Three-dimensional compressed sensing reconstruction method based on reset scanning and joint prior

Abstract

The invention discloses a three-dimensional compressed sensing reconstruction method based on reset scanning and joint prior. Comprising the following steps: sequentially performing forward scanning and reset scanning by using a compressed sensing three-dimensional imaging system to obtain a compressed frame image stack and an additional frame image stack of a sample, respectively performing pre-reconstruction by using an alternating multiplier method to obtain a compressed frame pre-reconstruction image stack and an additional frame pre-reconstruction image stack, and processing by using a cross-correlation method to obtain a dislocation amount of a reset process; and establishing a joint reconstruction enhanced compressed sensing three-dimensional reconstruction model according to the dislocation amount in the resetting process, and inputting a compressed frame and an additional frame image stack to obtain a reconstructed three-dimensional image. The method disclosed by the invention has good low cost, practicability and acquisition efficiency, the availability of the compressed sensing algorithm on a three-dimensional tissue imaging system such as a light sheet microscope is improved, and a new scheme is provided for low-cost and high-efficiency three-dimensional optical tomography microscopic imaging.

Description

Technical Field

[0001] This invention belongs to the fields of optical computational imaging and biological microscopy imaging, and specifically relates to a three-dimensional compressed sensing reconstruction method based on resetting scanning and joint priors. Background Technology

[0002] Rapid three-dimensional microscopic imaging and high-quality detailed image reconstruction of biological samples are crucial for biological research. In recent years, volumetric imaging technology, represented by light-sheet microscopy (LSFM), and high-speed wide-field image detection schemes have developed rapidly. By directly exciting a certain depth plane of the sample and recording the image signals emitted by the biological sample layer by layer, the microscopy system based on the surface detector has good imaging speed and optical tomography capabilities, and has been widely used in the three-dimensional imaging of large tissue samples.

[0003] Despite the significant success of various imaging techniques, one of the most important performance indicators of microscopes—imaging data throughput—remains limited by the image sensor frame rate and imaging field of view. Previously, the academic community introduced Snapshot Compressive Imaging (SCI) technology to enhance the throughput of imaging systems. As a computational imaging technique, SCI can increase the imaging speed of image sensors at a low cost, thereby improving the overall imaging data throughput. In principle, SCI applies B random masks that vary over time during the exposure process using an imaging encoding device, such as a digital micromirror device (DMD), within a single exposure. This allows for the simultaneous recording of image information at B different times within a single exposure. During reconstruction, SCI technology solves the inverse problem of the encoded exposure process to reconstruct several frames of images encoded from a compressed frame. The framework used in the reconstruction process is typically the Alternating Direction Method of Multipliers (ADMM) framework.

[0004] Currently, the combination of SCI technology and emerging microscopes has made it possible to construct a new generation of high-throughput microscopic imaging systems. However, existing SCI technology solutions still cannot achieve good reconstruction quality in depth space for three-dimensional imaging systems. This invention innovatively utilizes the time and information redundancy of the resetting process during imaging in existing biological microscopic imaging systems, introducing it a priori into the reconstruction process and applying it to imaging tools such as open-top light-sheet microscopes. This achieves efficient three-dimensional compressed sensing reconstruction of microscopic images, demonstrating its potential to further improve the high-throughput capabilities of three-dimensional microscopes. Summary of the Invention

[0005] To address the problems existing in the background art, the present invention provides a three-dimensional compressed sensing reconstruction method based on resetting scanning and joint prior.

[0006] The technical solution adopted in this invention is as follows:

[0007] I. A Three-Dimensional Compressed Sensing Reconstruction Method Based on Reset Scanning and Joint Priors

[0008] The reconstruction method includes the following steps:

[0009] Step 1: Use a compressed sensing 3D imaging system to perform forward scanning and resetting scanning in sequence to obtain the compressed frame image stack and the extra frame image stack of the samples required for 3D compressed sensing reconstruction.

[0010] Step 2: Use the alternating multiplier method to pre-reconstruct the compressed frame image stack and the extra frame image stack respectively, to obtain the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack;

[0011] Step 3: Use the cross-correlation method to process the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack to obtain the misalignment amount during the reset process;

[0012] Step 4: Construct a priori model for resetting and enhanced reconstruction based on the misalignment during the resetting process, and establish a joint reconstruction and enhanced compressed sensing three-dimensional reconstruction model based on the priori model for resetting and enhanced reconstruction.

[0013] Step 5: Input the compressed frame image stack and the additional frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model to obtain the reconstructed 3D image of the sample.

[0014] During both forward and reset scanning, the displacement stage moves along the z-axis in opposite directions. The excitation beam, the modulation method of the optical intensity encoder, and the frame rate of the image sensor remain constant. The scanning speed is uniform. The speed ratio between the reset scan and the forward scan is an integer value greater than 1. This results in the optical intensity encoder generating a sparser coding sequence in the sample depth direction during the reset process compared to the forward process.

[0015] In step 2, the alternating multiplier method is used, with the compressed frame image stack as input, to solve the following objective function to obtain the compressed frame pre-reconstructed image stack:

[0016]

[0017] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H represents the modulation mask, and y m Indicates the compressed frame image stack, x p_m Indicates the compressed frame pre-reconstructed image stack, x mThis represents the pre-reconstructed image stack of compressed frames to be optimized. This represents the x value corresponding to the minimum value of the function "·". m The value of α represents the regularization hyperparameter. This represents the square of the second norm;

[0018] In step 2, the alternating multiplier method is used, with the extra frame image stack as input, to solve the following objective function to obtain the extra frame pre-reconstructed image stack:

[0019]

[0020] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H * This represents the equivalent mask during the reset scanning process, y e Indicates the extra frame image stack, x p_e Indicates the extra frame pre-reconstructed image stack, x e This indicates the additional pre-reconstructed image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". e The value of α represents the regularization hyperparameter. This represents the square of the second norm.

