Measurement device, measurement method, method for creating an artifact reduction model for oblique CT images, and learning model for artifact reduction for oblique CT images.

Abstract

The objective is to provide a measurement device, a measurement method, a method for creating an oblique CT image artifact reduction model, and an oblique CT image artifact reduction learning model that can obtain oblique CT images with reduced artifacts. [Solution] The measuring device 10 includes a CT imaging device 14 that takes multiple oblique CT images of the same object to be measured by varying the installation angle with respect to the sample rotation axis that rotates the object to be measured, and an information processing device 12 that generates an oblique CT image with reduced artifacts using a convolutional neural network (CNN).

Description

Technical Field

[0001] The present invention relates to a measuring device, a measuring method, a method for creating a model for reducing artifacts in oblique CT images, and a learning model for reducing artifacts in oblique CT images.

Background Art

[0002] Patent Document 1 and Non-Patent Document 1 disclose a method for reducing artifacts in oblique CT images using a three-dimensional convolutional neural network.

[0003] Further, Non-Patent Documents 2 to 4 propose a method for reducing artifacts by combining normal CT images and oblique CT images. In each case, it is proposed to use the advantages of the high resolution of the vertical cross-section of the normal CT image and the high resolution of the horizontal cross-section of the oblique CT image respectively.

[0004] Also, Patent Document 2 discloses a method for reducing metal-derived artifacts using two CT data with different rotation directions. In Patent Document 2, one three-dimensional image with reduced metal artifacts is reconstructed from a plurality of normal CT measurement data.

[0005] Also, various techniques have been proposed for oblique CT measurement methods. For example, the techniques described in Patent Documents 3 and 4 have been proposed.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Non-Patent Documents

[0007] [Non-Patent Document 1] A.-K. Schnurr, et al., Z. Med. Phys. 29(2), 150-161 (2019) [Non-Patent Document 2] M. Zuber, et al., Scientific Reports 7, 41413 (2017) [Non-Patent Document 3] T. Jia, et al., Phys. Scr. 98, 105114 (2023) [Non-Patent Document 4] P. Ji, et al., NDT E Int. 141, 102991 (2024) [Overview of the project] [Problems that the invention aims to solve]

[0008] While various techniques exist to reduce artifacts, there is still room for improvement in further reducing artifacts in oblique CT images.

[0009] This disclosure is made in consideration of the above facts and aims to provide a measuring device, a measuring method, a method for creating an oblique CT image artifact reduction model, and an oblique CT image artifact reduction learning model that can obtain oblique CT images with reduced artifacts. [Means for solving the problem]

[0010] To achieve the above objective, the measuring device according to the first embodiment includes an imaging unit that captures multiple oblique CT images of the same object to be measured by varying the installation angle with respect to the sample rotation axis that rotates the object to be measured, and a generation unit that generates an oblique CT image with reduced artifacts using a convolutional neural network from the multiple oblique CT images captured by the imaging unit.

[0011] According to the first embodiment, by taking multiple oblique CT images at different installation angles with respect to the sample rotation axis, multiple oblique CT images with different missing information can be obtained, thus enabling the acquisition of oblique CT images with reduced artifacts.

[0012] The measuring device according to the second embodiment is an X-ray CT image in which the plurality of oblique CT images are constant angles between the X-ray and the rotation axis of the sample.

[0013] According to the second embodiment, a measuring device in which the angle between the X-ray and the sample rotation axis is constant can perform measurements efficiently because it can be used with a fixed arrangement of measuring devices, and is desirable for industrial applications such as in-line product inspection.

[0014] The measuring device according to the third embodiment is the measuring device according to the first embodiment, wherein the plurality of oblique CT images are two oblique CT images taken by changing the installation angle of the object to be measured so that it is at a constant positive or negative angle with respect to the rotation axis of the sample.

[0015] According to the third embodiment, for example, by using a seesaw-type mounting jig, it is possible to capture oblique CT images that are set at a constant positive and negative angle with respect to the sample rotation axis, and to efficiently capture multiple oblique CT images with different mounting angles with respect to the sample rotation axis.

