Neutron ray image super-resolution reconstruction method and device based on Real-ESRGAN and medium

Abstract

The invention discloses a neutron ray image super-resolution reconstruction method and device based on Real-ESRGAN and a medium, and relates to the technical field of image super-resolution. The method comprises the following steps: constructing a second-order degradation model, wherein the second-order degradation model is used for generating a low-resolution neutron ray image with gamma white spot noise; constructing an image pair training set based on the second-order degradation model and a sample data set; training a training set based on the image to obtain a Real-ESRGAN model suitable for the neutron ray image; and super-resolution reconstruction of the neutron ray image is realized based on the Real-ESRGAN model. According to the neutron ray image super-resolution reconstruction realized based on the method, the resolution of the neutron ray image can be effectively improved, and meanwhile, the degradation influence in the neutron ray imaging process is inhibited.

Description

Technical Field

[0001] The present application relates to the field of image super-resolution technology, and in particular to a neutron ray image super-resolution reconstruction method, device, and medium based on Real-ESRGAN. Background Art

[0002] Neutron radiography technology is widely used in the aerospace, materials, energy, and nuclear industries due to its four major advantages: strong penetration, sensitive detection of light elements, accurate composition identification, and strong anti-interference ability. Due to the difference in the reaction characteristics between X-rays and matter, neutron radiography technology is able to detect elements or structures that are difficult to detect with certain X-ray imaging technologies. After years of development, X-ray imaging technology can now produce high-resolution images. However, due to the limitations of the neutron source power, the structural characteristics of the collimator and detector, image degradation factors such as noise, geometric blur, and gamma white spot noise will result in low-resolution and visually unsatisfactory neutron radiography images, which will seriously hinder subsequent machine vision tasks such as defect detection or identification.

[0003] Currently, research on super-resolution reconstruction of neutron radiography images is still in its infancy. Traditional image super-resolution methods, including interpolation, sparse representation, and dictionary learning, can improve image restoration but are difficult to apply to complex scenes. With the development of deep learning methods, SRGAN, ESRGAN, and Real-ESRGAN have emerged, significantly improving image visual quality. In particular, the Real-ESRGAN super-resolution model can simulate complex degradation processes in the real world, improving image resolution in real scenes. However, the second-order degradation model in Real-ESRGAN cannot account for all image degradation factors in neutron radiography, particularly gamma-ray white noise.

[0004] Therefore, it is necessary to propose a neutron ray image super-resolution reconstruction method based on Real-ESRGAN. Summary of the Invention

[0005] In order to overcome the above-mentioned defects, the present application proposes a neutron radiography super-resolution reconstruction method based on Real-ESRGAN, which solves the technical problem that the second-order degradation model in Real-ESRGAN cannot cover all image degradation factors in neutron radiography technology, resulting in poor super-resolution reconstruction effect of neutron radiography images.

[0006] In a first aspect, the present application provides a neutron radiographic image super-resolution reconstruction method based on Real-ESRGAN, comprising:

[0007] Constructing a second-order degradation model, wherein the second-order degradation model is used to generate a low-resolution neutron radiographic image with gamma-ray white spot noise;

[0008] An image pair training set is constructed based on the second-order degradation model and a sample data set; the sample data set is a collection of high-resolution neutron radiation images, and the image pair training set includes multiple pairs of high-resolution neutron radiation images and low-resolution neutron radiation images;

[0009] Training the training set based on the images to obtain a Real-ESRGAN model suitable for neutron radiographic images;

[0010] Super-resolution reconstruction of neutron radiation images is achieved based on the Real-ESRGAN model.

[0011] In some embodiments, the second-order degradation model includes a first-order degradation processing module and a second-order degradation processing module;

[0012] The first-order degradation processing module is configured to blur, downsample and add simulated noise to the input high-resolution neutron ray image to generate a first-order degraded output image;

[0013] The second-order degradation processing module is configured to blur, downsample, add simulated noise, and add gamma white spot noise to the first-order degraded output image to generate a low-resolution neutron ray image corresponding to the high-resolution neutron ray image.

