Fast neutron photography system based on subpixel displacement super-resolution reconstruction

Abstract

This invention discloses a fast neutron radiography system based on subpixel displacement super-resolution reconstruction, aiming to obtain higher resolution and better quality imaging results to meet the requirements of higher precision imaging. The system includes a fast neutron radiography device and a multi-image super-resolution reconstruction algorithm module based on variance decision. The fast neutron radiography device comprises a neutron source, a collimator, and a detection system. The detection system consists of a precision positioning platform, a conversion screen fixed in a dark box, and an optical system. This device obtains multiple frames of neutron images with subpixel displacement by controlling the entire detection system to translate within a plane perpendicular to the beam. The multi-image super-resolution reconstruction algorithm module based on variance decision obtains the super-resolution reconstructed image through steps such as preliminary reconstruction, variance calculation to determine edges, and edge point pixel updates.

Description

Technical fields:

[0001] This invention relates to the fields of image processing and neutron imaging, specifically to a fast neutron radiography system based on subpixel displacement super-resolution reconstruction. Background technology:

[0002] Neutron radiography is an imaging technique based on the penetrating power of neutrons. It utilizes a neutron beam passing through the object under test and generating signals on a detector. These signals are then converted into images to reveal the object's internal structure. Compared to other non-destructive testing techniques, neutron radiography has unique advantages because the interaction mechanism of neutrons in matter differs from other radiation, making it particularly advantageous for the detection of lightweight materials. However, image resolution in neutron radiography is often low due to neutron beam scattering and absorption. While image quality can be further improved through hardware enhancements such as improving the quality and intensity of the radiographic neutron beam and refining the performance of the conversion screen and camera, these methods often require large-scale equipment upgrades, resulting in high costs. Therefore, utilizing existing equipment and image processing techniques to obtain higher-quality imaging results has become a viable solution. Super-resolution reconstruction is a method that improves image resolution by processing image data. It can convert low-resolution images into high-resolution images, thereby improving image quality and detail. Super-resolution reconstruction has wide applications in computer vision, medical imaging, astronomy, and other fields. Multi-image super-resolution reconstruction (MMR) is a type of super-resolution reconstruction technique that generates a high-resolution image by processing and combining multiple low-resolution images using specific algorithms and techniques. Many methods have emerged in the field of MMR, effectively improving reconstruction results. However, neutron radiography possesses unique characteristics such as noise, thus requiring specialized reconstruction techniques. This paper proposes a MMR method specifically for neutron radiography. This method employs sub-pixel displacement MMR to overcome the hardware limitations on imaging resolution, obtaining a more accurate and clearer image through post-image processing. Summary of the Invention:

[0003] The technical problem to be solved by the present invention is to provide a fast neutron radiography method and system based on subpixel displacement super-resolution reconstruction to obtain higher resolution and better imaging results.

[0004] The technical solution of the present invention is as follows: a fast neutron radiography system based on subpixel displacement super-resolution reconstruction, comprising a fast neutron radiography device and a multi-image super-resolution reconstruction algorithm module based on variance decision.

[0005] The fast neutron radiography device comprises three parts: a neutron source, a collimator, and a detection system. After being generated by the neutron source, the neutron beam is collimated by the collimator and then irradiates the sample. By detecting the spatial intensity distribution of transmitted neutrons penetrating the sample, internal imaging of the sample is achieved. The transmitted neutrons interact with a conversion screen to excite fluorescent photons, which are then focused onto a CCD camera by optical devices to form a digital image. The detection system consists of a conversion screen, a dark chamber, an optical system, and a precision positioning platform. The optical system is designed with a 45° mirror to reflect the fluorescence generated by the reaction of neutrons with the conversion screen to a lens. The lens focuses the fluorescence onto the CCD camera, which then collects the fluorescence for imaging. The optical system and conversion screen are fixed inside the dark chamber and placed on the precision positioning platform. The precision positioning platform allows for precise dual-axis movement of the entire detection system within a plane perpendicular to the beam. By moving the optical system, multiple frames of neutron images with sub-pixel displacement can be obtained.

