Differential holographic imaging method

Abstract

The invention discloses an imaging technology based on a differential holographic imaging method # imgabs0 #. Aiming at various challenges faced by holographic imaging and the limitation of the existing computing technology, system defects are incorporated into an imaging model, and a differentiable optimization technology is adopted. A relation between a hologram and a target object is established through a specific forward model, and complex field imaging based on a single coaxial hologram is realized by utilizing automatic differential optimization and special formulation for complex field imaging and combining corresponding constraint conditions and regularization technologies without additional hardware. Experimental results show that the imaging effect of the # imgabs1 # under different settings is better than that of the existing method, for example, the # imgabs1 # shows higher calculation efficiency and better reconstruction integrity in comparison with DCOD, can accurately reconstruct phase and amplitude in multiple sample imaging, and has the advantages of flexibility, robustness, wide applicability and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of holographic imaging, and specifically relates to a method based on microscopic holographic imaging. Imaging systems and technologies. Background Art

[0002] In many scientific research fields, such as physics, materials science, and biology, imaging objects of different sizes (from nanometers to millimeters) and studying the interaction between objects and light are extremely important. Holography has solved the problem of quantizing light waves to a certain extent through interferometry, but in the past century, holographic imaging has faced many challenges that have limited its widespread application. These challenges include the difficulty of separating unwanted terms (such as DC and twin images), the complexity of phase unwrapping, and the difficulty of achieving automatic focusing.

[0003] To overcome these challenges, computational techniques have been introduced into the field of holographic imaging, which has made some progress through optimization methods or deep learning algorithms. However, manual optimization methods usually rely on relatively simple and idealized imaging models, which may not accurately reflect the reality. Although learning-based methods use general but opaque black box models, they require a large amount of training data and take a lot of time to train, and often lack good generalization capabilities. In addition, both methods are sensitive to system-specific factors and cannot effectively compensate for the mismatch between numerical modeling and real-world conditions, such as imperfect factors in experiments (such as light sources, optical alignment, sensor quantization) and physical interference. Summary of the invention

[0004] The present invention proposes Inspired by the principle of differentiable imaging, the method innovatively incorporates system defects into the imaging model and uses differentiable optimization techniques to solve the inverse problem of the forward model. Specifically, the method uses defocus and illumination amplitude changes as additional variables in the model, enabling complex field imaging from a single in-line hologram without the need for additional hardware. The differentiable design of this imaging framework gives it a high degree of flexibility and robustness, enabling it to be widely used in a variety of different settings with minimal adjustments to the forward model.

[0005] The specific technical solution of the present invention is:

[0006] A microscopic holographic imaging method, characterized by comprising the following steps:

[0007] S1. The forward model f(x,θ) is used to establish the relationship between the intensity-only line-of-sight hologram y and the target object x, where θ contains unknown parameter conditions such as the combined effects of optical aberrations, light sources and sensors on the illumination wave;

[0008] S2. By minimizing Implement the inversion of the forward model, where ‖f(x,θ)-y‖ 2 To ensure data fidelity, R n (·) is the set of soft regularization terms, β n is the weight, Ω x,0 is the corresponding physical constraint;

[0009] S3. Using gradient descent to calculate the loss function The unconstrained part is optimized, the derivatives are calculated by reverse automatic differentiation, the conjugate Wirtinger derivative is used for complex numbers, and the approximate gradient is used for non-differentiable functions. The physical constraints are implemented by projecting the updated parameters into the physical domain, and the regularization of the complex numbers is applied to the amplitude and phase respectively.

[0010] S4. In complex field imaging, when x is a complex transmittance function, with the illumination amplitude a and the object-camera distance z as variables, minimize the following equation under the conditions of |t|≤1, a>0, z>0: where f(·) is a free-space wave propagator with a subsequent interferometer, is the norm regularization term, R TV (t) is the total variation norm regularization term. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0012] Figure 1 : Schematics showing various online holographic systems used for verification, including plane wave illumination, spherical wave illumination, lensless holography, and lensless holographic systems with fiber bundles, with components labeled.

[0013] Figure 2 : Presenting the target object in Experiment 1 and the reconstructed images under different conditions and related images of the target axial position.

[0014] Figure 3 :Holograms, multi-plane (MP) phase retrieval results and The resulting image of the reconstruction.

[0015] Figure 4 : Presenting holograms of cheek cell samples, DCOD reconstruction results and The resulting image of the reconstruction.

[0016] Figure 5 :Show holograms and The resulting image of the reconstruction.

[0017] Figure 6 : Demonstration of holograms and The resulting image of the reconstruction. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0019] On four different settings The method was fully validated.