[0021] The first three-dimensional denoising prior function adopts the three-dimensional total variation (3DTV) function.

[0022] Step 3 specifically involves: using the cross-correlation method, taking the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack obtained in step 2 as input, solving the following objective function to obtain the misalignment amount during the reset process:

[0023]

[0024] In the formula, d est Let d represent the misalignment amount during the reset process, and x represent the misalignment amount during the reset process to be optimized. p_m Indicates the compressed frame pre-reconstructed image stack, x p_e argmax represents the extra frame pre-reconstructed image stack. d(·) represents the value of d when the function "·" takes its minimum value, max() represents the function to take the maximum value, min() represents the function to take the minimum value, h represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, w represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, c represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, i represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, j represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, k represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, and ||||2 represents the second-order norm operation.

[0025] In step 4, the joint reconstruction enhanced compressed sensing 3D reconstruction model is set according to the following formula:

[0026]

[0027] In the formula, x est ′ represents the reconstructed 3D image stack, x est This represents the reconstructed 3D image stack to be optimized, y m This represents the compressed frame image stack, and Prior2(·) represents the second image denoising prior function. This represents the x value corresponding to the minimum value of the function "·". est The values ​​of REP(·) represent the reset enhancement reconstruction prior model, γ represents the first hyperparameter scalar corresponding to the prior hyperparameter, and ρ represents the second hyperparameter scalar corresponding to the additional compressed frame enhancement hyperparameter. This represents the square of the second norm.

[0028] In step 4, the prior model REP(x) is reset and enhanced for reconstruction. est Set it according to the following formula:

[0029]

[0030] In the formula, y e Indicates the extra frame image stack, x est H represents the reconstructed 3D image stack to be optimized. * D represents the equivalent mask during the reset scanning process. dest (·) represents a shift operation on the z-axis, indicating that the entire image stack "·" is shifted d along the z-axis.est pixels, d est D represents the misalignment during the reset process. dest (x est () represents the stack of reconstructed 3D images after misalignment. This represents the square of the second norm.

[0031] The second image denoising prior function is specifically a Hessian matrix first-order norm constraint prior.

[0032] In step 5, after inputting the compressed frame image stack and the extra frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model, the joint reconstruction enhanced compressed sensing 3D reconstruction model is solved by the alternating multiplier method to obtain the reconstructed 3D image of the sample.

[0033] II. A Three-Dimensional Compressed Sensing Reconstruction System

[0034] The three-dimensional compressed sensing reconstruction system includes:

[0035] The compressed sensing 3D imaging module employs a compressed sensing 3D imaging system with a sample reset or illumination scan backtracking process to acquire compressed frame image stacks and additional frame image stacks of the sample.

[0036] The calculation module is used to implement the steps of the above-mentioned three-dimensional compressed sensing reconstruction method.

[0037] The beneficial effects of this invention are:

[0038] (1) Improvement of axial quality in 3D reconstruction: On the one hand, the method of the present invention constructs a comprehensive 3D prior combination, which more specifically addresses the 3D compressed sensing reconstruction problem. On the other hand, by utilizing the additional information of the reset process, it increases the amount of effective information acquired, thus comprehensively improving the defect of poor axial quality in 3D reconstruction.

[0039] (2) Improved signal-to-noise ratio in 3D reconstruction: The method of this invention utilizes the additional information acquired during the reset process and forms a data pair with conventional compressed sensing data, which are in the same distribution. This data pair reduces the impact of image sensor noise on reconstruction during joint reconstruction, thereby improving the reconstruction signal-to-noise ratio;

[0040] (3) Improved imaging speed: The method of this invention is an enhanced method for compressed sensing reconstruction and can be applied to any compression ratio. Therefore, when the reconstruction quality is constant, the method of this invention is faster than ordinary three-dimensional compressed sensing imaging.

[0041] In summary, this invention effectively utilizes the time wasted during the reset process in existing tomographic 3D surface scanning microscope systems during compressed sensing measurement. By combining the compressed frame image stacks obtained from forward scanning and reset scanning with the additional frame image stacks, it achieves cost-effective and efficient compressed sensing reconstruction enhancement. This method has broad application scenarios and cost advantages, and can be applied to a series of light-sheet tomographic compressed sensing systems, such as open-top light sheets and volumetric orthogonal light sheets, thus effectively advancing the research of low-cost, high-throughput, and high-quality 3D imaging systems. Attached Figure Description

[0042] Figure 1 This is a flowchart of an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of the open-top light-sheet microscope system used in an embodiment of the present invention.

[0044] Figure 3 This is a schematic diagram of the timing relationship between the forward scan mask, compressed frame, additional scan mask, and additional compressed frame in this invention. The diagram shows the case where the reset speed ratio s = 2 and the compression ratio B = 4.

[0045] Figure 4 This diagram illustrates the difference between reset-scan compressed sensing reconstruction and ordinary compressed sensing reconstruction, showing the temporal relationship between reconstructed frames and extra frames during the reconstruction process. The diagram shows the case where the reset speed ratio s = 2 and the compression ratio b = 4.

[0046] Figure 5 This is a schematic diagram of the three-dimensional compressed sensing reconstruction results obtained in an embodiment of the present invention.

[0047] Figure 6 This is a comparative schematic diagram showing the results of three-dimensional compressed sensing reconstruction obtained from embodiments and comparative examples of the present invention.

