Image inversion method, device, electronic device and readable medium

Abstract

The present invention provides an image inversion method, device, electronic device and readable medium, the method comprising: sampling the observation signal by using a CUP-VISAR system to obtain a two-dimensional compressed observation image, based on the two-dimensional compressed observation image, using low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints, constructing an image inversion mathematical model based on a plug-and-play generalized alternating projection framework, decomposing the model into two optimization sub-problems about the value to be calculated and the auxiliary variables, alternately updating the value to be calculated and the auxiliary variables until the iteration termination condition is reached, and outputting the updated value to be calculated to obtain an interference fringe image. The present invention integrates the above multi-dimensional constraints to reconstruct the image, effectively suppresses noise interference and enhances the accuracy and robustness of image restoration, ensuring that the reconstructed image edge is clearer, more continuous, and has higher precision. And the inversion efficiency is improved based on the plug-and-play generalized alternating projection framework.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image inversion method, an image inversion device, an electronic device and a computer-readable medium. Background Art

[0002] Laser-driven inertial confinement fusion (ICF) is a very challenging research field in controlled nuclear fusion. There are many problems in ICF experiments, and diagnostic technology is an important research direction to promote progress. Among them, the imaging type velocity interferometer system for any reflector (VISAR) is the basic equipment for recording the shock wave information generated by ICF, and its imaging recording equipment is generally a streak camera or a framing camera. Among them, the 2D-VISAR using a framing camera only provides a few two-dimensional streak images with a low time resolution of ns scale, and cannot continuously record the entire process. Line imaging VISAR using a streak camera has a higher time resolution of ps scale, but due to the characteristics of the streak camera, if the slit of the streak camera is fully opened to record the 2D velocity, the streak images of different periods detected on the recording screen CCD will overlap, resulting in information errors. Therefore, it is usually used to detect the one-dimensional spatial information of the laser-driven shock wave velocity as a function of time. Although the interference fringes of the latest line imaging and view plane imaging velocity interferometers can be measured by streak cameras and framing cameras, it is still difficult to obtain two-dimensional information of shock wave velocity with high temporal resolution.

[0003] In 2019, someone proposed to obtain the CUP-VISAR system by coupling line imaging VISAR and compressed ultrafast photography (CUP). Its imaging method is to sample the original signal at a sampling rate significantly lower than the Nyquist sampling theorem to form a two-dimensional compressed image with high time resolution; further, the observation information is reconstructed through an inversion algorithm.

[0004] In recent years, some people have proposed a fast inversion method based on total variation (TV) constrained GAP based on the CUP-VISAR system. Among them, the mainstream inversion algorithms include the two-step iterative shrinkage / thresholding (TwIST) method based on total variation constraints, the generalized alternating projection (GAP) method based on total variation constraints, and the alternating direction method of multipliers (ADMM) based on total variation and low-rank constraints.

[0005] However, at present, the inversion quality and accuracy of CUP-VISAR two-dimensional compressed images are not high, and there are several main disadvantages:

[0006] a. There are factors such as information loss and noise pollution in the CUP-VISAR measurement process, which makes the high spatial resolution of the obtained observation compressed image poor;

[0007] b. The denoising process of noise pollution in the algorithm is conducive to enhancing the algorithm's noise resistance, but it is easy to lose image details;

[0008] c. The TV-TwIST (TwIST based on total variation constraints) algorithm is easily affected by the regularization parameters, resulting in unstable inversion results;

[0009] d. There is a correlation in the time dimension between each frame of the two-dimensional observation image, but previous algorithms did not extract and utilize this feature information. Summary of the invention

[0010] In view of the above problems, the present invention is proposed to provide an image inversion method and a corresponding image inversion device, an electronic device and a computer-readable medium that overcome the above problems or at least partially solve the above problems.

[0011] The present invention discloses an image inversion method, which comprises:

[0012] The CUP-VISAR system is used to sample the observation signal and obtain a two-dimensional compressed observation image;

[0013] According to the two-dimensional compressed observation image, a mathematical model of image inversion based on a plug-and-play generalized alternating projection framework is constructed with low-rank prior, total variation prior, Kalman prediction prior and inter-frame prediction prior as constraints;

[0014] The image inversion mathematical model is decomposed into two optimization sub-problems about the value to be calculated and the auxiliary variable. The value to be calculated and the auxiliary variable are updated alternately until the iteration termination condition is reached, then the alternating update is stopped, and the updated value to be calculated is output to obtain an interference fringe image.

[0015] Optionally, a CUP-VISAR system is used to sample the observation signal to obtain a two-dimensional compressed observation image, including:

[0016] The VISAR system carries the target surface information and generates interference to form an interference fringe image at the imaging plane, thereby obtaining the observation signal of the interference fringe image.

[0017] The observation signal of the interference fringe image is input into the CUP system, and the digital micromirror device in the CUP system encodes the interference fringe image to obtain an encoded image;

[0018] After the coded image enters the streak camera, the streak camera cuts and superimposes the coded image to generate a two-dimensional compressed observation image.

[0019] Optionally, the two-dimensional compressed observation image is expressed as:

[0020] Y=TSCI

[0021] Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

[0022] Optionally, according to the two-dimensional compressed observation image, a mathematical model of image inversion based on a plug-and-play generalized alternating projection framework is constructed with low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints, including:

[0023] The expression of the two-dimensional compressed observation image is converted into:

[0024]

[0025] stY=TSCI

[0026] Taking low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior as constraints, we get:

[0027] minλ||I|| TV +τR rank (I)

[0028] stY=TSCI、I=Γ kalman (I), I = Λ interframe (I)

[0029] Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ||I|| TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interframe (I) is the inter-frame prediction operator;

[0030] Applying the plug-and-play generalized alternating projection framework to the above equation and introducing the auxiliary variable u, we obtain the image inversion mathematical model based on the plug-and-play generalized alternating projection framework:

[0031]

[0032] Among them, argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ||·|| * represents the nuclear norm.

