A signal reconstruction method in a compressive sensing based computational ghost imaging system

Abstract

The application discloses a signal reconstruction method in a ghost imaging system based on compressed sensing, and comprises the following steps: S1, performing block processing on an original signal to obtain a pretreatment signal; S2, obtaining a sparse representation of the pretreatment signal and a measurement signal; S3, performing inner product of each atom in a sensing matrix and a current residual, and using L*t atoms with the maximum inner product to form a matrix A t ; S4, using a least square method to obtain a current sparse signal; S5, selecting t items with the maximum absolute value from the current sparse signal, and selecting corresponding atoms in the sensing matrix to form a new matrix A tL ; S6, calculating a residual of the measurement signal based on A tL ; if the residual is less than or equal to a threshold value, outputting the current sparse signal; if the residual is greater than the threshold value and greater than a residual of a last round of iteration, setting t=t+1 and jumping to S3; otherwise, directly jumping to S3; and S7, obtaining a reconstruction signal of the original signal based on the current sparse signal. In this way, the requirement of hardware in the sampling process can be reduced, and the signal recovery effect can be improved.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, and more specifically, relates to a signal reconstruction method in a computational ghost imaging system based on compressed sensing. Background Technology

[0002] Currently, next-generation information technologies such as the Internet of Things (IoT), cloud computing, mobile internet, big data, and artificial intelligence are booming globally, accelerating the refinement of global industrial division of labor and economic restructuring, reshaping the global economic competitive landscape, and becoming one of the core forces driving global industrial transformation. With the increasing maturity of various wireless communication technologies such as NB-IoT and 5G, people, things, and data are interconnected. The amount of digital information is also surging. In this context, researching efficient signal sampling and reconstruction technologies is in line with the development trend of modern massive signal and information processing technologies.

[0003] Classical signal sampling techniques are based on the Nyquist sampling theorem, which requires sampling followed by compression. The Nyquist theorem stipulates that signals must be sampled and processed at a rate more than twice the signal frequency to ensure lossless reconstruction. For high-frequency signals, classical sampling techniques place high demands on hardware, resulting in large, redundant digital signals that are unsuitable for real-time transmission, processing, and storage. Compressed sensing, through linear projection, performs both sampling and compression simultaneously, allowing sampling at rates far lower than those required by the Nyquist theorem while maintaining a high probability of reconstructing the original signal. Compressed sensing significantly reduces the amount of data to be stored and transmitted, making it crucial for applications such as information transmission, medical diagnosis, radar imaging, remote sensing, and image encryption.

[0004] However, existing signal reconstruction methods based on compressed sensing suffer from problems such as low efficiency, inability to remove selected atoms, and the need to estimate the sparsity of the signal in advance. These problems include poor reconstruction accuracy, poor versatility, and long processing time. Summary of the Invention

[0005] To address the shortcomings and improvement needs of existing technologies, this invention provides a signal reconstruction method for computational ghost imaging systems based on compressed sensing. This method can be applied when the hardware system provides only a small amount of original signal and the prior information of the original signal is unknown beforehand. It solves the problems of slow convergence speed and low reconstruction accuracy in the signal reconstruction process, thereby achieving the goal of high-precision and rapid signal reconstruction.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a signal reconstruction method in a computational ghost imaging system based on compressed sensing, comprising the following steps:

[0007] S1, the original signal is divided into blocks to obtain a preprocessed signal;

[0008] S2, the preprocessed signal is sparsely represented using the discrete cosine transform matrix; the preprocessed signal is measured using the measurement matrix to obtain the measurement signal;

[0009] S3, let each atom in the sensing matrix have an inner product with the current residual of the measurement signal, and use the L*t atoms with the largest inner product to form matrix A. t Wherein, the perception matrix is ​​the dot product of the measurement matrix and the discrete cosine transform matrix, each atom represents a row of the perception matrix, t is the number of iterations, and L is a constant greater than 1;

[0010] S4, calculate the measured signal and matrix A t The least squares value is used as the current sparse signal;

[0011] S5, select the term with the largest absolute value t from the current sparse signal, and select the corresponding atom from the perception matrix to form a new matrix A. tL ;

[0012] S6, based on the new matrix A tL Calculate the residual of the measured signal. If the residual is less than or equal to the threshold, output the current sparse signal. If the residual is greater than the threshold and greater than the residual of the previous iteration, let t = t + 1 and jump to S3. Otherwise, jump directly to S3.