[0048] Among them, 1. laser source, 2. first beam expander lens, 3. second beam expander lens, 4. cylindrical lens, 5. optical slit, 6. excitation objective lens, 7. sample slot, 8. displacement stage, 9. probe objective lens, 10. sleeve lens, 11. light intensity encoder, 12. relay lens one, 13. relay lens two, 14. image sensor. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0050] The first invention provides a three-dimensional compressed sensing reconstruction method based on resetting scanning and joint priors. For example... Figure 4As shown, the principle of the method of this invention is as follows: This method utilizes a three-dimensional optical tomographic imaging system such as a light sheet microscope to perform sequential high-speed light intensity encoding and illumination of a forward-moving three-dimensional tissue sample at different depths. The information excited by the illumination light of one encoding sequence is acquired through one image frame. Then, during the reverse reset process of the sample, the sample undergoes additional, sparser sequence encoding, and another image frame is acquired. The results of the two acquisitions are combined with an image denoising prior to enhance the quality of the three-dimensional image reconstruction. Specifically, the ordinary compressed frame obtained during the forward motion of the sample and the additional compressed frame obtained during the reset scan are pre-reconstructed separately. The initial results obtained from the pre-reconstruction are used for alignment to determine the misalignment during the reset process. Furthermore, the misalignment is used to construct an additional compressed frame as a prior, and combined with the image denoising prior, an alternating multiplier method reconstruction framework is introduced to complete the final enhanced three-dimensional reconstruction.

[0051] The method of the present invention specifically includes the following steps:

[0052] Step 1: Perform a forward scan using a compressed sensing 3D imaging system to obtain a stack of compressed frame images of the samples required for 3D compressed sensing reconstruction; then, perform a reversal scan using the compressed sensing 3D imaging system to obtain a stack of additional frame images of the samples.

[0053] During both forward and reset scanning, the displacement stage moves along the z-axis in opposite directions. The excitation beam, the modulation method of the optical intensity encoder, and the frame rate of the image sensor remain constant. The scanning speed is uniform. The speed ratio between the reset scan and the forward scan is an integer value greater than 1. This results in the optical intensity encoder generating a sparser coding sequence in the sample depth direction during the reset process compared to the forward process.

[0054] Specifically, the compressed sensing 3D imaging system employs a 3D light sheet compressed sensing microscopy system with a sample repositioning or illumination scan backtracking process.

[0055] Preferably, the three-dimensional light sheet compression sensing microscopy system includes, but is not limited to, open-top light sheet microscopy systems and orthogonal volumetric light sheet microscopy systems.

[0056] The imaging process using a compressed sensing 3D imaging system is as follows: A laser emitted from a laser source illuminates the sample, exciting a signal on a specific surface of the 3D tissue sample, which is then recorded using an area array image sensor. The image signal generated after the sample is excited first passes through a light intensity encoder, where it is modulated and reflected before being imaged again on the image sensor. During the forward scan, the stage moves uniformly along the z-axis at a preset forward scan speed v, and the data recorded during this process forms a compressed frame image stack y. mAfter the forward scan is completed, keeping the excitation beam and the intensity encoder modulation unchanged, the displacement stage performs a reverse reset motion along the z-axis. The image sensor records the image stack information at the same frame rate as during the forward scan, thus achieving a reset scan. The recorded data constitutes an additional frame image stack y. e During the reset scan, the reset scan speed is constant, and the speed ratio s between the reset scan and the forward scan is fixed and taken as an integer value greater than 1. That is, during the reset process, the optical intensity encoder generates a sparser coding sequence in the sample depth direction than in the forward process.

[0057] Specifically, the image stack (compressed frame image stack, extra frame image stack) obtained through the above imaging process consists of multiple two-dimensional images on the xy plane.

[0058] Specifically, the light intensity encoder used in a compressed sensing 3D imaging system refers to a device that can discretely or continuously modulate the intensity of passing light in a two-dimensional plane of a certain size, with pixels as the smallest unit. This includes, but is not limited to, digital micromirror devices (DMDs) and spatial light modulators (SLMs).

[0059] During the exposure process of the image sensor, the frequency of the light intensity encoder loading the modulation mask H is fixed relative to the recording frame rate of the image sensor, denoted as the compression ratio B. That is, for every B binary masks loaded by the light intensity encoder, after the sample motion scans through the corresponding spatial depth, the image sensor completes the signal exposure recording of one frame of image.

[0060] Specifically, the modulation mask H is composed of random, sequentially different binary masks, the number of which is equal to the compression ratio B. During forward scanning and reset scanning, the modulation mask H and its loading method remain unchanged. This ensures that the equivalent mask H during the reset scanning process... * It is the reverse and stretching of the modulation mask H. The modulation mask stretching ratio H* / H is equal to the speed ratio of the reset scan to the forward scan.

[0061] Step 2: Use the alternating multiplier method to pre-reconstruct the compressed frame image stack and the extra frame image stack respectively, to obtain the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack.

[0062] In step 2, the alternating multiplier method is used, with the compressed frame image stack as input, to solve the following objective function, resulting in the compressed frame pre-reconstructed image stack:

[0063]

[0064] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H represents the modulation mask loaded on the light intensity encoder, and y m Indicates the compressed frame image stack, x p_mIndicates the compressed frame pre-reconstructed image stack, x m This represents the pre-reconstructed image stack of compressed frames to be optimized. This represents the x value corresponding to the minimum value of the function "·". m The value of α represents the regularization hyperparameter. This represents the square of the second norm.

[0065] In step 2, the alternating multiplier method is used, with the extra frame image stack as input, to solve the following objective function, resulting in the extra frame pre-reconstructed image stack:

[0066]

[0067] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H * This represents the equivalent mask during the reset scanning process, y e Indicates the extra frame image stack, x p_e Indicates the extra frame pre-reconstructed image stack, x e This indicates the additional pre-reconstructed image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". e The value of α represents the regularization hyperparameter. This represents the square of the second norm.

[0068] Preferably, the first three-dimensional denoising prior function is a three-dimensional total variation (3DTV) function.

[0069] Step 3: Use the cross-correlation method to process the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack to obtain the misalignment amount during the reset process.