[0016] The measuring device according to the fourth embodiment is a measuring device according to the first embodiment in which the convolutional neural network is trained using the structural model of the object to be measured created on a computer, and the results of simulating multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant and the installation angle is different, as training data.

[0017] According to the fourth aspect, actual measurement sample data can be used as teacher data for learning. However, since measurement takes time, it is difficult to create a large amount of teacher data, and there is a possibility of unintended noise being added. By using teacher data obtained through simulation, it becomes possible to obtain high-quality correct images, prepare a large amount of data, and improve generalization by adding intentional noise.

[0018] The measuring device according to the fifth aspect is the measuring device according to the first aspect, wherein the convolutional neural network creates a structural model of the object to be measured on a computer, and uses the structural model to learn, as teacher data, the result of simulating oblique CT images of combinations in which the installation angle is a fixed angle with respect to the sample rotation axis, positive and negative.

[0019] According to the fifth aspect, actual measurement sample data can be used as teacher data for learning. However, since measurement takes time, it is difficult to create a large amount of teacher data, and there is a possibility of unintended noise being added. By using teacher data obtained through simulation, it becomes possible to obtain high-quality correct images, prepare a large amount of data, and improve generalization by adding intentional noise.

[0020] The measuring device according to the sixth aspect is the measuring device according to the first aspect, wherein the imaging unit has a jig capable of fixing the object to be measured in an arbitrarily inclined direction with respect to the direction of the sample rotation axis.

[0021] According to the sixth aspect, it is possible to efficiently capture a plurality of oblique CT images of the same object to be measured with different installation angles with respect to the sample rotation axis.

[0022] The measurement method according to the seventh aspect is such that a computer acquires a plurality of oblique CT images obtained by respectively imaging the same object to be measured at different installation angles with respect to a sample rotation axis for rotating the object to be measured, and performs a process of generating an oblique CT image with reduced artifacts using a convolutional neural network on the plurality of oblique CT images.

[0023] According to the seventh embodiment, by taking multiple oblique CT images at different installation angles with respect to the sample rotation axis, multiple oblique CT images with different missing information can be obtained, thus enabling the acquisition of oblique CT images with reduced artifacts.

[0024] The oblique CT image artifact reduction model creation method according to the eighth aspect is a method for creating an oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under test at different installation angles with respect to the sample rotation axis that rotates the object under test, wherein a structural model of the object under test is created on a computer, and the results of simulating multiple oblique CT images with different installation angles, where the angle between the X-ray and the sample rotation axis is constant, are created as training data and used for learning.

[0025] According to the eighth aspect, an oblique CT image artifact reduction model can be obtained that can acquire oblique CT images with reduced artifacts. Furthermore, by using training data obtained through simulation, it becomes possible to obtain high-quality ground truth images, prepare a large amount of data, and improve generalization by intentionally adding noise.

[0026] The oblique CT image artifact reduction model creation method according to the ninth aspect is a method for creating an oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under test at different installation angles with respect to the sample rotation axis that rotates the object under test, wherein a structural model of the object under test is created on a computer, and the results of simulating oblique CT images for combinations in which the installation angle is a constant positive or negative angle with respect to the sample rotation axis are used as training data for learning.

[0027] According to the ninth embodiment, an oblique CT image artifact reduction model can be obtained that can acquire oblique CT images with reduced artifacts. Furthermore, by using training data obtained through simulation, it becomes possible to obtain high-quality ground truth images, prepare a large amount of data, and improve generalization by intentionally adding noise.

[0028] The oblique CT image artifact reduction model according to the tenth embodiment is an oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under test at different installation angles with respect to the sample rotation axis that rotates the object under test, wherein a structural model of the object under test is created on a computer, and the model is trained by creating training data from the results of simulating multiple oblique CT images using the structural model, where the angle between the X-ray and the sample rotation axis is constant and the installation angles are different.

[0029] According to the tenth embodiment, an oblique CT image with reduced artifacts can be generated from multiple oblique CT images.

