Spread spectrum single photon counting target detection method based on three-dimensional physical kernel

Abstract

The invention discloses a spread spectrum single photon counting target detection method based on a three-dimensional physical kernel, which comprises the following steps: firstly, obtaining three-dimensional cross-correlation photon counting data of a target to be detected, then establishing a physical kernel model within a light propagation range according to characteristics of an imaging target and a detector, carrying out frequency domain correlation on the physical kernel and target data, and filtering noise, so as to detect the target to be detected. And generating a final three-dimensional image. Based on the overall characteristics of the three-dimensional photon arrival time point data, correlation and filtering are carried out in the frequency domain, the processing efficiency is high, the position of a hidden object can be reconstructed more carefully, and the imaging effect is better than that of a traditional imaging method.

Description

Technical Field

[0001] The invention belongs to the field of shielded target imaging, and in particular relates to a spread spectrum single photon counting target detection method based on a three-dimensional physical kernel. Background Art

[0002] In recent years, the research on single photon counting three-dimensional target imaging methods has been widely carried out. During the imaging process, the following problems exist:

[0003] 1) Low resolution: Due to multiple reflections and scattering of light, information is lost during the propagation process, resulting in low resolution of the final reconstructed image and unclear details.

[0004] 2) Poor signal-to-noise ratio: The reflected light becomes very weak after multiple attenuations, while the background noise is relatively strong. Therefore, it is difficult to obtain high-quality signals and the signal-to-noise ratio is low.

[0005] 3) Algorithm complexity: Complex mathematical models and optimization algorithms are required, which places higher demands on computing resources.

[0006] The existing solutions are as follows:

[0007] Most three-dimensional target reconstructions are based on obtaining depth and intensity information, filtering or optimizing in the time domain, such as (Shanshan, Shen, Chen qian, et al. Computational Depth Imaging Using the Fast Deconvolution Method, IEEE Access, 2019, 99, 1) using threshold filtering to remove noise. For example, (Maccarone A, Rocca FMD, McCarthy A, et al. Three-dimensional imaging of stationary and moving targets in turbid underwater environments using a single-photon detector array [J]. Optics express, 27 (20): 28437-28456.) further uses polynomial approximation for smoothing according to the depth and intensity distribution, and by finding an optimal polynomial function, the sum of the squared differences between the polynomial and the actual data points is minimized. For example, (Ding Y, Wu H, Gao X, et al. Coded-pulse-bunch-laser-based single-photon lidar for fast long-distance ranging [J]. Journal of the Optical Society of America. A, Optics, image science, and vision, 2022, 39 (2): 206-212.) uses the simulated annealing (SA) algorithm to optimize the cross-correlation waveform representing the depth and intensity distribution. The above methods mainly optimize the imaging resolution or calculation speed by combining pixel-by-pixel optimization with global two-dimensional image optimization to improve imaging efficiency. Summary of the invention

[0008] In view of the above problems, the purpose of the present invention is to provide a spread spectrum single photon counting target detection method based on a three-dimensional physical kernel. First, based on the range of light emission and reflection propagation paths, a physical kernel based on the physical characteristics of the target and the detector is constructed; secondly, the generated physical kernel is subjected to a three-dimensional Fourier transform, and frequency domain correlation is performed with the three-dimensional data after preprocessing such as resampling, and finally a Laplace-Gaussian filter is applied to filter out noise photons. This method takes into account the range of light emission and reflection propagation paths, combines the physical characteristics of the target and the detector, and has good adaptability to targets of different materials, shapes and sizes. It can be applied to target detection in a variety of complex scenes, and can capture more subtle target features, thereby providing higher imaging resolution. The Laplace-Gaussian filter is applied to filter out noise photons, effectively reducing the impact of background noise and improving the signal-to-noise ratio (SNR). This method performs correlation and filtering in the frequency domain, and has high processing efficiency. This method solves the problems existing in the above-mentioned prior art, and its imaging effect is better than that of traditional imaging methods.