[0070] Step 3 specifically involves using the cross-correlation method, taking the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack obtained in Step 2 as input, to solve the following objective function and obtain the misalignment amount during the reset process:

[0071]

[0072] In the formula, d est Let d represent the misalignment amount during the reset process, and x represent the misalignment amount during the reset process to be optimized. p_m Indicates the compressed frame pre-reconstructed image stack, x p_e argmax represents the extra frame pre-reconstructed image stack. d(·) represents the value of d when the function "·" takes its minimum value, max() represents the function to take the maximum value, min() represents the function to take the minimum value, h represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, w represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, c represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, i represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, j represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, k represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, and ||||2 represents the second-order norm operation.

[0073] Step 4: Construct a priori model for resetting and enhancing reconstruction based on the misalignment during the resetting process, and establish a joint reconstruction and enhanced compressed sensing 3D reconstruction model based on the priori model for resetting and enhancing reconstruction.

[0074] The joint reconstruction enhanced compressed sensing 3D reconstruction model is set according to the following formula:

[0075]

[0076] In the formula, x est ′ represents the reconstructed 3D image stack, x est This represents the reconstructed 3D image stack to be optimized, y m This represents the compressed frame image stack, and Prior2(·) represents the second image denoising prior function. This represents the x value corresponding to the minimum value of the function "·". est The values ​​of REP(·) represent the reset enhancement reconstruction prior model, γ represents the first hyperparameter scalar corresponding to the prior hyperparameter, and ρ represents the second hyperparameter scalar corresponding to the additional compressed frame enhancement hyperparameter. This represents the square of the second norm.

[0077] Among them, the prior model of repositioning and enhancement reconstruction, REP(x) est Set it according to the following formula:

[0078]

[0079] In the formula, y e Indicates the extra frame image stack, x est H represents the reconstructed 3D image stack to be optimized.* D represents the equivalent mask during the reset scanning process. dest (·) represents a shift operation on the z-axis, indicating that the entire image stack "·" is shifted d along the z-axis. est pixels, d est D represents the misalignment during the reset process. dest (x est () represents the stack of reconstructed 3D images after misalignment. This represents the square of the second norm.

[0080] Preferably, the second image denoising prior function is specifically a Hessian matrix first-order norm constrained prior function.

[0081] Step 5: Input the compressed frame image stack and the extra frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model to obtain the reconstructed 3D image of the sample.

[0082] Preferably, in step 5, after inputting the compressed frame image stack and the extra frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model, the joint reconstruction enhanced compressed sensing 3D reconstruction model is solved by the alternating multiplier method to obtain the reconstructed 3D image of the sample.

[0083] A second aspect of the present invention provides a three-dimensional compressed sensing reconstruction system.

[0084] The three-dimensional compressed sensing reconstruction system of the present invention includes:

[0085] The compressed sensing 3D imaging module employs a compressed sensing 3D imaging system with a sample reset or illumination scan backtracking process to acquire compressed frame image stacks and additional frame image stacks of the sample.

[0086] The calculation module is used to implement the steps of the above-mentioned three-dimensional compressed sensing reconstruction method.

[0087] Specifically, the samples are isolated biological tissue samples that have undergone transparency and fluorescent staining treatment.

[0088] Specific embodiments of the present invention are as follows:

[0089] Example

[0090] This embodiment employs an open-top light-sheet microscope system. For example... Figure 2 As shown, the open-top light-sheet microscope system includes a laser source 1, a first beam expander lens 2, a second beam expander lens 3, a cylindrical lens 4, an optical slit 5, an excitation objective lens 6, a sample slot 7, a displacement stage 8, a probe objective lens 9, a sleeve lens 10, a light intensity encoder 11, a relay lens 12, a relay lens 13, and an image sensor 14. In this embodiment, the light intensity encoder 11 is a digital micromirror device.

[0091] The laser emitted from laser source 1 passes sequentially through first beam expander lens 2, second beam expander lens 3, cylindrical lens 4, optical slit 5, excitation objective lens 6, and sample slot 7 before illuminating the sample. The sample is excited to emit fluorescence, which is first collected by detection objective lens 9, and then sequentially passes through sleeve lens 10, light intensity encoder 11, relay lens one 12, and relay lens two 13, finally forming an image on image sensor 14. First beam expander lens 2 and second beam expander lens 3 form a 4F system. The parallel laser emitted from laser source 1 remains parallel after passing through first beam expander lens 2 and second beam expander lens 3, but its diameter is increased several times. Excitation objective lens 6 and cylindrical lens 4 form a 4F system in the focusing direction of cylindrical lens. Cylindrical lens 4 shapes the laser into a sheet-like beam, and excitation objective lens 6 focuses the sheet-like beam onto the sample. Optical slit 5 is located between cylindrical lens 4 and excitation objective lens 6, cutting off the edge portion of the beam and shaping the beam.

[0092] The detection objective 9 and the sleeve lens 10 form a 4F system. The excited part of the object and the light intensity encoder 11 are on the front and rear focal planes of the 4F system, respectively. Therefore, after the detection objective 9 collects the fluorescence emitted by the sample, it is imaged on the reflective surface of the light intensity encoder 11 through the sleeve lens 10. The light intensity encoder 11 can perform pixel-level modulation of the light beam by using the reflection of a tiny mirror. At the same time, the light intensity encoder 11 rotates the light beam by 12° and reflects it to the relay lens 12. The relay lens 12 and the relay lens 2 13 form a 4F system. The reflective surface of the light intensity encoder 11 and the target surface of the image sensor 14 are on the front and rear focal planes of the 4F system, respectively. Therefore, the image of the object formed on the light intensity encoder 11 is imaged on the target surface of the image sensor 14 after passing through the relay lens 12 and the relay lens 2 13.

