Image three-dimensional reconstruction algorithm applied to energy resolution neutron imaging

Abstract

The invention relates to an image three-dimensional reconstruction algorithm applied to energy resolution neutron imaging, which comprises the following steps: pulse neutron beams are generated through spallation reaction, the time for generating the pulse neutron beams each time is taken as a trigger signal of TPX3Cam and recorded as T0, and information collected and recorded by TPX3Cam comprises coordinates (x, y), time of arrival (ToA) and intensity (ToT) of each response pixel, the data corresponding to each pixel is called as a photon case; photon case data collected by the TPX3Cam are reconstructed into neutron cases, and the method mainly comprises two parts of clustering and neutron case reconstruction; and then drawing a neutron image. A neutron image shot by the TPX3Cam camera is processed, photons belonging to the same neutron are clustered, and the more accurate position of the neutron is found; in the clustering process, not only is the plane feature of photons considered, but also the time feature of photons is considered, so that clustering is more accurate, optimization of the energy-resolved neutron image is realized, and the resolution of the energy-resolved neutron image is improved.

Description

Technical Field

[0001] The present invention relates to the field of neutron imaging, and in particular to a three-dimensional image reconstruction algorithm applied to energy-resolved neutron imaging. Background Art

[0002] Energy-resolved neutron imaging is an important direction of current neutron imaging development. By resolving neutrons simultaneously in time and space, it can be used to analyze the stress distribution, crystal structure, internal microstructure, etc. of materials. In pulsed spallation neutron sources, the time-of-flight method (ToF) is a method of measuring the flight speed of neutrons based on the flight distance of neutrons. The time-of-flight method can be used to achieve full-energy scanning neutron imaging and Bragg edge imaging of samples in one experiment. The imaging process requires the use of a high-speed camera TPX3Cam with a built-in Timepix3 chip. During the imaging process, the spot formed after the neutron hits the scintillating screen will be further enlarged after being magnified by the optical lens and enhanced by the image intensifier, resulting in the inability to find the exact location of the neutron. Figure 1 The characteristics of the pixels recorded by TPX3Cam are shown. Normally, for a neutron signal, TPX3Cam will have multiple pixels responding simultaneously; a signal with only one pixel responding can be considered as noise. By processing the photons in the image through the reconstruction algorithm, the true position of the neutron can be better reflected. The reconstructed position information is converted into a neutron image, and the resulting neutron image has higher spatial resolution.

[0003] Currently, there is no dedicated reconstruction algorithm in this field. The DBSCAN algorithm is used to directly process the data. This method ignores the temporal features in the data. Photon data that are close in plane features but far apart in temporal features may be classified as the same neutron, resulting in inaccurate image reconstruction and the inability to further improve the resolution of neutron images. Summary of the invention

[0004] The purpose of the present invention is to provide an image 3D reconstruction algorithm for energy-resolved neutron imaging, which processes the neutron image taken by the TPX3Cam camera and finds the more accurate position of the neutron by clustering the photons belonging to the same neutron. In the clustering process, not only the planar characteristics of the photons will be considered, but also the temporal characteristics of the photons will be considered, making the clustering more accurate, so as to achieve the optimization of the energy-resolved neutron image and improve its resolution.

[0005] In order to achieve the above objectives, the technical solution adopted by the present invention is: an image three-dimensional reconstruction algorithm applied to energy-resolved neutron imaging, comprising the following steps:

[0006] Step 1: Generate a pulsed neutron beam through spallation reaction. The time of each pulsed neutron beam generation is used as the trigger signal of TPX3Cam, denoted as T0. The information collected and recorded by TPX3Cam includes the coordinates (x, y), arrival time (ToA) and intensity (ToT) of each response pixel. The data corresponding to each pixel is called a photon event.

[0007] Step 2: Reconstruct the photon event data collected by TPX3Cam into neutron events, which mainly includes two parts: clustering and neutron event reconstruction; then draw the neutron image.

[0008] Preferably, in the event clustering process, the coordinates (x, y) and ToF of the photon event are used to describe its position in space. By clustering the photon events, the photon events that may be stimulated by the same neutron event are aggregated into a cluster. The photon events within the cluster are considered to be caused by the same neutron, and those single photon events that have no direct correlation with other photon events in space or time are considered to be noise and are not classified into any cluster.

