Multi-constraint fast inversion method based on double-channel random coding, medium and equipment

Abstract

The application discloses a kind of based on double-channel random coding multi-constraint fast inversion method, medium and equipment, method includes the steps of double-channel coding to original signal using double-channel random coding CUP system, obtain double-channel two-dimensional compressed image;According to the double-channel two-dimensional compressed image, with low rank priori as constraint, construct inversion mathematical model based on generalized alternating projection framework;The inversion mathematical model of the application is decomposed into two optimization sub-problems about to be solved value and auxiliary variable, and the to-be-solved value and auxiliary variable are alternately updated until the iteration termination condition is reached, then the alternately updating is stopped, and the updated to-be-solved value is output, to obtain three-dimensional dynamic image.The application reconstructs by the multi-coding imaging method of double channel, applies low rank to sub-problem for solving, and constructs CUP-VISAR double-channel, low rank constraint inversion model framework by generalized alternating projection framework, can effectively improve inversion efficiency and stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a multi-constraint fast inversion method based on double-channel random coding, a medium and an equipment. BACKGROUND

[0002] Laser-driven inertial confinement fusion (ICF) is considered as one of the most promising ways to achieve controllable nuclear fusion. At present, the mainstream diagnostic instrument for ICF shock wave velocity measurement is the velocity interferometer system for any reflector (VISAR), and the framing camera and the streak camera are the main imaging recording equipment of ICF. The framing camera has the advantages of being able to directly record a certain number of two-dimensional image information with a certain time sequence relationship, and the disadvantage is that its time resolution is relatively low; the streak camera has the advantage of being the highest time resolution imaging recording equipment in the world, and the disadvantage is that it can only record one-dimensional spatial information.

[0003] In recent years, the compressed ultrafast photography (CUP) based on compressed sensing has realized the imaging frame rate up to 1011 frames / s and the two-dimensional image with a spatial resolution of 100 μm by completely opening the slit of the streak camera. On this basis, some scholars proposed a CUP-VISAR method based on compressed ultrafast photography (CUP) technology and VISAR technology, and successfully inverted multiple two-dimensional shock wave information images with picosecond level from the CUP recorded two-dimensional compressed shock wave images.

[0004] The existing CUP-VISAR system adopts a single-channel imaging mode, but due to the information loss caused by coding, the sampling rate is low, and there is a high data compression ratio, which leads to low compression imaging quality of CUP-VISAR. In addition, the existing inversion algorithm adopts a single-constraint TV-TwIST algorithm, which cannot comprehensively and effectively describe the characteristics of the shock wave fringe image, thereby affecting the inversion accuracy of the compressed image. Therefore, how to improve the sampling rate and design the inversion algorithm is very important.

[0005] At present, the inversion quality of the CUP-VISAR two-dimensional compressed image is not high, and the precision is low, which mainly has the following shortcomings:

[0006] (1) The single-channel method in the CUP-VISAR coding process has the characteristics of low sampling rate and high data compression ratio, which easily leads to insufficient temporal and spatial resolution of the imaging image;

[0007] (2) The TV-TwIST inversion algorithm has a relatively good image restoration ability, but it cannot fully and effectively describe the characteristics of shock wave stripe images.

[0008] (3) Noise removal in the algorithm can enhance the algorithm’s noise resistance, but it is easy to lose image details. Summary of the Invention

[0009] To overcome the above-mentioned technical defects, the present invention provides a multi-constraint fast inversion method, medium and device based on dual-channel random coding, which can effectively improve inversion efficiency and stability.

[0010] To solve the above problems, the present invention is implemented according to the following technical solution:

[0011] In a first aspect, the present invention provides a fast inversion method based on dual-channel random coding with multiple constraints, comprising the following steps:

[0012] The original signal was encoded in two channels using a dual-channel random coding CUP system to obtain a two-dimensional compressed image in two channels.

[0013] Based on the dual-channel two-dimensional compressed image, an inversion mathematical model based on the generalized alternating projection framework is constructed using low-rank prior terms as constraints.