[0093] In this system, the stage 8 moves the tissue sample horizontally, and the excitation beam excites the sample from a direction at 45° to the horizontal. The detector objective 9 collects the fluorescence from a direction at 45° to the horizontal, with an angle of 90° between it and the excitation beam. The sample holder 7 is a custom-made transparent glass dish. The sample is placed into the sample holder from above and immersed in the liquid within the sample holder. The excitation beam enters the sample holder from one side, and the fluorescence emitted by the sample exits from the other side of the sample holder and is collected by the detector objective 9.

[0094] During 3D scanning, the displacement stage 8 moves the sample, allowing it to pass uniformly through the light beam. The image sensor 14 then records a time-continuous tomographic signal, and the stack of these signals forms a 3D image. A crucial detail in the imaging process is the modulation of the passing fluorescence signal by the intensity encoder 11. The intensity encoder 11 utilizes the reflection of tiny mirrors to perform pixel-level modulation of the light beam, and by loading a random binary mask, it achieves compressed sensing modulation of the recorded signal.

[0095] In this embodiment, a mouse eye blood vessel sample is used as an example. The imaging and three-dimensional compressed sensing reconstruction process is as follows: Figure 1 As shown, specifically:

[0096] A. Imaging Stage

[0097] Step 1: Use a compressed sensing 3D imaging system to perform forward scanning and resetting scanning in sequence to obtain the compressed frame image stack and the extra frame image stack of the samples required for 3D compressed sensing reconstruction.

[0098] Specifically, the following steps are included:

[0099] 1.1 Initial state: The displacement stage 8 is in the initial "0" position. The sample to be imaged is placed on the sample stage, and the excitation source 1 emits a laser beam.

[0100] 1.2. Start-up and braking: The displacement platform 8 begins to accelerate until it reaches the working speed vm / s, then transitions to a constant speed. The typical value of v is 1×10. -4 After entering a constant speed state, the image sensor 14 and the light intensity encoder 11 are activated simultaneously.

[0101] 1.3 Forward Scanning Imaging: The stage 8 moves at a constant speed. The image sensor 14 records images at a maximum frame rate of f1Hz, where f1 is typically 100Hz. The recorded images form a compressed frame image stack y. m The light intensity encoder 11 plays a binary mask H with a period of B at f2Hz, where B is typically 8 and f2 is typically 800Hz. During imaging, the fluorescence signal generated by the fluorescent dye in the transparent sample, excited by the sheet-like light beam, is transmitted to the light intensity encoder 11 via the probe objective lens 9 and the sleeve lens 10. The modulated signal generated by the light intensity encoder 11 is recorded by the image sensor 14 via the relay lens 12 and the relay lens 13.

[0102] 1.4 Stop Braking: After the displacement stage 8 reaches the designated position, all image sensors 14 stop working, and the light intensity encoder 11 plays the periodic mask normally. The displacement stage 8 enters a uniform deceleration process until it stops moving forward. After stopping, the displacement stage 8 accelerates in the opposite direction until it reaches the designated reset speed (s×v) m / s, and then enters a constant speed.

[0103] 1.5 Reset Scan Imaging: The stage 8 resumes constant speed and restarts the image sensor 14. The image sensor 14 continues to record images at the maximum frame rate f1Hz, where f1 is typically 100Hz. The recorded images form an additional frame image stack y. e Its light propagation process is the same as the aforementioned ordinary compressed sensing imaging process, until the displacement stage 8 returns to the initial 0 position.

[0104] B. Reconstruction Process

[0105] Step 2, Pre-reconstruction stage: The compressed frame image stack and the extra frame image stack are pre-reconstructed using the alternating multiplier method to obtain the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack.

[0106] The specific process is as follows:

[0107] 2.1 Construct a pre-reconstruction model, which is set according to the following formula:

[0108]

[0109] In the formula, H represents the modulation mask loaded on the optical intensity encoder 11, and y m Indicates the compressed frame image stack, x p_m Indicates the compressed frame pre-reconstructed image stack, x m This represents the pre-reconstructed image stack of compressed frames to be optimized. This represents the x value corresponding to the minimum value of the function "·". m The possible values ​​of ;

[0110] H * This represents the equivalent mask during the reset scanning process, y e Indicates the extra frame image stack, x p_e Indicates the extra frame pre-reconstructed image stack, x e This indicates the additional pre-reconstructed image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". e The possible values ​​of ;

[0111] Prior1(·) represents the first 3D denoising prior function, and α represents the regularization hyperparameter. This represents the square of the second norm.

[0112] In the pre-reconstruction model, both objective functions include a fidelity term and a prior term. The prior term is represented by the first three-dimensional denoising prior function Prior1(·). In this embodiment, a three-dimensional total variation (3DTV) prior is used.

[0113] In the pre-reconstruction model, the modulation mask H is a random sequence of binary masks, and the number of binary masks is equal to the compression ratio B. Equivalent mask H * The mask is a scaled mask that is in reverse order of the modulation mask H, with the scaling factor being equal to the reset speed ratio s, and the typical value of the reset speed ratio s is 2.

[0114] When the reset speed ratio s = 2 and the compression ratio B = 4, the modulation mask H * The difference from the modulation mask H is as follows Figure 3 As shown. Figure 3In the diagram, measurement images A, B, and C each represent a two-dimensional image frame in the compressed frame image stack. Measurement images A' and B' each represent a two-dimensional image frame in the extra frame image stack.

[0115] 2.2 Using the alternating multiplier method, with the compressed frame image stack as input, the first objective function in the pre-reconstruction model is solved to obtain the compressed frame pre-reconstruction image stack.