[0013] S7, obtain the reconstructed signal of the original signal based on the current sparse signal.

[0014] Further, S1 includes: uniformly dividing the original signal into several sub-signals, and then performing one-dimensional processing on each sub-signal to obtain a preprocessed signal.

[0015] Furthermore, in S3, matrix A is formed using L*t atoms with the largest inner product. t This includes: performing regularization screening on the L*t atoms corresponding to the largest inner product, and selecting t atoms to form matrix A. t .

[0016] Further, S4 includes:

[0017] The current sparse signal θ is calculated using the following formula. t :

[0018]

[0019] Where y represents the measurement signal. A represents t The transpose of .

[0020] Furthermore, in S6, based on the new matrix A tL Calculate the residual of the measured signal, including:

[0021] Calculate the current sparse signal θ using the following formula. t The residual r of the corresponding measurement signal tnew :

[0022]

[0023] in, A represents tL The transpose of .

[0024] In a second aspect, the present invention provides a signal reconstruction apparatus for a computational ghost imaging system based on compressed sensing, comprising:

[0025] The first processing module is used to divide the original signal into blocks to obtain a preprocessed signal;

[0026] The second processing module is used to perform sparse representation of the preprocessed signal using a discrete cosine transform matrix; and to measure the preprocessed signal using a measurement matrix to obtain a measurement signal.

[0027] The atom matrix update module is used to perform an inner product between each atom in the sensing matrix and the current residual of the measurement signal, and then use the L*t atoms with the largest inner products to form matrix A. t Wherein, the sensing matrix is ​​the dot product of the measurement matrix and the discrete cosine transform matrix, each atom represents a row of the sensing matrix, t is the iteration number, and L is a constant greater than 1; and, the measurement signal and matrix A are calculated. t The least squares value of is used as the current sparse signal; and the term with the largest absolute value t is selected from the current sparse signal, and the corresponding atom is selected from the perception matrix to form a new matrix A. tL ;

[0028] The judgment module is used to determine the new matrix A. tL Calculate the residual of the measured signal. If the residual is less than or equal to a threshold, output the current sparse signal. If the residual is greater than the threshold and greater than the residual of the previous iteration, let t = t + 1, and then execute the operation of the atomic matrix update module. Otherwise, directly execute the operation of the atomic matrix update module.

[0029] The signal reconstruction module is used to obtain a reconstructed signal of the original signal based on the current sparse signal.

[0030] Furthermore, the first processing module is also used to divide the original signal into several sub-signals evenly, and then perform one-dimensional processing on each sub-signal to obtain a preprocessed signal.

[0031] Furthermore, the atom matrix update module is also used to perform regularization filtering on the L*t atoms corresponding to the largest inner product, selecting t atoms to form matrix A. t .

[0032] Thirdly, the present invention provides an electronic device, comprising:

[0033] Memory, used to store computer programs;

[0034] A processor for executing the computer program to implement the steps of the signal reconstruction method in a computational ghost imaging system based on compressed sensing as described in the first aspect.

[0035] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the signal reconstruction method in the computational ghost imaging system based on compressed sensing as described in the first aspect.

[0036] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0037] (1) This invention uses continuous iteration to make the sparsity estimated by the algorithm approximate the sparsity of the real signal. Each iteration involves two selections: the first is to take the inner product of each atom in the sensing matrix with the current residual of the measured signal, and then use the L*t atoms with the largest inner products to form matrix A. t And based on matrix A t The first step is to obtain the current sparse signal; the second step is to select the term with the largest absolute value t from the current sparse signal, select the corresponding atom in the perception matrix, and form a new matrix A. tL And based on the new matrix A tL The residual of the measured signal is calculated. Thus, this invention achieves good signal reconstruction results without presetting sparsity or the maximum number of iterations.

[0038] (2) In the image acquisition stage, the present invention performs image segmentation to improve the computing speed and reduce the requirements of the hardware system and reduce costs.