[0014] The inversion mathematical model is decomposed into two optimization subproblems concerning the unsolved value and the auxiliary variable. The unsolved value and the auxiliary variable are updated alternately until the iteration termination condition is met, at which point the alternating updates stop, and the updated unsolved value is output, resulting in a three-dimensional dynamic graph.

[0015] As an improvement to the above scheme, the original signal is encoded in two channels using a dual-channel random coding CUP system to obtain a dual-channel two-dimensional compressed image, including the following steps;

[0016] The original signal is encoded using different random codes to obtain two different encoded copies of the original signal;

[0017] The two different encoded copies are time-shifted respectively, and the two different encoded copies after time shifting are imaged to obtain a dual-channel two-dimensional compressed image.

[0018] As an improvement to the above scheme, the dual-channel two-dimensional compressed image is represented as follows:

[0019]

[0020] Wherein, C1 and C2 are spatial coding operators, S is time shearing operator, T is spatiotemporal integration operator, x is the original interference fringe, and y1 and y2 are two-dimensional compressed images with two channels.

[0021] As an improvement to the above scheme, the step of constructing an inversion mathematical model based on the generalized alternating projection framework using the dual-channel two-dimensional compressed image and low-rank prior terms as constraints includes the following steps:

[0022] The expression for the two-dimensional compressed image is transformed by adding prior constraints to be evaluated, resulting in:

[0023] min x Ф(x) sty=Ox (2)

[0024] in, Represents a compressed image matrix; Represents the observation matrix;

[0025] Substituting equation (2) into the generalized alternating projection framework, and introducing low-rank constraints and auxiliary variables u, we obtain the inversion mathematical model based on the generalized alternating projection framework:

[0026]

[0027] Where η is the adjustment parameter, β and λ are Lagrange parameters (β≥0, λ≥0), D(x) is the denoising factor, D(x) can be the TV regularization operator, and rank(u) is the low-rank constraint.

[0028] As an improvement to the above scheme, the decomposition of the inversion mathematical model into two optimization sub-problems concerning the value to be solved and the auxiliary variables is expressed as follows:

[0029]

[0030]

[0031] Where k is the number of iterations, Equation (4) is the optimized expression for the value to be determined, and Equation (5) is the optimized expression for the auxiliary variable.

[0032] As an improvement to the above scheme, the alternating update of the value to be calculated and the auxiliary variable includes:

[0033] The expression for the subproblem to be evaluated is obtained by breaking it down into its components:

[0034]

[0035]

[0036] x is obtained from the Euclidean projection onto the linear manifold v. (k+0.5) :

[0037] x (k+0.5) =u k +P T (OO T ) -1 (yO·u k (8)

[0038] Among them, O T Represents the transpose of matrix O;

[0039] According to x (k+0.5) The value of is then updated using the soft threshold shrinkage formula to obtain the updated value of the desired result:

[0040]

[0041] As an improvement to the above scheme, the alternating update of the value to be calculated and the auxiliary variable further includes:

[0042] Based on the expression for the subproblem concerning the auxiliary variable, the rank function is approximated by the nuclear norm and solved as follows:

[0043]

[0044] In the formula, ||·|| = ∑ i σ i (·) represents the nuclear norm, σ i (·) represents the i-th singular value;

[0045] Based on the principle of proximal mapping, the update of the auxiliary variable is obtained by solving formula (10):

[0046]

[0047] Among them, u (k+1) =Uσ ∑ V T , which represents singular value decomposition.