[0020] S1. In Experiment 1, we used Figure 1 The setup shown in a is verified on an object with known real conditions. The object is a highly transparent silicon wafer with a phase profile etched on it. A laser with a wavelength of 532 nm is used as the illumination source. The reconstruction can accurately reconstruct the phase target, the phase value is close to the theoretical value, and the target axial position can converge well. Without it, the reconstructed image is severely blurred and the loss convergence is poor, which fully demonstrates the importance of system parameters in the reconstruction process.

[0021] S2. In Experiment 2, the results were verified on the Tilia root microscope samples. The reconstructed amplitude and phase match well with the multi-plane (MP) phase retrieval results and show better resolution in the magnified part. Excellent performance when dealing with microscopy samples with realistic phase and amplitude distributions.

[0022] S3. In Experiment 3, the proposed method is compared with the advanced learning complex field imaging method (DCOD) based on single-shot online holograms of physical models. The validation results on cheek cell samples show that It outperforms DCOD in both computational efficiency and reconstruction completeness. The reconstructed phase image has a clear background and is able to complete 25,000 iterations in 46 seconds, while DCOD takes 40 minutes to complete 30,000 iterations on an Nvidia Tesla k80 GPU. In addition, DCOD fails to reconstruct some parts of the object (especially near the edge of the image), while The entire image can be reconstructed clearly.

[0023] S4. In Experiment 4, a wafer with a high-transmittance pattern etched on its surface was imaged under a spherical wave illumination setting. The height of the pattern on the wafer surface was 200 nm, and the corresponding phase difference was 0.3174 under light with a wavelength of 630 nm. The reconstructed pattern phase value is approximately, which is consistent with the design, indicating that The image can be accurately reconstructed under this setting.

[0024] S5. In Experiment 5, a positive resolution chart is imaged in a fiber bundle lensless imaging system. The experimental results show that The reconstructed image is clear and has no twin images, which indicates that The image can be efficiently reconstructed in this lens-free setup.

[0025] Compared with optimization-based compression holography and its variants, It has the following significant advantages:

[0026] Adhering to the strict theoretical formula of holographic imaging ensures the reliability and accuracy of the reconstruction process.

[0027] Optimizing system defects as additional variables can better compensate for system imperfections and thus improve imaging performance.

[0028] The framework is differentiable, which facilitates efficient optimization and can introduce gradient-based algorithms to further improve imaging effects.

[0029] It can reconstruct complex fields and provide more comprehensive and detailed imaging results to meet the needs of different application scenarios.

[0030] Compared with neural networks, It has the following features and advantages:

[0031] It can handle multiple specific variables, just like a neural network, but unlike the black box model of traditional neural networks, The experiment-theory mismatch problem was solved.

[0032] A mechanism is provided to incorporate system parameters into the imaging model, allowing it to adapt to a variety of data types acquired from different experimental setups.

[0033] It does not require a large amount of training data, which is a great advantage in practical applications, especially when it is difficult to obtain a large amount of training data.

Claims

1. A differentiable holographic imaging method, characterized in that The following steps are involved: S1. Use the forward model f(x,θ) to establish the relationship between the coaxial hologram y and the target object x, where θ contains unknown parameter conditions such as optical aberration, the combined influence of the light source and the sensor on the illumination wave; S2. By minimizing Implement the inversion of the forward model, where ‖f(x,θ)-y‖ 2 To ensure data fidelity, R n (·) is the soft regularization term, β n is the weight, Ω x,0 is the corresponding physical constraint; S3. Using gradient descent to calculate the loss function The unconstrained part is optimized, the derivatives are calculated by reverse automatic differentiation, the conjugate Wirtinger derivative is used for complex numbers, and the approximate gradient is used for non-differentiable functions. The physical constraints are implemented by projecting the updated parameters into the physical domain, and the regularization of the complex numbers is applied to the amplitude and phase respectively. S4. In complex field imaging, when x is a complex transmittance function, the illumination light wave amplitude a and the object-camera distance z are used as variables, and under the conditions of |t|≤1, a>0, z>0, minimize where f(·) is a free-space wave propagator with a subsequent interferometer, is the regularization term, R TV (t) is the TV regularization term.

2. The differentiable holographic imaging method according to claim 1, characterized in that The method described enables complex-field imaging from a single on-axis hologram without the need for additional hardware.

3. The differentiable holographic imaging method according to claim 1, characterized in that The imaging framework is differentiable and can be applied to various settings with minimal adjustments to the forward model.

4. The differentiable holographic imaging method according to claim 1, characterized in that The method solves the problem of mismatch between experimental data and theoretical models by incorporating system defects into the imaging model.

Images