[0009] Preferably, during the reconstruction of the neutron event, the center of mass method is used to approximately reconstruct the position of the neutron event; the longer the electrons accelerated in the microchannel plate last, the more energy they obtain, and thus the ToT value recorded by the Timepix3 chip will be higher; electrons with higher energy are more likely to be generated at positions closer to the neutron event, and for photon events within the same cluster, the center of mass method is used to calculate the corresponding weights based on the ToT values ​​of different photon events to calculate the coordinates of the neutron event corresponding to the cluster, and convert the position information into a neutron image.

[0010] The technical effects of the present invention are:

[0011] 1. Process the neutron images taken by the TPX3Cam camera, and find the more accurate position of the neutron by clustering the photons belonging to the same neutron; the clustering process will not only consider the planar characteristics of the photons, but also the temporal characteristics of the photons, making the clustering more accurate, so as to achieve the optimization of the energy-resolved neutron image and improve its resolution.

[0012] 2. When processing the image data of TPX3Cam, the image 3D reconstruction algorithm adds the temporal characteristics of the data into the clustering consideration process, instead of directly applying the DBSCAN algorithm for processing. This method can effectively improve the accuracy of the clustering process in the algorithm, thereby more accurately reconstructing the position of neutrons and further improving the resolution of neutron images.

[0013] 3. After processing the neutron images taken by the TPX3Cam camera using the image 3D reconstruction algorithm, the spatial resolution of the neutron images can be effectively improved. It was observed that the spatial resolution increased from 150um / line width to 100um / line width. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Feature map of pixels recorded for TPX3Cam.

[0015] Figure 2 is the original neutron image without algorithm processing.

[0016] Figure 3 is a neutron image processed by this application.

[0017] Figure 4 Schematic diagram of an energy-resolving neutron imaging detector.

[0018] Figure 5 This is the working principle diagram of the image intensifier.

[0019] Figure 6 This is a data feature diagram of the photon events collected by TPX3Cam.

[0020] Figure 7 This is the characteristic diagram of the neutron signal measured under the same experimental conditions.

[0021] Figure 8 A 2D image of a representative data set from selected austenitic steel experiments.

[0022] Fig. 9 A 3D image of a representative data set from selected austenitic steel experiments.

[0023] Fig.10 This is a two-dimensional classification result diagram after using the case clustering algorithm.

[0024] Fig.11 This is the three-dimensional classification result diagram after using the case clustering algorithm.

[0025] Fig.12 Bragg edge plotted for the austenite experiment data set.

[0026] Fig.13 for Fig.12 The middle part of the Gaussian curve. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings. The description in this part is only exemplary and explanatory and should not have any limiting effect on the protection scope of the present invention.

[0028] Detector principle

[0029] Schematic diagram of an energy-resolving neutron imaging detector Figure 4 As shown; a pulsed neutron beam is generated through spallation reaction with a repetition frequency of 25Hz. The time of each pulsed neutron beam generation is used as the trigger signal of TPX3Cam, recorded as T0. After passing through the sample, the pulsed neutron beam is absorbed by the scintillator and converted into a light beam, which is deflected by a reflector placed at a 45° angle and enters the optical lens. The Timepix3 chip has a high detection threshold. Each pixel needs about 1,000 photons to be hit before it can be excited. Therefore, an image intensifier is used to amplify the light signal. Except for the final imaging device, the experimental principle is set up similarly to a traditional neutron imaging detector using a scintillator. Through this experimental setup, information about the path of neutron propagation in the sample and its interaction with the sample can be obtained.

[0030] Figure 5 The working principle of the image intensifier is demonstrated. The image intensifier is a vacuum electron multiplier device, which is usually used to amplify weak light signals. It consists of a photocathode, a microchannel plate, a fluorescent screen and a relay lens. The photons generated by the neutrons excited by the scintillator are focused on the photocathode through the optical lens and then converted into photoelectrons. When the photoelectrons pass through the microchannel plate, the inner wall of the channel will emit more secondary electrons, which will then hit the fluorescent screen after cascade multiplication to generate an amplified light signal. Generally speaking, the amplification gain of the image intensifier can reach thousands of times.

