Fast Inversion Method and System for Super-Resolution Three-Dimensional Images of Low- and Medium-Level Radioactive Solid Waste Barrels

Abstract

The present invention provides a method and system for fast super-resolution three-dimensional image inversion applicable to low- and medium-level radioactive solid waste barrels, including: Step S1: Simulate and generate waste barrels with non-uniform density distributions; Step S2: Perform γ scans on each waste barrel to obtain the counting rate, and combine an iterative algorithm to obtain the three-dimensional density image of the waste barrel; Step S3: Use the three-dimensional density image of the waste barrel to train a preset convolutional neural network CNN to obtain the trained preset convolutional neural network CNN; Step S4: Use the trained preset convolutional neural network CNN to obtain a super-resolution image. The purpose of the present invention is to use CNN to extract and identify iterative image features, so as to improve the measurement accuracy; achieve a higher image resolution with fewer measurement times, so as to reduce the measurement time, thereby realizing the practical application of the TGS technology.

Description

Technical Field

[0001] The present invention relates to the technical field of low and intermediate level radioactive solid waste measurement, and specifically, to a method and system for fast super-resolution three-dimensional image inversion applicable to low and intermediate level radioactive solid waste barrels. Background Art

[0002] For a 1 million-kilowatt nuclear power unit, about 13,000 cubic meters of low and intermediate level radioactive waste will be generated during 60 years of operation. After solidification or compression, it is prepared into 200L or 400L waste barrels, with a total volume of about 3,000 cubic meters. Currently, the situation of waste disposal is urgent. Before final disposal, it is necessary to measure the types, activities, and surface dose rates of radionuclides in these waste barrels. The surface dose rate is generally measured by a dose rate meter, and the types and activities of radionuclides are generally measured by non-destructive gamma scanning. Non-destructive gamma scanning is generally divided into sectional gamma scanning (SGS) and tomographic gamma scanning (TGS). SGS divides the waste barrel into many segments and assumes that the medium and radionuclides are uniformly distributed within each segment. This method has high accuracy for uniform waste such as liquid waste and resin, but for cement-fixed waste such as waste filters and waste equipment, the measurement error reaches 500%. Therefore, the TGS technology needs to be adopted. TGS combines the technologies of CT and PET-CT to perform gamma scanning on the waste barrel, thus improving the measurement accuracy. The result of TGS density reconstruction is used as an attenuation factor to correct the result of activity reconstruction. Therefore, the accuracy of density reconstruction directly determines the accuracy of activity reconstruction.

[0003] Currently, all density reconstruction methods in the world adopt the method of combining low-resolution grids with iterative algorithms. Because the measurement time is long, and at the same time, there are grid artifacts and noises in the reconstruction results, with large errors. Due to the limitation of measurement time, TGS cannot be actually applied in nuclear power.

[0004] Patent document CN112132920A (application number: 202010953443.8) discloses a method and system for density reconstruction of radioactive waste barrels based on deep learning. However, this method has the following several disadvantages: a) The measurement object is two-dimensional, that is, only the density distribution of each plane can be obtained. However, three-dimensional reconstruction of the waste barrel is required, and the above patent does not solve the problem of low vertical resolution, resulting in measurement errors; b) The most important limiting factor for the inapplicability of TGS is the measurement time. The above patent adopts the same measurement method as the traditional measurement method, with improved accuracy, but the measurement time is not shortened.

[0005] The objective of the present invention is to use CNN to extract and identify iterative image features, so as to improve the measurement accuracy; achieve a higher image resolution with fewer measurement times, reduce the measurement time, and thus realize the practical application of the TGS technology. This patent can promote the progress of low- and intermediate-level radioactive solid waste measurement technology, improve the waste measurement and management level in China, and ensure the independence and controllability of radioactive waste measurement technology in the nuclear power and national defense fields. Summary of the Invention

[0006] Aiming at the defects in the prior art, the objective of the present invention is to provide a super-resolution three-dimensional image rapid inversion method and system applicable to low- and intermediate-level radioactive solid waste barrels.

[0007] A super-resolution three-dimensional image rapid inversion method applicable to low- and intermediate-level radioactive solid waste barrels provided by the present invention includes:

[0008] Step S1: Simulate and generate waste barrels with uneven density distribution;

[0009] Step S2: Perform γ scanning on each waste barrel to obtain the counting rate, and combine it with the iterative algorithm to obtain the three-dimensional density image of the waste barrel;

[0010] Step S3: Use the three-dimensional density image of the waste barrel to train the preset convolutional neural network CNN to obtain the trained preset convolutional neural network CNN;

[0011] Step S4: Use the trained preset convolutional neural network CNN to obtain the super-resolution image.