[0009] The specific technical solution for achieving the purpose of the present invention is:

[0010] A spread spectrum single photon counting target detection method based on a three-dimensional physical core comprises the following steps:

[0011] Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M

[0012] Step 2: Based on the detector parameters and the target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform;

[0013] Step 3: Correlate the physical kernel with the target data, and use a Laplace Gaussian filter to filter the correlated data to obtain filtered three-dimensional volume data;

[0014] Step 4: Reconstruct the filtered three-dimensional volume data to obtain the reconstructed three-dimensional volume data v o , complete target detection.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The solution of the present invention uses Fourier transform to perform inverse filtering operation on the physical kernel, which greatly improves the calculation efficiency and can effectively restore image details while enhancing the signal-to-noise ratio. Data cropping and rearrangement of data dimensions simplify the amount of calculation. The Laplace-Gaussian filter further enhances the edge information while reducing noise interference, making the reconstructed image sharper and more realistic with high imaging accuracy.

[0017] The present invention is further described below in conjunction with specific implementation modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the spread spectrum single photon counting target detection method based on three-dimensional physical kernel in this scheme. DETAILED DESCRIPTION

[0019] A spread spectrum single photon counting target detection method based on a three-dimensional physical core comprises the following steps:

[0020] Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M

[0021] In practice, a scanning galvanometer can be used to move the light signal emitted by the collimator horizontally and vertically. The computer controls the movement of the scanning galvanometer through the serial port, thereby controlling the movement of the light spot and scanning point by point to obtain a two-dimensional depth image.

[0022] Specifically, the single-point photon counting data is obtained, and the single-point cross-correlation photon counting data x of the target is calculated based on the one-dimensional fast Fourier transform method. e ∈R 1×M , based on the scanning galvanometer, the target is scanned by N rows and N columns to obtain N×∈ single-point cross-correlated photon counting data x e , finally forming the three-dimensional cross-correlation photon counting data m s ∈R N×N×M ;

[0023] Step 2: Based on the detector parameters and target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform:

[0024] Step 2-1, based on the time interval resolution, light speed and time series length of the detector, the detector first obtains the three-dimensional cross-correlation photon counting data of the target, then performs point time offset and cropping processing based on the single-point cross-correlation data, and finally performs fast Fourier transform, linear interpolation, and inverse fast Fourier transform operations to generate the final three-dimensional image; this application makes all maximum measurement ranges based on point time offset:

[0025] r a =M×c×b in

[0026] Where M represents the time resolution parameter, b in represents the time interval resolution of the SPAD detector;

[0027] Step 2-2, based on the detector, first obtain the target's three-dimensional cross-correlation photon counting data, then perform point time offset and cropping based on the single-point cross-correlation data, and finally perform fast Fourier transform, linear interpolation, and inverse fast Fourier transform operations to generate the final three-dimensional image. This application determines the physical core based on point time offset so that all maximum measurement ranges are:

[0028] p f =|(4×w a ÷2÷r a ) 2 ×(g x 2 +g y 2 )-g z |

[0029] Among them, g x g y g z is the grid coordinate in the constructed three-dimensional coordinate system, xyz is the coordinate number of each axis in the three-dimensional coordinate system, x=[-1,1]∈R 1×2N y=[-1,1]∈R 1×2N z=[0,2]∈R 1×2M , g x ∈R 2M×2N×2N , g y ∈R 2M×2N×2N , g z ∈R 2M×2N×2N , N is the spatial resolution parameter of the detector, w a represents the area of ​​the target region, p f ∈R 2N ×2N×2M , r a Interpolate position parameters for the detector in the frequency domain;

[0030] Step 2-3: Perform three-dimensional discrete Fourier transform on the physical kernel:

[0031]

[0032] Among them, F P (u,v,w)∈R N×N×M , u, v, w represent the frequency components in three directions, corresponding to the Fourier frequencies in the x, y, and z directions, u∈[0,N-1], v∈[0,N-1], w∈[0,M-1];

[0033] Step 2-4: Calculate the conjugate of the physical kernel after the three-dimensional discrete Fourier transform:

[0034]

[0035] Step 3: Associate the physical kernel with the target data, and use the Laplace Gaussian filter to filter the associated data to obtain filtered three-dimensional volume data:

[0036] Step 3-1: Construct a diagonal matrix And the sparse matrix resampled on the time axis Get the scaled sparse matrix mtx':

[0037] Where: D is: M 2 is the maximum number of non-zero elements in the matrix;

