H-Net-based dual-channel synchronous Fresnel incoherent correlation holographic imaging method

Abstract

The invention discloses a dual-channel synchronous Fresnel incoherent correlation holographic imaging method based on H-Net, and the method comprises the following steps: 1), synchronously collecting a 0 phase shift hologram and a pi / 2 phase shift hologram of object reflection or self-illumination under single exposure through a dual-channel polarized light path system; 2) inputting the two holograms acquired in the step 1) into a pre-trained H-Net neural network to generate a corresponding pi phase shift prediction hologram and a corresponding 3pi / 2 phase shift prediction hologram; 3) combining the four holograms of 0, pi / 2, pi and 3pi / 2, and calculating a complex amplitude hologram of the object through a four-step phase shift algorithm; and 4) based on the complex amplitude hologram, reconstructing a three-dimensional image of the object by using an angular spectrum back propagation algorithm. According to the method, the H-Net prediction map and the four-step phase shift reconstruction algorithm are combined, artifact interference can be effectively eliminated, the reconstruction quality of the image is improved, the method is excellent in performance on evaluation indexes such as PSNR and SSI M, and the method is suitable for real-time and high-quality FI NCH imaging requirements.

Description

Technical Field

[0001] This invention belongs to the field of imaging technology, and particularly relates to a dual-channel synchronous Fresnel incoherent correlation holographic imaging method based on H-Net. Background Technology

[0002] Fresnel incoherent correlation holography (FINCH) is a holographic imaging technique based on incoherent light sources. It can record the three-dimensional information of objects without scanning, showing great application potential in 3D imaging under outdoor natural light and microscopic fluorescence 3D imaging. The core advantage of this technology is that it does not rely on an external illumination source; it can directly utilize the self-emission or ambient light reflection characteristics of objects to achieve three-dimensional imaging of any object, and the imaging resolution can exceed the optical diffraction limit, with a theoretical lateral resolution twice that of coherent imaging. However, traditional FINCH systems are coaxial systems, which are easily affected by background components and twin images during image reconstruction. Multi-step phase-shifting techniques are usually required to eliminate these interferences. Since acquiring multiple phase-shifted holograms reduces the temporal resolution of FINCH, there is an urgent need to design a method to achieve high-quality single-exposure FINCH imaging.

[0003] To overcome the limitation of traditional phase-shifting strategies, which require multiple exposures to acquire multiple holograms sequentially, researchers have proposed various single-exposure techniques. One such technique is single-pulse phase-shifting incoherent digital holography, which utilizes spatial segmentation multiplexing and light wave polarization states to achieve parallel phase-shifting holography. This technique acquires an incoherent hologram and reconstructs a 3D object image in a single exposure. However, this method requires attention to the sampling theorem and higher light intensity to obtain high-quality images. Another approach is to load a checkerboard phase grating onto a spatial light modulator (SLM) to achieve spatial parallel phase-shifting. However, this method reduces the field of view of the detectable hologram, and the diffraction and refraction dispersion of the grating exacerbates the hologram blurring problem caused by broadband light sources. Summary of the Invention

[0004] The purpose of this invention is to provide a dual-channel synchronous Fresnel incoherent correlation holographic imaging method based on H-Net to solve the technical problems mentioned in the background.

[0005] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0006] In some embodiments of this application, a dual-channel synchronous Fresnel incoherent correlation holographic imaging method based on H-Net is provided, comprising the following steps:

[0007] Step 1) Using a dual-channel polarization optical path system, simultaneously acquire 0-phase-shift holograms and π / 2-phase-shift holograms of object reflection or self-emission under a single exposure;

[0008] Step 2) Input the two holograms acquired in Step 1 into the pre-trained H-Net neural network to generate the corresponding π phase shift prediction hologram and 3π / 2 phase shift prediction hologram;

[0009] Step 3) Combine the four holograms (0, π / 2, π, 3π / 2) and calculate the complex amplitude hologram of the object using a four-step phase-shifting algorithm;

[0010] Step 4) Based on the complex amplitude hologram, reconstruct the three-dimensional image of the object using the angular spectrum backpropagation algorithm.

[0011] In some embodiments of this application, the dual-channel polarization optical path system in step 1 includes:

[0012] An incoherent light source, a digital micromirror device (DMD), and a spatial light modulator (SLM) are arranged sequentially along the optical path.

[0013] Polarization modulation assembly: includes a half-wave plate (HWP) with its fast axis at an angle of 22.5° to the horizontal direction, a quarter-wave plate (QWP) with its fast axis at an angle of 45° to the horizontal direction, a first polarizer (P2) with a transmission axis of 0°, and a second polarizer (P3) with a transmission axis of 90°.