[0030] The oblique CT image artifact reduction model according to the 11th embodiment is an oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under test at different installation angles with respect to the sample rotation axis that rotates the object under test, wherein a structural model of the object under test is created on a computer, and the model is trained using the results of simulations of oblique CT images with a fixed positive and negative angle with respect to the sample rotation axis using the structural model.

[0031] According to the 11th embodiment, an oblique CT image with reduced artifacts can be generated from multiple oblique CT images. [Effects of the Invention]

[0032] As described above, the present invention provides a measuring device, a measuring method, a method for creating an oblique CT image artifact reduction model, and an oblique CT image artifact reduction learning model that can obtain oblique CT images with reduced artifacts. [Brief explanation of the drawing]

[0033] [Figure 1] This figure shows the schematic configuration of the measuring device according to this embodiment. [Figure 2] This is a front view of a jig for taking oblique CT images with a different placement angle of the measurement sample. [Figure 3] This is a side view of a jig for taking oblique CT images with a different placement angle of the measurement sample. [Figure 4] This is an illustrative diagram of an actual measurement using a jig (a seesaw-type jig tilted 15 degrees upstream of the X-ray beam). [Figure 5] This is an illustrative diagram of an actual measurement using a jig (a seesaw-type jig tilted 15 degrees downstream of the X-ray beam). [Figure 6] This is a functional block diagram showing the functional configuration of an information processing device. [Figure 7] This is a diagram to specifically explain the learning process conducted by the learning department. [Figure 8] This diagram specifically illustrates the generation of oblique CT images with reduced artifacts by the estimation unit. [Figure 9] This is a flowchart illustrating an example of the learning process flow. [Figure 10] This is a flowchart illustrating an example of the image reconstruction process. [Figure 11] This is a flowchart illustrating an example of the estimation process flow. [Figure 12] This is a schematic diagram showing an example of a test sample. [Figure 13] This figure shows the results of three-dimensional reconstruction of two types of oblique CT images, as well as an example of an oblique CT image with reduced artifacts. [Modes for carrying out the invention]

[0034] Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings. Figure 1 is a diagram showing the schematic configuration of the measuring device according to this embodiment.

[0035] The measuring device 10 according to this embodiment captures oblique CT images with varying installation angles of the measurement sample and generates oblique CT images with reduced artifacts.

[0036] As shown in Figure 1, the measuring device 10 includes a CT image acquisition device 14 as an example of an imaging unit and an information processing device 12 as an example of an information generation unit.

[0037] The CT imaging device 14 is equipped with a radiation source 16 and a radiation detector 18. The radiation source 16 irradiates the sample to be measured with X-rays, and the radiation detector 18 detects the X-rays that have passed through the sample. A CT image is taken by rotating the sample using a jig described later. In this embodiment, the CT imaging device 14 takes multiple oblique CT images of the same object to be measured by varying the installation angle of the object with respect to the sample rotation axis that rotates the object to be measured.

[0038] Here, we will describe a jig for taking oblique CT images with a changed installation angle of the measurement sample. Figure 2 is a front view of the jig 20 for taking oblique CT images with a changed installation angle of the measurement sample, and Figure 3 is a side view of the jig 20 for taking oblique CT images with a changed installation angle of the measurement sample.

[0039] The jig 20 is of the seesaw type, as shown in Figure 2. The material of the jig 20 must allow X-rays to penetrate sufficiently; for example, resin or aluminum may be used.

[0040] The jig 20 includes a seesaw section 22 on which the measurement sample 60 is placed and whose installation angle can be changed to any desired angle, and a rotary motor 24. The seesaw section 22 can be freely rotated relative to the rotation axis 24A of the rotary motor 24, which serves as the sample rotation axis, but it is fixed when taking oblique CT images. Figures 4 and 5 show an image of the actual imaging process.

[0041] By using the seesaw-type jig 20, multiple oblique CT images with different installation angles relative to the sample rotation axis can be efficiently acquired for the same object being measured.