[0038] Step 3-2: Select all odd rows in mtx' to form a matrix All even-numbered rows form a matrix Get the sampled forward transformation matrix mtx" = (j + o) / 2;

[0039] Transpose mtx' to get Select all odd columns in mtxi' to form a matrix All even columns form a matrix Get the inverse transformation matrix after sampling mtxi" = (j i +o i ) / 2;

[0040] Step 3-3: Acquire the three-dimensional cross-correlation photon counting data m of the target s ∈R N×N×M The dimension order is rearranged to obtain the time flight measurement data d t ∈R M×N×N , for the three-dimensional matrix d t Flatten to obtain a two-dimensional matrix

[0041] d t (i,j=1,k) is the two-dimensional matrix d when j=1 t1 ∈R M×N , d t (i,j=2,k) is the two-dimensional matrix d when j=2 t2 ∈R M×N , until d t (i,j=N,k) is the two-dimensional matrix d when j=2 tN ∈R M×N , where ij k are the row index, column index and depth index of the three-dimensional matrix, 1≤i≤M, 1≤j≤N, 1≤k≤N, and so on, we get

[0042] Step 3-4: Convert the two-dimensional matrix After processing with the forward transformation matrix mtx" = (j + o) / 2, reshape to obtain the reshaped three-dimensional array t∈R M×N×N :

[0043] The three-dimensional array t∈R M×N×N Assign to all zero three-dimensional array preprocessed frequency domain data t d ∈R 2M×2N×2N the first half of

[0044] More specifically, these include:

[0045] Step 3-4-1: Convert the two-dimensional matrix Processing with the forward transformation matrix mtx" = (j + o) / 2 gives the matrix m:

[0046]

[0047] Step 3-4-2, decompose the matrix m into sub-matrices m1 m2 m3...m N :

[0048] Where m(i=M,j=[1:N]) is the two-dimensional matrix m1∈R when j=1 to j=N M×N , m(i=M,j=[N+1:2N]) is the two-dimensional matrix m2∈R when j=N+1 to j=2N M×N , and so on, until m N (i=M,j=[(N-1)N+1:N 2 ]) is j=(N-1)N+1 to j=N 2 The two-dimensional matrix m N ∈R M×N , 1≤i≤M 1≤j≤N 2 ;

[0049]

[0050] Step 3-4-3, assign the submatrix m1 to the first layer depth t(i,j,k=1)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=1)∈R M×N×N For a three-dimensional matrix when k = 1, the submatrix m2 is assigned to the second layer depth t(i, j, k = 2) ∈ R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=2)∈R M×N×N For a three-dimensional matrix when k = 2, the submatrix m3 is assigned to the third layer depth t(i, j, k = 3)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=3)∈R M×N×Nis a three-dimensional matrix when k=3, and so on. N Assign to the Nth layer depth t(i,j,k=3)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=N)∈R M×N×N is a three-dimensional matrix when k=N, 1≤i≤M, 1≤j≤N, 1≤k≤N, and finally the three-dimensional matrix t∈R M×N×N ;

[0051] Step 3-5: frequency domain data t d ∈R 2M×2N×2N Processing is performed to obtain the three-dimensional volume data associated with the physical nucleus t v ∈R M×N×N :

[0052] Step 3-5-1: Frequency domain data t d ∈R 2M×2N×2N Perform a 3D fast Fourier transform and perform physical kernel correlation:

[0053]

[0054] t vf =t df ⊙p i

[0055] Among them, ⊙ represents the Hadamard product, that is, the corresponding elements are multiplied;

[0056] Step 3-5-2, take array t vf The first half of the elements in each dimension are obtained by vf '∈R M×N×N ;

[0057] Step 3-5-3, for t vf 'Inverse Fourier transform of data

[0058]

[0059] Get the three-dimensional data t after physical core correlation v ∈R M×N×N ;

[0060] Step 3-6: 3D data t associated with the physical core v ∈R M×N×N Flatten it and process it using the Laplace Gaussian filter to obtain the filtered reconstructed volume data v l ∈R M×N×N :