[0048] As an improvement to the above scheme, the alternating update of the value to be calculated and the auxiliary variable further includes:

[0049] Adaptive adjustment of the Euclidean projection of formula (8) is expressed as:

[0050] x (k+0.5) =u k +O T (OO T ) -1 (y k -O·u k (12)

[0051] In the formula, O T Represents the transpose of matrix O;

[0052] in:

[0053] y (k+1) =y k +(yO·u k (13)

[0054] For dual-channel data coupling in CUP-VISAR, the following operation is performed in each iteration:

[0055]

[0056] Among them, (C) i ) T Let (TS) be the transpose of matrix Ci. T Let TS be the transpose of matrix S, where S is the time shearing operator, T is the spatiotemporal integration operator, and i represents the i-th imaging channel. In dual-channel imaging, y1 and y2, which carry different encoded information, are subjected to the inverse operation of encoding shearing and integration to obtain x1 and x2, which carry different information. Data coupling is then performed on them to obtain x with richer and more accurate information.

[0057] In a second aspect, the present invention provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the multi-constraint fast inversion method based on dual-channel random coding as described in the first aspect.

[0058] Thirdly, the present invention provides an apparatus comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the multi-constraint fast inversion method based on dual-channel random coding as described in the first aspect.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] This application employs a dual-channel, multi-coding imaging method for reconstruction, applies low-rank to solve sub-problems, and constructs a CUP-VISAR dual-channel, low-rank constrained inversion model framework through a generalized alternating projection framework. Dual-channel sampling improves the image sampling rate and significantly enhances noise resistance compared to single-channel ultrafast compressed image inversion methods. Furthermore, multiple sampling improves the signal-to-noise ratio (SNR), resulting in an increased SNR in each iteration and thus improving the final image quality. Meanwhile, the low-rank constraint preserves image details during the inversion process, effectively enhancing the quality of the inverted image. Attached Figure Description

[0061] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, wherein:

[0062] Figure 1 This is a flowchart illustrating a multi-constraint fast inversion method based on dual-channel random coding in one embodiment of this application.

[0063] Figure 2 This is a flowchart illustrating step S10 as described in some embodiments of this application;

[0064] Figure 3 This is a schematic diagram of the CUP imaging system described in some embodiments of this application;

[0065] Figure 4 This is a schematic diagram of dual-channel random coding imaging in some embodiments of this application;

[0066] Figure 5 This is a schematic diagram of the overall process of the dual-channel random coding multi-constraint fast inversion method in some embodiments of this application;

[0067] Figure 6 This is a comparison chart of peak signal-to-noise ratio results under the experimental condition where Gaussian noise is 0 in the embodiments of this application;

[0068] Figure 7 This is a comparison diagram of the structural similarity results under the experimental state where Gaussian noise is 0 in the embodiments of this application;

[0069] Figure 8 This is a comparison chart of peak signal-to-noise ratio results under the experimental condition of Gaussian noise of 0.1 in the embodiments of this application;

[0070] Figure 9 This is a comparison diagram of the structural similarity results under the experimental condition of Gaussian noise of 0.1 in the embodiments of this application;

[0071] Figure 10 This is a comparison chart of peak signal-to-noise ratio results under the experimental condition of Gaussian noise of 0.5 in the embodiments of this application;

[0072] Figure 11 This is a comparison diagram of the structural similarity results under the experimental condition of Gaussian noise of 0.5 in the embodiments of this application;

[0073] Figure 12 This is a comparison image of single-channel and dual-channel schemes under noise conditions in the embodiments of this application. Detailed Implementation

[0074] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0075] It should be noted that the serial numbers mentioned in this article, such as S1, S2, etc., are merely used to distinguish between steps and do not mean that these steps must be strictly performed in the order of the serial numbers.

[0076] In one embodiment, such as Figure 1 As shown, a fast inversion method with multiple constraints based on dual-channel random coding is provided, including the following steps:

[0077] S10: The original signal is encoded in two channels using a dual-channel random coding CUP system to obtain a dual-channel two-dimensional compressed image;

[0078] Specifically, the dual-channel random coding (CUP) system is a technique for dual-channel encoding of raw signals, applied in data transmission, communication, and signal processing to increase signal robustness and reliability. In a dual-channel random coding (CUP) system, the raw signal is encoded through two independent coding channels. Each channel generates a coding sequence using an uncorrelated random pattern. These random patterns are typically highly complex sequences generated by a pseudo-random number generator, ensuring no correlation between the coding sequences of different channels. The purpose of the coding channels is to convert the raw signal into a coded signal, enhancing its distinguishability and robustness. The coded signal can be generated by applying the coding sequence to the raw signal through direct multiplication, addition, or other transformations. The two channels can use different coding sequences, thus achieving dual-channel encoding.