[0033] Optionally, the image inversion mathematical model is decomposed into two optimization sub-problems about the value to be evaluated and the auxiliary variable and expressed as:

[0034] The optimized expression to be evaluated is:

[0035]

[0036] Optimization expression of auxiliary variables:

[0037]

[0038] Optionally, the Kalman prediction prior term is determined as follows:

[0039] Expand each frame of the two-dimensional compressed observation image of the current iteration number by columns to generate a one-dimensional signal of each frame of the two-dimensional compressed observation image;

[0040] Performing Kalman prediction on the one-dimensional signal of each frame of the two-dimensional compressed observation image to obtain a one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number;

[0041] The one-dimensional signal prediction value of each frame of two-dimensional compressed observation image at the next iteration is reconstructed into a two-dimensional compressed observation image predicted by Kalman.

[0042] Optionally, the process of determining the inter-frame prediction prior item is:

[0043] By analyzing the phase difference of each line of stripes in the two frames of two-dimensional compressed observation images in the same iteration, the displacement pixel difference of each line of stripes in the two frames of two-dimensional compressed observation images is determined.

[0044] Fitting each row of fringes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of fringes of the previous frame two-dimensional compressed observation image;

[0045] Generate a predicted fringe sequence by using the cosine function of each row of fringe of the previous frame two-dimensional compressed observation image and the displacement pixel difference of each row of fringe of the previous and next two frames of two-dimensional compressed observation image;

[0046] According to the length of each row of stripes in the rear frame two-dimensional compressed observation image, each row of stripes in the rear frame two-dimensional compressed observation image is extracted from the predicted pixel value sequence and spliced ​​into a predicted rear frame two-dimensional compressed observation image.

[0047] The present invention also discloses an image inversion device, which comprises:

[0048] A two-dimensional compressed observation image acquisition module is used to sample the observation signal using the CUP-VISAR system to obtain a two-dimensional compressed observation image;

[0049] An image inversion mathematical model construction module is used to construct an image inversion mathematical model based on a plug-and-play generalized alternating projection framework according to the two-dimensional compressed observation image, with low-rank prior terms, total variation prior terms, Kalman prediction prior terms, and inter-frame prediction prior terms as constraints;

[0050] The model solving module is used to decompose the image inversion mathematical model into two optimization sub-problems about the value to be solved and the auxiliary variable, alternately update the value to be solved and the auxiliary variable until the iteration termination condition is reached, then stop the alternating update, output the updated value to be solved, and obtain the interference fringe image.

[0051] Optionally, the two-dimensional compressed observation image acquisition module includes:

[0052] The observation signal generation submodule is used for the VISAR system to carry the target surface information to interfere and form an interference fringe image at the imaging plane, and obtain the observation signal of the interference fringe image;

[0053] The encoding submodule is used to input the observation signal of the interference fringe image into the CUP system, and the digital micromirror device in the CUP system encodes the interference fringe image to obtain a coded image;

[0054] The two-dimensional compressed observation image generation submodule is used for the streak camera to cut and superimpose the coded image after the coded image enters the streak camera to generate a two-dimensional compressed observation image.

[0055] Optionally, the two-dimensional compressed observation image is expressed as:

[0056] Y=TSCI

[0057] Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

[0058] Optionally, the image inversion mathematical model construction module includes:

[0059] The to-be-evaluated expression constructing submodule is used to convert the expression of the two-dimensional compressed observation image into the following by adding the prior constraints to be evaluated:

[0060]

[0061] stY=TSCI

[0062] The prior constraint submodule is used to use low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints to obtain:

[0063] minλ||I|| TV +τR rank (I)

[0064] stY=TSCI、I=Γ kalman (I), I = Λ interframe (I)

[0065] Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ||I|| TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interframe (I) is the inter-frame prediction operator;

[0066] The image inversion mathematical model construction submodule is used to apply the plug-and-play generalized alternating projection framework to the above formula and introduce the auxiliary variable u to obtain the image inversion mathematical model based on the plug-and-play generalized alternating projection framework:

[0067]

[0068] Among them, argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ‖·‖ * represents the nuclear norm.

[0069] Optionally, the image inversion mathematical model is decomposed into two optimization sub-problems about the value to be evaluated and the auxiliary variable and expressed as:

[0070] The optimized expression to be evaluated is:

[0071]

[0072] Optimization expression of auxiliary variables:

[0073]

[0074] Optionally, the image inversion device further includes:

[0075] A one-dimensional signal conversion module, used for expanding each frame of the two-dimensional compressed observation image of the current iteration number by column to generate a one-dimensional signal of each frame of the two-dimensional compressed observation image;

[0076] A one-dimensional signal prediction module is used to perform Kalman prediction on the one-dimensional signal of each frame of the two-dimensional compressed observation image to obtain a one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number;

[0077] The Kalman prediction image reconstruction module is used to reconstruct the one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number into a two-dimensional compressed observation image predicted by Kalman.

[0078] Optionally, the image inversion device further includes:

[0079] The displacement pixel difference determination module is used to determine the displacement pixel difference of each row of stripes in the two previous and next frames of two-dimensional compressed observation images by analyzing the phase difference of each row of stripes in the two previous and next frames of two-dimensional compressed observation images in the same iteration.

[0080] A fitting module is used to fit each row of stripes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of stripes of the previous frame two-dimensional compressed observation image;

[0081] A subsequent frame prediction module, used to generate a predicted stripe sequence by using the cosine function of each stripe row of the two-dimensional compressed observation image of the previous frame and the displacement pixel difference of each stripe row of the two-dimensional compressed observation images of the previous and next frames;

[0082] The post-frame generation module is used to extract each row of stripes of the post-frame two-dimensional compressed observation image from the predicted pixel value sequence according to the length of each row of stripes of the post-frame two-dimensional compressed observation image, and splice them into the predicted post-frame two-dimensional compressed observation image.