[0042] In both Figures 4 and 5, the rotation axis 24A is angled 30 degrees relative to the direction perpendicular to the X-ray incidence. This allows for efficient measurement by enabling imaging with a fixed arrangement of measuring devices 10, which is desirable for industrial applications such as in-line product inspection.

[0043] In Figure 4, the seesaw-type jig 20 is tilted 15 degrees upstream of the X-ray beam, and in Figure 5, it is tilted 15 degrees downstream. By rotating each jig 360 degrees while taking images, oblique CT images with different artifact orientations can be obtained. Note that when rotated 180 degrees, Figures 4 and 5 appear to overlap, but because their internal structures are different, different information can be obtained.

[0044] In this way, by using the seesaw-type jig 20, it is possible to capture oblique CT images set at a constant positive and negative angle with respect to the sample rotation axis, and to efficiently capture multiple oblique CT images with different installation angles relative to the sample rotation axis. Note that the angle between the X-ray and the sample rotation axis is not limited to a constant positive or negative angle, as long as it is a constant angle. Also, the above angle is just an example, and other angles may be applied.

[0045] The information processing device 12 acquires oblique CT images obtained by the CT imaging device 14 and processes them to generate oblique CT images with reduced artifacts using AI (Artificial Intelligence). Specifically, it processes them to generate oblique CT images with reduced artifacts using a convolutional neural network (CNN).

[0046] The information processing device 12 includes a CPU (Central Processing Unit) 12A as an example of a processor, a ROM (Read Only Memory) 12B, a RAM (Random Access Memory) 12C, storage 12D, an input unit 12E, a display unit 12F, and a communication interface (communication I / F) 12G. Each component is connected to the others via a bus 12H so as to be able to communicate with each other.

[0047] The CPU 12A is the central processing unit, which executes various programs and controls various components. Specifically, the CPU 12A reads programs from ROM 12B or storage 12D and executes them using RAM 12C as the working area. The CPU 12A controls each of the above components and performs various calculations according to the programs recorded in ROM 12B or storage 12D.

[0048] ROM12B stores various programs and data. RAM12C temporarily stores programs or data as a working area. Storage12D is composed of an HDD (Hard Disk Drive) or SSD (Solid State Drive), etc., and stores various programs including the operating system and various data. In this embodiment, ROM12B or storage12D stores information processing programs 13 such as a program that generates training data for generating oblique CT images with reduced artifacts and trains a learning model, an imaging and reconstruction program that performs imaging and reconstruction of oblique CT images, and a trained model 15.

[0049] The input unit 12E includes a pointing device such as a mouse and a keyboard, and is used for various types of input. The display unit 12F is, for example, a liquid crystal display and displays various types of information. The display unit 12F may also function as the input unit 12E by employing a touch panel system.

[0050] The communication interface 12G is an interface for communicating with other devices such as the CT imaging device 14, and standards such as FDDI and Wi-Fi (registered trademark) can be used. In this embodiment, the information processing device 12 and the CT imaging device 14 are connected wirelessly, but they may also be connected by wire.

[0051] Next, the functional configuration of the information processing device 12 will be described. Figure 6 is a functional block diagram showing the functional configuration of the information processing device 12.

[0052] As shown in Figure 6, the information processing device 12 is configured as follows: a training data generation unit 30, a training data storage unit 32, a learning unit 34, a model storage unit 36, a reception unit 38, and an estimation unit 40. Each functional configuration is realized by the CPU 12A reading and executing a program stored in the ROM 12B or storage 12D.

[0053] The training data generation unit 30 generates training data for training a learning model to reduce oblique CT artifacts. While actual sample data can be used as training data, it is difficult to create large quantities due to the complexity of the measurement, and there is a possibility of unintended noise being introduced. Therefore, a structural model that simulates the structure of the object to be measured is created on a computer, and training data is created by performing a simulation using this structural model. In detail, the training data generation unit 30 creates multiple oblique CT image simulation data with different installation angles from the simulation data that will be the ground truth. For example, multiple oblique CT image simulation data with randomly changed installation angles are created. The advantages of the simulation data include the ability to obtain high-quality ground truth images, prepare a large amount of data, and improve generalizability by intentionally adding noise.