[0014] Two CMOS cameras receive beams modulated by P2 and P3 respectively, enabling the synchronous acquisition of 0 and π / 2 phase-shifted holograms.

[0015] In some embodiments of this application, the SLM is loaded with a dual-lens mode, and the focal lengths of the two lenses are f0 and f1 respectively. d1 and f d2 Furthermore, one of the lenses has a certain phase angle θ; the complex amplitude of the optical field on the front plane of the SLM can be expressed as:

[0016]

[0017] The two beams propagate freely over an additional distance z. h The interference fringes are formed when the signals reach the CCD, and the amplitude distribution on the CCD plane is as follows:

[0018]

[0019] In the formula It is a complex constant; L s The linear phase function is expressed as: λ is the center wavelength of the illuminating light; Q(b) is a quadratic phase function, expressed as: Q(b) = exp[iπbλ] -1 (x 2 +y 2 )]; * indicates a two-dimensional spatial convolution operation;

[0020] Assuming the modulation axis of the SLM is horizontal, and the angle between polarizer P1 and the horizontal direction is α, then the horizontal polarization component of the object light is loaded onto the SLM with a focal length of f. a The phase difference is introduced by spherical phase modulation. The optical field on the back plane of the SLM is modulated, while the vertical polarization component is not modulated; It can be represented in Jones matrix form:

[0021]

[0022] The dual-channel synchronous phase-shifting device includes a half-wave plate (HWP), a quarter-wave plate (QWP), a non-polarizing beam splitter (BS), two polarizers (P), and two identical cameras. The fast axis of the QWP makes an angle of 45° with the horizontal direction, and the polarizers P2 and P3 make angles of 0° and 90° with the horizontal direction, respectively. For the QWP, when the fast axis makes an angle of θ with the horizontal direction, its Jones matrix is...

[0023]

[0024] For horizontal and vertical polarizers, their Jones matrices are respectively... and

[0025] In some embodiments of this application, the H-Net neural network is a deep learning model based on the U-Net architecture, including:

[0026] The encoder includes five convolutional blocks and four 2×2 max pooling layers, with the convolutional blocks and pooling layers connected alternately.

[0027] The decoder adopts a mirror architecture that is symmetrical to the encoder. The core of the architecture is to convert the encoder's downsampling process into upsampling reconstruction.

[0028] The input consists of 0 and π / 2 phase-shifted holograms, which are independently encoded by encoder 1 and encoder 2, respectively.

[0029] The output is the prediction results of the π and 3π / 2 phase shift holograms.

[0030] In some embodiments of this application, the training method for H-Net includes:

[0031] Training data: 3000 sets of real-world four-step phase-shift holograms, each set containing 0, π / 2, π, and 3π / 2 phase-shift images;

[0032] Input label pairing: Use 0 and π / 2 phase shift maps as input, and π and 3π / 2 phase shift maps as training targets;

[0033] Loss function: Mean squared error optimizes network weights.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows. First, a dual-channel incoherent digital holographic imaging system is built. Through the collaborative modulation of polarization optical components, two holograms with phase shifts of 0 and π / 2 respectively are synchronously acquired during a single exposure process. Based on these two experimentally measured holograms, two holograms with phase shifts of π and 3π / 2 are predicted by using the self-designed H-Net neural network to construct a complete four-step phase shift data set. After obtaining four holograms with different phase shifts, a complex amplitude hologram is calculated through a four-step phase shift algorithm, and then combined with the angular spectrum backpropagation algorithm to achieve high-precision reconstruction of the hologram. The H-Net network forms a stable mapping relationship by learning the phase correlation features between different phase shift holograms, and has excellent reconstruction accuracy and generalization ability. The trained network can be used as a virtual phase shifter to generate the target complex amplitude hologram from the single-exposure data of the camera without physical phase shift devices or equipment, greatly reducing the time cost. Through systematic network performance testing and optical experiment verification, this method is outstanding in aspects such as hologram generation accuracy, three-dimensional reconstruction quality, and real-time processing efficiency. It can provide an efficient and feasible new technical solution for fast incoherent digital holographic imaging. Description of the Drawings

[0035] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0036] Figure 1 Schematic diagram of the principle of dual-channel Fresnel incoherent correlation holography provided by an embodiment of the present invention;

[0037] Figure 2 Schematic diagram of the process of using H-Net to implement a dual-channel FINCH single-exposure real-time imaging system provided by an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the experimental device of a dual-channel Fresnel incoherent correlation holographic system provided by an embodiment of the present invention;