[0006] The multi-image super-resolution reconstruction algorithm module based on variance decision-making includes the following steps:

[0007] Step 1: Perform preliminary super-resolution reconstruction of multiple frames of neutron images with sub-pixel displacement using bilinear interpolation: First, expand the pixel grid. Then, based on the positional relationship of the acquired multiple frames with certain sub-pixel offsets, insert the pixel values ​​of the acquired images into the corresponding grid positions. Finally, calculate the pixel values ​​of the remaining grid points using bilinear interpolation. Bilinear interpolation is a method that estimates the target pixel value using nearest-neighbor pixels. Its basic idea is to perform linear interpolation on each row (or column) of the image matrix near the target pixel, and then perform further linear interpolation in the column (or row) direction on the interpolated matrix to obtain the pixel value of the target grid point.

[0008] Step 2: Calculate the mean local variance of the preliminary super-resolution reconstructed image. The local variance is calculated centered at each interpolation position. The formula for calculating local variance is as follows:

[0009]

[0010] in:

[0011] I p Indicates the grayscale value of the target pixel;

[0012] I q This represents the grayscale value of each neighboring pixel surrounding the target pixel;

[0013] S represents the number of neighboring pixels around the target pixel;

[0014] This formula involves squaring the difference between the target pixel and each of its neighboring pixels, summing the squares of all the differences, and finally dividing the sum by the number of neighboring pixels to obtain the average of the squared differences between the target pixel and its surrounding pixels.

[0015] The mean of local variance This refers to the value obtained by calculating the local variance centered on each pixel in the image, and then averaging all the local variances.

[0016] Step 3: Calculate the local variance for each region. with global variance mean In comparison, when At that time, the value of the pixel is updated using a bilateral filtering operator.

[0017] Furthermore, the formula for updating the pixel value using the bilateral filtering operator in step S3 is as follows:

[0018]

[0019] in:

[0020]

[0021] O p This represents the updated pixel value of the target pixel.

[0022] I p Indicates the original pixel value of the target pixel;

[0023] P represents the position of the target pixel;

[0024] q represents the position of other pixels within the filter window;

[0025] I q Indicates the pixel value of other pixels;

[0026] S represents the filtering window involved in the bilateral filtering operator;

[0027] ||pq|| represents the distance between other pixels and the target pixel. and These represent the spatial kernel function and the intensity kernel function, respectively. The spatial kernel function measures the impact of the distance between pixel locations on the weights, while the intensity kernel function measures the impact of the difference in pixel intensity on the weights. Both are Gaussian functions with parameter σ. r and σ s These are the standard deviations of the spatial kernel function and the intensity kernel function, respectively, obtained by adjusting the parameter σ. s and σ r Adjusting the kernel can control the weight distribution of the kernel function;

[0028] w p The normalization factor is used to normalize the weights of the spatial kernel function and the intensity kernel function, ensuring that their sum is 1 during the filtering process. Attached image description:

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0030] Figure 1 This is a schematic diagram of the fast neutron radiography device of the present invention.

[0031] Figure 2 This is a flowchart of the steps in the multi-image super-resolution reconstruction algorithm module based on variance decision-making of the present invention.

[0032] Figure 3 The images shown are the results of multi-image super-resolution reconstruction based on variance decision using the fast neutron radiography device of this invention. Detailed implementation method:

[0033] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0034] Example: The present invention provides a technical solution, a fast neutron radiograph system based on subpixel displacement super-resolution reconstruction. The system includes a fast neutron radiograph device and a multi-image super-resolution reconstruction algorithm module based on variance decision.