[0116] During the iterative solution process using the alternating multiplier method, the specific iterative formula for compressed frame pre-reconstruction is as follows:

[0117]

[0118] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H represents the modulation mask, and y m This represents the compressed frame image stack; i represents the iteration round. Let x represent the stack of intermediate compressed frame pre-reconstructed images obtained in the i-th iteration. p_m This represents the pre-reconstructed image stack of compressed frames to be optimized. This represents the x value corresponding to the minimum value of the function "·". p_m The value of , This represents the result of the first prior constraint obtained in the i-th iteration. Let v denote the square of the second norm. (i) Let represent the first convergence auxiliary parameter obtained in the i-th iteration, ρ represent the first convergence step size hyperparameter, and γ represent the second convergence step size hyperparameter. The typical value of both convergence step size hyperparameters is 1. The typical value of the preset total number of iterations is 30.

[0119] 2.3 Using the alternating multiplier method, with the extra frame image stack as input, the second objective function in the pre-reconstruction model is solved to obtain the extra frame pre-reconstruction image stack.

[0120] During the iterative solution process using the alternating multiplier method, the specific iterative formula for extra frame pre-reconstruction is as follows:

[0121]

[0122] In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H * This represents the equivalent mask during the reset scanning process, y e This represents the stack of additional frame images; i represents the iteration round. Let x represent the stack of intermediate extra frames pre-reconstructed images obtained in the i-th iteration. p_e This indicates the additional pre-reconstructed image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". p_eThe value of , Denotes the square of the second norm. w represents the second prior constraint result obtained in the i-th iteration. (i) Let represent the second convergence auxiliary parameter obtained in the i-th iteration, ρ represent the first convergence step size hyperparameter, and γ represent the second convergence step size hyperparameter. The typical value of both convergence step size hyperparameters is 1. The typical value of the preset total number of iterations is 30.

[0123] Step 3, Misalignment Estimation Stage: For the compressed frame pre-reconstructed image stack x obtained in Step 2 p_m and additional frame pre-reconstructed image stack x p_e The dimensions of the 3D image stack along the x-axis, y-axis, and z-axis are known to be h, w, and c, respectively. During the misalignment estimation stage, the dimensions of the pre-reconstructed image stack x in the compressed frame are... p_m and additional frame pre-reconstructed image stack x p_e A cross-correlation algorithm is used on the z-axis, and the peak position of the cross-correlation spectrum is determined to identify the misalignment amount of the reset scan and forward scan at the frame level, thereby determining the misalignment estimate d. est .

[0124] Specifically, using the cross-correlation method, with the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack obtained in step 2 as input, the following objective function is solved to obtain the misalignment amount during the reset process:

[0125]

[0126] In the formula, d est Let d represent the misalignment amount during the reset process, and x represent the misalignment amount during the reset process to be optimized. p_m Indicates the compressed frame pre-reconstructed image stack, x p_e argmax represents the extra frame pre-reconstructed image stack. d(·) represents the value of d when the function "·" takes its minimum value, max() represents the function to take the maximum value, min() represents the function to take the minimum value, h represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, w represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, c represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, i represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the x-axis, j represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the y-axis, k represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack (compressed frame pre-reconstructed image stack and extra frame pre-reconstructed image stack) on the z-axis, and ||||2 represents the second-order norm operation.

[0127] Step 4, Reconstruction Stage: Construct a priori model for resetting and enhancing reconstruction based on the misalignment amount during the resetting process, and establish a joint reconstruction and enhanced compressed sensing 3D reconstruction model based on the priori model for resetting and enhancing reconstruction.

[0128] During the reconstruction phase, this embodiment combines the compressed frame image stack and the additional frame image stack to complete the reconstruction. The specific joint reconstruction enhanced compressed sensing 3D reconstruction model is as follows:

[0129]

[0130] In the formula, x est ' represents the reconstructed 3D image stack, y m This represents the compressed frame image stack, Prior2(·) represents the second image denoising prior function, and x est This represents the reconstructed 3D image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". est The values ​​of REP(·) represent the reset enhancement reconstruction prior model, γ represents the first hyperparameter scalar, corresponding to the prior hyperparameter, ρ represents the second hyperparameter scalar, corresponding to the additional compressed frame enhancement hyperparameter, and y e Indicates the extra frame image stack H * D represents the equivalent mask during the reset scanning process. dest (·) represents a shift operation on the z-axis, indicating that the entire image stack "·" is shifted d along the z-axis. est pixels, d est D represents the misalignment during the reset process. dest (x est () represents the stack of reconstructed 3D images after misalignment. This represents the square of the second norm.

[0131] Step 5, Reconstruction Stage: After inputting the compressed frame image stack and the extra frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model, the solution is obtained using the alternating multiplier method according to the following iterative formula:

[0132]

[0133] v (i+1) =v (i) +γ(x est (i) -u est (i) )

[0134] w (i+1) =w (i) +ρ(x est (i) -p est (i) )

[0135] In the formula, y m This represents the compressed frame image stack. Let x represent the stack of reconstructed 3D images obtained in the i-th iteration. est This represents the reconstructed 3D image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". est The value of y. e This represents the stack of extra frame images. u represents the additional frame constraint result obtained in the i-th iteration. est This represents the reconstructed 3D image stack to be optimized. This represents the value of u when the function "·" reaches its minimum. est The value of v. (i) w represents the first convergence auxiliary parameter obtained in the i-th iteration. (i)Let represent the second convergence auxiliary parameter obtained in the i-th iteration, ρ represent the first convergence step size hyperparameter, γ represent the second convergence step size hyperparameter, and Prior2(·) be the specific image denoising prior. Here, the first norm constraint of the Hessian matrix is ​​usually chosen as the prior. The principle of this prior was first given in the paper (Huang, X., Fan, J., Li, L. et al. Fast, long-term, super-resolution imaging with Hessian structured illumination microscopy. Nat Biotechnol 36, 451–459 (2018). https: / / doi.org / 10.1038 / nbt.4115). In short, the principle of this prior is to constrain the first norm value of the Hessian matrix of the input image. dest (·) refers to the z-axis misalignment process, which is equivalent to shifting the image along the z-axis by d. est Pixels. Where the misalignment estimate d... est Obtained from the misalignment estimation stage. ρ and γ are both hyperparameter scalars, corresponding to the prior hyperparameter and the additional compressed frame enhancement hyperparameter, respectively, and their typical values ​​are both 1.

