconstruction optimization model trained for 2300 epochs.

[0094] In sub-image (g), the red curve represents the pixel intensity change of the red dashed line in sub-image (c), and the blue curve represents the pixel intensity change of the blue dashed line in sub-image (f).

[0095] The holographic reconstruction optimization model trained for 200 rounds refers to the holographic reconstruction optimization model composed of the first optimization network, the first reconstruction unit, and the second optimization network extracted from the basic model trained for 200 rounds; and so on.

[0096] from Figure 10 It can be seen that the holographic reconstruction optimization model achieved excellent reconstruction results after 2300 training rounds. Subgraph (g) shows that the reconstructed hologram obtained by optimizing the hologram Ori using the second optimization network and then combining it with PSF1 is better than that without optimizing Ori, with a significant increase in the strength of the reconstructed hologram. This is because, although the second optimization network is not trained, it uses the same network structure as the first optimization network. Even with randomly initialized parameters, the second optimization network can map the image features of the hologram Ori and the optimized PSF1 features to the same space, thus facilitating the first reconstruction unit to better process and fuse the features of PSF1 and Ori for reconstruction.

[0097] In this embodiment, after 2300 training rounds, the incoherent hologram shown in Figure 11(a) was reconstructed using both a contrast model and a holographic reconstruction optimization model. The reconstructed hologram output by the contrast model is shown in Figure 11(b), and the reconstructed hologram output by the proposed holographic reconstruction optimization model is shown in Figure 11(c). It is evident that the model of this invention achieves better reconstruction results and more completely preserves the information of the original image.

[0098] 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.

[0099] 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.

[0100] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. A training method for a digital holographic reconstruction optimization model based on cycle consistency adaptive optimization, characterized in that, First, we construct the basic model and dataset {(PSF, Pri); Ori}; PSF represents the calibration image before spatial modulation, Pri represents the test object image before spatial modulation, and Ori represents the incoherent hologram of Pri after spatial modulation. The basic model includes: a first optimization network, a first reconstruction unit, a second optimization network, an angular spectrum propagation unit, and a second reconstruction unit. The first optimization network optimizes the calibration image PSF, the second optimization network optimizes the hologram Ori, and the first reconstruction unit combines the optimized calibration image PSF1 output by the first optimization network and the optimized hologram Ori1 output by the second optimization network to reconstruct the holographic reconstructed image R. The holographic reconstructed image R is propagated through angular spectrum to obtain a cyclic hologram Cyc. The cyclic hologram Cyc and the optimized calibration image PSF1 are reconstructed by the second reconstruction unit to obtain a cyclic reconstructed image CR. The second optimized network has the same structure as the first optimized network, and the parameters of the second optimized network are randomly initialized and fixed; the base model is trained on the dataset, and the first optimized network is updated with the goal of minimizing the loss function during the training process; After the basic model is trained, the connection structure of the first optimization network, the first reconstruction unit and the second optimization network is extracted to form a holographic reconstruction optimization model. Loss function L=λ1L f +λ2L b +λ3L c Or, L=λ1L f +λ2L b L f L represents the mean squared error loss between the reconstructed image R and the test image Pri; b L represents the mean squared error loss of the hologram Ori and the cyclic hologram Cyc. c λ represents the mean squared error loss between the cyclic reconstruction image CR and the test object image Pri; λ1, λ2, and λ3 all represent weighting coefficients that take values ​​in the interval (0,1).

2. The training method for the digital holographic reconstruction optimization model based on cycle consistency adaptive optimization as described in claim 1, characterized in that, The first and second optimized networks use fully connected networks.

3. The training method for the digital holographic reconstruction optimization model based on cycle consistency adaptive optimization as described in any one of claims 1-2, characterized in that, The first and second reconstruction units use convolutional networks.

4. The training method for the digital holographic reconstruction optimization model based on cycle consistency adaptive optimization as described in claim 1, characterized in that, The construction of the dataset includes the following steps: A holographic acquisition system is constructed, which modulates the light wave carrying the test object information through a second spatial modulator. The phase mask loaded on the second spatial modulator adopts a phase-constrained amplitude with a concentric ring structure, so that the output light of the second spatial modulator forms multiple focal points on the optical axis. The electronic coupler moves on the output optical axis of the second spatial modulator and outputs holograms Ori at different focal points. These are combined with the calibration image PSF used to modulate the holographic acquisition system, the test object image Pri, and the holograms Ori to form a sample.

5. The training method for the digital holographic reconstruction optimization model based on cycle consistency adaptive optimization as described in claim 4, characterized in that, The holographic acquisition system includes: a white light source, a first polarizer, a first beam splitter, a first spatial modulator, a second polarizer, a collimating lens, a second beam splitter, a second spatial modulator, and an electronic coupler; The white light emitted from the light source is polarized by the first polarizer and then redirected by the first beam splitter before being projected onto the first spatial modulator. The first spatial modulator loads the test object image Pri. The reflected light from the first spatial modulator carries the test object information. This reflected light is polarized by the second polarizer, calibrated by the collimating lens, and then redirected by the beam splitter before being projected onto the second spatial modulator. The second spatial modulator reflects the light onto the electronic coupler, which outputs a hologram.

6. A digital holographic reconstruction optimization method employing the training method of the digital holographic reconstruction optimization model based on cycle consistency adaptive optimization as described in any one of claims 1-5, characterized in that, First, the holographic reconstruction optimization model, the holographic acquisition system, and the calibration image PSF are obtained. After calibrating the holographic acquisition system using the calibration image, the test object image Pri is input into the holographic acquisition system. The holographic acquisition system spatially modulates the test object image Pri and outputs the hologram Ori through the electronic coupler CCD. The PSF, Pri, and Ori are then input into the holographic reconstruction optimization model to obtain the holographic reconstruction image R output by the holographic reconstruction optimization model.

7. The digital holographic reconstruction optimization method as described in claim 6, characterized in that, Holograms Ori at different focal points are acquired using a holographic system. Each hologram Ori is reconstructed using a holographic reconstruction optimization model. Then, a three-dimensional hologram of the test object is constructed by combining the holographic reconstruction image R corresponding to each focal point.

8. A digital holographic reconstruction and optimization system, characterized in that, It includes a memory and a processor. The memory stores a computer program, and the processor is connected to the memory. The processor is used to execute the computer program to implement the digital holographic reconstruction optimization method as described in claim 6 or 7.

9. A storage medium, characterized in that, The system contains a computer program that, when executed, is used to implement the digital holographic reconstruction optimization method as described in claim 6 or 7.

Images