[0039] (3) The present invention introduces a regularization module between the first screening and the second screening to perform regularized screening on the atoms selected in the first screening, thereby selecting the combination of atoms with the highest energy, the most representative signal, and a relatively balanced energy distribution. Attached Figure Description

[0040] Figure 1 A schematic flowchart illustrating the signal reconstruction method in a computational ghost imaging system based on compressed sensing provided by this invention;

[0041] Figure 2 A schematic diagram of the original image segmentation process provided by the present invention;

[0042] Figure 3 This invention provides a schematic diagram of one-dimensional processing of a block matrix;

[0043] Figure 4 A schematic diagram of the sparse representation process of compressed sensing signals provided by the present invention;

[0044] Figure 5 A schematic diagram of the compressed sensing signal measurement process provided by this invention;

[0045] Figure 6 The flowchart of the signal reconstruction algorithm provided by this invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0047] In this invention, the terms "first," "second," etc. (if present) in the invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0048] See Figure 1 , combined Figures 2 to 6 The present invention provides a signal reconstruction method in a computational ghost imaging system based on compressed sensing, the method comprising operations S1 to S7.

[0049] Operation S1 divides the original signal into blocks to obtain a preprocessed signal.

[0050] It should be noted that the signal reconstruction method provided by this invention is applicable to different types of signals, such as image signals and speech signals. This embodiment takes an image signal as an example, and a complete image can be reconstructed from the original signal through signal reconstruction.

[0051] In this embodiment, operation S1 includes sub-operations S11 to S13.

[0052] In sub-operation S11, an image of size N*N is arbitrarily selected, and the image is first converted into a grayscale image.

[0053] Typically, a color image consists of three components: R, G, and B, representing red, green, and blue colors, respectively. Grayscale conversion is the process of converting the three components of a color image into a single component. In a grayscale image, each pixel has only one color (or brightness), representing a multi-level color depth between black and white. Pixels with smaller grayscale values ​​are darker, and vice versa. The smallest pixel value is 0, representing black; the largest pixel value is 255, representing white.

[0054] The advantage of converting color images to grayscale images is that it significantly reduces computational load. Based on the varying sensitivities of the human eye to different colors, this invention employs a weighted average method for grayscale conversion.

[0055] Gray=R×0.299+G×0.587+B×0.114 (1)

[0056] The signal after grayscale conversion is called the original signal. The weighted average coefficients of the R, G, and B components sum to 1, with G having the largest coefficient, followed by R, and B having the smallest. This is because the human eye is most sensitive to green and least sensitive to blue.

[0057] In sub-operation S12, the image is divided into blocks.

[0058] For example, the image is divided into several image blocks of the same size, with each block having a size of n*n. All blocks are independent, ultimately dividing the complete grayscale image into... Each block, such as Figure 2 As shown.

[0059] In sub-operation S13, the segmented image is processed to be one-dimensional.

[0060] For example, for an image of size n*n, one-dimensionalization specifically involves: accumulating each column of data sequentially, such as placing the first column of data on top, the second column below the first, the third column below the second, and so on. This yields a data matrix (or column vector) of size n*n rows and one column, which is called the preprocessed signal. Figure 3 As shown.

[0061] Operation S2 involves using the discrete cosine transform matrix to perform a sparse representation of the preprocessed signal; then, the measurement matrix is ​​used to measure the preprocessed signal to obtain a measurement signal.

[0062] In this embodiment, the preprocessed signal is sparsely represented using the discrete cosine transform matrix, specifically as follows:

[0063] The prerequisite for reconstructing a signal using compressed sensing is that the signal itself can be represented in a sparse manner. For a vector of size N rows and 1 column, if only K of its N elements are non-zero, where K << N, then the vector is said to be K-sparse.

[0064] Therefore, before reconstructing the preprocessed signal using a signal reconstruction algorithm, it needs to be sparsely represented, that is, let...

[0065] x = Ψ DCT ·θ (2)

[0066] Then θ is a signal that satisfies the sparsity condition, which is called the sparse representation of the preprocessed signal x. In this embodiment, the discrete cosine transform matrix is ​​used as the sparse matrix. θ can be used in signal reconstruction algorithms to complete signal reconstruction and recover the original signal. Equation (2) can be visualized as follows: Figure 4 express.