[0079] S20: Based on the dual-channel two-dimensional compressed image, and using low-rank prior terms as constraints, construct an inversion mathematical model based on the generalized alternating projection framework;

[0080] Specifically, since the two-dimensional striped images recorded by CUP-VISAR have extremely strong low-rank properties, and the low-rank constraint can effectively guarantee image details and low-rank properties during the inversion process, the low-rank constraint is applied to the CUP-VISAR inversion process to make the algorithm noise-resistant while ensuring that image details are not lost.

[0081] For image inversion solutions, algorithms such as GAP, AL, and ADMM can be used. Considering fast solution, this embodiment only considers the GAP framework. The Generalized Alternating Projection (GAP) algorithm has the advantages of convergence at any time and high computational efficiency. Implementing this algorithm as the iterative solution framework for the CUP-VISAR inversion process can effectively improve inversion efficiency and stability.

[0082] S30: Decompose the inversion mathematical model into two optimization sub-problems concerning the unsolved value and the auxiliary variable, and alternately update the unsolved value and the auxiliary variable until the iteration termination condition is met, then stop the alternating update, output the updated unsolved value, and obtain a three-dimensional dynamic graph.

[0083] In some embodiments, such as Figure 2 As shown, the process of using a dual-channel random coding CUP system to encode the original signal in two channels to obtain a dual-channel two-dimensional compressed image includes the following steps;

[0084] S11: Encode the original signal using different random codes to obtain two different encoded copies of the original signal;

[0085] S12: Time-shift the two different encoded copies respectively, and image the two different encoded copies after time shifting to obtain a dual-channel two-dimensional compressed image.

[0086] Specifically, the optical path required for the experiment is set up, and the original signal is input into a dual-channel random coding CUP system. In this embodiment, the original signal is an interference fringe. After the interference fringe enters the dual-channel random coding CUP system, the ultrafast dynamic scene is copied into two copies. Each copy is then encoded using a different random code, and finally, the CCD images each copy after the time shift.

[0087] In some embodiments, such as Figure 3 The diagram shows a schematic of an ultrafast compressed imaging system for compressed sensing. During information recording, the CUP first uses optical devices such as lenses to image the object onto a digital micromirror device (DMD). The DMD consists of several tiny mirrors, each capable of deflection in two directions (±12°), but only one direction of reflection is utilized. The DMD encodes the image of the object, and the encoded image is reflected by the DMD and enters a streak camera. The streak camera shifts the encoded image according to different times, and all the encoded and shifted images are finally superimposed on the external CCD detector of the streak camera.

[0088] In this context, the mathematical operator C represents the encoding process, the mathematical operator S represents the offset process, and the mathematical operator T represents the temporal-spatial superposition process. Let the original dynamic scene be denoted as x, and the final encoded offset superposition yields the information y. Therefore, a two-dimensional compressed image can be represented as:

[0089] y = T·S·C·x = O·x (0)

[0090] In this embodiment, as Figure 4As shown, because different random codes are used for encoding and dual-channel processing, the CCD finally images each copy obtained after the time offset. Formula (2) can be further expressed in matrix form as follows:

[0091]

[0092] Wherein, C1 and C2 are spatial coding operators, S is time shearing operator, T is spatiotemporal integration operator, x is the original interference fringe, and y1 and y2 are two-dimensional compressed images with two channels.