[0012] Preferably, in step S2, the detector and the radiation source are respectively placed on both sides of the waste barrel, and γ scanning is sequentially performed on the waste barrel from different angles using grids with different resolutions.

[0013] Preferably, the preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolution layer, and an output layer, and symmetric skip connections are adopted;

[0014] The convolutional layer is used to extract features;

[0015] The deconvolution layer is used to restore the three-dimensional image size.

[0016] Preferably, the iterative algorithm includes the MLEM algorithm.

[0017] A super-resolution three-dimensional image rapid inversion system applicable to low- and intermediate-level radioactive solid waste barrels provided by the present invention includes:

[0018] Module M1: Simulate and generate waste barrels with uneven density distribution;

[0019] Module M2: Perform γ-scanning on each waste bin to obtain the counting rate, and combine with an iterative algorithm to obtain the three-dimensional density image of the waste bin;

[0020] Module M3: Train a preset convolutional neural network CNN using the three-dimensional density image of the waste bin to obtain the trained preset convolutional neural network CNN;

[0021] Module M4: Obtain a super-resolution image using the trained preset convolutional neural network CNN.

[0022] Preferably, Module M2 adopts: Place the detector and the radiation source on both sides of the waste bin respectively, and perform γ-scanning on the waste bin from different angles in sequence using grids with different resolutions.

[0023] Preferably, the preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolutional layer, and an output layer, and adopts symmetric skip connections;

[0024] The convolutional layer is used to extract features;

[0025] The deconvolutional layer is used to restore the three-dimensional image size.

[0026] Preferably, the iterative algorithm includes the MLEM algorithm.

[0027] According to a computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the above-mentioned method are implemented.

[0028] According to a super-resolution three-dimensional image fast inversion device for low and intermediate level radioactive waste bins provided by the present invention, it includes: a controller;

[0029] The controller includes the computer-readable storage medium storing the computer program described above, or the controller includes the super-resolution three-dimensional image fast inversion system for low and intermediate level radioactive waste bins described above.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. By optimizing the neural network structure and establishing the mapping function relationship between the iterative image and the real image, the measurement accuracy is improved; under the same measurement time: compared with the traditional maximum likelihood expectation method, the mean square error (MSE) of CNN is reduced by more than 52%, the peak signal-to-noise ratio (PSNR) is increased by more than 21%, and the structural similarity (SSIM) is increased by more than 12%;

[0032] 2. Use CNN to extract features and identify iterative images, establish a mapping between the coarse-grid iterative images and the fine-grid real images, so as to improve the resolution and reduce noise interference; at the same measurement time: the resolution is increased by 28 times, effectively eliminating the artifact noise problems caused by the iterative algorithm and the low-resolution grid;

[0033] 3. Based on the mapping from a low-resolution grid with fewer measurement times to a high-resolution grid, achieve the resolution of iterative images with more measurement times, thereby reducing the measurement time; when the measurement time is reduced by 89%: compared with the MLEM method, the MSE is reduced by more than 25%, the PSNR is increased by more than 8%, the SSIM is increased by more than 2%, and the greater the Gaussian noise, the more the performance is improved, that is, excellent robustness; BRIEF DESCRIPTION OF THE DRAWINGS

[0034] By reading the following detailed description of the non-limiting embodiments with reference to the accompanying drawings, other features, objects, and advantages of the present invention will become more apparent:

[0035] Figure 1 It is a flowchart for CNN training and testing.

[0036] Figure 2 It is a schematic diagram of the waste bin scanning process.

[0037] Figure 3 It is a schematic diagram of the preset convolutional neural network CNN structure.

[0038] Figure 4 It is a schematic diagram of the sensitivity analysis of key parameters.

[0039] Figure 5 It is a schematic diagram of performance analysis.

[0040] Figure 6 It is a schematic diagram of the comparison of test performance. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0042] Example 1

[0043] According to a super-resolution three-dimensional image fast inversion method applicable to low- and intermediate-level radioactive waste bins provided by the present invention, as Figure 1 shown, it includes:

[0044] Step S1: Generate a waste bin with non-uniform density distribution through simulation; for example, a waste bin with non-uniform density distribution can be generated by numerical calculation methods; specifically, refer to the numerical method for detection efficiency in nuclear waste bin detection published by "Qian Yalan, Wang Dezhong, Gu Weiguo, Xiong Jiemei; School of Mechanical and Power Engineering, Shanghai Jiao Tong University".