[0136] Last iteration This is the result of the reconstruction.

[0137] In step 5 of this embodiment, the typical value for the number of iterations is 40.

[0138] In this embodiment, the reconstruction results of the fluorescence microscopic images of the blood vessels in the mouse eyeball are as follows: Figure 5 The image stack for reconstructing the proposed scheme is shown in the figure.

[0139] Comparative Example 1

[0140] This comparative example is based on existing solutions in the relevant field. The difference between this comparative imaging process and the embodiment can be briefly summarized as follows: the resetting scan imaging process is removed, and no additional frame image stacks are acquired. The difference between this comparative reconstruction process and the embodiment can be briefly summarized as follows: the pre-reconstruction stage is removed, the misalignment estimation stage is removed, and the reconstruction stage is simplified.

[0141] This comparative example uses the open-top light-sheet microscope system described in the previous embodiment to acquire a compressed frame image stack of the sample. The imaging process in this comparative example is consistent with the previous embodiment in terms of initial state, start-up braking, and imaging. Specifically:

[0142] A. Imaging process:

[0143] 1. Initial state: The displacement stage is at position 0, and the sample to be imaged is placed on the sample stage.

[0144] The excitation path excites the light source to emit a laser beam.

[0145] 2. Start-up braking: The displacement table begins to accelerate until it reaches the working speed vm / s, then transitions to a constant speed. The typical value of v is 1×10. -4 After entering a constant speed state, the image sensor and light intensity encoder 11 are activated simultaneously.

[0146] 3. Imaging: The stage moves at a constant speed. Image sensor 14 records images at a maximum frame rate of f1Hz, with a typical value of 100Hz. The recorded series of images constitutes a compressed frame stack y. Light intensity encoder 11 plays a binary mask H with a period of B at f2Hz, where a typical value of B is 8 and a typical value of f2 is 800Hz. During imaging, the fluorescence signal generated by the fluorescent dye in the transparent sample, excited by the sheet-like light beam, passes through the probe objective lens 9 and the sleeve lens 10 to the light intensity encoder 11. The modulated signal generated by the light intensity encoder 11 is recorded by image sensor 14 via relay lens 12 and relay lens 13.

[0147] 4. Stop braking: After reaching the designated position, both the optical intensity encoder 11 and the image sensor 14 stop working. The displacement stage 8 enters a uniform deceleration process until it stops moving forward.

[0148] 5. Reset: After the displacement stage 8 stops moving forward, it accelerates in the opposite direction to vm / s and then enters a constant speed until the displacement stage 8 returns to the initial 0 position. The typical value of v is 1×10. -4 m / s, waiting for the operator to replace the sample or adjust the sample scanning depth.

[0149] B. Reconstruction Process:

[0150] The reconstruction uses the alternating multiplier method, and the model is constructed as follows:

[0151]

[0152] In the above formula, Prior*(·) represents a certain image prior, y represents the compressed frame stack, and i represents the iteration round. Let x represent the reconstructed image stack obtained in the i-th iteration, and let argmax represent the reconstructed image stack to be optimized. x (·) represents the value of x when the function "·" takes its minimum value. This represents the square of the second norm. w represents the prior constraint result obtained in the i-th iteration. (i) This represents the convergence auxiliary parameter obtained in the i-th iteration, where ρ represents the convergence step size hyperparameter, with a typical value of 1. The preset typical value for the total number of iterations is 40.

[0153] Last iteration This is the result of the reconstruction.

[0154] Comparative Example 2

[0155] This comparative example uses the method of Comparative Example 1, and applies the Prior*(·) prior from Comparative Example 1 as a WNNM (Weighted Nuclear Norm Minimization) prior. Based on this, fluorescence microscopy images of mouse eye vessels were reconstructed, and the results are as follows: Figure 6 As shown in the "WNNM scheme".

[0156] Comparative Example 3

[0157] This comparative example uses the method of Comparative Example 1, and applies the Prior*(·) prior from Comparative Example 1 as a combined prior of 2DTV (Two-dimensional Total Variation) and IDR (Iterative Data Refinement) priors. Based on this, fluorescence microscopic images of mouse eye vessels were reconstructed, and the results are as follows: Figure 6 The “2DTV+IDR scheme” is shown in the figure.

[0158] Comparative Example 4

[0159] This comparative example uses the method of Comparative Example 1, and applies the Prior*(·) prior from Comparative Example 1 as a 3DTV (3D Total Variation prior) prior. Based on this, fluorescence microscopic images of mouse eye vessels were reconstructed, and the results are as follows: Figure 6 The “3DTV solution” is shown in the document.

[0160] Comparative Example 5

[0161] This comparative example uses the method of Comparative Example 1, and applies the Prior*(·) prior from Comparative Example 1 as a first-order norm-constrained prior of the Hessian matrix. Based on this, fluorescence microscopic images of blood vessels in the mouse eyeball were reconstructed, and the results are as follows: Figure 5 The proposed scheme for removing the reverse scan reconstructed image stack is shown in the figure.

[0162] like Figure 5 As shown, comparing the reconstruction results obtained from the embodiment of the present invention with those from Comparative Example 5, it can be seen that the peak signal-to-noise ratio (PSNR) of this embodiment is improved by 8.212 compared to Comparative Example 5.

[0163] like Figure 6As shown, comparing the reconstruction results obtained by the embodiments of the present invention with those obtained by Comparative Examples 2 to 4, the present embodiment achieves a maximum improvement of 17.497 in terms of peak signal-to-noise ratio (PSNR).

