Sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology

Abstract

The invention discloses a sparse view NCT three-dimensional image reconstruction method based on a neutron imaging technology, and relates to the technical field of neutron imaging, the sparse view NCT three-dimensional image reconstruction method is applied to a neutron computer imaging system, the neutron computer imaging system comprises a ray generation subsystem and a neutron imaging subsystem, and the method comprises the following steps: the ray generation subsystem generates a target ray; the neutron imaging subsystem acquires a two-dimensional original image based on the target ray; and the neutron imaging subsystem performs three-dimensional image reconstruction on the two-dimensional original image by using a sparse view NCT three-dimensional reconstruction algorithm to obtain a sparse view NCT three-dimensional image reconstruction result. According to the Os-SART-PDTV algorithm, the advantages of the Os-SART iterative algorithm and the first-order prime-dual algorithm with total variation denoising are combined, the two algorithms have partial complementarity, noise can be more effectively removed, artifacts can be more effectively inhibited, the detail structure of the image can be more effectively reserved in sparse view NCT three-dimensional reconstruction, and the reconstructed image is clearer.

Description

Technical Field

[0001] The present invention relates to the field of neutron imaging technology, and in particular to a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology. Background Art

[0002] Computed tomography (NCT), a non-invasive measurement technique, has been widely used in fields such as nuclear engineering, thermal hydraulics, and cultural heritage. However, the technology has significant limitations. Low neutron source intensity results in extended scanning times, which severely impacts the quality of projected images. Therefore, appropriate image processing algorithms must be selected to meet the demands of practical applications.

[0003] Among the existing image reconstruction algorithms, they are mainly divided into three categories: analytical algorithms, iterative algorithms and deep learning algorithms. Among them, the application of analytical algorithms is subject to strict conditional restrictions. It requires complete projection data, which makes it unable to be used normally in harsh environments such as strong radiation, high temperature and high pressure; deep learning algorithms have high requirements for data and models. Not only do they need a large amount of data as support, but they also rely on complex models, which makes their calculations extremely difficult and has poor interpretability, and is subject to many limitations in practical applications.

[0004] Therefore, a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology is urgently needed to address the shortcomings of the existing technology. Summary of the Invention

[0005] The purpose of the present invention is to propose a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology, which can effectively remove noise and suppress artifacts in the case of sparse views while preserving the edge structure of the image to achieve high-quality three-dimensional image reconstruction.

[0006] To achieve the above objectives, the present invention provides a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology, which is applied to a neutron computer imaging system. The neutron computer imaging system includes a ray generation subsystem and a neutron imaging subsystem. The method includes:

[0007] The ray generation subsystem generates target rays;

[0008] The neutron imaging subsystem acquires a two-dimensional original image based on the target ray;

[0009] The neutron imaging subsystem uses a sparse view NCT three-dimensional reconstruction algorithm to perform three-dimensional image reconstruction on the two-dimensional original image to obtain a sparse view NCT three-dimensional image reconstruction result.

[0010] Optionally, the ray generating subsystem includes a collimator and a rotating sample stage;

[0011] The ray generation subsystem generates target rays, including:

[0012] The rotating sample stage is used to generate collimated neutron rays generated by the collimator to irradiate the sample to be detected at multiple angles to obtain target rays.

[0013] Optionally, the neutron imaging subsystem includes a two-dimensional neutron image detector, a conversion screen and a controller;

[0014] The neutron imaging subsystem acquires a two-dimensional original image based on the target ray, including:

[0015] The conversion screen converts the target ray into a light signal of the target ray;

[0016] The two-dimensional neutron image detector acquires a two-dimensional original image based on the optical signal of the target ray;

[0017] The controller receives the two-dimensional original image sent by the two-dimensional neutron image detector.

[0018] Optionally, the sparse view NCT three-dimensional reconstruction algorithm adopts an Os-SART iterative algorithm and a first-order primitive-dual algorithm with total variation denoising.

[0019] Optionally, the objective function of the sparse view NCT 3D reconstruction algorithm is:

[0020]

[0021] Among them, f * is the image after 3D reconstruction, arg min is the minimum value of the function, f represents the image to be reconstructed, TV is the total variation, A is the system matrix, b is the real projection data or observation data, and ε is a non-negative threshold parameter.