[0014] In some embodiments, the first-order degradation processing module includes a first blur processing unit, a first downsampling unit, and a first noise unit;

[0015] The first fuzzy processing unit is configured to perform fuzzy processing on the input high-resolution neutron ray image to obtain a first output image;

[0016] The first downsampling unit is configured to perform a downsampling operation on the first output image to obtain a second output image;

[0017] The first noise unit is configured to generate simulated noise and add the simulated noise to the second output image to obtain a first noise image.

[0018] In some embodiments, the second-order degradation processing module includes a second blur processing unit, a second downsampling unit, a second noise unit, and a third noise unit;

[0019] The second blur processing unit is configured to perform blur processing on the first noise image to obtain a third output image;

[0020] The second downsampling unit is configured to perform a downsampling operation on the third output image to obtain a fourth output image;

[0021] The second noise unit is configured to generate simulated noise and add the simulated noise to the fourth output image to obtain a second noise image;

[0022] The third noise unit is configured to extract gamma white spot noise according to the real neutron ray image and add the gamma white spot noise to the second noise image to obtain the low-resolution neutron ray image.

[0023] In some embodiments, generating simulated noise specifically includes: using Poisson noise and Gaussian noise models to simulate neutron microscopic quantum noise, photoelectron noise and electronic noise in the neutron imaging process as the simulated noise.

[0024] In some embodiments, extracting gamma white spot noise from the real neutron ray image and adding it to the second noise image to obtain the low-resolution neutron ray image specifically includes:

[0025] Acquire a real neutron ray image, perform thresholding and dethresholding operations on it to obtain a first image and a second image;

[0026] Generate a random matrix and multiply the matrix with the first image to obtain the gamma white spot noise;

[0027] The low-resolution neutron ray image is obtained by adding a result of multiplying the second image and the second noise image to the gamma white spot noise.

[0028] In some embodiments, acquiring a real neutron ray image specifically includes: performing data enhancement processing on the real neutron ray image to obtain an expanded data set, and randomly selecting a neutron ray image from the expanded data set.

[0029] In some embodiments, the network structure of the Real-ESRGAN model includes a generator and a discriminator, wherein the generator adopts the same generator as the ESRGAN model, and the discriminator adopts a U-Net discriminator with spectral normalization.

[0030] In a second aspect, the present application provides an electronic device comprising at least one processor and at least one memory, wherein the memory is suitable for storing multiple program codes, and the program codes are suitable for being loaded and run by the processor to execute the neutron radiation image super-resolution reconstruction method based on Real-ESRGAN as described in any one of the technical solutions of the above-mentioned neutron radiation image super-resolution reconstruction method based on Real-ESRGAN.

[0031] In a third aspect, the present application provides a computer-readable storage medium storing a plurality of program codes, wherein the program codes are suitable for being loaded and run by a processor to execute the neutron ray image super-resolution reconstruction method based on Real-ESRGAN as described in any one of the technical solutions of the above-mentioned neutron ray image super-resolution reconstruction method based on Real-ESRGAN.

[0032] The above one or more technical solutions of this application have at least one or more of the following beneficial effects:

[0033] In implementing the technical solution of this application, the second-order degradation model in Real-ESRGAN is improved by adding gamma white spot noise to the model, making it more consistent with the degradation process of neutron radiography, thereby improving the visual quality and effect of neutron radiography. High-resolution and low-resolution neutron radiography images obtained based on the improved second-order degradation model are then used to train a super-resolution model. The trained Real-ESRGAN model is then used for super-resolution reconstruction of neutron radiography images, effectively improving the resolution of neutron radiography images while simultaneously suppressing the degradation effects during the neutron radiography imaging process. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The disclosure of this application will be more easily understood with reference to the accompanying drawings. Those skilled in the art will readily appreciate that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this application. Furthermore, similar numbers in the figures represent similar components, where:

[0035] Figure 1 1 is a flow chart of the main steps of a neutron radiographic image super-resolution reconstruction method based on Real-ESRGAN provided in an embodiment of the present application;

[0036] Figure 2 This is a schematic block diagram of a second-order degradation model provided in an embodiment of the present application;

[0037] Figure 3 is based on Figure 2 Schematic diagram of the image degradation processing process of the second-order degradation model shown. DETAILED DESCRIPTION

[0038] Some embodiments of the present application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and are not intended to limit the scope of protection of the present application.

[0039] In the description of this application, "module" and "processor" may include hardware, software, or a combination of both. A module may include hardware circuitry, various suitable sensors, communication ports, and memory. It may also include software components, such as program code, or a combination of software and hardware. A processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. A processor has data and / or signal processing capabilities. A processor may be implemented in software, hardware, or a combination of both. Non-transitory computer-readable storage media include any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" refers to all possible combinations of A and B, such as only A, only B, or both A and B. The terms "at least one of A or B" or "at least one of A and B" have similar meanings to "A and / or B" and may include only A, only B, or both A and B. The singular forms "a" and "the" may also include the plural forms.

[0040] Real-ESRGAN: It is an image super-resolution model based on deep learning. It is an extension of ESRGAN (Enhanced Super-Resolution Generative Adversarial Network). Its core idea is to restore a high-definition version of a blurred image by simulating the process of converting a high-resolution image into a low-resolution one.

[0041] This application proposes a super-resolution reconstruction method for neutron radiography based on Real-ESRGAN. The second-order degradation model in Real-ESRGAN is improved by adding gamma white spot noise to the model to make it more consistent with the degradation process of neutron radiography. This method can effectively improve the resolution of neutron radiography and suppress the degradation effect during the neutron radiography imaging process.

[0042] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart of the main steps of a neutron ray image super-resolution reconstruction method based on Real-ESRGAN according to an embodiment of the present application. Figure 1 As shown, the neutron ray image super-resolution reconstruction method based on Real-ESRGAN in the embodiment of the present application mainly includes the following steps S11 to S14.

[0043] Step S11: constructing a second-order degradation model, wherein the second-order degradation model is used to generate a low-resolution neutron radiographic image with gamma white spot noise;

[0044] The second-order degradation model in this embodiment can degrade the input high-resolution (HR) neutron image and output a corresponding low-resolution (LR) neutron image with gamma white spot noise. This second-order degradation model encompasses the main degradation types of neutron images, making the generated LR neutron image closer to a true neutron image. The second-order degradation model in this embodiment can be understood as a second-order degradation model obtained by improving the second-order degradation model of Real-ESRGAN. The improvement mainly retains the blurring, downsampling, and noise degradation processes implemented by the second-order degradation model of Real-ESRGAN, and newly adds gamma white spot noise. The improved second-order degradation model can achieve a degradation process that is more consistent with neutron image degradation. Specifically, during neutron radiography, the collimation ratio and sample-detector distance can cause geometric unsharpness, while photon divergence at the neutron-photon conversion screen contributes to inherent unsharpness in the imaging system. In these blur degradations, particles diverge isotropically, but neutron scattering tends to be biased toward high-scattering areas of the sample, resulting in anisotropic neutron divergence. Therefore, the Gaussian filter in the second-order degradation model is retained to simulate both isotropic and anisotropic blurring. The decay of radioactive elements in a neutron radiation source produces a particle stream with inherent statistical fluctuations. Under specific conditions, this random fluctuation can be mathematically modeled using either a Gaussian or Poisson distribution. Therefore, Gaussian and Poisson noise are retained in the second-order degradation model to simulate the inherent noise of the electronic system, additive interference in the signal processing process, the randomness of neutron-matter interactions, and the statistical fluctuations of detector counts, respectively. The spatial resolution of neutron images is limited by the detector pixel size (e.g., the pixel size of the scintillator or CCD), so the downsampling operation in the second-order degradation model is retained. Gamma photons generated during the imaging process are incident on the CCD camera and concentrated, forming white spots of varying sizes, shapes, and intensities, known as gamma white spot noise. Among the various degradation factors present in neutron radiation images, gamma white spot noise is the most obvious, seriously affecting the quality of neutron radiation images. Therefore, the improved second-order degradation model of this embodiment introduces gamma white spot noise.