[0083] The present invention also discloses an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0084] The memory is used to store computer programs;

[0085] The processor is used to implement the image inversion method of the present invention when executing the program stored in the memory.

[0086] The present invention also discloses one or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the image inversion method according to the present invention.

[0087] The present invention includes the following advantages:

[0088] The image inversion method of the present invention obtains a two-dimensional compressed observation image by sampling the observation signal using a CUP-VISAR system, and constructs an image inversion mathematical model based on a plug-and-play generalized alternating projection framework based on the two-dimensional compressed observation image with low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior as constraints, decomposes the image inversion mathematical model into two optimization sub-problems about the value to be evaluated and the auxiliary variable, alternately updates the value to be evaluated and the auxiliary variable, stops alternating updating until the iteration termination condition is reached, outputs the updated value to be evaluated, and obtains an interference fringe image. The present invention skillfully integrates multi-dimensional constraints such as low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior to reconstruct the image, effectively suppresses noise interference and enhances the accuracy and robustness of image restoration, and ensures that the reconstructed image edge is clearer and more continuous with higher precision. In addition, the image inversion mathematical model for reconstructing the image is built based on a plug-and-play generalized alternating projection framework, which makes the image inversion mathematical model have plug-and-play characteristics, effectively improves the inversion efficiency, and can also flexibly adjust the prior terms and constraints according to specific needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 is a flowchart of the steps of an image inversion method provided by an embodiment of the present invention;

[0090] Figure 2 It is a schematic diagram of the inter-frame prediction concept provided by an embodiment of the present invention;

[0091] Figure 3 is an inter-frame prediction flow chart provided by an embodiment of the present invention;

[0092] Figure 4 is a schematic diagram of CUP-VISAR system imaging provided by an embodiment of the present invention;

[0093] Figure 5 is a schematic diagram of compressed imaging provided by an embodiment of the present invention;

[0094] Figure 6 It is a flow chart of solving the image inversion mathematical model based on plug-and-play generalized alternating projection provided by an embodiment of the present invention;

[0095] FIG. 7( a ) is a simulated original interference fringe diagram of a simulation experiment of the present invention;

[0096] FIG. 7( b ) is a simulated original interference fringe diagram of a simulation experiment of the present invention;

[0097] FIG7( c ) is a noise observation image of a simulation experiment of the present invention;

[0098] Figure 8 It is the interference fringe sequence reconstructed by the ADMM-TVLR algorithm in the simulation experiment;

[0099] Fig. 9 It is the interference fringe sequence reconstructed by the simulation experiment through the variable acceleration GAP-TVLR algorithm;

[0100] Fig.10 It is the interference fringe sequence reconstructed by the PnP-GAP-TLKI algorithm of the present invention in the simulation experiment;

[0101] FIG11( a ) is an interference fringe image obtained by reconstructing the interference fringe arrangement using the ADMM-TVLR algorithm in a simulation experiment;

[0102] FIG11( b ) is an interference fringe image obtained by reconstructing the interference fringe arrangement using the variable acceleration GAP-TVLR algorithm in the simulation experiment;

[0103] FIG. 11( c ) is an interference fringe image obtained by reconstructing the interference fringe arrangement using the PnP-GAP-TLKI algorithm of the present invention in a simulation experiment;

[0104] Figure 12(a) is a comparison chart of peak signal-to-noise ratio (PSNR) of ADMM-TVLR, GAP-TVLR, and PnP-GAP-TLKI;

[0105] Figure 12(b) is a structural similarity SSIM comparison diagram of ADMM-TVLR, GAP-TVLR, and PnP-GAP-TLKI;

[0106] Fig.13 It is a structural block diagram of an image inversion device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0107] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0108] Reference Figure 1 , shows a flowchart of a base image inversion method provided in an embodiment of the present invention, which may specifically include the following steps:

[0109] Step 101, sampling the observation signal using the CUP-VISAR system to obtain a two-dimensional compressed observation image;

[0110] Step 102, according to the two-dimensional compressed observation image, using low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints, constructing an image inversion mathematical model based on a plug-and-play generalized alternating projection framework;

[0111] Step 103, decomposing the image inversion mathematical model into two optimization sub-problems about the value to be calculated and the auxiliary variable, alternately updating the value to be calculated and the auxiliary variable until the iteration termination condition is reached, then stopping the alternating update, outputting the updated value to be calculated, and obtaining the interference fringe image.

[0112] In the embodiment of the present invention, the CUP-VISAR system is first used to sample the observation signal to obtain a two-dimensional compressed observation image, which provides basic data for subsequent image inversion. Then, based on the obtained two-dimensional compressed observation image, a mathematical model of image inversion based on the Plug-and-Play Generalized Alternating Projection (PnP-GAP) framework is constructed with low-rank priors, total variation (TV) priors, Kalman prediction priors, and inter-frame prediction priors as constraints, so that the mathematical model of image inversion can make full use of the advantages of the PnP-GAP algorithm, decompose the complex large-scale optimization problem into two optimization sub-problems about the value to be calculated and the auxiliary variable, and gradually approach the optimal solution by alternately solving these two optimization sub-problems, so as to reconstruct the interference fringe image. And the mathematical model of image inversion using the Plug-and-Play Generalized Alternating Projection (PnP-GAP) framework has high convergence and convergence speed, ensuring the stability and efficiency of the inversion process.