[0164] In summary, the 3D compressed sensing reconstruction method provided by this invention can simultaneously improve the axial quality and signal-to-noise ratio of the reconstructed 3D image. Furthermore, as an enhancement method for compressed sensing reconstruction, this invention can be applied to any compression ratio. Therefore, when the reconstruction quality is constant, this method is faster than ordinary 3D compressed sensing imaging.

[0165] The above specific embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

[0166] The above description is only a preferred embodiment of the present invention. Therefore, all equivalent changes or modifications made to the structure, features and principles described in the claims of this patent application are included in the scope of this patent application.

Claims

1. A three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior, characterized in that, Includes the following steps: Step 1: Use a compressed sensing 3D imaging system to perform forward scanning and reset scanning in sequence to obtain the compressed frame image stack and the extra frame image stack of the sample. Step 2: Use the alternating multiplier method to pre-reconstruct the compressed frame image stack and the extra frame image stack respectively, to obtain the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack; Step 3: Use the cross-correlation method to process the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack to obtain the misalignment amount during the reset process; Step 4: Construct a priori model for resetting and enhanced reconstruction based on the misalignment during the resetting process, and establish a joint reconstruction and enhanced compressed sensing three-dimensional reconstruction model based on the priori model for resetting and enhanced reconstruction. Step 5: Input the compressed frame image stack and the additional frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model to obtain the reconstructed 3D image.

2. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 1, characterized in that: During both forward and reset scanning, the stage moves along the z-axis in opposite directions, and the scanning speed is constant. The speed ratio between the reset scan and the forward scan is an integer value greater than 1.

3. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 1, characterized in that: In step 2, the alternating multiplier method is used, with the compressed frame image stack as input, to solve the following objective function to obtain the compressed frame pre-reconstructed image stack: In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H represents the modulation mask, and y m Indicates the compressed frame image stack, x p_m Indicates the compressed frame pre-reconstructed image stack, x m This represents the pre-reconstructed image stack of compressed frames to be optimized. This represents the x value corresponding to the minimum value of the function "·". m The value of α represents the regularization hyperparameter. This represents the square of the second norm; In step 2, the alternating multiplier method is used, with the extra frame image stack as input, to solve the following objective function to obtain the extra frame pre-reconstructed image stack: In the formula, Prior1(·) represents the first three-dimensional denoising prior function, H * This represents the equivalent mask during the reset scanning process, y e x represents the extra frame image stack. p_e Indicates the extra frame pre-reconstructed image stack, x e This indicates the additional pre-reconstructed image stack to be optimized. This represents the x value corresponding to the minimum value of the function "·". e The value of α represents the regularization hyperparameter. This represents the square of the second norm.

4. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 3, characterized in that: The first three-dimensional denoising prior function adopts a three-dimensional total variation function.

5. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 1, characterized in that: Step 3 specifically involves: using the cross-correlation method, taking the compressed frame pre-reconstructed image stack and the extra frame pre-reconstructed image stack obtained in step 2 as input, solving the following objective function to obtain the misalignment amount during the reset process: In the formula, d est Let d represent the misalignment amount during the reset process, and x represent the misalignment amount during the reset process to be optimized. p_m Indicates the compressed frame pre-reconstructed image stack, x p_e argmax represents the extra frame pre-reconstructed image stack. d (·) represents the value of d when the function "·" takes its minimum value, max() represents the function to take the maximum value, min() represents the function to take the minimum value, h represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack on the x-axis, w represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack on the y-axis, c represents the size of the 3D image matrix corresponding to the pre-reconstructed image stack on the z-axis, i represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack on the x-axis, j represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack on the y-axis, k represents the index of the 3D image matrix corresponding to the pre-reconstructed image stack on the z-axis, and |||2 represents the second-order norm operation.

6. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 1, characterized in that: In step 4, the joint reconstruction enhanced compressed sensing 3D reconstruction model is set according to the following formula: In the formula, x est ′ represents the reconstructed 3D image stack, x est This represents the reconstructed 3D image stack to be optimized, y m This represents the compressed frame image stack, and Prior2(·) represents the second image denoising prior function. This represents the x value corresponding to the minimum value of the function "·". est The values ​​of are given by REP(·), which represents the prior model for resetting and enhancing reconstruction, γ, which represents the first hyperparameter scalar, and ρ, which represents the second hyperparameter scalar. This represents the square of the second norm.

7. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 6, characterized in that: In step 4, the prior model REP(x) is reset and enhanced for reconstruction. est Set it according to the following formula: In the formula, y e x represents the extra frame image stack. est H represents the reconstructed 3D image stack to be optimized. * D represents the equivalent mask during the reset scanning process. dest (·) indicates that the image stack "·" is translated d along the z-axis. est pixels, d est D represents the misalignment during the reset process. dest (x est () represents the stack of reconstructed 3D images after misalignment. This represents the square of the second norm.

8. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 6, characterized in that: The second image denoising prior function is specifically a Hessian matrix first-order norm constraint prior.

9. The three-dimensional compressed sensing reconstruction method based on resetting scan and joint prior as described in claim 6, characterized in that: In step 5, after inputting the compressed frame image stack and the extra frame image stack into the joint reconstruction enhanced compressed sensing 3D reconstruction model, the joint reconstruction enhanced compressed sensing 3D reconstruction model is solved by the alternating multiplier method to obtain the reconstructed 3D image of the sample.

10. A three-dimensional compressed sensing reconstruction system, characterized in that, include: The compressed sensing 3D imaging module employs a compressed sensing 3D imaging system with a sample reset or illumination scan backtracking process to acquire compressed frame image stacks and additional frame image stacks of the sample. A computation module is used to implement the steps of the three-dimensional compressed sensing reconstruction method as described in any one of claims 1 to 8.

Images