[0093] For details, please continue to refer to [the website / information]. Figure 5 The target object's image is first transmitted to the imaging point IP using optical devices such as lenses, and then split into two beams by a beam splitter at IP. The first image of the object is imaged onto DMD1 through the first set of convex lenses, and combined to generate a coded image. DMD consists of several tiny mirrors, each of which can be deflected in two directions (±12°), but only one direction of the mirror is used. The image encoded by DMD1 is imaged onto the slit position of the streak camera through the second set of 4f optical paths. The streak camera shifts the encoded image accordingly at different times, and all the images after the shift are finally superimposed on the external CCD1 detector of the streak camera to form a two-dimensional image y1. The second image of the object is imaged onto DMD2 through the third set of convex lenses, and combined to generate a coded image. The image encoded by DMD2 is imaged onto the slit position of the streak camera through the fourth set of 4f optical paths. After the streak camera shifts, all the images are finally superimposed on the external CCD2 detector of the streak camera to form a two-dimensional image y2.

[0094] Furthermore, in some embodiments, such as Figure 5 The flowchart shown is a multi-constraint fast inversion method based on dual-channel random coding. The inversion process is the process of restoring the original dynamic scene x from the two-dimensional compressed image information y, which is the solution of the inverse problem of formula (2). However, the amount of data contained in the three-dimensional original dynamic scene information x is much larger than that in the two-dimensional compressed image information y, so the solution of the inverse problem of formula (2) is an underdetermined problem. Therefore, by adding constraints, the underdetermined problem is transformed into a constrained problem, and formula (2) can be expressed as:

[0095] min x Ф(x) sty=Ox (2)

[0096] in, Represents a compressed image matrix; Represents the observation matrix;

[0097] Substituting equation (2) into the generalized alternating projection framework, and introducing low-rank constraints and auxiliary variables u, we obtain the inversion mathematical model based on the generalized alternating projection framework:

[0098]

[0099] Where η is the adjustment parameter, β and λ are Lagrange parameters (β≥0, λ≥0), D(x) is the denoising factor, D(x) can be the TV regularization operator, and rank(u) is the low-rank constraint.

[0100] Furthermore, the problem described above is broken down into problems of finding x and u, with x and u being updated alternately. The inversion mathematical model is decomposed into two optimization sub-problems concerning the value to be solved and the auxiliary variable, as follows:

[0101]

[0102]

[0103] Where k is the number of iterations, Equation (4) is the optimized expression for the value to be determined, and Equation (5) is the optimized expression for the auxiliary variable.

[0104] Furthermore, based on the optimized expression for the value to be solved and the optimized expression for the auxiliary variable, the value to be solved x and the auxiliary variable u are updated alternately. Specifically, updating the value to be solved x includes the following steps:

[0105] In the optimized expression (4) for the value to be determined, since A minimum value exists, but D(x) is a non-differentiable term and cannot be solved by differential optimization. Therefore, the expression for the subproblem to be evaluated is broken down into:

[0106]

[0107]

[0108] Furthermore, given u, x (k+0.5) Through the Euclidean projection of u onto the linear manifold. (k+0.5) The iteration can be expressed as:

[0109] x (k+0.5) =u k +P T (OO T ) -1 (yO·u k (8)

[0110] Among them, O T Represents the transpose of matrix O;

[0111] Finally, based on x (k+0.5) The value of is then updated using the soft threshold shrinkage formula to obtain the updated value of the desired result:

[0112]

[0113] Specifically, updating the auxiliary variable u includes the following steps:

[0114] Due to the discrete nature of the rank, solving formula (5) is an NP-hard problem. The rank function is usually approximated by the nuclear norm, which can be expressed as:

[0115]

[0116] Where, ||·|| = ∑ i σ i (·) represents the nuclear norm, σ i (·) represents the i-th singular value;

[0117] Specifically, singular value decomposition (SVD) is a matrix factorization method commonly used in fields such as data dimensionality reduction, matrix compression, and recommender systems. It decomposes a matrix into the product of three matrices, including an orthogonal matrix, a diagonal matrix, and the transpose of another orthogonal matrix.