[0113] In the image inversion mathematical model based on the Plug-and-Play Generalized Alternating Projection (PnP-GAP) framework, the low-rank prior is used to describe the low-rank characteristics of the stripe image, the TV total variation prior is used to protect the image edge texture details while denoising, and the Kalman prediction prior is used to make full use of the correlation information between the same frame images in each iteration. This can be done by extracting the correlation information of the same frame image in the previous and next iterations and performing Kalman prediction. The inter-frame prediction prior is used to make full use of the correlation information between the previous and next frames. This can be done by mining the correlation information of the previous frame image to the next frame image in the same iteration and performing inter-frame prediction.

[0114] The image inversion method of the present invention cleverly integrates multi-dimensional constraints such as low-rank prior, total variation prior, Kalman prediction prior and inter-frame prediction prior to reconstruct the image, effectively suppresses noise interference and enhances the accuracy and robustness of image restoration, ensuring that the reconstructed image edge is clearer and more continuous with higher precision. In addition, the image inversion mathematical model of the reconstructed image is constructed based on a plug-and-play generalized alternating projection framework, so that the image inversion mathematical model has a plug-and-play feature, effectively improves the inversion efficiency, and can also flexibly adjust the prior terms and constraints according to specific needs.

[0115] In one embodiment of the present invention, the process of determining the Kalman prediction prior term is as follows:

[0116] Expand each frame of the two-dimensional compressed observation image of the current iteration number by columns to generate a one-dimensional signal of each frame of the two-dimensional compressed observation image;

[0117] Performing Kalman prediction on the one-dimensional signal of each frame of the two-dimensional compressed observation image to obtain a one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number;

[0118] The one-dimensional signal prediction value of each frame of two-dimensional compressed observation image at the next iteration is reconstructed into a two-dimensional compressed observation image predicted by Kalman.

[0119] In the iteratively reconstructed fringe image, as the number of iterations increases, the iterations of the same frame image have a higher similarity. Therefore, relevant information can be extracted from the perspective of the previous and next iterations of the same frame image. The changes in the same frame image with the increase in the number of iterations can be regarded as the changes in the same frame image over time. The embodiment of the present invention uses the Kalman prediction algorithm to extract relevant prior information.

[0120] Considering the correlation between the frames of the two-dimensional observation image in the time dimension, Kalman prediction is introduced into the inversion algorithm. Based on the characteristics of Kalman prediction, the basic idea of ​​Kalman prediction in this method is: expand the two-dimensional observation fringe image by column to generate a one-dimensional signal, perform Kalman prediction on the one-dimensional signal, and finally reconstruct the predicted one-dimensional signal into a two-dimensional fringe image. The specific steps are as follows:

[0121] 1) The i-th frame image of the k-1 iteration Expand to a column vector X(k-1|k-1).

[0122] 2) According to the column vector of the i-th frame image in the k-1th iteration, estimate the column vector X(k|k-1) of the i-th frame image in the k-th iteration:

[0123]

[0124] In the formula, is the state transition matrix corresponding to the k-1th iteration to the k-th iteration, and W(k-1) is the preset environmental noise distribution of the k-1th iteration.

[0125] 3) Based on the error covariance matrix of the i-th frame image column vector in the k-1th iteration, estimate the error covariance matrix P(k|k-1) of the i-th frame image column vector in the k-th iteration:

[0126]

[0127] Where M T represents the transposed matrix of matrix M, P(k-1|k-1) represents the error covariance of the column vector of the i-th frame image after updating in the k-1th iteration, and Q represents the state error covariance. P(0|0)=E[((X(0)-μ0)((X(0)-μ0) T ]=P0, E(M) represents the mean of matrix M.

[0128] 4) Update the Kalman gain in the kth iteration:

[0129]

[0130] Among them, H is the observation matrix corresponding to the current iteration, and R is the measurement noise covariance matrix.

[0131] 5) Use the Kalman gain to update the i-th frame image column vector in the k-th iteration:

[0132] X(k|k)=X(k|k-1)+K(k)[X(k-1|k-1)-HX(k|k-1)](4)

[0133] 6) Update error covariance:

[0134] P(k|k)=[EK(k)H]P(k|k-1)(5)

[0135] Where E is the identity matrix.

[0136] 7) Reshape the updated i-th frame image column vector into a two-dimensional compressed observation image

[0137] In one embodiment of the present invention, the process of determining the inter-frame prediction prior term is as follows:

[0138] By analyzing the phase difference of each line of stripes in the two frames of two-dimensional compressed observation images in the same iteration, the displacement pixel difference of each line of stripes in the two frames of two-dimensional compressed observation images is determined.

[0139] Fitting each row of fringes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of fringes of the previous frame two-dimensional compressed observation image;

[0140] Generate a predicted fringe sequence by using the cosine function of each row of fringe of the previous frame two-dimensional compressed observation image and the displacement pixel difference of each row of fringe of the previous and next two frames of two-dimensional compressed observation image;

[0141] According to the length of each row of stripes in the rear frame two-dimensional compressed observation image, each row of stripes in the rear frame two-dimensional compressed observation image is extracted from the predicted pixel value sequence and spliced ​​into a predicted rear frame two-dimensional compressed observation image.

[0142] In the VISAR part of the data acquisition process, the clipping part can be regarded as a moving window on a single two-dimensional observation image Y, and the latter image is essentially an offset of the previous image. Therefore, there is a certain close correlation between the previous and next frames. Since the two-dimensional fringe image has obvious fringe characteristics, it can provide the model with inter-frame prior information from the perspective of fringe properties. The upper and lower frame images can be regarded as cosine functions with the same period, the same maximum value, and different initial phases. By fitting the cosine function for each fringe image, the upper and lower frames of the image can be regarded as two-dimensional fringe images composed of different intervals on the same cosine function image.