[0039] Figure 4 Schematic diagrams of three groups of results in the test data provided by an embodiment of the present invention, where (a) shows the samples of "one", (b) shows the samples of "I", and (c) shows the samples of "8": the holograms with phase shifts of π and 3π / 2 collected, the holograms with phase shifts of π and 3π / 2 output by the network, and the intensity comparison diagram of the 336th column of the label and the network output result. The scale bar in the figure is 300 μm;

[0040] Figure 5Schematic diagram of the reconstruction results of "one", "I", and "8" in the dataset provided by the embodiments of the present invention. Among them, the first column is the result of true value reconstruction; the second column is the method proposed in this paper; the third column is the reconstruction results of the two-phase shift interference patterns of 0 and π; the fourth column is the difference map between the method proposed in this paper and the true value reconstruction; the fifth column is the difference map between the two-step phase shift reconstruction and the true value reconstruction. The scale bar in the figure is 300μm;

[0041] Figure 6 Schematic diagram for comparing the reconstruction results provided by the embodiments of the present invention. Among them, (a) is the three-step phase shift reconstruction; (b) is the four-step phase shift reconstruction; (c) is the method proposed in this paper. The scale bar in the figure is 300μm;

[0042] Figure 7 Schematic diagram of the multi-depth three-dimensional object imaging results provided by the embodiments of the present invention. Among them, (a) is a multi-depth three-dimensional object composed of objects at different axial distances; (b) is the imaging results at different reconstruction distances, where: (1) the reconstruction distance z r = 70mm; (2) the reconstruction distance z r = 80mm; (3) the reconstruction distance z r = 90mm; (4) the reconstruction distance z r = 100mm. The reconstruction distance is focused on the object indicated by the white arrow. The scale bar in the figure is 300μm. Detailed implementation manners

[0043] The following will further describe in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0044] In order to better understand the purpose, structure and function of the present invention, the following will further describe the present invention in detail in conjunction with the accompanying drawings. <0,000134>Refer to the attached<,000135>As shown, according to the embodiments of the present application, a dual-channel synchronous Fresnel incoherent correlation holographic imaging method based on H-Net is provided, including the following steps:

[0046] (1) System setup: The present invention designs a set of such as Figure 3The dual-channel self-interference digital holographic system is shown. A digital micromirror device (DMD, Fldiscovery F4300, 1920×1080 pixels, 10.8μm pixel size) is used to load the sample. Light emitted from an LED (THORLABS M625L4-C1, 630mW, λ=625nm) is collimated and then illuminates the DMD. The light reflected from the DMD passes through a collimating lens L3 and a polarizer P1 before reaching an SLM (Hamamatsu, X15213-16, 1280×1024 pixels, 12.5μm pixel pitch) for modulation. Then, a half-wave plate (HWP), a beam splitter (BS), a quarter-wave plate (QWP), and a polarizer are used to achieve synchronous phase shifting of the dual channels. The two orthogonally polarized modulated beams are simultaneously acquired by two identical CMOS cameras, generating two holograms with phase shift differences, achieving the ability to acquire dual-channel phase-shifted images in a single exposure.

[0047] (2) Dataset Acquisition and Preprocessing: Based on the aforementioned optical path system, this invention acquired 3000 sets of holographic datasets, each set containing four phase-shifting interferometric images (phase shift angles of 0, π / 2, π, and 3π / 2, respectively). The 0 and π / 2 phase-shifting images were directly and synchronously acquired by the dual-channel system, while the π and 3π / 2 phase-shifting images served as training target samples for the H-Net network. To ensure the stability and generalization ability of the model training, the dataset was divided into training, validation, and test sets in an 8:1:1 ratio.

[0048] (3) H-Net Network Architecture Design and Training: To achieve intelligent generation of two additional phase-shifted holograms (π and 3π / 2), an H-Net deep learning network based on the U-Net architecture was designed. This network employs a symmetrical encoder-decoder structure, using a cascade of multiple convolutional layers, pooling layers, and upsampling layers to achieve hierarchical extraction and reconstruction of holographic features. The network input consists of 0 and π / 2 phase-shifted holograms acquired through dual channels, and the output consists of predicted π and 3π / 2 phase-shifted holograms. The mean square error (MSE) is used as the loss function to optimize prediction accuracy. After training, this network can act as a virtual phase shifter, generating two additional phase-shifted holograms from the two phase-shifted holograms acquired in a single camera exposure.