[0118] Given an m×n real matrix A, singular value decomposition represents it as a product of the following form:

[0119] A=U*Σ*V T

[0120] Where U is an m×m orthogonal matrix, Σ is an m×n diagonal matrix, and V is the transpose of an n×n orthogonal matrix. The diagonal elements of the diagonal matrix Σ are called singular values, which are usually arranged in descending order.

[0121] Based on the principle of proximal mapping, the solution to equation (10) is as follows:

[0122]

[0123] Among them, u (k+1) =Uσ ∑ V T , which represents singular value decomposition.

[0124] In some embodiments, to improve the algorithm iteration speed, the Euclidean projection of formula (10) is adaptively adjusted, as follows:

[0125] x (k+0.5) =u k +O T (OO T ) -1 (y k -O·u k (12)

[0126] In the formula, O TRepresents the transpose of matrix O;

[0127] in,

[0128] y (k+1) =y k +(yO·u k (13)

[0129] In this embodiment, after dual-channel imaging, two initial data sets, y1 and y2, are obtained. These two data sets can be inverted to obtain two x sets. Data coupling is then performed on these two sets to obtain more information about x. For dual-channel data coupling in CUP-VISAR, the following operations are performed in each iteration:

[0130]

[0131] Among them, (C) i ) T Let (TS) be the transpose of matrix Ci. T Let be the transpose of matrix TS, where S is the time shearing operator, T is the spatiotemporal integration operator, and i represents the i-th imaging channel. The entire iterative process will eventually continue until ||f(x)||. (k+1) -x k ||2≤δ, where δ is a pre-set threshold, or the iteration can stop when the number of iterations reaches a set number K.

[0132] The effectiveness of the multi-constraint fast inversion method based on dual-channel random coding provided by this invention will be verified next:

[0133] like Figures 6 to 11 As shown, simulation experiments were conducted using MATLAB, with a single-channel GAP and low-rank constraint as the control group. Using the actual shock wave velocity as a reference, 25 frames of dynamic simulation fringe patterns were provided as the raw data. Considering the presence of noise in the actual environment, multiple control experiments were conducted with Gaussian noise levels of 0, 0.1, and 0.5 to test the method's noise resistance. Finally, peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) were used as evaluation metrics for the inversion results.

[0134] Table 1 Comparison of Average PSNR and SSIM for Single-Channel and Dual-Channel Schemes under Noise Conditions

[0135]

[0136] Table 1 shows a comparison of average PSNR and SSIM for single-channel and dual-channel schemes under noise conditions. Figures 6 to 7As shown in Table 1, under noise-free conditions, the overall inversion quality of the single-channel, low-rank constrained GAP method is basically on par with that of the dual-channel, low-rank constrained GAP method. The multi-constraint fast inversion method based on dual-channel random coding provided by this invention has a PSNR 0.2455 dB lower and an SSIM 0.0043 dB lower than the single-channel, low-rank constrained GAP method.

[0137] Depend on Figures 8-11 and Figure 12 It can be seen that the multi-constraint fast inversion method based on dual-channel random coding provided by this invention has stronger noise resistance than the single-channel, low-rank constraint GAP method. Through Figure 12 It can be seen that as the noise increases, both the single-channel, low-rank constrained GAP method and the dual-channel, low-rank constrained GAP method can guarantee good fringe profiles, but the dual-channel, low-rank constrained GAP method has higher fringe profile quality.

[0138] In some embodiments, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the multi-constraint fast inversion method based on dual-channel random coding provided in the first aspect.

[0139] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer-readable storage media (or non-transitory media) and communication media (or transient media).

[0140] As is known to those skilled in the art, the term computer-readable storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer-readable storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0141] For example, the computer-readable storage medium may be an internal storage unit of the network management device described in the foregoing embodiments, such as the hard drive or memory of the network management device. The computer-readable storage medium may also be an external storage device of the network management device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the network management device.