[0031] The information recorded by TPX3Cam includes the coordinates (x, y), arrival time (ToA) and intensity (ToT) of each response pixel. The data corresponding to each pixel is called a photon event. The flight time of each photon event can be calculated by the take-off time T0 signal of the neutron with a period of 40ms and the arrival time of the photon event. The spot size of the light spot formed after the neutron hits the scintillator screen will be further enlarged after being magnified by the optical lens and enhanced by the image intensifier. Normally, for a neutron signal, TPX3Cam will have multiple pixels responding simultaneously; signals with only one pixel responding can be considered as noise; based on the time and space information recorded by TPX3Cam, the image reconstruction algorithm can be used to determine the incident position and arrival time of a single neutron event, thereby improving the spatial resolution and energy resolution of the detector.

[0032] Image reconstruction method

[0033] The photon event data collected by TPX3Cam is reconstructed into neutron events, which mainly includes clustering and neutron event reconstruction. In the event clustering process, the coordinates (x, y) and ToF of the photon event can be used to describe its position in space. By clustering the photon events, the photon events that may be excited by the same neutron event are aggregated into a cluster, and the photon events in the cluster can be regarded as caused by the same neutron. Individual photon events that have no direct correlation with other photon events in space or time are considered noise and are not classified into any cluster. In the process of neutron event reconstruction, the centroid method can be used to approximate the position of the neutron event. The longer the electrons accelerated in the microchannel plate last, the more energy they obtain. Therefore, the ToT value recorded by the Timepix3 chip will be higher. Electrons with higher energy are more likely to be generated at a position closer to the neutron event. For photon events in the same cluster, the centroid method is used to calculate the corresponding weights according to the ToT values ​​of different photon events to calculate the coordinates of the neutron event corresponding to the cluster. The reconstructed event coordinates can better reflect the true position of the neutron. Finally, the position information is converted into a neutron image, and the drawn neutron image has higher spatial resolution.

[0034] Introduction to Density Clustering

[0035] Event clustering requires identifying photon events that belong to the same neutron event. Cluster analysis can accomplish this goal. Cluster analysis is an important unsupervised learning method that divides data into a certain number of clusters based on the similarity between samples, so that the similarity between sample points in the same cluster is greater, and the similarity between samples in different clusters is smaller. Density clustering in cluster analysis helps identify clustering patterns between photon events, thereby more accurately reconstructing the location information of neutron events; DBSCAN (density-based spatial clustering application) is a commonly used density clustering algorithm.

[0036] An overview of the DBSCAN algorithm is as follows: There is a set of data points in the plane, and for each data point, the algorithm calculates the number of data points within a small distance near it. This distance is represented by the radius of the circle formed with the data point as the center, mathematically called ε (εilon). This area is called the ε neighborhood of the data point. If a data point's ε neighborhood contains at least min_points data points (including itself), then the data point is considered a core data point. In other words, a core data point refers to a data point located in a high-density area, and min_points is the minimum number of data points within the radius ε. In the neighborhood of a core data point, all data points are classified into the same cluster, which may include other core data points, so a series of adjacent core data points form a cluster. Any data point that is not a core data point and has no core data points in its neighborhood is considered a noise value.

[0037] The previous method of processing photon events is to observe the characteristics of neutron signals through neutron signal analysis experiments, and then select an appropriate value Photon_Time as the duration of the neutron signal. Next, the data will be segmented according to Photon_Time, that is, the third-dimensional data ToF will be compressed. Then DBSCAN clustering is performed on the two-dimensional data to find photon event clusters belonging to the same neutron event. Although this dimensionality reduction method improves efficiency, it will lose clustering accuracy because a neutron signal may be segmented into multiple parts.

[0038] Three-dimensional DBSCAN case clustering algorithm

[0039] In order to further improve the clustering accuracy to improve the resolution of neutron images, this study will adopt an event clustering method based on improved DBSCAN. The advantage of TPX3Cam over traditional cameras is that it records the ToF information of each event data. The clustering algorithm regards ToF as the third dimension of information in clustering and takes it into consideration for clustering. This strategy aims to distinguish between photon events belonging to the same neutron event with a close front-to-back distance in the ToF dimension, so as to more accurately identify photon events belonging to the same neutron event. Through the event clustering algorithm, neutron events can be identified more accurately, avoiding the division of a neutron signal into multiple parts, thereby improving the accuracy of image reconstruction and thus improving the resolution of neutron images.