[0045] Step S2: Perform γ-scanning on each waste bin to obtain the counting rate, and combine with the iterative algorithm (MLEM algorithm) to obtain the three-dimensional density image of the waste bin;

[0046] Specifically, step S2 adopts: as Figure 2 shown, including: Figure 2 Left vertical direction scanning; Figure 2 Right horizontal direction scanning; Place the detector and the radiation source on both sides of the waste bin respectively, and perform γ-scanning on the waste bin from different angles in turn using grids with different resolutions. For example: 4*24 grids and 4*8 grids;

[0047] Step S3: Use the three-dimensional density image of the waste bin to train a preset convolutional neural network CNN to obtain the trained preset convolutional neural network CNN;

[0048] The preset convolutional neural network CNN extracts features through convolution and restores the image resolution through deconvolution. To solve the problem of gradient disappearance / explosion caused by too deep convolutional layers and the loss of original image information, symmetric skip connections are adopted.

[0049] Specifically, use the three-dimensional density image (iterative image) and the real image of each waste bin as a set of samples, and form a training set with a large number of samples to train the CNN, where the input is the iterative image and the output is the real image; when the loss function curve converges, the CNN network training is completed.

[0050] Specifically, the preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolutional layer, and an output layer. At the same time, to solve the problem of gradient disappearance / explosion caused by too deep convolutional layers and the loss of original image data, symmetric skip connections are adopted; as Figure 3 shown; the convolutional layer is used to extract features; the deconvolutional layer is used to restore the three-dimensional image size.

[0051] As Figure 4 shown, perform sensitivity analysis on the key parameters of the network including: convolutional layer (as Figure 4 a), number of channels (as Figure 4 b), optimization function (as Figure 4 c), and learning rate (as Figure 4 d) to obtain the optimal network parameters; the main parameters of the network structure are shown in Table 1:

[0052] Table 1

[0053]

[0054] Step S4: Obtain a super-resolution image by using the trained preset convolutional neural network CNN. That is, in practical applications, directly input the iteratively reconstructed three-dimensional density image of the waste bin into the trained CNN network to obtain a super-resolution density distribution image.

[0055] A super-resolution three-dimensional image fast inversion system applicable to low- and intermediate-level radioactive waste bins according to the present invention includes:

[0056] Module M1: Generate a waste bin with uneven density distribution through simulation; for example, a waste bin with uneven density distribution can be generated through numerical calculation; specifically, reference can be made to the numerical method for detection efficiency in nuclear waste bin detection published by "Qian Yalan, Wang Dezhong, Gu Weiguo, Xiong Jiemei; School of Mechanical and Power Engineering, Shanghai Jiao Tong University".

[0057] Module M2: Perform γ scanning on each waste bin to obtain a count rate, and combine it with an iterative algorithm (MLEM algorithm) to obtain a three-dimensional density image of the waste bin;

[0058] Specifically, the module M2 adopts: as Figure 2 shown, including: Figure 2 Left vertical direction scanning; Figure 2 Right horizontal direction scanning; Place the detector and the radiation source on both sides of the waste bin respectively, and perform γ scanning on the waste bin from different angles in sequence using grids with different resolutions. For example: 4*24 grids and 4*8 grids;

[0059] Module M3: Train a preset convolutional neural network CNN using the three-dimensional density image of the waste bin to obtain a trained preset convolutional neural network CNN;

[0060] The preset convolutional neural network CNN extracts features through convolution and restores the image resolution through deconvolution. To solve the problem of gradient disappearance / explosion caused by too deep convolutional layers and loss of original image information, symmetric skip connections are adopted.

[0061] Specifically, use the three-dimensional density image (iterative image) and the real image of each waste bin as a set of samples, and form a training set with a large number of samples to train the CNN, where the input is the iterative image and the output is the real image; when the loss function curve converges, the CNN network training is completed.

[0062] Specifically, the preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolutional layer, and an output layer. At the same time, to solve the problem of gradient disappearance / explosion caused by too deep convolutional layers and loss of original image data, symmetric skip connections are adopted; asFigure 3 As shown; the convolutional layer is used to extract features; the deconvolutional layer is used to restore the three-dimensional image size.

[0063] As Figure 4 shown, the network includes: a convolutional layer (such as Figure 4 a), the number of channels (such as Figure 4 b), the optimization function (such as Figure 4 c) and the learning rate (such as Figure 4 d) for sensitivity analysis of key parameters to obtain the optimal network parameters; the main parameters of the network structure are shown in Table 1:

[0064] Table 1

[0065]

[0066] Module M4: Obtain the super-resolution image using the trained preset convolutional neural network CNN. That is, in practical applications, the three-dimensional density image reconstructed by iterative waste bins is directly input into the trained CNN network to obtain the super-resolution density distribution image.