[0142] In some embodiments, an apparatus is provided, including a processor and a memory, the memory being used to store a computer program; the processor being used to execute the computer program and, when executing the computer program, to implement the multi-constraint fast inversion method based on dual-channel random coding provided in the first aspect of the present invention.

[0143] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0144] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A fast inversion method based on dual-channel random coding with multiple constraints, characterized in that, Includes the following steps: The original signal was encoded in two channels using a dual-channel random coding CUP system to obtain a two-dimensional compressed image in two channels. Based on the dual-channel two-dimensional compressed image, an inversion mathematical model based on the generalized alternating projection framework is constructed using low-rank prior terms as constraints. The inversion mathematical model is decomposed into two optimization sub-problems concerning the unsolved value and the auxiliary variable. The unsolved value and the auxiliary variable are updated alternately until the iteration termination condition is met, at which point the alternating update stops, the updated unsolved value is output, and a three-dimensional dynamic graph is obtained. The dual-channel two-dimensional compressed image is represented as follows: (1) Wherein, C1 and C2 are spatial coding operators, S is time shearing operator, T is spatiotemporal integration operator, x is the original interference fringe, and y1 and y2 are two-dimensional compressed images with two channels; The step of constructing an inversion mathematical model based on the generalized alternating projection framework using the dual-channel two-dimensional compressed image and low-rank prior terms as constraints includes the following steps: The expression for the two-dimensional compressed image is transformed by adding prior constraints to be evaluated, resulting in: （2） in, , represents the compressed image matrix; , representing the observation matrix; Substituting equation (2) into the generalized alternating projection framework, and introducing low-rank constraints and auxiliary variables u, we obtain the inversion mathematical model based on the generalized alternating projection framework: （3） in, To adjust the parameters, , Lagrange parameters , , For noise reduction factor, It can be a TV regularization operator, where rank(u) is a low-rank constraint; The decomposition of the inversion mathematical model into two optimization subproblems concerning the unsolved value and the auxiliary variable is expressed as follows: （4） （5） Where k is the number of iterations, equation (4) is the optimized expression for the value to be determined, and equation (5) is the optimized expression for the auxiliary variable; The alternating update of the value to be determined and the auxiliary variable includes: The expression for the subproblem to be evaluated is obtained by breaking it down into its components: （6） （7） According to linear manifold The Euclidean projection on is obtained : （8） in, Representative matrix Transpose of; according to The value of is then updated using the soft threshold shrinkage formula to obtain the updated value of the desired result: （9）； The alternating update of the value to be determined and the auxiliary variable also includes: Based on the expression for the subproblem concerning the auxiliary variable, the rank function is approximated by the nuclear norm and solved as follows: （10） In the formula, Represents the nuclear norm. Represents the i-th singular value; Based on the principle of proximal mapping, the update of the auxiliary variable is obtained by solving formula (10): （11） in, , indicating singular value decomposition; The alternating update of the value to be determined and the auxiliary variable also includes: Adaptive adjustment of the Euclidean projection of formula (8) is expressed as: （12） In the formula, Representative matrix Transpose of; in, （13） For dual-channel data coupling in CUP-VISAR, the following operation is performed in each iteration: （14） in, This is the transpose of matrix Ci. For matrix The transpose of , where S is the time shear operator and T is the spacetime integral operator. Indicates the first One imaging channel.

2. The multi-constraint fast inversion method based on dual-channel random coding according to claim 1, characterized in that, The process of using a dual-channel random coding CUP system to encode the original signal in two channels to obtain a dual-channel two-dimensional compressed image includes the following steps; The original signal is encoded using different random codes to obtain two different encoded copies of the original signal; The two different encoded copies are time-shifted respectively, and the two different encoded copies after time shifting are imaged to obtain a dual-channel two-dimensional compressed image.

3. A computer device, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the multi-constraint fast inversion method based on dual-channel random coding as described in any one of claims 1 to 2.

4. A computer-readable storage medium, characterized in that, The readable storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the multi-constraint fast inversion method based on dual-channel random coding as described in any one of claims 1 to 2.

Images