[0049] (4) Network performance testing and verification: The trained H-Net model was applied to an independent test set. After inputting 0 and π / 2 phase shift holograms, π and 3π / 2 phase shift predicted holograms were generated. By comparing the predicted holograms with the actual acquired phase shift images, the reconstruction accuracy of the network was evaluated from the dimensions of intensity distribution consistency and structural feature restoration degree, and the model's ability to map different phase features and its generalization performance were verified.

[0050] (5) Comprehensive verification of imaging effect: The π and 3π / 2 phase shift maps generated by H-Net and the acquired 0 and π / 2 phase shift maps are used to recover the complex amplitude hologram of the object using a four-step phase shift reconstruction algorithm, and then the angular spectrum propagation algorithm is used to complete the image reconstruction. By visually comparing and quantitatively analyzing the reconstruction results with those of two-step phase shift, traditional three-step and four-step phase shift, the feasibility of the phase shift map generated by the network in actual imaging is verified, which fully demonstrates that the present invention has the advantages of both high-precision reconstruction and real-time performance in FINCH imaging.

[0051] Figure 1 This is the schematic diagram of a dual-channel fine-pass filter. The first half shows the fine-pass system. For any one of them... A point light source, located in front of the collimating lens L. s Among them The focal length of the lens is f0. In the experiment, an incoherent light source illuminates the object. The light wave is modulated by the collimating lens and reaches the SLM after a distance d. The SLM is in dual-lens mode, and the focal lengths of the two lenses are f0 and f0, respectively. d1 and f d2 Furthermore, one of the lenses has a certain phase angle θ. The complex amplitude of the optical field on the front plane of the SLM can be expressed as:

[0052]

[0053] Then the two beams propagate freely an additional distance z. h The interference fringes are formed when the signals reach the CCD, and the amplitude distribution on the CCD plane is as follows:

[0054]

[0055] In the formula It is a complex constant; L s The linear phase function is expressed as: λ is the center wavelength of the illuminating light; Q(b) is a quadratic phase function, expressed as: Q(b) = exp[iπbλ] -1 (x 2 +y 2 )]; * indicates a two-dimensional spatial convolution operation.

[0056] Assuming the modulation axis of the SLM is horizontal, and the angle between polarizer P1 and the horizontal direction is α, then the horizontal polarization component of the object light is loaded onto the SLM with a focal length of f. a The phase difference is introduced by spherical phase modulation. The optical field on the back plane of the SLM is modulated, while the vertical polarization component is not modulated. It can be represented in Jones matrix form:

[0057]

[0058] The dual-channel synchronous phase-shifting device includes a half-wave plate (HWP), a quarter-wave plate (QWP), a non-polarizing beam splitter (BS), two polarizers (P), and two identical cameras, such as... Figure 1 The dashed box in the figure

[25] shows that the fast axis of the QWP is at an angle of 45° to the horizontal direction, and the angles between polarizers P2 and P3 and the horizontal direction are 0° and 90°, respectively. For the QWP, when the fast axis is at an angle of θ to the horizontal direction, its Jones matrix is

[0059]

[0060] For horizontal and vertical polarizers, their Jones matrices are respectively... and

[0061] Based on the Jones matrices of typical polarization optical elements, we describe the polarization modulation process of the imaging light wave in a traditional FI NCH system using the Jones matrix form of a dual-channel spatial phase-shifting device module. In this process, only the changes in the polarization state and phase delay of the imaging beam caused by the polarization optical elements are considered, neglecting the spatial phase change of the complex amplitude in the transverse plane during the diffraction propagation of the imaging beam. Therefore, to simplify the derivation, the transverse coordinate is omitted in the formula. Finally, Figure 1 The vector fields of the two camera recording planes can be expressed as follows:

[0062]

[0063] Two holograms with a phase shift of π / 2 can be obtained simultaneously from two cameras. DC1 with I DC2 Their intensities are expressed as follows:

[0064]

[0065] The technical effects achieved by the above technical solution in the embodiments of this application are as follows:

[0066] Traditional FI NCH reconstruction requires acquiring multiple holograms with different phase shifts, necessitating multiple exposures and reducing temporal resolution. This invention designs a dual-channel optical path system to simultaneously acquire two holograms with different phase shifts in a single exposure, and uses a deep learning network to complete the acquisition of two additional holograms, achieving one-time acquisition of the images required for four-step phase-shift reconstruction, significantly improving imaging quality and temporal resolution. Traditional two-step phase-shift reconstruction is prone to leaving conjugate images and background terms, affecting image quality, while traditional three-step phase-shift methods have limitations in noise suppression. Although three-step phase-shifting can theoretically eliminate background noise and twin image noise, its actual suppression effect does not meet theoretical expectations. This invention combines H-Net prediction maps with a four-step phase-shift reconstruction algorithm, effectively eliminating artifact interference and improving image reconstruction quality. It exhibits excellent performance in evaluation metrics such as PSNR and SSIM, making it suitable for real-time, high-quality FI NCH imaging requirements.