[0061] Three-dimensional data associated with physical nuclei v ∈R M×N×N Flatten to obtain volume data in two-dimensional form Then multiply the inverse transformation matrix mtxi' by t to adjust the obtained volume data into a three-dimensional array v∈R M×N×N , use the Laplace-Gaussian filter to process the three-dimensional array v:

[0062] First: Create a Gaussian kernel w(x,y,z)∈R h×h×h , where x, y and z represent spatial coordinates, σ 2 is the square of the standard deviation;

[0063] Second: construct the Laplacian operator: w1(x,y,z)∈R h ×h×h ;

[0064] w'(x,y,z)∈R h×h×h , where h is the side length of the filter window, h 3 Represents the total number of elements in the entire filter window;

[0065] Perform a three-dimensional convolution on v and w' to obtain the reconstructed volume v l ∈R M×N×N :

[0066]

[0067] Among them, i, j, k are coordinates in three-dimensional space, a, b, c are the coordinates of the convolution kernel, w'(a, b, c) is the three-dimensional convolution kernel, and v(x+a, y+b, z+c) is the corresponding offset of the input data.

[0068] Step 4: Reconstruct the filtered volume data of the target to obtain the reconstructed three-dimensional volume data v l , complete target detection:

[0069] For the reconstruction volume v l Rearrange the dimensions and adjust the three-dimensional order to [width, height, depth], respectively for v l Flip vertically (second dimension) and horizontally (third dimension), that is, v o ∈R M×N×N

[0070] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0071] Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M

[0072] Step 2: Based on the detector parameters and the target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform;

[0073] Step 3: Correlate the physical kernel with the target data, and use a Laplace Gaussian filter to filter the correlated data to obtain filtered three-dimensional volume data;

[0074] Step 4: Reconstruct the filtered three-dimensional volume data to obtain the reconstructed three-dimensional volume data v o , complete target detection.

[0075] The present invention also provides a computer storable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0076] Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M

[0077] Step 2: Based on the detector parameters and the target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform;

[0078] Step 3: Correlate the physical kernel with the target data, and use a Laplace Gaussian filter to filter the correlated data to obtain filtered three-dimensional volume data;

[0079] Step 4: Reconstruct the filtered three-dimensional volume data to obtain the reconstructed three-dimensional volume data v o , complete target detection.

[0080] Example

[0081] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an kind" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0083] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be considered as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so that once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0084] A spread spectrum single photon counting target detection method based on a three-dimensional physical core comprises the following steps:

[0085] Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M :

[0086] In practice, a scanning galvanometer can be used to move the light signal emitted by the collimator horizontally and vertically. The computer controls the movement of the scanning galvanometer through the serial port, thereby controlling the movement of the light spot and scanning point by point to obtain a two-dimensional depth image.

[0087] The single-point cross-correlation waveform X∈R of the target is obtained by using the fast Fourier transform method. 1×256 ;

[0088] By scanning 512×512 pixels, 512×512 cross-correlation waveform measurement data are obtained, denoted as m s ∈R 512 ×512×256 ;

[0089] Step 2: Based on the detector parameters and the target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform;

[0090] The time interval resolution b of the SPAD detector in this embodiment in =3.2×10 -11 ps, speed of light c = 3 × 10 8 m / s, spatial resolution parameter N = 512, time resolution parameter M = 256, frequency domain interpolation position parameter r a =2.4567;

[0091] Step 2-1: Determine the maximum measurement range of the detector based on the detector's time interval resolution, light speed, and time resolution parameters:

[0092] r a =M×c×b in =2.4567

[0093] Where M represents the time resolution parameter, b in represents the time interval resolution of the SPAD detector;

[0094] Step 2-2: Determine the physical nucleus based on the maximum measurement range of the detector:

[0095] p f =|(4×w a ÷2÷r a ) 2 ×(g x 2 +g y 2 )-g z |

[0096] Among them, g x g y g z is the grid coordinate in the constructed three-dimensional coordinate system, and the three-dimensional grid g x g y g z It is used to simulate the light propagation characteristics under active imaging conditions. xyz is the coordinate number of each axis in the three-dimensional coordinate system, x = [-1, 1] ∈ R 1×1024 , y=[-1,1]∈R 1×1024 , z=[0,2]∈R 1×512 , g x ∈R 1024×1024×512 , g y ∈R 1024×1024×512 , g z ∈R 1024×1024×512 , p f ∈R 2N×2N×2M , r a Interpolate position parameters for the detector in the frequency domain;