[0045] Step S12: constructing an image pair training set based on the second-order degradation model and the sample data set;

[0046] In this embodiment, the sample data set is a set of high-resolution (HR) neutron radiation images, and the image pair training set includes multiple pairs of high-resolution neutron radiation images and low-resolution neutron radiation images.

[0047] Specifically, in this step, images generated by a neutron radiography device driven by a large neutron source can be used as HR neutron radiation images to form the sample data set, and each HR neutron radiation image in the sample data set is input into the second-order degradation model to output a low-resolution (LR) neutron radiation image paired with the HR neutron radiation image, thereby constructing the image pair training set based on the HR neutron radiation image and the corresponding LR neutron radiation image.

[0048] Step S13: training the training set based on the image to obtain a Real-ESRGAN model suitable for neutron X-ray images;

[0049] The network structure of the Real-ESRGAN model in this embodiment includes a generator and a discriminator. The generator adopts the same generator (SR network) as the ESRGAN model, that is, a deep network containing multiple residual-in-residual dense blocks (RRDB). The discriminator adopts a U-Net discriminator with spectral normalization (SN), mainly to improve the discrimination ability of the discriminator and stabilize the training process. It should be understood that the network structure of the ESRGAN model is a prior art and will not be introduced in detail here. The model training process of the Real-ESRGAN model in this embodiment is as follows: first, a model with peak signal-to-noise ratio (PSNR) as the target is trained, and the L1 loss (i.e., absolute error loss) function is used for optimization; then, this trained PSNR-guided model is used as the initialization of the generator, and the Real-ESRGAN model is trained in combination with L1 loss, perceptual loss (Perceptual Loss) and generative adversarial network (GAN) loss. It should be understood that the model training process is a process of iteratively training the Real-ESRGAN model according to the above training process until the model converges and obtains model parameters. The Real-ESRGAN model configured according to the obtained model parameters is the obtained Real-ESRGAN model suitable for neutron radiographic images.

[0050] Step S14: super-resolution reconstruction of neutron ray images is achieved based on the Real-ESRGAN model.

[0051] In a specific implementation of step S11 above, the second-order degradation model improves the traditional degradation process, such as Figure 2The figure shows a schematic diagram of the composition architecture of the second-order degradation model proposed in an embodiment of the present application, which mainly includes a first-order degradation processing module 100 and a second-order degradation processing module 200, wherein the first-order degradation processing module 100 can be configured as a traditional degradation process, wherein the second-order degradation processing module 200 implements an improved degradation process.

[0052] Specifically in this embodiment, the first-order degradation processing module 100 is configured to blur, downsample and add simulated noise to the input high-resolution neutron ray image to generate a first-order degraded output image; the second-order degradation processing module 200 is configured to blur, downsample, add simulated noise and add gamma white spot noise to the first-order degraded output image to generate a low-resolution neutron ray image corresponding to the high-resolution neutron ray image.

[0053] In a specific implementation, the first-order degradation processing module 100 includes a first fuzzy processing unit 101 , a first downsampling unit 102 , and a first noise unit 103 ;

[0054] The first fuzzy processing unit 101 is configured to perform fuzzy processing on the input high-resolution neutron ray image to obtain a first output image;

[0055] The first downsampling unit 102 is configured to perform a downsampling operation on the first output image to obtain a second output image;

[0056] The first noise unit 103 is configured to generate simulated noise and add it to the second output image to obtain a first noise image.