[0022] Optionally, the neutron imaging subsystem uses a sparse view NCT three-dimensional reconstruction algorithm to perform three-dimensional image reconstruction on the two-dimensional original image to obtain a sparse view NCT three-dimensional image reconstruction result, including:

[0023] The controller initializes first target parameters, wherein the first target parameters include an initial image, an iterative relaxation parameter, a number of iterations of a sub-cycle, a number of ordered subsets, a weight parameter of a primal-dual TV regularization, and a number of iterations of a primal-dual TV sub-cycle;

[0024] The controller performs outer loop processing on the two-dimensional original image based on the first target parameter to obtain an outer loop processed image of the two-dimensional original image and an image update amount corresponding to the outer loop processed image;

[0025] The controller determines whether the image update amount of the outer loop processed image is less than a first target threshold value. If so, the iteration stops; otherwise, the Os-SART iterative small loop is performed on the outer loop processed image of the two-dimensional original image to obtain an iterative small loop processed image of the outer loop processed image;

[0026] The controller performs primal-dual TV regularization sub-cycle processing on the iterative small-cycle processing image of the outer-loop processing image to obtain an output result of the TV sub-cycle;

[0027] The controller uses the output result of the TV sub-loop as a new starting point for outer iteration, and returns to the controller to perform outer loop processing based on the first target parameter according to the two-dimensional original image until a sparse view NCT three-dimensional image reconstruction result is obtained.

[0028] Optionally, the controller performs primal-dual TV regularization sub-loop processing on the iterative small-loop processing image of the outer-loop processing image to obtain an output result of the TV sub-loop, including:

[0029] The controller initializes a second target parameter;

[0030] The controller performs primal-dual TV regularized sub-loop processing based on the second target parameter according to the iterative small-loop processing image of the outer-loop processing image, and obtains an image update amount of the primal-dual TV regularized sub-loop processing image of the iterative small-loop processing image and the corresponding primal-dual TV regularized sub-loop processing image;

[0031] The controller determines whether the image update amount of the primal-dual TV regularized sub-loop processing image is less than the second target threshold. If so, the iteration stops; otherwise, the primal-dual TV regularized sub-loop processing image based on the iterative small loop processing image performs dual variable update, primal variable update and convergence check in sequence to obtain the output result of the TV sub-loop.

[0032] Compared with the closest prior art, the present invention has the following beneficial effects:

[0033] Compared with deep learning algorithms, the method of the present invention does not require a large amount of data and complex model support, the computational difficulty is relatively low, and it has better interpretability. Compared with analytical algorithms, the method of the present invention does not require complete projection data and can perform image reconstruction in the case of sparse views. Therefore, it is suitable for harsh environments such as strong radiation, high temperature and high pressure. The Os-SART-PDTV algorithm of the present invention combines the advantages of the Os-SART iterative algorithm and the first-order primitive dual algorithm with total variation denoising, and the two algorithms are partially complementary. They can more effectively remove noise and suppress artifacts in sparse view NCT three-dimensional reconstruction, retain the detailed structure of the image, and make the reconstructed image clearer. In addition, the present invention uses an iterative algorithm to reconstruct images for sparse view neutron computed tomography (NCT) three-dimensional reconstruction, which has advantages in removing noise, suppressing artifacts, and retaining edge structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 This is a flowchart of a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of a neutron computer imaging system proposed in an embodiment of the present invention;

[0037] Figure 3 This is a flow chart of the Os-SART-PDTV algorithm proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, 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 making creative efforts are within the scope of protection of the present invention.

[0039] The terms used in the embodiments of the present invention are only used to explain the specific embodiments of the present invention and are not intended to limit the present invention.

[0040] This paper mainly focuses on the sparse view NCT 3D reconstruction algorithm (Os-SART-PDTV algorithm), which is a new sparse view NCT 3D image reconstruction algorithm based on the ordered subset synchronous algebraic reconstruction technology (Os-SART) iterative algorithm and the first-order element-primal-dual algorithm with total variation denoising. Figure 1 The present invention describes a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology.