[0097] In this embodiment, p f ∈R 1024×1024×512 ;

[0098] x and y are the horizontal and vertical coordinates on the two-dimensional plane, respectively, and 2×N sample points are generated from -1 to 1 at equal intervals. x Each row of is a constant value of the x-coordinate, g y Each column is a constant value of the y coordinate; z is the vertical coordinate, which generates 2×M sample points at equal intervals from 0 to 2. z Each layer of is a constant value for the z coordinate.

[0099] Step 2-3: Perform three-dimensional discrete Fourier transform on the physical kernel:

[0100]

[0101] Get F P ∈R 512×512×256 ;

[0102] Among them, for p f (x,y,z) is transformed by Fourier to get F P (u,v,w)∈R N×N×M , represents the frequency spectrum of the point in the frequency domain, u, v, w represent the frequency components in three directions, corresponding to the Fourier frequencies in the x, y, z directions, u∈[0,511], v∈[0,511], w∈[0,255], is a complex exponential function, representing the conversion between the spatial domain and the frequency domain. u, v, w are frequency variables, which correspond to the Fourier transform of the spatial coordinates x, y, z;

[0103] Step 2-4: Calculate the conjugate of the physical kernel after the three-dimensional discrete Fourier transform:

[0104] That is, p i ∈R 512×1024×1024

[0105] Step 3: Associate the physical kernel with the target data, and use the Laplace Gaussian filter to filter the associated data to obtain the filtered volume data of the target:

[0106] Step 3-1: Construct a diagonal matrix D∈R 65536×65536 And the sparse matrix resampled on the time axis

[0107]

[0108] mtx∈R 65536×256 , get the scaled sparse matrix mtx': D×mtx=mtx'∈R 65536×256 ;

[0109] 65536 is the maximum number of non-zero elements in a matrix;

[0110] Step 3-2: Select all odd rows in mtx' to form a matrix j∈R 128×128 , all even rows form a matrix o∈R 128×128 , get the sampled forward transformation matrix mtx" = (j + o) / 2;

[0111] Transpose mtx' to get Select all odd columns in mtxi' to form matrix j i ∈R 128 ×128 , all even columns form the matrix o i ∈R 128×128 , get the inverse transformation matrix after sampling mtxi" = (j i +o i ) / 2;

[0112] Step 3-3: Single-point cross-correlation waveform m of the target's two-dimensional depth image s ∈R 512×512×256 The dimension order is rearranged to obtain the time flight measurement data d t ∈R 256×512×512 , for the three-dimensional matrix d t Flatten to obtain a two-dimensional matrix

[0113] d t (i,j=1,k) is the two-dimensional matrix d when j=1 t1 ∈R 256×512 , d t (i,j=2,k) is the two-dimensional matrix d when j=2 t2 ∈R 256×512 , until d t (i,j=N,k) is the two-dimensional matrix d when j=512 t512 ∈R 256×512 , where ij k are the row index, column index and depth index of the three-dimensional matrix, 1≤i≤256, 1≤j≤512, 1≤k≤512, and so on, we get

[0114] Step 3-4: Convert the two-dimensional matrix After processing with the forward transformation matrix mtx" = (j + o) / 2, reshape to obtain the reshaped three-dimensional array t∈R 256×512×512 :

[0115] The three-dimensional array t∈R 256×512×512 Assign to all zero three-dimensional array preprocessed frequency domain data t d ∈R 516×516×1024 the first half of

[0116] More specifically, these include:

[0117] Step 3-4-1: Convert the two-dimensional matrix Processing with the forward transformation matrix mtx" = (j + o) / 2 gives the matrix m:

[0118]

[0119] Step 3-4-2, decompose the matrix m into sub-matrices m1 m2 m3...m N :

[0120]

[0121] m(i=256,j=[1:512]) is the two-dimensional matrix m1∈R when j=1 to j=512 256×512 , m(i=256,j=[513:1024]) is the two-dimensional matrix m2∈R when j=512 to j=1024 256×512 , until m 512 (i=256,j=[261632:512 2 ]) for j=261632 to j=512 2 The two-dimensional matrix m 512 ∈R 256×512 , 1≤i≤2561≤j≤512 2 ;