[0057] The second-order degradation processing module 200 includes a second fuzzy processing unit 201, a second downsampling unit 202, a second noise unit 203 and a third noise unit 204;

[0058] The second blur processing unit 201 is configured to perform blur processing on the first noise image to obtain a third output image;

[0059] The second downsampling unit 202 is configured to perform a downsampling operation on the third output image to obtain a fourth output image;

[0060] The second noise unit 203 is configured to generate simulated noise and add it to the fourth output image to obtain a second noise image;

[0061] The third noise unit 204 is configured to extract gamma white spot noise according to the real neutron ray image and add it to the second noise image to obtain the low-resolution neutron ray image.

[0062] The first blur processing unit 101 and the second blur processing unit 201 in this embodiment can be implemented using a filter based on Gaussian distribution or a filter based on Cauchy distribution. For example, the first blur processing module and the second blur processing module are both configured to use Gaussian filters to simulate isotropic and anisotropic blur degradation. For example, the configuration parameters of the first blur processing module are as follows: the probability of isotropy and anisotropy is [0.8, 0.2], and the variance range is [0.1, 5]; the configuration parameters of the second blur processing module are as follows: the probability of isotropy and anisotropy is [0.8, 0.2], and the variance range is [0.1, 3]. The above configuration can make the blur processing process closer to the blur degradation of real neutron radiation images. It can be understood that the Gaussian blur kernel generated by the Gaussian filter under the above configuration parameters is convolved with the image to be processed to achieve blur processing of the image.

[0063] In this embodiment, the first down-sampling unit 102 and the second down-sampling unit 202 may adopt down-sampling methods such as bicubic interpolation, bilinear interpolation, and regional interpolation.

[0064] In this embodiment, the simulated noise generated by the first noise unit 103 and the second noise unit 203 can specifically utilize Poisson noise and Gaussian noise models to simulate the neutron microscopic quantum noise, photoelectron noise, and electronic noise generated during the neutron imaging process. For example, the probability of occurrence of Gaussian noise and Poisson noise is set to [0.5, 0.5], the Gaussian variance range is [10, 30], and the Poisson noise scaling factor range is [0.05, 3]. Based on the above settings, the generated Gaussian noise and Poisson noise can simulate the neutron microscopic quantum noise, photoelectron noise, and electronic noise during the neutron imaging process.

[0065] The second noise module 104 is configured to extract gamma white spot noise from a real neutron ray image and add it to the noise image to obtain a low-resolution neutron ray image.

[0066] In this embodiment, the third noise unit 204 extracts gamma white spot noise from a real neutron ray image and adds it to the noise image. A specific implementation step includes the following steps a1-a3:

[0067] a1: Obtain a real neutron ray image, perform thresholding and dethresholding operations on it to obtain the first image and the second image;

[0068] a2: Generate a random matrix and multiply it with the first image to obtain gamma white spot noise;

[0069] a3: Adding a result of multiplying the second image by the noise image to the gamma white spot noise to obtain the low-resolution neutron ray image.

[0070] Acquiring a real neutron radiation image in step a1 can further specifically include performing data augmentation processing on a real neutron radiation image to obtain an expanded dataset, and randomly selecting an image from the expanded dataset as the acquired real seed radiation image. The data augmentation processing includes at least one of flipping, rotating, and cyclic translation.

[0071] The following combination Figure 3 The degradation process of generating corresponding LR neutron radiation images by degrading HR neutron radiation images using the second-order degradation model is described in detail.