[0143] According to the CUP-VISAR data acquisition process, there is a certain displacement relationship and image data correlation between the previous frame and the next frame of the compressed image. At the same time, considering the general characteristics of the fringe image, the basic idea of ​​inter-frame prediction is as follows:

[0144] 1) The displacement pixel difference between the two frames is obtained by the phase difference between the two frames in the same iteration.

[0145] 2) Fit each row of the previous frame of the compressed image to The cosine curve is used to determine the parameter value in the cosine function based on the cosine curve, thereby determining the cosine function expression.

[0146] Where b is the background light intensity, a is the amplitude intensity, T c is the image period, is the initial phase of the current frame image.

[0147] 3) Generate a predicted stripe sequence by using the cosine function of each stripe row of the previous frame two-dimensional compressed observation image and the displacement pixel difference of each stripe row of the previous and next two-dimensional compressed observation images.

[0148] 4) According to the length of each row of stripes in the subsequent frame two-dimensional compressed observation image, each row of stripes in the subsequent frame two-dimensional compressed observation image is extracted from the predicted pixel value sequence, and the stripes are spliced ​​into the predicted subsequent frame two-dimensional compressed observation image.

[0149] The basic idea and prediction process of the above inter-frame prediction are as follows Figure 2 , 3 As shown in the figure, i represents the i-th frame image.

[0150] In one embodiment of the present invention, a CUP-VISAR system is used to sample an observation signal to obtain a two-dimensional compressed observation image, including:

[0151] The VISAR system carries the target surface information and generates interference to form an interference fringe image at the imaging plane, thereby obtaining the observation signal of the interference fringe image.

[0152] The observation signal of the interference fringe image is input into the CUP system, and the digital micromirror device in the CUP system encodes the interference fringe image to obtain an encoded image;

[0153] After the coded image enters the streak camera, the streak camera cuts and superimposes the coded image to generate a two-dimensional compressed observation image.

[0154] In the embodiment of the present invention, an experimental optical path is constructed to collect observation signals and output a two-dimensional compressed observation image. Specifically:

[0155] CUP-VISAR system imaging Figure 4 As shown, it mainly includes VISAR system and CUP system. Figure 5 , which is a schematic diagram of the compression imaging principle. In the VISAR system, after the probe light source reaches the target surface TSF through the optical path, the probe reflected light carrying the Doppler frequency shift information reaches two optical branches with interferometers, and finally the two branches interfere with each other to form interference fringes at the imaging plane IP2. In the CUP system, the incoming interference fringes enter the digital micromirror device DMD from the beam splitter (BS4) optical path for encoding, and the encoded image enters the corresponding streak camera (steak camera2) for shearing and superposition, and finally generates a two-dimensional compression observation image of the shock wave. At the same time, another light beam at BS4 reaches the streak camera (steak camera1) through the optical path when the additional branch slit is not fully opened, generating a row of streak images. Finally, a 1D-VISAR image is obtained at streak camera 1 (steak camera1), and a two-dimensional compression observation image is obtained at streak camera (steak camera2).

[0156] In one embodiment of the present invention, the two-dimensional compressed observation image is represented as:

[0157] Y=TSCI

[0158] Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

[0159] The expression of the two-dimensional compressed observation image obtained by CUP-VISAR system imaging can be expressed as:

[0160] Y=TSCI(6)

[0161] Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

[0162] In one embodiment of the present invention, according to the two-dimensional compressed observation image, a mathematical model of image inversion based on a plug-and-play generalized alternating projection framework is constructed with low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints, including:

[0163] The expression of the two-dimensional compressed observation image is converted into:

[0164]

[0165] stY=TSCI

[0166] Taking low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior as constraints, we get:

[0167] minλ||||I| TV +τR rank (I)

[0168] stY=TSCI、I=Γ kalman (I), I = Λ interframe (I)

[0169] Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ||I|| TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interframe (I) is the inter-frame prediction operator;

[0170] Applying the plug-and-play generalized alternating projection framework to the above equation and introducing the auxiliary variable u, we obtain the image inversion mathematical model based on the plug-and-play generalized alternating projection framework:

[0171]

[0172] Among them, argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ||·|| * represents the nuclear norm.

[0173] In the embodiment of the present invention, based on the collected two-dimensional compressed observation image, the specific process of constructing the image inversion mathematical model based on the plug-and-play generalized alternating projection framework is as follows, by integrating the prior constraints such as low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior:

[0174] First, add the prior constraints to be evaluated to convert the expression of the two-dimensional compressed observation image into:

[0175]

[0176] Since the number of unknowns is much greater than the number of equations, it is a problem of solving ill-conditioned equations. By using low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints, we can convert it into an optimization problem, which can be expressed as:

[0177]

[0178] Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ||I|| TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interfram e(I) is the inter-frame prediction operator;

[0179] The optimization problem corresponding to formula (8) is transformed into an unconstrained optimization problem for solution, which is:

[0180]

[0181] Where argmin{·} is the variable value when the target area is minimized. rank (I) is an NP-hard problem, so R rank (I) Perform nuclear norm approximation, ||·|| * represents the nuclear norm, that is, the sum of all singular values ​​of the matrix, so formula (9) is transformed into:

[0182]

[0183] In order to effectively solve the proposed optimization problem, the PnP-GAP algorithm is used for alternating update iterations. The auxiliary variable u is introduced, and we have:

[0184]

[0185] Among them, argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ||·|| * represents the nuclear norm.