[0122] Step 3-4-3, assign the submatrix m1 to the first layer depth t(i,j,k=1)∈R of the three-dimensional matrix t 256 ×512×512 , that is, t(i,j,k=1)∈R 256×512×512 For a three-dimensional matrix when k = 1, the submatrix m2 is assigned to the second layer depth t(i, j, k = 2) ∈ R of the three-dimensional matrix t 256×512×512 , that is, t(i,j,k=2)∈R 256×512×512 For a three-dimensional matrix when k = 2, the submatrix m3 is assigned to the third layer depth t(i, j, k = 3)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=3)∈R 256×512×512 is a three-dimensional matrix when k=3, and so on. N Assign to the Nth layer depth t(i,j,k=N)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=N)∈R 256×512×512 is a three-dimensional matrix when k=512, 1≤i≤256 1≤j≤5121≤k≤512, and finally the three-dimensional matrix t∈R 256×512×512 ;

[0123] Step 3-5: frequency domain data t d ∈R 512×1024×1024 Processing is performed to obtain the three-dimensional data associated with the physical nucleus v ∈R M×N×N :

[0124] Step 3-5-1: Frequency domain data t d ∈R 512×1024×1024Perform a 3D fast Fourier transform and associate it with the physical kernel:

[0125]

[0126] t vf =t df ⊙p i , t vf ∈R 512×1024×1024

[0127] Among them, ⊙ represents the Hadamard product, that is, the corresponding elements are multiplied;

[0128] Step 3-5-2, take array t vf The first half of the elements in each dimension are obtained by vf '∈R 256×512×512 ;

[0129] Step 3-5-3, for t vf 'Inverse Fourier transform of data

[0130]

[0131] Get the three-dimensional data after physical core correlation v ∈R 256×512×512 ;

[0132] Step 3-6: t after physical core correlation v ∈R 256×512×512 Flatten it and process it using the Laplace Gaussian filter to obtain the filtered reconstructed volume data v l ∈R M×N×N :

[0133] After the physical nucleus correlation, t v ∈R 256×512×512 Flatten to obtain volume data in two-dimensional form Then multiply the inverse transformation matrix mtxi' by t to adjust the obtained volume data into a three-dimensional array v∈R 256×512×512 , use the Laplace-Gaussian filter to process the three-dimensional array v:

[0134] First: Create a Gaussian kernel w(x,y,z)∈R h×h×h , where x, y and z represent spatial coordinates, σ 2 is the square of the standard deviation;

[0135] Second: construct the Laplacian operator: w1(x,y,z)∈R h ×h×h ;

[0136] w'(x,y,z)∈R h×h×h , where h is the side length of the filter window, h 3 Represents the total number of elements in the entire filter window;

[0137] Perform a three-dimensional convolution on v and w' to obtain the reconstructed volume v l ∈R M×N×N :

[0138]

[0139] Among them, i, j, k are coordinates in three-dimensional space, a, b, c are the coordinates of the convolution kernel, w'(a, b, c) is the three-dimensional convolution kernel, and v(x+a, y+b, z+c) is the corresponding offset of the input data.

[0140] Step 4: Reconstruct the filtered volume data of the target to obtain the reconstructed three-dimensional volume data v l , complete target detection:

[0141] For the reconstruction volume v l Rearrange the dimensions and adjust the three-dimensional order to [width, height, depth], respectively for v l Flip vertically (second dimension) and horizontally (third dimension), that is, v o ∈R 256×512×512 .

[0142] The solution of the present invention uses Fourier transform to perform inverse filtering operation on the physical kernel, which greatly improves the calculation efficiency, and can effectively restore image details while enhancing the signal-to-noise ratio. Data cropping and rearrangement of data dimensions simplify the amount of calculation. The Laplace-Gaussian filter further enhances edge information while reducing noise interference, making the reconstructed image sharper and more realistic, and also improving the final imaging details to achieve the imaging effect.