[0072] like Figure 3 As shown in (a), the second-order degradation process includes the first-order process and the second-order process. The blur, downsampling and noise in the first-order process and the second-order process can be processed in the same way, such as using isotropic and anisotropic Gaussian filters to achieve blur, using bicubic interpolation, bilinear interpolation and regional interpolation methods to achieve downsampling, and using Poisson noise and Gaussian noise to achieve noise. As shown in the figure, the first-order process includes the HR image I HR Perform fuzzy processing to obtain I blur1 , to I blur1 Perform downsampling operation to obtain I resize1 , in I resize1 Add Poisson noise and Gaussian noise to get I noise1 ; The second-order process includes I noise1 Perform fuzzy processing to obtain I blur2 , to I blur2 Perform downsampling operation to obtain I resize2 , in I resize2 Add Poisson noise and Gaussian noise to get I noise2 , in I noise2 Gamma white spot noise is added to generate LR neutron ray image I LR , that is, by adding gamma white spot noise, the reaction between gamma rays and CCD camera during neutron imaging is simulated. noise2 For details, see (b) gamma white spot noise. First, a batch of real neutron ray images are collected and augmented by data enhancement. Then, a real neutron ray image is randomly selected from the augmented data. white Perform thresholding and dethresholding operations respectively to obtain I mask and I inverse, generate a random matrix Random(w,1.0) in the range [w,1.0] and sum it with I mask Multiply, I inverse and I noise2 Multiply; add the two multiplication results to generate the LR neutron ray image I LR . It should be understood that the shape and size of the random matrix are consistent with the neutron ray image to be processed. Taking the neutron ray image as an 8-bit image as an example, the pixel intensity of the generated gamma white spot noise is [w, 1.0] * 255. In this embodiment, the value range of w is preferably [0.7, 0.8]. Exemplarily, the data enhancement in this embodiment can be performed by flipping, rotation, and cyclic translation. For example, flipping can be horizontal flipping, vertical flipping, etc., rotation can be 90° clockwise rotation, 180° clockwise rotation, etc., and cyclic translation can be 13 units to the right and 43 units upward, or 20 units to the right and 9 units upward, etc. It should be understood that the above-mentioned image enhancement process can be one or more operations of flipping, rotation, and cyclic translation multiple times in succession.

[0073] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effect of the present application, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present application.

[0074] It will be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment of the present application can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code.

[0075] Furthermore, the present application also provides an electronic device. In an electronic device embodiment according to the present application, the electronic device includes at least one processor and at least one memory. The memory can be configured to store a program for executing the Real-ESRGAN-based neutron radiation image super-resolution reconstruction method of the above-mentioned method embodiment. The processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the Real-ESRGAN-based neutron radiation image super-resolution reconstruction method of the above-mentioned method embodiment. For ease of explanation, only the parts related to the embodiment of the present application are shown. For specific technical details not disclosed, please refer to the method section of the embodiment of the present application.

[0076] In the embodiment of the present application, the electronic device may be a control device device formed by various devices. In some possible implementations, the electronic device may include multiple memories and multiple processors. The program for executing the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above method embodiment can be divided into multiple subroutines, and each subroutine can be loaded and run by the processor to execute different steps of the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above method embodiment. Specifically, each subroutine can be stored in different memories, and each processor can be configured to execute the programs in one or more memories to jointly implement the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above method embodiment, that is, each processor executes different steps of the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above method embodiment to jointly implement the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above method embodiment.

[0077] The aforementioned multiple processors may be processors deployed on the same device. For example, the aforementioned electronic device may be a high-performance device composed of multiple processors, and the aforementioned multiple processors may be processors configured on the high-performance device. Furthermore, the aforementioned multiple processors may also be processors deployed on different devices. For example, the aforementioned electronic device may be a server cluster, and the aforementioned multiple processors may be processors on different servers in the server cluster.

[0078] Furthermore, the present application also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present application, the computer-readable storage medium can be configured to store a program for executing the Real-ESRGAN-based neutron ray image super-resolution reconstruction method of the above-mentioned method embodiment, and the program can be loaded and run by the processor to implement the above-mentioned Real-ESRGAN-based neutron ray image super-resolution reconstruction method. For ease of explanation, only the parts related to the embodiment of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present application. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present application is a non-temporary computer-readable storage medium.

[0079] Thus far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of the present application is obviously not limited to these specific embodiments. Without departing from the principles of the present application, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present application.