[0186] In one embodiment of the present invention, the image inversion mathematical model is decomposed into two optimization sub-problems about the value to be evaluated and the auxiliary variable, which are expressed as follows:

[0187] The optimized expression to be evaluated is:

[0188]

[0189] Optimization expression of auxiliary variables:

[0190]

[0191] After the introduction of auxiliary variables, the mathematical model of image inversion can be solved by decomposing the mathematical model into two optimization sub-problems about the value to be evaluated and the auxiliary variables. Specifically, in the iteration process, variable u is first fixed to update I, and then variable I is fixed to update u. Therefore, in this optimization problem, formula (11) can be divided into two iterative sub-formulas:

[0192]

[0193]

[0194] Therefore, the multi-constrained optimization problem based on the PnP-GAP framework has the following iterative steps:

[0195] In the k+1th iteration:

[0196] 1) Fix the u variable and update the I variable:

[0197] because is a differentiable term, ||I|| TV It is a non-differentiable term, so it is processed as follows:

[0198]

[0199] Where δ1 and δ2 are weight parameters. Update by Euclidean projection on the linear manifold u Then, the proximal mapping method is used to update Therefore, formula (14) and formula (15) can be converted into the following form:

[0200]

[0201] In the formula, let TSC = A, A is the observation matrix, prox TV {} is the proximal mapping operator with respect to total variation TV.

[0202] 2) Fix the I variable and update the u variable:

[0203] Since ||u|| * is the nuclear norm, and the singular value thresholding (SVT) method is used to minimize the nuclear norm, and the update of u is converted into the following form:

[0204]

[0205] In the formula, is u k The singular value decomposition of The representative threshold is The soft threshold operator is

[0206] In order to speed up the optimization solution of I and make the projection of the linear manifold adaptive, formula (17) is modified as follows:

[0207]

[0208] In the formula, Y1=Y.

[0209] Noise is an important factor affecting the inversion quality. The reason is that when there is noise in the inversion data, as the number of iterations increases, Y k+1 will show a trend of gradual increase, making Y k+1 -A·u k The error of accumulating continuously, when the error accumulates to a certain extent, will have a significant impact on the optimization solution, and eventually lead to the decline of inversion quality. Therefore, the embodiment of the present invention performs variable acceleration processing on GAP, sets the error threshold ρ to adjust the projection weight parameter N. When (Y k -A·u k )≤ρ, N is set to 0, the weight is 1, and the information contained in y is fully utilized to converge quickly; when (Y k -A·u k )>ρ, cancel the acceleration and increase Reducing the projection weight effectively reduces Y that carries too much error information k Therefore, formula (19) can be rewritten as:

[0210]

[0211] Among them, ρ represents whether to adopt the acceleration strategy (Y k-A·u k ) error threshold.

[0212] Reference Figure 6 Finally, through the iterative solution of the above steps, when the number of iterations k+1 is greater than the maximum number of iterations iter max or Y k+1 -A.I k When the error is less than the error threshold ε, the iteration stops and the reconstructed interference fringe image I is output. k+1 .

[0213] Next, the effectiveness of the image inversion method provided by the present invention is verified:

[0214] In order to intuitively demonstrate the effectiveness of the proposed method in actual experiments, the reconstructed image quality is higher. The image inversion method based on plug-and-play generalized alternating projection (PnP-GAP-TLKI) proposed in the present invention is compared with the alternating direction multiplier method ADMM-TVLR based on total variation and low-rank constraints and the variable accelerated generalized alternating projection method GAP-TVLR in real experimental reconstruction images. The experiment uses peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) as evaluation indicators to measure image quality.

[0215] The simulation experiment uses a simulated original interference fringe pattern with a size of 790(x)×330(y)×25(t), as shown in Figure 7(a). The interference fringe pattern data is representative, and its shock wave has experienced stationary, acceleration, deceleration and other processes along the time dimension t, which contains rich information. In order to be closer to the real complex environment, the simulated DMD uses a coding pattern with a coding aperture of 8pixel×8pixel, as shown in Figure 7(b). In order to reflect that the image inversion quality of the method of the present invention is higher when the noise cannot be ignored, a larger noise is introduced in the simulation experiment and the inversion images of the three methods are compared and analyzed. Gaussian noise with a mean of 0 and a variance of 0.5 is introduced into the fringe image. The noise observation image obtained by simulating the data acquisition process of the CUP-VISAR system is shown in Figure 7(c), with a pixel size of 790pixel×389pixel.

[0216] The fringe sequence reconstructed by the ADMM-TVLR algorithm is as follows Figure 8 As shown, the stripe sequence reconstructed by variable acceleration GAP-TVLR is as follows Fig. 9 As shown, the proposed method (PnP-GAP-TLKI) reconstructs the stripe sequence as Fig.10 shown.

[0217] like Figure 8 , 9As shown in Figures 10, compared with ADMM-TVLR and variable acceleration GAP-TVLR, the fringe image reconstructed by the method proposed in the present invention has clearer and more continuous edges. By arranging each row of reconstructed fringe images in chronological order, some new images as shown in Figure 11 are obtained, 11(a) is the interference fringe image reconstructed by the ADMM-TVLR algorithm, 11(b) is the interference fringe image reconstructed by the variable acceleration GAP-TVLR algorithm, and 11(c) is the interference fringe image reconstructed by the PnP-GAP-TLKI algorithm.

[0218] In order to quantitatively analyze the quality of the reconstructed image, Figure 12 (a) (b) shows the comparison of PSNR and SSIM. The PSNR distribution of the ADMM-TVLR reconstruction algorithm is 11.94dB~15.01dB, and the SSIM distribution is 47.97%~62.54%. The PSNR distribution of the variable acceleration GAP-TVLR reconstruction algorithm is 12.06dB~19.64dB, and the SSIM distribution is 55.11%~81.98%. The PSNR distribution of the reconstruction algorithm of the method proposed in the present invention (PnP-GAP-TLKI) is 14.55dB~22.68dB, and the SSIM distribution is 68.11%~90.54%. It can be seen that the PnP-GAP-TLKI algorithm has obtained a higher numerical range in both PSNR and SSIM indicators, indicating that this method can reconstruct images more accurately while maintaining more structural information and details, and is significantly better than the ADMM-TVLR and variable acceleration GAP-TVLR reconstruction algorithms in terms of reconstructed image quality.