[0054] The training data storage unit 32 stores a large amount of training data generated by the training data generation unit 30.

[0055] The learning unit 34 constructs a convolutional neural network model as a learning model for estimating an oblique CT image with artifacts reduced, using the training data stored in the training data storage unit 32 as input to the oblique CT image. Hereinafter, the trained convolutional neural network model may be referred to as the trained model. For the learning model, for example, U-ResNet is used, which combines U-Net (Ronneberger, O.; et al. in Proc. MICCAI, 234 (2015)) and ResNet (He, K.; et al. In Proc. IEEE CVPR, 770 (2016)), which are commonly used in image processing.

[0056] Specifically, as shown in Figure 7, the learning unit 34 inputs multiple oblique CT simulation data 42 generated by the training data generation unit 30 into a convolutional neural network (CNN) 44 to generate inference data 46. It calculates the error 48 between the inference data 46 and the ground truth image and modifies the parameters of the convolutional neural network 44. This process from inference to parameter modification is considered one learning cycle, and hundreds of learning cycles are performed to create a trained model 15 with reduced artifacts.

[0057] The model storage unit 36 ​​stores the trained model learned by the learning unit 34.

[0058] The reception unit 38 receives the CT images taken by the CT imaging device 14. The CT images are oblique CT images obtained by changing the installation angle of the measurement sample 60 using the jig 20 described above.

[0059] The estimation unit 40 uses the trained model stored in the model storage unit 36 ​​to estimate an oblique CT image with artifacts reduced from the oblique CT image received by the reception unit 38.

[0060] Specifically, as shown in Figure 8, the estimation unit 40 inputs oblique CT image data 50, which is a CT image taken by the CT imaging device 14, into a convolutional neural network 44, which is a trained model 15 for reducing artifacts learned by the learning unit 34, thereby generating an oblique CT image with reduced artifacts as inference data 52.

[0061] Next, we will explain the specific processes performed by the information processing device 12 configured as described above.

[0062] The measurement device 10 according to this embodiment, configured as described above, consists of three stages: creating a learning model to reduce artifacts in oblique CT images by training a convolutional neural network; acquiring and reconstructing multiple oblique CT images with different tilt angles; and generating artifact-reduced oblique CT images using the trained model.

[0063] First, let's explain the generation of the learning model. In the information processing device 12, the CPU 12A reads the learning program, which is part of the information processing program 13, from the ROM 12B or storage 12D, loads it into the RAM 12C, and executes it, thereby performing the learning process shown in Figure 9. Figure 9 is a flowchart showing an example of the flow of the learning process.

[0064] In step 100, the CPU 12A generates training data and proceeds to step 102. Specifically, the training data generation unit 30 generates training data for training a learning model to reduce oblique CT artifacts. In particular, the training data generation unit 30 creates multiple oblique CT image simulation data sets with different installation angles from the simulation data that will be the ground truth. For example, it creates multiple oblique CT image simulation data sets with randomly changed installation angles.

[0065] In step 102, the CPU 12A stores the generated training data in the storage 12D and proceeds to step 104. That is, the training data storage unit 32 stores the large amount of training data generated by the training data generation unit 30.

[0066] In step 104, the CPU 12A learns the model and completes the series of learning processes. Specifically, the learning unit 34 constructs a convolutional neural network model for estimating an oblique CT image with artifacts reduced, based on the training data stored in the training data storage unit 32, and stores it in the model storage unit 36.

[0067] Next, we will describe the acquisition and reconstruction of multiple oblique CT images with varying tilt angles. In the information processing device 12, the CPU 12A reads the imaging and reconstruction program, which is part of the information processing program 13, from the ROM 12B or storage 12D, loads it into the RAM 12C, and executes it, thereby performing the imaging and reconstruction process shown in Figure 10. Figure 10 is a flowchart showing an example of the flow of the imaging and reconstruction process.