[0067] The formula for the two-dimensional discrete cosine transform is:

[0068]

[0069] in,

[0070] Therefore, the discrete cosine transform matrix Ψ can be obtained. DCT for:

[0071]

[0072] In this embodiment, the preprocessed signal is measured using a measurement matrix to obtain a measurement signal, specifically as follows:

[0073] First, randomly generate a measurement matrix of size M rows and n*n columns, following a Gaussian distribution. The method is to ensure that each element in the matrix Φ independently follows a mean of 0 and a variance of . The Gaussian distribution, that is:

[0074]

[0075] in

[0076]

[0077] Right now

[0078]

[0079] The measured signal and the preprocessed signal have the following relationship:

[0080] y = Φ M，n×n ·x n×n，1 (8)

[0081] Where x is the preprocessed signal, Φ M，n×n For the measurement matrix. Equation (8) can be visualized as follows: Figure 5 express.

[0082] Furthermore, by combining equations (2) and (8), we can obtain:

[0083] y = Φ M，n×n ·Ψ DCT ·θ n×n，1 (9)

[0084] By inputting y, Φ M，n×n Ψ DCT In the signal reconstruction algorithm, θ can be reconstructed. n×n，1 Then, by using equation (2), the preprocessed signal x can be obtained through calculation, and then the original signal can be obtained.

[0085] The essence of the compressed sensing reconstruction algorithm mentioned in this invention is to reconstruct the signal by iteratively making the sparsity estimated by the algorithm approximate the sparsity of the real signal, thereby obtaining a more accurate signal estimate. Therefore, it eliminates the need to preset the sparsity and the maximum number of iterations, resulting in a more accurate signal estimate and high-quality signal reconstruction. The initial sparsity estimate is set to 1, and then after each iteration, the estimated sparsity value is incremented by one, i.e., the estimated sparsity value equals the iteration number t. The compressed sensing reconstruction algorithm includes operations S3 to S6.

[0086] Operation S3 involves taking the inner product of each atom in the sensing matrix with the current residual of the measurement signal, and then using the L*t atoms with the largest inner products to form matrix A. t Wherein, the perception matrix is ​​the dot product of the measurement matrix and the discrete cosine transform matrix, each atom represents a row of the perception matrix, t is the number of iterations, and L is a constant greater than 1.

[0087] In this embodiment, it is necessary to Φ M，n×n ·Ψ DCT (to Φ) M，n×n With Ψ DCT The product of the two is called the perception matrix, denoted as A). To find the atom that best represents the preprocessed signal x, the specific method is to take the inner product of each atom in the perception matrix with the initial residual, and search for the atoms corresponding to the L*t largest inner product values ​​(where L should generally be greater than 1; after repeated testing and calculation, L is preferably 2 in this embodiment). These atoms can most fully represent the preprocessed signal. At this point, it is also necessary to record the specific value of the largest inner product and the position of that atom in the dictionary. The process of selecting the atom set based on the inner product size is called the first screening. The inner product is calculated as follows:

[0088] U = abs(A T ·r t-1(10)

[0089] Where abs(·) represents taking the absolute value of the value within the parentheses, A T This represents the transpose of matrix A, which means interchanging the rows and columns of a matrix to obtain a new matrix; r t-1 This represents the residual of the measured signal obtained in the t-th iteration. The initial residual is generally set to the measured signal y.

[0090] Furthermore, a regularized selection module can be used to perform regularized selection on the atoms selected in the first round, selecting t atoms to obtain the atom set A. t The regularization module has the following properties: Given the initial set of atoms, the module selects a new subset of atoms (i.e., a specific subset of the initial set); this subset satisfies the following two properties: ① The energy of any atom in the subset is no greater than twice the energy of any other atom; the energy of an atom is defined as a certain norm of that atom; ② For all subsets satisfying condition ①, the subset with the largest sum of the squares of the energies of all atoms is selected. Specifically, this invention uses a first norm. The advantage of regularization is that it allows the selection of the atom combination with the highest energy, the most representative signal, and a relatively balanced energy distribution.

[0091] Operation S4, calculate the measured signal and matrix A t The least squares value is used as the current sparse signal.

[0092] In this embodiment, the least squares method is used to solve for the current estimated signal, which is the input measurement result y and the matrix A selected in the current iteration. t Solve for ||yA t θ t ||Minimum θ t , as the current sparse signal.