[0041] like Figure 1 As shown, an embodiment of the present invention provides a sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology, which is applied to a neutron computer imaging system. The neutron computer imaging system includes a ray generation subsystem and a neutron imaging subsystem. The method includes:

[0042] S1, the ray generation subsystem generates the target ray;

[0043] In this embodiment, a neutron beam is emitted from a neutron source and confined by a collimator to form a targeted beam with specific directionality. After the neutrons interact with the sample (through absorption and scattering), their energy attenuation characteristics are directly related to the sample's material composition and structural thickness. This allows the targeted beam to carry physical information within the sample, providing the data foundation for subsequent imaging. This step ensures the beam's directionality and physical information-carrying capacity, a prerequisite for non-invasive internal structure detection.

[0044] S2. The neutron imaging subsystem acquires a two-dimensional original image based on the target ray;

[0045] In this embodiment, target radiation passing through the sample is converted into optical signals by a neutron conversion screen, which are then captured by a charge-coupled device (CCD) camera to form a two-dimensional raw image. Multi-angle illumination (sparse viewing) is achieved by rotating the sample stage, allowing for the acquisition of projection data from different viewing angles. This step converts the invisible neutron signal into a visible image and, under sparse viewing conditions, reduces scanning time and neutron source dependency. Furthermore, the multi-view data provides the necessary projection dimensions for three-dimensional reconstruction, balancing imaging efficiency and data volume requirements.

[0046] S3. The neutron imaging subsystem uses a sparse view NCT 3D reconstruction algorithm to perform 3D image reconstruction on the 2D original image to obtain a sparse view NCT 3D image reconstruction result;

[0047] In this embodiment, the sub-imaging subsystem uses the sparse view NCT 3D reconstruction algorithm (Os-SART-PDTV algorithm) to reconstruct 3D images from the original 2D images. This algorithm can achieve 3D image reconstruction under sparse view conditions with low neutron source intensity and limited scanning angles without the need for complete projection data. This effectively overcomes the limitation of analytical algorithms that rely on complete data and is suitable for harsh environments such as strong radiation, high temperature and high pressure. By combining the ordered subset iterative acceleration mechanism of the ordered subset simultaneous algebraic reconstruction technique (Os-SART) with the total variation regularization of the total variation denoising primal-dual algorithm (PDTV), the algorithm suppresses noise and artifacts while ensuring the consistency of projection data, avoiding the dependence of deep learning on large amounts of data and complex models, with low computational complexity and strong interpretability. In addition, the algorithm prioritizes image edges and structural details through gradient constraints and back-projection update strategies, making the reconstructed 3D images under sparse views clear and achieving a balanced optimization between noise and detail, significantly enhancing the practical application value of NCT imaging.

[0048] Furthermore, the ray generating subsystem includes a collimator and a rotating sample stage.

[0049] As a possible implementation, in the above embodiment, step S1 may specifically include the following steps:

[0050] The rotating sample stage is utilized, and the collimator generates collimated neutron rays to irradiate the sample to be detected at multiple angles to obtain target rays.

[0051] Furthermore, the neutron imaging subsystem includes a two-dimensional neutron image detector, a conversion screen and a controller.

[0052] As a possible implementation, in the above embodiment, step S2 may specifically include the following steps:

[0053] S2-1, the conversion screen converts the target ray into a light signal of the target ray;

[0054] S2-2, the two-dimensional neutron image detector acquires a two-dimensional original image based on the optical signal of the target ray;

[0055] S2-3. The controller receives the two-dimensional original image sent by the two-dimensional neutron image detector.

[0056] like Figure 2As shown in Figure 1, the neutron computed tomography (NCT) system primarily consists of a ray generation subsystem and a neutron imaging subsystem. The ray generation subsystem includes a collimator, a shield, and a rotating sample stage. The neutron imaging subsystem comprises a two-dimensional neutron image detector, a conversion screen, and a controller, which utilizes three-dimensional reconstruction software. The collimator is used to collimate the neutron ray (neutron source) to form a beam in a specific direction; the shield shields stray neutrons and reduces interference; the rotating sample stage is used to place the sample to be tested and can rotate the sample to obtain projection data from different angles; and the conversion screen converts the neutron ray signal into an optical signal that can be recognized and captured by the CCD camera.