[0035] Fast neutron radiography device such as Figure 1 As shown, the fast neutron radiography device consists of three parts: a neutron source, a collimator, and a detection system. After being generated by the neutron source, the neutron beam is collimated by the collimator and then irradiates the sample. By detecting the spatial intensity distribution of the transmitted neutrons penetrating the sample, internal imaging is achieved. The transmitted neutrons interact with the conversion screen to excite fluorescent photons, which are then focused onto a CCD camera by optical devices to form a digital image. The detection system consists of a conversion screen, a dark chamber, an optical system, and a precision positioning platform. The optical system is designed with a 45° mirror to reflect the fluorescence generated by the reaction of neutrons with the conversion screen to a lens. The lens focuses the fluorescence onto the CCD camera, which then collects the fluorescence for imaging. The optical system and conversion screen are fixed inside the dark chamber and placed on the precision positioning platform. The precision positioning platform can control the movement of the entire detection system along the X and Y axes in a plane perpendicular to the beam. By moving the position of the detection system, multiple frames of neutron images with sub-pixel displacement can be obtained.

[0036] The multi-image super-resolution reconstruction process based on variance decision-making is as follows: Figure 2 As shown: First, the same object is photographed multiple times using a fast neutron radiograph. During each photograph, the detector moves to a certain extent and records its position information, resulting in multiple low-resolution images. These low-resolution images are then processed using a multi-image super-resolution reconstruction algorithm based on variance decision-making, mainly including the following steps:

[0037] Step 1: Perform preliminary super-resolution reconstruction of multiple frames of neutron images with sub-pixel displacement using bilinear interpolation: First, expand the pixel grid. Then, based on the positional relationship of the acquired multiple frames with certain sub-pixel offsets, insert the pixel values ​​of the acquired images into the corresponding grid positions. Finally, calculate the pixel values ​​of the remaining grid points using bilinear interpolation. Bilinear interpolation is a method that estimates the target pixel value using nearest-neighbor pixels. Its basic idea is to perform linear interpolation on each row (or column) of the image matrix near the target pixel, and then perform further linear interpolation in the column (or row) direction on the interpolated matrix to obtain the pixel value of the target grid point.

[0038] Step 2: Calculate the mean local variance of the preliminary super-resolution reconstructed image. The local variance is calculated centered at each interpolation position. The formula for calculating local variance is as follows:

[0039]

[0040] in:

[0041] I p Indicates the grayscale value of the target pixel;

[0042] I q This represents the grayscale value of each neighboring pixel surrounding the target pixel;

[0043] S represents the number of neighboring pixels around the target pixel;

[0044] This formula involves squaring the difference between the target pixel and each of its neighboring pixels, summing the squares of all the differences, and finally dividing the sum by the number of neighboring pixels to obtain the average of the squared differences between the target pixel and its surrounding pixels.

[0045] The mean of local variance This refers to the value obtained by calculating the local variance centered on each pixel in the image, and then averaging all the local variances.

[0046] Step 3: Calculate the local variance for each region. with global variance mean In comparison, when At that time, the value of the pixel is updated using a bilateral filtering operator.

[0047] Furthermore, the formula for updating the pixel value using the bilateral filtering operator in step S3 is as follows:

[0048]

[0049] in:

[0050]

[0051] O p This represents the updated pixel value of the target pixel.

[0052] I p Indicates the original pixel value of the target pixel;

[0053] P represents the position of the target pixel;

[0054] q represents the position of other pixels within the filter window;

[0055] I q Indicates the pixel value of other pixels;

[0056] S represents the filtering window involved in the bilateral filtering operator;

[0057] ||pq|| represents the distance between other pixels and the target pixel. and These represent the spatial kernel function and the intensity kernel function, respectively. The spatial kernel function measures the impact of the distance between pixel locations on the weights, while the intensity kernel function measures the impact of the difference in pixel intensity on the weights. Both are Gaussian functions with parameter σ. r and σ s These are the standard deviations of the spatial kernel function and the intensity kernel function, respectively, obtained by adjusting the parameter σ. s and σ r Adjusting the kernel can control the weight distribution of the kernel function;

[0058] w p The normalization factor is used to normalize the weights of the spatial kernel function and the intensity kernel function, ensuring that their sum is 1 during the filtering process.