[0143] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A spread spectrum single photon counting target detection method based on three-dimensional physical kernel, characterized in that: The following steps are involved: Step 1: Obtain the three-dimensional cross-correlation photon counting data m of the target to be imaged s ∈R N×N×M ; Step 2: Based on the detector parameters and the target parameters, determine the physical core of the target and its corresponding three-dimensional Fourier transform; Step 3: Correlate the physical kernel with the target data, and use a Laplace Gaussian filter to filter the correlated data to obtain filtered three-dimensional volume data; Step 4: Reconstruct the filtered three-dimensional volume data to obtain the reconstructed three-dimensional volume data v o , complete target detection.

2. The spread spectrum single photon counting target detection method based on three-dimensional physical core according to claim 1 is characterized in that: The photon counting data m in step 1 s ∈R N×N×M The acquisition process is: Obtain single-point photon counting data, and calculate the single-point cross-correlation photon counting data x of the target based on the one-dimensional fast Fourier transform method e ∈R 1×M , based on the scanning galvanometer, the target is scanned by N rows and N columns to obtain N×N single-point cross-correlated photon counting data x e , finally forming the three-dimensional cross-correlation photon counting data m s ∈R N×N×M .

3. The method for target detection based on three-dimensional physical core spread spectrum single photon counting according to claim 1 is characterized in that: The calculation of the physical kernel and its corresponding three-dimensional Fourier transform in step 2 are specifically: Step 2-1: Determine the maximum measurement range of the detector based on the detector's time interval resolution, light speed, and time resolution parameters: r a =M×c×b in Where M represents the time resolution parameter, b in represents the time interval resolution of the SPAD detector; Step 2-2: Determine the physical nucleus based on the maximum measurement range of the detector: p f =|(4×w a ÷2÷r a ) 2 ×(g x 2 +g y 2 )-g z | Among them, g x g y g z is the grid coordinate in the constructed three-dimensional coordinate system, xyz is the coordinate number of each axis in the three-dimensional coordinate system, x=[-1,1]∈R 1×2N y=[-1,1]∈R 1×2N z=[0,2]∈R 1×2M , g x ∈R 2M×2N×2N , g y ∈R 2M ×2N×2N , g z ∈R 2M×2N×2N , N is the spatial resolution parameter of the detector, w a represents the area of ​​the target region, p f ∈R 2N ×2N×2M , r a Interpolate position parameters for the detector in the frequency domain; Step 2-3: Perform three-dimensional discrete Fourier transform on the physical kernel: Among them, F P (u,v,w)∈R N×N×M , u, v, w represent the frequency components in three directions, corresponding to the Fourier frequencies in the x, y, and z directions, u∈[0,N-1], v∈[0,N-1], w∈[0,M-1]; Step 2-4: Calculate the conjugate of the physical kernel after the three-dimensional discrete Fourier transform:

4. The spread spectrum single photon counting target detection method based on three-dimensional physical core according to claim 1 is characterized in that: The filtering process using the Laplace Gaussian filter in step 3 is specifically as follows: Step 3-1: Construct a diagonal matrix And the sparse matrix resampled on the time axis Get the scaled sparse matrix mtx': Where: D is: Step 3-2: Select all odd rows in mtx' to form a matrix All even-numbered rows form the matrix Get the sampled forward transformation matrix mtx" = (j + o) / 2; Transpose mtx' to get Select all odd columns in mtxi' to form a matrix All even columns form a matrix Get the inverse transformation matrix after sampling mtxi" = (j i +o i ) / 2; Step 3-3: Acquire the three-dimensional cross-correlation photon counting data m of the target s ∈R N×N×M The dimension order is rearranged to obtain the time flight measurement data d t ∈R M×N×N , for the three-dimensional matrix d t Flatten to obtain a two-dimensional matrix Step 3-4: Convert the two-dimensional matrix After processing with the forward transformation matrix mtx" = (j + o) / 2, reshape to obtain the reshaped three-dimensional array t∈R M×N×N ; The three-dimensional array t∈R M×N×N Assign to all zero three-dimensional array preprocessed frequency domain data t d ∈R 2M×2N×2N the first half of Step 3-5: frequency domain data t d ∈R 2M×2N×2N Processing is performed to obtain the three-dimensional data associated with the physical nucleus v ∈R M×N×N ; Step 3-6: 3D data t associated with the physical core v ∈R M×N×N Flatten it and filter it with Laplace Gaussian filter to obtain the filtered volume data v l ∈R M×N×N .