[0040] Figure 6 The data characteristics of the photon events collected by TPX3Cam are displayed. Through observation, photon clusters can be intuitively identified, and single photon events that are not directly related to other photon events in space or time can be clearly observed. These photon events are mainly caused by B rays and afterglow. It should be added that for photon events in the same T0 signal, TPX3Cam's data reading will be carried out in ascending order according to the ToF of each photon event.

[0041] The event clustering algorithm integrates the third dimension data ToF for clustering, and uses Manhattan distance for distance measurement in the two-dimensional data space. That is, when there is a photon event α, another photon event β satisfies both: 1) the ToF value is less than ΔT from the ToF value of the photon event α, 2) the Manhattan distance between the coordinates (x, y) and the coordinates of the photon event α is less than ε, then the photon event β is in the field of the photon event α. The parameter ΔT is added to the event clustering algorithm, which conforms to the logic that the photon events belonging to the same neutron event are similar in ToF. At the same time, when a photon event calculates how many other photon events exist in its neighborhood, it only needs to calculate the photon events whose ToF values ​​differ from it by less than ΔT, which reduces a lot of calculations for the event clustering algorithm. By using the comprehensive judgment of the two-dimensional plane and ToF, and setting the parameters reasonably, the event clustering algorithm can more accurately identify the photon events belonging to the same neutron event.

[0042] In the event clustering algorithm, the photon event data belonging to the same T0 signal will be stored in order from small to large ToF, and the data collected by a T0 signal is called a frame. The event clustering algorithm processes each frame of data, and divides each frame of 0 data into small data groups according to a fixed time interval (50μs), which can effectively reduce the number of distance calculations between each data point and other data points. In the nearest neighbor search process, it is only necessary to search within the same data group, avoiding the search of all data in a frame.

[0043] Algorithm parameter optimization

[0044] In practical applications, event clustering algorithms usually need to adjust three key parameters, namely ΔT, ε, and min_points. The distribution characteristics of photons may be different under different neutron fluxes and different scintillator conditions. The study selected ZnF scintillator, set the gain of the image intensifier to 3400, and the experimental sample was austenitic steel (304 steel). The experimental data is large in scale, totaling about 60GB, including about 44,000 frames of imaging with objects (austenitic steel + straight beam) and without objects (straight beam).

[0045] In order to optimize the effect of case clustering algorithm, it is necessary to find the best parameter combination. Figure 7 The neutron signal characteristics measured under the same experimental conditions are shown, and the duration of the neutron is about 200ns. Since photons have afterglow, TPXCam may capture the same photon signal twice at the same time, so the parameter ΔT should be appropriately relaxed. Figure 8 , 9 Shown are 2D and 3D images of a representative data set selected from experiments on austenitic steel. Figure 8It can be observed that the distance between two clusters that are close to each other is about 5, so ε should be set no greater than 5 to prevent clusters that should be classified into two categories from being classified into one. Fig. 9 It can be observed that there are cases where the photon cluster has only two data points, so min_points should be set to 2.

[0046] The data set in the austenite steel experiment was processed using an event clustering algorithm, and different parameter combinations were used, and the neutron images were reconstructed by the centroid method. By comparing the resolution of the neutron images under different parameter combinations, the optimal parameter combination can be intuitively obtained. However, the above experiment was carried out under conditions of low neutron flux, and the resolution difference of the neutron images under different parameter combinations was not obvious, and a good comparison effect could not be obtained.

[0047] The Bragg edge neutron imaging of crystal samples is an important application direction of the detector based on TPX3Cam. Neutrons will undergo Bragg diffraction on the crystal. By comparing the sharpness of the Bragg edge, the quality of the algorithm can be judged. Fig.12 This is a Bragg edge image drawn using the data set of the austenite experiment. The Bragg edge of the Bragg edge image after using the case reconstruction algorithm is sharper. The Gaussian curve is obtained by differential operation on the position of the Bragg edge. Fig.13 As shown in the figure, the standard value of the Gaussian curve is obtained, and the effects of the case clustering algorithm under different parameter combinations can be intuitively compared through the data.