[0059] like Figure 3The image shows the results of a lead block with grooves of varying widths and depths, captured by a fast neutron radiograph and reconstructed using a variance-based multi-image super-resolution algorithm. Figure (a) shows a neutron transmission photograph of the lead block (6cm wide, 4cm high, 2cm thick) with grooves of different widths and depths, captured by the fast neutron radiograph (Figure (b)). Figure (c) shows the reconstruction result using the variance-based multi-image super-resolution algorithm. The comparison of the imaging results shows that the variance-based multi-image super-resolution algorithm significantly improves the resolution and quality of the reconstructed image compared to the directly obtained neutron radiograph image. The fast neutron radiograph system based on sub-pixel displacement super-resolution reconstruction achieves excellent super-resolution reconstruction results by utilizing multiple frames of neutron images captured at different locations with certain sub-pixel offsets, combined with a variance-based reconstruction algorithm module.

[0060] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fast neutron radiography system based on subpixel displacement super-resolution reconstruction, characterized in that: The system includes a fast neutron radiograph device and a multi-image super-resolution reconstruction algorithm module based on variance decision. The fast neutron radiograph device consists of three parts: a neutron source, a collimator, and a detection system, used to acquire multiple digital images formed at different detection locations. The multi-image super-resolution reconstruction algorithm module based on variance decision performs the following steps: Step 1: Perform preliminary super-resolution reconstruction of multiple frames of neutron images with sub-pixel displacement using bilinear interpolation: First, expand the pixel grid. Then, based on the positional relationship of the acquired multiple frames with certain sub-pixel offsets, insert the pixel values ​​of the acquired images into the corresponding grid positions. Finally, use bilinear interpolation to calculate the pixel values ​​of the remaining grid points. Bilinear interpolation is a method that uses nearest neighbor pixels to estimate the target pixel value. Its basic idea is to perform linear interpolation on each row or column of the image matrix near the target pixel, and then perform linear interpolation in the column or row direction on the interpolated matrix to obtain the pixel value of the target grid point. Step 2: Calculate the mean local variance of the preliminary super-resolution reconstructed image. The local variance is calculated centered at each interpolation position. The formula for calculating local variance is as follows: in: I p Indicates the grayscale value of the target pixel; I q This represents the grayscale value of each neighboring pixel surrounding the target pixel; S represents the number of neighboring pixels around the target pixel; The mean of local variance This refers to the value obtained by calculating the local variance using each pixel on the image as the center, and then averaging all the local variances. Step 3: Calculate the local variance for each region. with local variance mean In comparison, when At that time, the value of the pixel is updated using the bilateral filtering operator; Furthermore, the formula for updating the pixel value using the bilateral filtering operator in step 3 is as follows: in: Op represents the updated pixel value of the target pixel; I p Indicates the original pixel value of the target pixel; P represents the position of the target pixel; q represents the position of other pixels within the filter window; I q Indicates the pixel value of other pixels; S represents the filtering window involved in the bilateral filtering operator; ||pq|| represents the distance between other pixels and the target pixel. and These represent the spatial kernel function and the intensity kernel function, respectively. The spatial kernel function measures the impact of the distance between pixel locations on the weights, while the intensity kernel function measures the impact of the difference in pixel intensity on the weights. Both are Gaussian functions with parameter σ. r and σ s These are the standard deviations of the spatial kernel function and the intensity kernel function, respectively, obtained by adjusting the parameter σ. s and σ r Adjusting the kernel can control the weight distribution of the kernel function; w p The normalization factor is used to normalize the weights of the spatial kernel function and the intensity kernel function, ensuring that their sum is 1 during the filtering process.

2. The fast neutron radiography system according to claim 1, characterized in that: The fast neutron radiography device includes a conversion screen, a dark box, an optical system, and a precision positioning platform. The optical system is designed to use a 45° reflector to reflect the fluorescence generated by the reaction between the neutron and the conversion screen onto a lens. The lens focuses the fluorescence onto a CCD camera, which then collects the fluorescence for imaging. The optical system and conversion screen are fixed inside the dark box and placed on the precision positioning platform. The precision positioning platform can control the entire detection system to move precisely in two axes within a plane perpendicular to the beam.

Images