[0057] Specifically, due to the interaction between neutrons and matter, such as absorption and scattering, the energy of neutrons will change after passing through the matter. The change in energy is usually related to the material of the matter and the length of the neutron beam passing through the matter. Therefore, the internal information of the detected matter can be obtained by the change in energy before and after the neutron beam. First, the collimated neutron beam is obtained through a collimator, and the collimated neutron beam is used to irradiate the sample to be detected. After irradiation with neutron beams at a single angle, only two-dimensional data of the sample can be obtained, and the distribution of the sample in three-dimensional space cannot be obtained. Therefore, based on the two-dimensional data of the sample, the sample to be detected is placed on a rotating table and irradiated from multiple angles to obtain the rays after passing through the sample to be detected.

[0058] After passing through the sample, the radiation passes through a neutron conversion screen (also known as a conversion screen). Because cameras cannot recognize neutron radiation signals, the conversion screen converts them into light signals, which are then captured by a CCD camera. This produces a set of two-dimensional raw images, which can then be reconstructed into a three-dimensional image using a computer.

[0059] As a possible implementation, in the above embodiment, step S3 may specifically include the following steps:

[0060] S3-1. The controller initializes first target parameters, wherein the first target parameters include an initial image, an iterative relaxation parameter, the number of iterations of a sub-loop, the number of ordered subsets, a weight parameter of a primal-dual TV regularization, and the number of iterations of a primal-dual TV sub-loop;

[0061] S3-2, the controller performs outer loop processing on the two-dimensional original image based on the first target parameter, and obtains an outer loop processed image of the two-dimensional original image and an image update amount corresponding to the outer loop processed image;

[0062] S3-3, the controller determines whether the image update amount of the outer loop-processed image is less than a first target threshold value, and if so, stops the iteration; otherwise, performs an Os-SART iterative small loop on the outer loop-processed image of the two-dimensional original image to obtain an iterative small loop-processed image of the outer loop-processed image;

[0063] S3-4, the controller performs primal-dual TV regularization sub-loop processing on the iterative small-loop processing image of the outer-loop processing image to obtain an output result of the TV sub-loop;

[0064] S3-5. The controller uses the output result of the TV sub-loop as a new starting point for outer iteration, and returns to the controller to perform outer loop processing based on the first target parameter according to the two-dimensional original image until the sparse view NCT three-dimensional image reconstruction result is obtained.

[0065] As a possible implementation, in the above embodiment, step S3-4 may specifically include the following steps:

[0066] S3-4-1, the controller initializes the second target parameter;

[0067] S3-4-2, the controller performs primal-dual TV regularized sub-loop processing based on the second target parameter according to the iterative small-loop processing image of the outer-loop processing image, and obtains the image update amount of the primal-dual TV regularized sub-loop processing image of the iterative small-loop processing image and the corresponding primal-dual TV regularized sub-loop processing image;

[0068] S3-4-3. The controller determines whether the image update amount of the primal-dual TV regularized sub-loop processing image is less than the second target threshold. If so, the iteration stops. Otherwise, the primal-dual TV regularized sub-loop processing image based on the iterative small loop processing image performs dual variable update, primal variable update and convergence check in sequence to obtain the output result of the TV sub-loop.

[0069] Specifically, computed tomography (CT) image reconstruction requires the use of mathematical models to synthesize a collection of 2D projection images into a 3D volume. Image reconstruction involves two phases: forward projection and backprojection. Forward projection maps the target onto the projection domain, allowing the calculation of model measurements corresponding to actual measurements that match the physical properties. Backprojection, on the other hand, is the inverse of forward projection. These two operations largely determine the accuracy of image reconstruction.

[0070] Conversion between mathematical formulas and computer languages: Mathematical formulas are usually based on continuous space (such as differentiation and integration), while computers process discrete data. Numerical methods are needed to achieve conversion, for example:

[0071] Gradient calculation is mathematically defined as partial derivatives, which are approximated by finite differences in computers. The gradient of a 3D image can be calculated by the difference between adjacent pixel values:

[0072]

[0073] Among them, ∇f is the gradient vector of the three-dimensional image, f i+1,j,k is the point with coordinates (i+1, j, k) in the three-dimensional image, f i,j,k is the point with coordinates (i, j, k) in the three-dimensional image, f i,j+1,k is the point with coordinates (i, j+1, k) in the three-dimensional image, f i,j,k+1 is the point with coordinates (i, j, k+1) in the three-dimensional image.