[0093] Least squares is a mathematical optimization technique that seeks the optimal function fit for data by minimizing the sum of squared errors. It allows for the easy acquisition of unknown data while guaranteeing that the sum of squared errors between the obtained data and the actual data is minimized. The general form of least squares is:

[0094]

[0095] Where L i (x) is the residual function. Let y and A... t θ t Substituting into equation (11), θ can be obtained. t for:

[0096]

[0097] Operation S5 involves selecting the term with the largest absolute value t from the current sparse signal, and then selecting the corresponding atom from the perception matrix to form a new matrix A. tL .

[0098] The current sparse signal θ is obtained by operation S4. t Next, a second screening stage is required, in which the sparse signal θ is selected. t Select the terms with the largest absolute value t from θ, and record the values ​​of these terms in θ. t The sequence number is used to select the corresponding atom in the perception matrix A to form a new matrix, denoted as A. tL .

[0099] Operation S6, based on the new matrix A tL Calculate the residual of the measured signal. If the residual is less than or equal to the threshold, output the current sparse signal. If the residual is greater than the threshold and greater than the residual of the previous iteration, let t = t + 1 and jump to S3. Otherwise, jump directly to S3.

[0100] In this embodiment, the current residual is obtained by equation (13):

[0101]

[0102] At this point, if ①: the residual satisfies the iteration stopping condition, that is, it satisfies equation (14):

[0103] r tnew ≤ε (14)

[0104] The iteration then stops, and the signal reconstruction algorithm outputs the sparse representation θ of the preprocessed signal calculated in the last iteration. t ;

[0105] If ②: the residual does not meet the iteration stopping condition, and the residual is greater than the residual of the previous iteration, it indicates that the sparsity estimate is too low at this time, and the selected atoms cannot accurately represent the preprocessed signal. In this case, increment the iteration number by one, that is, let:

[0106] t = t + 1 (15)

[0107] Proceed to the next iteration;

[0108] If ③: the residual is less than the residual of the previous iteration, but does not satisfy equation (14), then do not enter the next iteration, keep t unchanged, substitute the current residual into equation (10), and continue the current iteration.

[0109] The allowable residual value ε represents the degree to which the recovered signal approximates the preprocessed signal (i.e., how closely the recovered signal approximates the original signal). The allowable residual value is a marker to stop the iterative process of the signal reconstruction algorithm; when the residual value is less than or equal to the allowable residual value, the signal reconstruction algorithm iteration ends. Generally, the allowable residual value is less than 10. -3 And it is often set to 10 -6 .

[0110] It is understandable that the number of iterations t is the number of times the signal reconstruction algorithm runs. The more iterations there are, the closer the sparse representation of the signal is to the preprocessed signal. At the same time, the total time consumed will also be longer.

[0111] The essence of the compressed sensing reconstruction algorithm mentioned in this invention is to reconstruct the signal by iterating continuously, making the sparsity estimated by the algorithm close to the sparsity of the real signal, thereby obtaining a more accurate signal estimate. Therefore, it does not require presetting the sparsity or the maximum number of iterations, resulting in a more accurate signal estimate and high-quality signal reconstruction. The initial sparsity estimate is set to 1, and then after each iteration, the estimated sparsity value is incremented by one, meaning the estimated sparsity value equals the iteration number t.

[0112] The flowchart of the specific signal reconstruction algorithm is as follows: Figure 6 As shown.

[0113] Operation S7: Obtain the reconstructed signal of the original signal based on the current sparse signal.

[0114] After operations S1 to S6, θ was calculated. t Furthermore, the preprocessed signal x can be calculated from equation (2), i.e., x = Ψ DCT ·θ t .

[0115] Furthermore, x, which has a size of n*n rows and 1 column, n×n，1 Transform it into a matrix of size n*n, that is, x n×n，1 To restore the original data to its original blocks, the specific steps are: x n×n，1 By decomposing the data sequentially from top to bottom, such as taking the highest n data values ​​from top to bottom as the first column, the (n+1)th to 2nth data values ​​as the second column, and so on, a matrix of size n*n can be obtained.

[0116] For all The original signal can be successfully recovered by performing operations S3 to S6 on each block in sequence and then combining them according to the original arrangement.