[0068] In step 200, the CPU 12A displays a predetermined shooting conditions setting screen on the display unit 12F and proceeds to step 202.

[0069] In step 202, the CPU 12A determines whether or not a shooting instruction has been given. This determination determines whether or not the setting of the shooting conditions has been completed and a shooting instruction has been given. The system waits until this determination is affirmative and then proceeds to step 204.

[0070] In step 204, the CPU 12A drives the rotary motor 24 to perform imaging and then proceeds to step 206. Specifically, the rotary motor 24 is driven to rotate the sample 60, and X-rays are irradiated onto the sample 60 from the radiation source 16 to capture a CT image using the radiation detector 18.

[0071] In step 206, the CPU 12A creates an image and proceeds to step 208. Specifically, it creates an image by reconstructing the image using the raw data obtained from the CT image acquisition by the radiation detector 18.

[0072] In step 208, CPU 12A stores the created image in storage 12D and proceeds to step 210.

[0073] In step 210, the CPU 12A determines whether the installation angle has been changed. This determination is made by operating the seesaw unit 22 to determine whether the installation angle of the measurement sample 60 has been changed. The system waits until this determination is confirmed and then proceeds to step 212. If an actuator or the like is provided to drive the seesaw unit 22, in step 210 the actuator is driven to change the installation angle of the measurement sample 60.

[0074] In step 212, the CPU 12A drives the rotary motor 24 to perform imaging and then proceeds to step 214. Specifically, the rotary motor 24 is driven to rotate the sample 60, and X-rays are irradiated onto the sample 60 from the radiation source 16 to capture a CT image using the radiation detector 18.

[0075] In step 214, the CPU 12A creates an image and proceeds to step 216. That is, it creates an image by reconstructing the image using the raw data obtained from the CT image acquisition by the radiation detector 18.

[0076] In step 216, the CPU 12A stores the created image in the storage 12D and completes the series of image reconstruction processes.

[0077] Next, in the information processing device 12, the CPU 12A reads the estimation program, which is part of the information processing program 13, from the ROM 12B or storage 12D, loads it into the RAM 12C, and executes it, thereby performing the estimation process shown in Figure 11. Figure 11 is a flowchart showing an example of the estimation process flow.

[0078] In step 300, the CPU 12A receives the oblique CT image captured by the CT imaging device 14 and proceeds to step 302. That is, the receiving unit 38 receives the CT image captured by the CT imaging device 14.

[0079] In step 302, the CPU 12A generates an artifact-reduced oblique CT image and completes the series of estimation processes. That is, the estimation unit 40 generates an artifact-reduced CT image by using the trained model stored in the model storage unit 36 ​​to estimate an artifact-reduced oblique CT image from the oblique CT image received by the reception unit 38.

[0080] In this way, by taking multiple oblique CT images at different installation angles relative to the sample rotation axis and inputting them into a trained model, it is possible to obtain multiple oblique CT images with different missing information, thereby obtaining oblique CT images with reduced artifacts.

[0081] Next, we will explain a specific example of generating an oblique CT image with reduced artifacts by acquiring an oblique CT image using the measurement sample 60.

[0082] As an example, we will explain the measurement of the solder material used to join the substrate and the element as the measurement sample 60 for evaluation.

[0083] The results of three-dimensional reconstruction of the acquisition data of two types of oblique CT images are shown in Figure 13 (top and center), as shown by the dotted line in Figure 12, and are cross-sectional views cut from the measurement sample 60. Figure 13 shows the results of three-dimensional reconstruction of the acquisition data of two types of oblique CT images, and an example of an oblique CT image with reduced artifacts.

[0084] The measurement sample 60 has an element 62, solder material 64, and a substrate 66. Originally, it is a sandwich-like structure with the solder material 64 sandwiched between the element 62 and the substrate 66, but as shown in the upper and center of Figure 13, the overall outline is not clear, and as indicated by the arrows, streak-like artifacts characteristic of oblique CT images are occurring in different directions.