[0074] Divergence calculation (div g): Divergence is the inverse operation of gradient. When discretizing, the gradient field needs to be summed up by reverse differences. For example:

[0075]

[0076] in, is the gradient field value of the point with coordinates (i, j, k) along the x direction, is the gradient field value along the x direction at the point with coordinates (i-1, j, k), is the gradient field value along the y direction at the point with coordinates (i, j, k), is the gradient field value along the y direction at the point with coordinates (i, j-1, k), is the gradient field value of the point with coordinates (i, j, k) along the z direction, is the gradient field value along the z direction at the point with coordinates (i, j, k-1).

[0077] Projection and backprojection (A and Aᵀ): The system matrix A represents the path integral of the ray passing through the voxel. The actual calculation adopts ray-driven or voxel-driven methods, accumulating pixel values (forward projection) or distributing errors inversely (backprojection) by traversing the ray path.

[0078] Orthographic projection provides the basis for the mathematical model. This operator constructs an idealized scene in which all rays are assumed to have infinitesimal widths, to reach the center of the detector pixel exactly along a straight path, and to have uniform energy distribution. The most basic is the ray-voxel intersection method, where the path of a ray through an object can be expressed as:

[0079]

[0080] Where l(i,j,k) represents the length of the ray passing through the voxel, and ρ(i,j,k) represents the voxel density of the object;

[0081] Backprojection is an operator that updates an image using information from the projection data. A common backprojection algorithm is the voxel-driven algorithm, which builds a path from the ray source to the voxel center, then to the detector, and updates the voxel value based on the sampled value on the detector.

[0082] The iterative part is broken down into computer loops and conditional branches, which mainly include data segmentation, error calculation, back-projection update, etc.

[0083] The three-dimensional reconstruction algorithm can be mainly divided into analytical method and iterative method. This embodiment mainly introduces the Os-SART-PDTV algorithm based on the iterative method.

[0084] The derivation of analytical algorithms is mainly based on the assumption of continuous rays, but in practical applications, only limited projection data can usually be obtained, which has led to the emergence of iterative reconstruction methods. Algebraic image reconstruction (AIR) technology is a non-statistical image reconstruction technology. Unlike traditional analytical reconstruction algorithms, it can better simulate the geometry of the acquisition process, thereby producing better reconstruction effects in the case of incomplete projections (such as sparse views and limited angle data). AIR algorithms can be divided into three categories: algebraic reconstruction technology (ART), simultaneous algebraic reconstruction technology (SART), and simultaneous iterative reconstruction technology (SIRT). Mathematically, in order to reconstruct an M×N image from N pixels using M projections, the line integral of the attenuation coefficient is given by a set of linear equations, namely:

[0085]

[0086] Among them, I i0 and I i are the incident particles and the particles detected by the i-th detector, respectively, a i,j is the contribution of pixel j to the i-th detector, 0≤a i,j ≤1, , μ j is the attenuation coefficient of pixel j, y i is the measured sinusoidal curve.

[0087] The SART algorithm considers only one projection at a time, but adds a relaxation parameter 𝛽 to the update term to reduce overcorrection of noise and artifacts, which speeds up convergence. In SART, the attenuation coefficient is estimated after processing all rays in a projection, which greatly improves the convergence speed of the stable solution.

[0088] Os-SART algorithm: SART iterates all rays with the same projection angle by calculating, and Os-SART is a combination of ordered subset and SART algorithms. The formula is as follows:

[0089]

[0090] in, represents the value of the j-th voxel at the k+1-th iteration, represents the value of the jth voxel at the kth iteration, S t is the subset currently being processed, is the relaxation factor, p i is the actual measured projection value of the i-th ray, M is the current number of voxels, which means that all voxels are traversed during calculation, m is a constant, and w im is the contribution weight of the i-th ray to the m-th voxel, is the value of the i-th ray to the m-th voxel in the k-th iteration, w ij is the contribution weight of the i-th ray to the j-th voxel, i represents the i-th ray at a certain projection angle, j represents the pixel value in the i-th ray, and k represents the number of iterations.