[0117] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A signal reconstruction method in a computational ghost imaging system based on compressed sensing, characterized in that, Includes the following steps: S1, the original signal is divided into blocks to obtain a preprocessed signal; S2, using the discrete cosine transform matrix to perform sparse representation on the preprocessed signal; The preprocessed signal is measured using a measurement matrix to obtain a measurement signal; S3, let each atom in the sensing matrix have an inner product with the current residual of the measurement signal, and use the L*t atoms with the largest inner product to form matrix A. t Wherein, the perception matrix is ​​the dot product of the measurement matrix and the discrete cosine transform matrix, each atom represents a row of the perception matrix, t is the number of iterations, and L is a constant greater than 1; S4, calculate the measured signal and matrix A t The least squares value is used as the current sparse signal; S5, select the term with the largest absolute value t from the current sparse signal, and select the corresponding atom from the perception matrix to form a new matrix A. tL ; S6, based on the new matrix A tL Calculate the residual of the measured signal. If the residual is less than or equal to the threshold, output the current sparse signal. If the residual is greater than the threshold and greater than the residual of the previous iteration, let t = t + 1 and jump to S3. Otherwise, jump directly to S3. S7, obtain the reconstructed signal of the original signal based on the current sparse signal.

2. The signal reconstruction method in a computational ghost imaging system based on compressed sensing according to claim 1, characterized in that, S1 includes: dividing the original signal into several sub-signals evenly, and then performing one-dimensional processing on each sub-signal to obtain a preprocessed signal.

3. The signal reconstruction method in a computational ghost imaging system based on compressed sensing according to claim 1, characterized in that, In S3, matrix A is composed of L*t atoms with the largest inner product. t This includes: performing regularization screening on the L*t atoms corresponding to the largest inner product, and selecting t atoms to form matrix A. t .

4. The signal reconstruction method in a computational ghost imaging system based on compressed sensing according to any one of claims 1 to 3, characterized in that, S4 includes: The current sparse signal θ is calculated using the following formula. t : Where y represents the measurement signal. A represents t The transpose of .

5. The signal reconstruction method in a computational ghost imaging system based on compressed sensing according to claim 4, characterized in that, In S6, based on the new matrix A tL Calculate the residual of the measured signal, including: Calculate the current sparse signal θ using the following formula. t The residual r of the corresponding measurement signal tnew : in, A represents tL The transpose of .

6. A signal reconstruction device in a computational ghost imaging system based on compressed sensing, characterized in that, include: The first processing module is used to divide the original signal into blocks to obtain a preprocessed signal; The second processing module is used to perform sparse representation of the preprocessed signal using the discrete cosine transform matrix. The preprocessed signal is measured using a measurement matrix to obtain a measurement signal; The atom matrix update module is used to perform an inner product between each atom in the sensing matrix and the current residual of the measurement signal, and then use the L*t atoms with the largest inner product to form a matrix A. t Wherein, the sensing matrix is ​​the dot product of the measurement matrix and the discrete cosine transform matrix, each atom represents a row of the sensing matrix, t is the iteration number, and L is a constant greater than 1; and, the measurement signal and matrix A are calculated. t The least squares value of is used as the current sparse signal; and the term with the largest absolute value t is selected from the current sparse signal, and the corresponding atom is selected from the perception matrix to form a new matrix A. tL ; The judgment module is used to determine the value of the new matrix A. tL Calculate the residual of the measured signal. If the residual is less than or equal to a threshold, output the current sparse signal. If the residual is greater than the threshold and greater than the residual of the previous iteration, let t = t + 1, and then execute the operation of the atomic matrix update module. Otherwise, directly execute the operation of the atomic matrix update module. The signal reconstruction module is used to obtain a reconstructed signal of the original signal based on the current sparse signal.

7. The signal reconstruction device in the computational ghost imaging system based on compressed sensing according to claim 6, characterized in that, The first processing module is further configured to divide the original signal into several sub-signals evenly, and then perform one-dimensional processing on each sub-signal to obtain a preprocessed signal.

8. The signal reconstruction device in the computational ghost imaging system based on compressed sensing according to claim 6, characterized in that, The atomic matrix update module is also used to perform regularization filtering on the L*t atoms corresponding to the largest inner product, and select t atoms to form matrix A. t .

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the signal reconstruction method in a computational ghost imaging system based on compressed sensing as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the signal reconstruction method in a computational ghost imaging system based on compressed sensing as described in any one of claims 1 to 5.

Images