[0219] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0220] Reference Figure 3 , shows a structural block diagram of an image inversion device provided in an embodiment of the present invention, which may specifically include the following modules:

[0221] A two-dimensional compressed observation image acquisition module 1301 is used to sample the observation signal using the CUP-VISAR system to obtain a two-dimensional compressed observation image;

[0222] An image inversion mathematical model construction module 1302 is used to construct an image inversion mathematical model based on a plug-and-play generalized alternating projection framework according to the two-dimensional compressed observation image, using low-rank prior terms, total variation prior terms, Kalman prediction prior terms, and inter-frame prediction prior terms as constraints;

[0223] The model solving module 1303 is used to decompose the image inversion mathematical model into two optimization sub-problems about the value to be calculated and the auxiliary variable, alternately update the value to be calculated and the auxiliary variable until the iteration termination condition is reached, then stop the alternating update, output the updated value to be calculated, and obtain the interference fringe image.

[0224] Optionally, the two-dimensional compressed observation image acquisition module includes:

[0225] The observation signal generation submodule is used for the VISAR system to carry the target surface information to interfere and form an interference fringe image at the imaging plane, and obtain the observation signal of the interference fringe image;

[0226] The encoding submodule is used to input the observation signal of the interference fringe image into the CUP system, and the digital micromirror device in the CUP system encodes the interference fringe image to obtain a coded image;

[0227] The two-dimensional compressed observation image generation submodule is used for the streak camera to cut and superimpose the coded image after the coded image enters the streak camera to generate a two-dimensional compressed observation image.

[0228] Optionally, the two-dimensional compressed observation image is expressed as:

[0229] Y=TSCI

[0230] Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

[0231] Optionally, the image inversion mathematical model construction module includes:

[0232] The to-be-evaluated expression constructing submodule is used to convert the expression of the two-dimensional compressed observation image into the following by adding the prior constraints to be evaluated:

[0233]

[0234] stY=TSCI

[0235] The prior constraint submodule is used to use low-rank priors, total variation priors, Kalman prediction priors, and inter-frame prediction priors as constraints to obtain:

[0236] minλ||I|| TV +TR rank (I)

[0237] stY=TSCI、I=Γ kalman (I), I = Λ interframe (I)

[0238] Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ||I|| TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interframe (I) is the inter-frame prediction operator;

[0239] The image inversion mathematical model construction submodule is used to apply the plug-and-play generalized alternating projection framework to the above formula and introduce the auxiliary variable u to obtain the image inversion mathematical model based on the plug-and-play generalized alternating projection framework:

[0240]

[0241] Among them, argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ||·|| * represents the nuclear norm.

[0242] Optionally, the image inversion mathematical model is decomposed into two optimization sub-problems about the value to be evaluated and the auxiliary variable and expressed as:

[0243] The optimized expression to be evaluated is:

[0244]

[0245] Optimization expression of auxiliary variables:

[0246]

[0247] Optionally, the image inversion device further includes:

[0248] A one-dimensional signal conversion module, used for expanding each frame of the two-dimensional compressed observation image of the current iteration number by column to generate a one-dimensional signal of each frame of the two-dimensional compressed observation image;

[0249] A one-dimensional signal prediction module is used to perform Kalman prediction on the one-dimensional signal of each frame of the two-dimensional compressed observation image to obtain a one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number;

[0250] The Kalman prediction image reconstruction module is used to reconstruct the one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number into a two-dimensional compressed observation image predicted by Kalman.

[0251] Optionally, the image inversion device further includes:

[0252] The displacement pixel difference determination module is used to determine the displacement pixel difference of each row of stripes in the two previous and next frames of two-dimensional compressed observation images by analyzing the phase difference of each row of stripes in the two previous and next frames of two-dimensional compressed observation images in the same iteration.

[0253] A fitting module is used to fit each row of stripes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of stripes of the previous frame two-dimensional compressed observation image;

[0254] A subsequent frame prediction module, used to generate a predicted stripe sequence by using the cosine function of each stripe row of the two-dimensional compressed observation image of the previous frame and the displacement pixel difference of each stripe row of the two-dimensional compressed observation images of the previous and next frames;

[0255] The post-frame generation module is used to extract each row of stripes of the post-frame two-dimensional compressed observation image from the predicted pixel value sequence according to the length of each row of stripes of the post-frame two-dimensional compressed observation image, and splice them into the predicted post-frame two-dimensional compressed observation image.

[0256] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0257] In addition, an embodiment of the present invention further provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus.

[0258] Memory, used to store computer programs;

[0259] The processor is used to implement the image inversion method described in the above embodiment when executing the program stored in the memory.

[0260] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0261] The communication interface is used for communication between the above terminal and other devices.

[0262] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0263] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0264] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the image inversion method described in the above embodiment.

[0265] In another embodiment provided by the present invention, a computer program product including instructions is also provided, and when the computer is run on the computer, the computer executes the communication inversion method described in the above embodiment.