[0085] The results of inputting this data into the pre-trained model 15, a convolutional neural network 44, are shown in the lower part of Figure 13.

[0086] As shown in the lower part of Figure 13, it can be seen that the streak-like artifacts are reduced compared to the upper and center parts of Figure 13, and the contours of the internal structure of the solder material 64 are also clarified.

[0087] In the above embodiment, a CPU was given as an example of a processor, but the term "processor" refers to a broader type of processor, including general-purpose processors (such as CPUs) and specialized processors (such as GPUs: Graphics Processing Units, ASICs: Application Specific Integrated Circuits, FPGAs: Field Programmable Gate Arrays, and programmable logic devices).

[0088] Furthermore, the operation of the processor in the above embodiments may not be performed by a single processor, but may be performed by multiple processors located in physically separate locations working together. Also, the order of the processor operations is not limited to the order described in each of the above embodiments, but may be changed as appropriate.

[0089] Furthermore, although the "system" in this embodiment is described as being composed of multiple devices as an example, it may also be composed of a single device that has some of the functions of multiple devices.

[0090] Furthermore, the processing performed by the information processing device 12 according to the above embodiment may be software-based processing, hardware-based processing, or a combination of both.

[0091] The program of this application can be provided as a program product. A program product includes any form of product for providing a program. For example, a program product includes a program provided via a network such as the Internet, and non-temporary computer-readable recording media such as CD-ROMs and DVDs on which the program is stored.

[0092] Furthermore, the present invention is not limited to the above, and it is of course possible to implement it in various modified forms without departing from its spirit.

[0093] This disclosure may adopt the following embodiments: (1) A scanning unit that captures multiple oblique CT images of the same object to be measured by varying the installation angle with respect to the sample rotation axis that rotates the object to be measured, A measuring device comprising: a generation unit that generates an oblique CT image with reduced artifacts using a convolutional neural network from the plurality of oblique CT images captured by the imaging unit;

[0094] (2) The measurement apparatus according to (1), wherein the plurality of oblique CT images are X-ray CT images in which the angle between the X-ray and the rotation axis of the sample is constant.

[0095] (3) The measurement apparatus according to (1), wherein the plurality of oblique CT images are two oblique CT images taken by changing the installation angle of the object to be measured so that it is at a constant positive and negative angle with respect to the rotation axis of the sample.

[0096] (4) The convolutional neural network is trained using the structural model of the object to be measured on a computer, and the results of simulating multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant and the installation angle is different, as training data, as described in any one of (1) to (3).

[0097] (5) The convolutional neural network is trained using the structural model of the object to be measured on a computer, and the results of simulating oblique CT images for combinations in which the installation angle is a constant positive or negative angle with respect to the rotation axis of the sample as training data, as described in any one of (1) to (3).

[0098] (6) The measuring device according to any one of (1) to (5), wherein the imaging unit has a jig capable of fixing the object to be measured in an arbitrary inclined direction with respect to the direction of the sample rotation axis.

[0099] (7) Computers For the same object to be measured, multiple oblique CT images are acquired, each taken at a different installation angle with respect to the sample rotation axis that rotates the object to be measured. A measurement method that performs a process to generate an oblique CT image with reduced artifacts using a convolutional neural network on the aforementioned multiple oblique CT images.

[0100] (8) A method for creating an oblique CT image artifact reduction model, which generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A method for creating an oblique CT image artifact reduction model, comprising creating a structural model of the object to be measured on a computer, and using the structural model to simulate multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant and the installation angle is different, and then creating training data from the results of this simulation and learning from it.

[0101] (9) A method for creating an oblique CT image artifact reduction model, which generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A method for creating an oblique CT image artifact reduction model, comprising creating a structural model of the object to be measured on a computer, and using the structural model to simulate oblique CT images for combinations where the installation angle is a constant positive or negative angle with respect to the sample rotation axis, and learning the results of this simulation as training data.