[0048] Results Analysis

[0049] This study uses an improved 3D DBSCAN case clustering algorithm to achieve image reconstruction of energy-resolved neutron imaging. The case clustering algorithm adds ToF to the clustering based on 2D clustering to form 3D clustering. The effect of the case clustering algorithm is analyzed. Fig.10 , Fig.11 The classification results after using the case clustering algorithm are shown. The classification results on the right show the labels to which each cluster belongs. The number of the label only indicates that these data points belong to a cluster. There are 9 labels in total, of which 0 represents noise and does not belong to any cluster. The red data points are considered noise points. Fig. 9 All the data points that are relatively isolated in space are classified as category 0. Although the two clusters corresponding to label 4 and label 5 are close in two dimensions, they are far apart in the ToF dimension, so they are divided into two clusters.

[0050] The austenitic steel experimental data contains 1991201659 photon events, which are divided into 257727013 clusters by the event clustering algorithm. About 7% of the events are not assigned to any cluster, are regarded as noise and are removed from the final clustering results. Fig.11We demonstrate that the algorithm can successfully group photon events into clusters without relying on any prior knowledge of the data.

[0051] The optimal parameters of the case clustering algorithm can be found by reconstructing the image with different parameter combinations. The min_points parameter is set to 2, and ΔT and ε are set to different values. Gaussian fitting is performed on the curve of the Bragg edge, and the standard value corresponding to the Gaussian curve can quantify the sharpness of the Bragg edge. The following table shows the standard values ​​obtained by different parameter combinations. The algorithm uses the parameters of ΔT of 1.5, ε of 2, and min_points of 2, and the standard value of the Gaussian fitting curve of the reconstructed Bragg edge is the lowest. It is proved that under this experimental condition, the optimal parameters of the case clustering algorithm can be obtained.

[0052]

[0053] The case clustering algorithm is used to analyze the data of the energy-resolved neutron imaging detector measuring the star-shaped test target. The star-shaped test target experimental data includes 48,000 frames of images with objects (space star-shaped test target + direct beam) and 48,000 frames of images without objects (direct beam). The neutron beam intensity of the star-shaped test target experiment is the same as that of the austenitic steel experiment, and both use ZnF as the scintillator. The spatial resolution test results obtained by the experiment with a magnification of 0.27 are shown in Figure 2. Figure 2 The result after image reconstruction is shown in Figure 3 As shown in the figure, the spatial resolution is improved from 400μm / line width to 200μm / line width. The results show that the image reconstruction method using the case clustering algorithm can effectively improve the image resolution of energy-resolved neutron imaging.

[0054] It should be noted that, in this article, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0055] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. The above is only a preferred implementation of the present invention. It should be pointed out that due to the limitations of textual expression and the objective existence of infinite specific structures, ordinary technicians in this technical field can make several improvements, modifications or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the inventive concept and technical solution to other occasions without improvement, should be regarded as the protection scope of the present invention.

Claims

1. A three-dimensional image reconstruction algorithm for energy-resolved neutron imaging, characterized in that: The steps include: Step 1: Generate a pulsed neutron beam through spallation reaction. The time of each pulsed neutron beam generation is used as the trigger signal of TPX3Cam, denoted as T0. The information collected and recorded by TPX3Cam includes the coordinates (x, y), arrival time (ToA) and intensity (ToT) of each response pixel. The data corresponding to each pixel is called a photon event. Step 2: Reconstruct the photon event data collected by TPX3Cam into neutron events, which mainly includes two parts: clustering and neutron event reconstruction; then draw the neutron image.

2. The three-dimensional image reconstruction algorithm for energy-resolved neutron imaging according to claim 1, characterized in that: In the event clustering process, the coordinates (x, y) and ToF of the photon event are used to describe its position in space. By clustering the photon events, the photon events that may be stimulated by the same neutron event are aggregated into a cluster. The photon events in the cluster are considered to be caused by the same neutron, while those single photon events that have no direct correlation with other photon events in space or time are considered as noise and are not classified into any cluster.

3. The three-dimensional image reconstruction algorithm for energy-resolved neutron imaging according to claim 2, characterized in that: During the reconstruction of neutron events, the centroid method is used to approximately reconstruct the position of the neutron event; the longer the electrons accelerated in the microchannel plate last, the more energy they obtain, so the ToT value recorded by the Timepix3 chip will also be higher; electrons with higher energy are more likely to be generated at a position closer to the neutron event. For photon events in the same cluster, the centroid method is used to calculate the corresponding weights according to the ToT values ​​of different photon events to calculate the coordinates of the neutron event corresponding to the cluster, and convert the position information into a neutron image.

Images