[0266] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0267] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0268] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0269] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. An image inversion method, characterized in that: The method comprises: The CUP-VISAR system is used to sample the observation signal and obtain a two-dimensional compressed observation image; According to the two-dimensional compressed observation image, a mathematical model of image inversion based on a plug-and-play generalized alternating projection framework is constructed with low-rank prior, total variation prior, Kalman prediction prior and inter-frame prediction prior as constraints; Decomposing the image inversion mathematical model into two optimization sub-problems about the value to be evaluated and the auxiliary variable, alternately updating the value to be evaluated and the auxiliary variable until the iteration termination condition is reached, then stopping the alternate updating, outputting the updated value to be evaluated, and obtaining an interference fringe image; The process of determining the inter-frame prediction prior is: By analyzing the phase difference of each line of stripes of the two previous and next frames of two-dimensional compressed observation images in the same iteration, the displacement pixel difference of each line of stripes of the two previous and next frames of two-dimensional compressed observation images is determined; Fitting each row of fringes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of fringes of the previous frame two-dimensional compressed observation image; Generate a predicted fringe sequence by using the cosine function of each row of fringe of the previous frame two-dimensional compressed observation image and the displacement pixel difference of each row of fringe of the previous and next two frames of two-dimensional compressed observation image; According to the length of each row of stripes in the rear frame two-dimensional compressed observation image, each row of stripes in the rear frame two-dimensional compressed observation image is extracted from the predicted pixel value sequence and spliced ​​into a predicted rear frame two-dimensional compressed observation image.

2. The method according to claim 1, characterized in that The CUP-VISAR system is used to sample the observation signal to obtain a two-dimensional compressed observation image, including: The VISAR system carries the target surface information and generates interference to form an interference fringe image at the imaging plane, thereby obtaining the observation signal of the interference fringe image. The observation signal of the interference fringe image is input into the CUP system, and the digital micromirror device in the CUP system encodes the interference fringe image to obtain an encoded image; After the coded image enters the streak camera, the streak camera cuts and superimposes the coded image to generate a two-dimensional compressed observation image.

3. The method according to claim 1, characterized in that The two-dimensional compressed observation image is expressed as: Y=TSCI Where I is the interference fringe image with a pixel size of m×n×k, Y is the two-dimensional compressed observation image recorded by the streak camera, whose size is (m+k-1)×n, T is the space-time integration operator, S is the time shear operator, and C is the spatial code mask operator.

4. The method according to claim 3, characterized in that According to the two-dimensional compressed observation image, a mathematical model of image inversion based on a plug-and-play generalized alternating projection framework is constructed with low-rank prior, total variation prior, Kalman prediction prior and inter-frame prediction prior as constraints, including: The expression of the two-dimensional compressed observation image is converted into: stY=TSCI Taking low-rank prior, total variation prior, Kalman prediction prior, and inter-frame prediction prior as constraints, we get: minλ‖I‖ TV +τR rank (I) s.t.Y=TSCI、I=Γ kalman (I)、I=Λ interframe (I) Where I represents the interference fringe image, which is the value to be evaluated; λ and τ represent Lagrange multipliers greater than 0; ‖I‖ TV is the total variation denoising operator, R rank (I) is a low-rank function operator, Γ kalman (I) is the Kalman prediction operator, Λ interframe (I) is the inter-frame prediction operator; Applying the plug-and-play generalized alternating projection framework to the above equation and introducing the auxiliary variable u, we obtain the image inversion mathematical model based on the plug-and-play generalized alternating projection framework: Where argmin{·} is the variable value when the target area is minimized, η is the adjustment parameter, ‖·‖ * represents the nuclear norm.

5. The method according to claim 4, characterized in that The image inversion mathematical model is decomposed into two optimization sub-problems about the value to be evaluated and the auxiliary variable, which are expressed as follows: The optimized expression to be evaluated is: Optimization expression of auxiliary variables:

6. The method according to claim 1, characterized in that The process of determining the Kalman prediction prior is: Expand each frame of the two-dimensional compressed observation image of the current iteration number by columns to generate a one-dimensional signal of each frame of the two-dimensional compressed observation image; Performing Kalman prediction on the one-dimensional signal of each frame of the two-dimensional compressed observation image to obtain a one-dimensional signal prediction value of each frame of the two-dimensional compressed observation image at the next iteration number; The one-dimensional signal prediction value of each frame of two-dimensional compressed observation image at the next iteration is reconstructed into a two-dimensional compressed observation image predicted by Kalman.

7. An image inversion device, characterized in that: The device comprises: A two-dimensional compressed observation image acquisition module is used to sample the observation signal using the CUP-VISAR system to obtain a two-dimensional compressed observation image; An image inversion mathematical model construction module is used to construct an image inversion mathematical model based on a plug-and-play generalized alternating projection framework according to the two-dimensional compressed observation image, with low-rank prior terms, total variation prior terms, Kalman prediction prior terms, and inter-frame prediction prior terms as constraints; A model solving module is used to decompose the image inversion mathematical model into two optimization sub-problems about the value to be solved and the auxiliary variable, alternately update the value to be solved and the auxiliary variable, and stop the alternate updating until the iteration termination condition is reached, output the updated value to be solved, and obtain the interference fringe image; The image inversion device also includes: A displacement pixel difference determination module is used to determine the displacement pixel difference of each row of stripes of the two-dimensional compressed observation images of the previous and next two frames in the same iteration by analyzing the phase difference of each row of stripes of the two-dimensional compressed observation images of the previous and next two frames; A fitting module is used to fit each row of stripes of the previous frame two-dimensional compressed observation image to a cosine curve to obtain a cosine function of each row of stripes of the previous frame two-dimensional compressed observation image; A subsequent frame prediction module, used to generate a predicted stripe sequence by using the cosine function of each stripe row of the two-dimensional compressed observation image of the previous frame and the displacement pixel difference of each stripe row of the two-dimensional compressed observation images of the previous and next frames; The post-frame generation module is used to extract each row of stripes of the post-frame two-dimensional compressed observation image from the predicted pixel value sequence according to the length of each row of stripes of the post-frame two-dimensional compressed observation image, and splice them into the predicted post-frame two-dimensional compressed observation image.

8. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the image inversion method according to any one of claims 1 to 6 when executing the program stored in the memory.

9. One or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the image inversion method according to any one of claims 1 to 6.

Images