[0067] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0068] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0069] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dual-channel synchronous Fresnel incoherent holographic imaging method based on H-Net, characterized in that, Includes the following steps: Step 1) Using a dual-channel polarization optical path system, simultaneously acquire 0-phase-shift holograms and π / 2-phase-shift holograms of object reflection or self-emission under a single exposure; Step 2) Input the two holograms acquired in Step 1 into the pre-trained H-Net neural network to generate the corresponding π phase shift prediction hologram and 3π / 2 phase shift prediction hologram; Step 3) Combine the four holograms (0, π / 2, π, 3π / 2) and calculate the complex amplitude hologram of the object using a four-step phase-shifting algorithm; Step 4) Based on the complex amplitude hologram, reconstruct the three-dimensional image of the object using the angular spectrum backpropagation algorithm.

2. The dual-channel synchronous Fresnel incoherent holographic imaging method based on H-Net according to claim 1, characterized in that, The dual-channel polarization optical path system in step 1 includes: An incoherent light source, a digital micromirror device (DMD), and a spatial light modulator (SLM) are arranged sequentially along the optical path. Polarization modulation assembly: includes a half-wave plate (HWP) with its fast axis at an angle of 22.5° to the horizontal direction, a quarter-wave plate (QWP) with its fast axis at an angle of 45° to the horizontal direction, a first polarizer (P2) with a transmission axis of 0°, and a second polarizer (P3) with a transmission axis of 90°. Two CMOS cameras receive beams modulated by P2 and P3 respectively, enabling the synchronous acquisition of 0 and π / 2 phase-shifted holograms.

3. The dual-channel synchronous Fresnel incoherent holographic imaging method based on H-Net according to claim 2, characterized in that, SLM loading dual-lens mode, the focal lengths of the two lenses are f_1 and f_2 respectively. d1 and f d2 Furthermore, one of the lenses has a certain phase angle θ; the complex amplitude of the optical field on the front plane of the SLM can be expressed as: The two beams propagate freely over an additional distance z. h The interference fringes are formed when the signals reach the CCD, and the amplitude distribution on the CCD plane is as follows: In the formula It is a complex constant; L s The linear phase function is expressed as: λ is the center wavelength of the illuminating light; Q(b) is a quadratic phase function, expressed as: Q(b) = exp[iπbλ] -1 (x 2 +y 2 )]; * indicates a two-dimensional spatial convolution operation; Assuming the modulation axis of the SLM is horizontal, and the angle between polarizer P1 and the horizontal direction is α, then the horizontal polarization component of the object light is loaded onto the SLM with a focal length of f. a The phase difference is introduced by spherical phase modulation. The optical field on the back plane of the SLM is modulated, while the vertical polarization component is not modulated; It can be represented in Jones matrix form: The dual-channel synchronous phase-shifting device includes a half-wave plate (HWP), a quarter-wave plate (QWP), a non-polarizing beam splitter (BS), two polarizers (P), and two identical cameras. The fast axis of the QWP makes an angle of 45° with the horizontal direction, and the polarizers P2 and P3 make angles of 0° and 90° with the horizontal direction, respectively. For the QWP, when the fast axis makes an angle of θ with the horizontal direction, its Jones matrix is... For horizontal and vertical polarizers, their Jones matrices are respectively... and 4. The dual-channel synchronous Fresnel incoherent holographic imaging method based on H-Net according to claim 1, characterized in that, The H-Net neural network is a deep learning model based on the U-Net architecture, including: The encoder includes five convolutional blocks and four 2×2 max pooling layers, with the convolutional blocks and pooling layers connected alternately. The decoder adopts a mirror architecture that is symmetrical to the encoder. The core of the architecture is to convert the encoder's downsampling process into upsampling reconstruction. The input consists of 0 and π / 2 phase-shifted holograms, which are independently encoded by encoder 1 and encoder 2, respectively. The output is the prediction results of the π and 3π / 2 phase shift holograms.

5. The dual-channel synchronous Fresnel incoherent holographic imaging method based on H-Net according to claim 1, characterized in that, The training methods for H-Net include: Training data: 3000 sets of real-world four-step phase-shift holograms, each set containing 0, π / 2, π, and 3π / 2 phase-shift images; Input label pairing: Use 0 and π / 2 phase shift maps as input, and π and 3π / 2 phase shift maps as training targets; Loss function: Mean squared error optimizes network weights.

Images