[0067] Performance analysis:

[0068] The flowchart of the performance analysis is as Figure 5 shown. First, the tested bins are divided into two categories, namely no noise interference and noise interference (10% increase in the counting rate error). Two different γ-scanning methods are used to obtain two MLEM iterative images (the scanning methods are 15×4×24 and 5×4×8) respectively.

[0069] The reconstruction results of different methods are as Figure 6 shown. Among them, Figure 6 (a) is the top view of the true distribution of the bin and Figure 6 (b) is the side view of the true distribution of the bin. By comparing Figure 6 (a1) the MLEM32 result (top view) and Figure 6 (a2) the CNN32 result (top view) and Figure 6 (a3) the MLEM96 result (top view) and Figure 6 (a4) the CNN96 result (top view), it can be found that whether it is CNN with fewer measurement times (32×5) or CNN with more measurement times (96×15), compared with MLEM, the artifact noise is effectively eliminated. By comparing Figure 6 (b1) the MLEM32 result (side view) and Figure 6 (b2) the CNN32 result (side view) and by comparing Figure 6 (b3) the MLEM96 result (side view) and Figure 6(b4) The results of CNN96 (side view) also have the same effect. This is the first time to achieve high-resolution reconstruction in the vertical direction of the waste bin. Since the grid resolution is increased from 96×15 to 45×30×30, the spatial resolution is increased by 28 times.

[0070] Table 2 shows the numerical comparison of the reconstruction results. It can be found that, compared with MLEM, the reconstruction results of CNN have a 52% reduction in MLEM, a 21% increase in PSNR, and a 12% increase in SSIM. Among them, when comparing MLEM-A and CNN-B, it can be found that the performance of CNN is still better than that of MLEM results while the measurement time is reduced by 89%.

[0071] Therefore, the present invention can shorten the measurement time and improve the measurement accuracy.

[0072] Table 2 Comparison of Reconstruction Results

[0073]

[0074] Those skilled in the art know that in addition to implementing the systems, devices, and their respective modules provided by the present invention in the form of pure computer-readable program codes, the method steps can be logically programmed to enable the systems, devices, and their respective modules provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same program. Therefore, the systems, devices, and their respective modules provided by the present invention can be regarded as a kind of hardware component, and the modules included therein for implementing various programs can also be regarded as the structures within the hardware component; the modules for implementing various functions can also be regarded as either software programs for implementing the methods or the structures within the hardware component.

[0075] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A fast super-resolution three-dimensional image inversion method applicable to low and intermediate level radioactive solid waste barrels, characterized in that, Including: Step S1: Simulate and generate waste bins with non-uniform density distribution; Step S2: Perform γ scanning on each waste bin to obtain the count rate, and combine with an iterative algorithm to obtain the three-dimensional density image of the waste bin; Step S3: Use the three-dimensional density image of the waste bin to train a preset convolutional neural network CNN to obtain the trained preset convolutional neural network CNN; Step S4: Use the trained preset convolutional neural network CNN to obtain a super-resolution image; In the said Step S2, it is adopted that: the detector and the radiation source are respectively placed on both sides of the waste bin, and γ scanning is performed on the waste bin from different angles in turn using grids with different resolutions; The said preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolutional layer and an output layer, and symmetric skip connections are adopted; The convolutional layer is used to extract features; The deconvolutional layer is used to restore the three-dimensional image size; The iterative algorithm includes the MLEM algorithm.

2. A fast super-resolution three-dimensional image inversion system applicable to low and intermediate-level radioactive waste barrels, characterized in that Including: Module M1: Simulate and generate waste bins with non-uniform density distribution; Module M2: Perform γ scanning on each waste bin to obtain the count rate, and combine with an iterative algorithm to obtain the three-dimensional density image of the waste bin; Module M3: Use the three-dimensional density image of the waste bin to train a preset convolutional neural network CNN to obtain the trained preset convolutional neural network CNN; Module M4: Use the trained preset convolutional neural network CNN to obtain a super-resolution image; In the said Module M2, it is adopted that: the detector and the radiation source are respectively placed on both sides of the waste bin, and γ scanning is performed on the waste bin from different angles in turn using grids with different resolutions; The said preset convolutional neural network CNN includes: an input layer, a convolutional layer, a deconvolutional layer and an output layer, and symmetric skip connections are adopted; The convolutional layer is used to extract features; The deconvolutional layer is used to restore the three-dimensional image size; The iterative algorithm includes the MLEM algorithm.

3. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in claim 1.

4. A rapid super-resolution three-dimensional image inversion device applicable to low- and medium-level radioactive waste barrels, characterized in that, Including: A controller; The controller includes the computer-readable storage medium storing the computer program described in claim 3, or the controller includes the super-resolution three-dimensional image fast inversion system for low- and medium-level radioactive solid waste bins described in claim 2.

Images