Claims

1. A neutron radiographic image super-resolution reconstruction method based on Real-ESRGAN, characterized in that: include: Constructing a second-order degradation model, wherein the second-order degradation model is used to generate a low-resolution neutron radiographic image with gamma-ray white spot noise; An image pair training set is constructed based on the second-order degradation model and a sample data set; the sample data set is a collection of high-resolution neutron radiation images, and the image pair training set includes multiple pairs of high-resolution neutron radiation images and low-resolution neutron radiation images; Training the training set based on the images to obtain a Real-ESRGAN model suitable for neutron radiographic images; Super-resolution reconstruction of neutron radiation images is achieved based on the Real-ESRGAN model.

2. The neutron radiographic image super-resolution reconstruction method based on Real-ESRGAN according to claim 1, characterized in that: The second-order degradation model includes a first-order degradation processing module and a second-order degradation processing module; The first-order degradation processing module is configured to blur, downsample and add simulated noise to the input high-resolution neutron ray image to generate a first-order degraded output image; The second-order degradation processing module is configured to blur, downsample, add simulated noise, and add gamma white spot noise to the first-order degraded output image to generate a low-resolution neutron ray image corresponding to the high-resolution neutron ray image.

3. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 2, characterized in that: The first-order degradation processing module includes a first fuzzy processing unit, a first downsampling unit and a first noise unit; The first fuzzy processing unit is configured to perform fuzzy processing on the input high-resolution neutron ray image to obtain a first output image; The first downsampling unit is configured to perform a downsampling operation on the first output image to obtain a second output image; The first noise unit is configured to generate simulated noise and add the simulated noise to the second output image to obtain a first noise image.

4. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 3, characterized in that: The second-order degradation processing module includes a second fuzzy processing unit, a second downsampling unit, a second noise unit and a third noise unit; The second blur processing unit is configured to perform blur processing on the first noise image to obtain a third output image; The second downsampling unit is configured to perform a downsampling operation on the third output image to obtain a fourth output image; The second noise unit is configured to generate simulated noise and add the simulated noise to the fourth output image to obtain a second noise image; The third noise unit is configured to extract gamma white spot noise according to the real neutron ray image and add the gamma white spot noise to the second noise image to obtain the low-resolution neutron ray image.

5. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 3 or 4, characterized in that: The generating of simulated noise specifically includes: using Poisson noise and Gaussian noise models to simulate and generate neutron microscopic quantum noise, photoelectron noise and electronic noise in the neutron imaging process as the simulated noise.

6. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 4, characterized in that: The step of extracting gamma white spot noise from the real neutron ray image and adding the noise to the second noise image to obtain the low-resolution neutron ray image specifically includes: Acquire a real neutron ray image, perform thresholding and dethresholding operations on it to obtain a first image and a second image; Generate a random matrix and multiply the matrix with the first image to obtain the gamma white spot noise; The low-resolution neutron ray image is obtained by adding a result of multiplying the second image and the second noise image to the gamma white spot noise.

7. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 6, characterized in that: The acquiring of a real neutron ray image specifically includes: performing data enhancement processing on the real neutron ray image to obtain an expanded data set, and randomly selecting a neutron ray image from the expanded data set.

8. The method for super-resolution reconstruction of neutron radiographic images based on Real-ESRGAN according to claim 1, characterized in that: The network structure of the Real-ESRGAN model includes a generator and a discriminator. The generator adopts the same generator as the ESRGAN model, and the discriminator adopts a U-Net discriminator with spectral normalization.

9. An electronic device comprising at least one processor and at least one memory, wherein the memory is adapted to store a plurality of program codes, wherein: The program code is suitable for being loaded and run by the processor to execute the Real-ESRGAN-based neutron radiation image super-resolution reconstruction method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the Real-ESRGAN-based neutron radiation image super-resolution reconstruction method according to any one of claims 1 to 8.

Images