[0091] Using ordered subsets (OS) to divide the projections into groups or subsets and update the estimate for each group instead of updating the entire dataset further improves convergence. The smaller the number of projections per subset, the faster the convergence. However, as the number of subsets increases, overcorrection can lead to higher noise and artifacts.

[0092] The Os-SART-PDTV algorithm proposes a new image reconstruction algorithm (Os-SART-PDTV) for sparse view NCT 3D based on the Os-SART iterative algorithm and the first-order primitive-dual algorithm with total variation denoising.

[0093] The general expression of the sparse neutron CT 3D reconstruction problem is as follows:

[0094]

[0095] Among them, f * is the image after 3D reconstruction, f represents the image to be reconstructed, A is the system matrix, which represents the conversion between the projection data and the image to be reconstructed, and b is the actual projection data or observation data; is the data fidelity term, which represents the square error of the norm between the projection of the reconstructed image and the true projection data, ensuring the consistency between the observed data and the reconstructed image; R(f) is the regularization term, which is used to introduce prior knowledge of the image, such as smoothness, sparsity, etc.; λ is the regularization parameter, which is the weight for balancing the relationship between the regularization term and data fidelity.

[0096] In the above framework, the regularization term is introduced into the total variation model, and its expression is as follows:

[0097]

[0098] Among them, TV is the total variation, and ε is a non-negative threshold parameter, which can be considered as the upper limit of the allowable error.

[0099] In practical applications, it is also necessary to introduce non-negative constraints on images. The objective function of the Os-SART-PDTV algorithm becomes the following form:

[0100]

[0101] Where arg min is the minimum value of the function.

[0102]

[0103] Where ∇f is the gradient vector of f.

[0104] This algorithm combines the advantages of two algorithms, and the two algorithms are partially complementary and can be better used for image reconstruction.

[0105] The OS-SART-PDTV algorithm reconstructs the most detailed structures and produces the clearest images. This result not only validates the application value of the OSSART-PDTV algorithm in sparse-angle neutron CT 3D reconstruction, but also emphasizes its superior performance in accurately reconstructing image details and improving image quality.

[0106] The algorithm mainly includes the Os-SART-PDTV cycle, such as Figure 3 The specific steps are as follows:

[0107] (1) Introducing initialization parameters: f0=0, λ Os-SART , iter Os-SART , N Os-SART ,λ PDTV With N PDTV , where f0 is the initial image, which should be a three-dimensional image, and is an all-zero matrix, λ Os-SART Iterative relaxation parameter, controlling the update step size, iter Os-SART is the number of iterations of the sub-loop, N Os-SART is the number of ordered subsets, the projections are grouped to speed up the convergence, λ PDTV The weight parameter of the primal-dual TV regularization, N PDTV is the number of iterations of the primal-dual TV subcycle;

[0108] (2) Perform the outer loop, when the image update amount Stop when the tolerance is less than tol1 or the maximum number of iterations maxIter1 is reached;

[0109] (3) In the outer loop, perform the Os-SART iterative small loop, the goal is to minimize (error between the projected data and the current estimate);

[0110] (4) Perform the primal-dual TV regularization subcycle:

[0111] Initialization: v=0, , , where v is the inner iteration counter, is the initial input image, equal to f Os-SART , f Os-SART is the output of the outer loop, g 0 is the initialized dual variable, is the gradient of the image;

[0112] Stop when the tolerance tol2 is less than or the maximum number of iterations is reached;

[0113] Dual variable update: ,γ is the dual step size parameter, and satisfies τ , , where g v is the update result of the dual variable in the vth step iteration, g v-1 is the update result of the dual variable at the v-1th iteration, is the intermediate update result of the original variable in the v-1 step iteration, τ is the original variable parameter, which is used to control the amplitude of the image update, ▽ is the gradient sign, and f v-1 is the update result of the original variable at step v-1, f v-2 is the updated result of the original variable at step v-2;

[0114] Original variable update: ,in, is the updated result of the v-th iteration of the original image, is the updated result of the original image at step v-1, divg v is the calculated divergence field;

[0115] Convergence check: evaluate the image change after each iteration;

[0116] (5) Output of TV sub-loop As the new starting point f for the outer iteration n , continue the outer loop until the result is reconstructed;

[0117] (6) After the reconstruction is completed, output f Os-SART-PDTV , where f Os-SART-PDTV The reconstructed image.