5. The method for target detection based on three-dimensional physical core spread spectrum single photon counting according to claim 4 is characterized in that: The reshaped three-dimensional array t∈R in steps 3-4 M×N×N The process is: Step 3-4-1: Convert the two-dimensional matrix Processing with the forward transformation matrix mtx" = (j + o) / 2 gives the matrix m: m=mtx”×d, Step 3-4-2, decompose the matrix m into sub-matrices m1 m2 m3...m N ; Where m(i=M,j=[1:N]) is the two-dimensional matrix m1∈R when j=1 to j=N M×N , m(i=M,j=[N+1:2N]) is the two-dimensional matrix m2∈R when j=N+1 to j=2N M+N , and so on, until m N (i=M,j=[(N-1)N+1:N 2 ]) is j=(N-1)N+1 to j=N 2 The two-dimensional matrix m N ∈R M×N , 1≤i≤M 1≤j≤N 2 ; Step 3-4-3, assign the submatrix m1 to the first layer depth t(i,j,k=1)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=1)∈R M×N×N For a three-dimensional matrix when k = 1, the submatrix m2 is assigned to the second layer depth t(i, j, k = 2) ∈ R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=2)∈R M×N×N For a three-dimensional matrix when k = 2, the submatrix m3 is assigned to the third layer depth t(i,j,k=3)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=3)∈R M×N×N is a three-dimensional matrix when k=3, and so on. N Assign to the Nth layer depth t(i,j,k=N)∈R of the three-dimensional matrix t M×N×N , that is, t(i,j,k=N)∈R M×N×N is a three-dimensional matrix when k=N, 1≤i≤M, 1≤j≤N, 1≤k≤N, and finally the three-dimensional matrix t∈R M×N×N .

6. The method for target detection based on three-dimensional physical core spread spectrum single photon counting according to claim 4 is characterized in that: The three-dimensional volume data t of the target image after filtering in step 3-5 v ∈R M×N×N for: Step 3-5-1: Frequency domain data t d ∈R 2M×2N×2N Take a 3D fast Fourier transform and perform a physical kernel correlation: t df ∈R 2M×2N×2N t vf =t df ⊙p i Among them, ⊙ represents the Hadamard product; Step 3-5-2, take array t vf The first half of the elements in each dimension are obtained by vf '∈R M×N×N ; Step 3-5-3, for t vf 'Inverse Fourier transform of the data: Get the three-dimensional data t after physical core correlation v ∈R M×N×N .

7. The method for target detection based on three-dimensional physical core spread spectrum single photon counting according to claim 4, characterized in that: The filtered volume data v in step 3-6 l ∈R M×N×N for: After the physical nucleus correlation, t v ∈R M×N×N Flatten to obtain volume data in two-dimensional form Then multiply the inverse transformation matrix mtxi' by t to adjust the obtained volume data into a three-dimensional array v∈R M×N×N , use the Laplace-Gaussian filter to process the three-dimensional array v: First: Create a Gaussian kernel w(x,y,z)∈R h×h×h , where x, y and z represent spatial coordinates, σ 2 is the square of the standard deviation; Construct the Laplacian operator: w1(x,y,z)∈R h×h×h ; w'(x,y,z)∈R h×h×h , where h is the side length of the filter window, h 3 Represents the total number of elements in the entire filter window; Perform a three-dimensional convolution on v and w' to obtain the reconstructed volume v l ∈R M×N×N : Among them, i, j, k are coordinates in three-dimensional space, a, b, c are the coordinates of the convolution kernel, w'(a, b, c) is the three-dimensional convolution kernel, and v(x+a, y+b, z+c) is the corresponding offset of the input data.

8. The spread spectrum single photon counting target detection method based on three-dimensional physical core according to claim 1 is characterized in that: The step 4 of reconstructing the filtered volume data of the target is specifically as follows: For the reconstruction volume v l Rearrange the dimensions and adjust the three-dimensional order to [width, height, depth], respectively for v l Perform flip operations in the vertical and horizontal directions and output v o ∈R M×N×N .

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer storable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images