[0102] (10) A diagonal CT image artifact reduction model that generates a diagonal CT image with reduced artifacts from multiple diagonal CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A model for reducing oblique CT image artifacts is trained by creating a structural model of the object to be measured on a computer, and then using this structural model to simulate multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant, and the installation angle is different, and using these simulation results as training data.

[0103] (11) A diagonal CT image artifact reduction model that generates a diagonal CT image with reduced artifacts from multiple diagonal CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A diagonal CT image artifact reduction model is trained using the results of simulations of diagonal CT images for combinations where the installation angle is a constant positive or negative angle with respect to the sample rotation axis, which are created on a computer as a structural model of the object to be measured. [Explanation of symbols]

[0104] 10 Measuring devices 12 Information Processing Devices 13 Information Processing Programs 14 CT imaging device 15 Pre-trained models 16 Radiation source 18. Radiation detector 20 jigs 22 Seesaw section 24-speed motor 24A Rotating shaft 30 Training Data Generation Unit 34. Learning Department 38 Reception Department 40 Estimation part 42 Oblique CT Simulation Data 44 Convolutional Neural Networks 46, 52 Inference data 48 error 50 Oblique CT image data 60 measurement samples 62 elements 64 Soldering materials 66 Base

Claims

1. A shooting unit that captures multiple oblique CT images of the same object to be measured by varying the installation angle with respect to the sample rotation axis that rotates the object to be measured, A generation unit generates an oblique CT image with reduced artifacts using a convolutional neural network from the plurality of oblique CT images captured by the imaging unit, Measuring device including.

2. The measuring device according to claim 1, wherein the plurality of oblique CT images are X-ray CT images in which the angle between the X-ray and the rotation axis of the sample is constant.

3. The measuring device according to claim 1, wherein the plurality of oblique CT images are two oblique CT images taken by changing the installation angle of the object to be measured so that it is at a constant positive and negative angle with respect to the rotation axis of the sample.

4. The measurement device according to claim 1, wherein the convolutional neural network is trained using the structural model of the object to be measured on a computer, and the results of simulating multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant and the installation angle is different, as training data.

5. The measurement device according to claim 1, wherein the convolutional neural network is trained using the structural model of the object to be measured on a computer, and the results of simulating oblique CT images for combinations in which the installation angle is a constant positive or negative angle with respect to the rotation axis of the sample are used as training data.

6. The measuring device according to claim 1, wherein the imaging unit has a jig capable of fixing the object to be measured in an orientation arbitrarily inclined with respect to the direction of the sample rotation axis.

7. Computers For the same object to be measured, multiple oblique CT images are acquired, each taken at a different installation angle with respect to the sample rotation axis that rotates the object to be measured. A measurement method that performs a process to generate an oblique CT image with reduced artifacts using a convolutional neural network on the aforementioned multiple oblique CT images.

8. A method for creating an oblique CT image artifact reduction model, which generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A method for creating an oblique CT image artifact reduction model, comprising creating a structural model of the object to be measured on a computer, and using the structural model to simulate multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant and the installation angle is different, and then creating training data from the results of this simulation and learning the model.

9. A method for creating an oblique CT image artifact reduction model, which generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A method for creating an oblique CT image artifact reduction model, comprising creating a structural model of the object to be measured on a computer, and using the structural model to simulate oblique CT images for combinations where the installation angle is a constant positive or negative angle with respect to the sample rotation axis, and learning the results of such simulations as training data.

10. An oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A model for reducing oblique CT image artifacts is trained by creating a structural model of the object to be measured on a computer, and then using this structural model to simulate multiple oblique CT images in which the angle between the X-ray and the rotation axis of the sample is constant, and the installation angle is different, and using the simulation results as training data.

11. An oblique CT image artifact reduction model that generates an oblique CT image with reduced artifacts from multiple oblique CT images taken of the same object under measurement at different installation angles with respect to the sample rotation axis that rotates the object under measurement, A diagonal CT image artifact reduction model is trained using the results of simulations of diagonal CT images for combinations where the installation angle is a constant positive or negative angle with respect to the sample rotation axis, which are created on a computer as a structural model of the object to be measured.

Images