[0118] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology, characterized in that: Applied to a neutron computer imaging system, the neutron computer imaging system includes a ray generation subsystem and a neutron imaging subsystem, the method includes: The ray generation subsystem generates target rays; The neutron imaging subsystem acquires a two-dimensional original image based on the target ray; The neutron imaging subsystem uses a sparse view NCT three-dimensional reconstruction algorithm to perform three-dimensional image reconstruction on the two-dimensional original image to obtain a sparse view NCT three-dimensional image reconstruction result.

2. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 1 is characterized in that: The ray generating subsystem includes a collimator and a rotating sample stage; The ray generation subsystem generates target rays, including: The rotating sample stage is utilized, and the collimator generates collimated neutron rays to irradiate the sample to be detected at multiple angles to obtain target rays.

3. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 1 is characterized in that: The neutron imaging subsystem includes a two-dimensional neutron image detector, a conversion screen and a controller; The neutron imaging subsystem acquires a two-dimensional original image based on the target ray, including: The conversion screen converts the target ray into a light signal of the target ray; The two-dimensional neutron image detector acquires a two-dimensional original image based on the optical signal of the target ray; The controller receives the two-dimensional original image sent by the two-dimensional neutron image detector.

4. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 3 is characterized in that: The sparse view NCT 3D reconstruction algorithm adopts an Os-SART iterative algorithm and a first-order primitive-dual algorithm with total variation denoising.

5. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 4 is characterized in that: The objective function of the sparse view NCT 3D reconstruction algorithm is: , where f * is the image after 3D reconstruction, arg min is the minimum value of the function, f represents the image to be reconstructed, TV is the total variation, A is the system matrix, b is the real projection data or observation data, and ε is a non-negative threshold parameter.

6. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 5 is characterized in that: The neutron imaging subsystem reconstructs the two-dimensional original image into a three-dimensional image using a sparse view NCT three-dimensional reconstruction algorithm to obtain a sparse view NCT three-dimensional image reconstruction result, including: The controller initializes first target parameters, wherein the first target parameters include an initial image, an iterative relaxation parameter, a number of iterations of a sub-cycle, a number of ordered subsets, a weight parameter of a primal-dual TV regularization, and a number of iterations of a primal-dual TV sub-cycle; The controller performs outer loop processing on the two-dimensional original image based on the first target parameter to obtain an outer loop processed image of the two-dimensional original image and an image update amount corresponding to the outer loop processed image; The controller determines whether the image update amount of the outer loop processed image is less than a first target threshold value. If so, the iteration stops; otherwise, the Os-SART iterative small loop is performed on the outer loop processed image of the two-dimensional original image to obtain an iterative small loop processed image of the outer loop processed image; The controller performs primal-dual TV regularization sub-cycle processing on the iterative small-cycle processing image of the outer-loop processing image to obtain an output result of the TV sub-cycle; The controller uses the output result of the TV sub-loop as a new starting point for outer iteration, and returns to the controller to perform outer loop processing based on the first target parameter according to the two-dimensional original image until a sparse view NCT three-dimensional image reconstruction result is obtained.

7. The sparse view NCT three-dimensional image reconstruction method based on neutron imaging technology according to claim 6, characterized in that: The controller performs primal-dual TV regularization sub-cycle processing on the iterative small-cycle processing image of the outer-loop processing image to obtain an output result of the TV sub-cycle, including: The controller initializes a second target parameter; The controller performs primal-dual TV regularized sub-loop processing based on the second target parameter according to the iterative small-loop processing image of the outer-loop processing image, and obtains an image update amount of the primal-dual TV regularized sub-loop processing image of the iterative small-loop processing image and the corresponding primal-dual TV regularized sub-loop processing image; The controller determines whether the image update amount of the primal-dual TV regularized sub-loop processing image is less than the second target threshold. If so, the iteration stops; otherwise, the primal-dual TV regularized sub-loop processing image based on the iterative small loop processing image performs dual variable update, primal variable update and convergence check in sequence to obtain the output result of the TV sub-loop.

Images