Information processing device, information processing method, and information processing program

Abstract

The information processing apparatus acquires projection data corresponding to each of at least two imaging positions in which radiation is irradiated to an object to be inspected in different directions, and specifies a defect of the object to be inspected on the basis of the projection data, attenuation information of the object to be inspected, and shape information of the object to be inspected.

Description

Technical Field

[0001] The present invention relates to an information processing device, an information processing method and an information processing program. Background Art

[0002] As one of the non-destructive inspection techniques for detecting defects in objects such as mechanical parts and structures without destroying them, there is a well-known technique for detecting defects by irradiating the object with radiation, such as X-rays. In non-destructive inspection using radiation, it is desirable to pinpoint the three-dimensional location of defects. To achieve this, it is necessary to irradiate the object with radiation from multiple different directions to acquire projection data and generate a tomographic image of the object's internal structure.

[0003] In addition, a method is known, which obtains an X-ray image of an inspection object, obtains a simulated X-ray image that is closest to the shape of the inspection object from a plurality of simulated X-ray images having different shapes of the inspection object, and detects defects of the inspection object based on the difference between the X-ray image and the simulated X-ray image (see Japanese Patent Gazette No. 2014-16239). Summary of the Invention

[0004] Technical issues to be solved by the invention

[0005] To pinpoint the three-dimensional location of defects in an inspection object using radiation, it is necessary to irradiate the object with radiation from many different directions to acquire projection data, which can be time-consuming. Furthermore, depending on the inspection object, projection directions may be limited, making pinpointing the three-dimensional location of defects difficult. Furthermore, the technique described in Japanese Patent Application Laid-Open No. 2014-16239 is a method for detecting defects in projection data of an inspection object and cannot pinpoint the three-dimensional location of the defect.

[0006] The present invention provides an information processing device, an information processing method and an information processing program, which can identify defects with high precision and three-dimensional position based on projection data even with a small amount of projection data or projection data with limited projection direction, regardless of the state of the defect, etc.

[0007] Means for solving technical problems

[0008] The first embodiment of the present invention is an information processing device comprising at least one processor, which obtains projection data corresponding to at least two photographic positions having different irradiation directions of radiation on an inspection object, and identifies defects of the inspection object based on the projection data, attenuation information of the inspection object, and shape information of the inspection object.

[0009] An information processing apparatus according to a second aspect of the present invention is the information processing apparatus according to the first aspect, wherein the attenuation information may include attenuation information corresponding to at least two components included in the inspection object.

[0010] An information processing device according to a third aspect of the present invention is the information processing device according to the first aspect, wherein the attenuation information may include attenuation information of air.

[0011] An information processing device according to a fourth aspect of the present invention is the information processing device according to the first aspect, wherein the attenuation information may include attenuation information in at least two energy bands.

[0012] An information processing device according to a fifth aspect of the present invention is the information processing device according to the first aspect, wherein the processor acquires attenuation information of the defect and specifies the defect based on the attenuation information of the defect.

[0013] The information processing device of the sixth mode of the present invention is based on the information processing device of the first mode, and can also be that the processor distinguishes between areas where parts of the inspection object do not exist and areas where parts of the inspection object exist in the distribution of the inspection object in three-dimensional space based on shape information to identify defects.

[0014] The information processing device of the seventh embodiment of the present invention is based on the information processing device of the first embodiment, and can also be that the processor generates temporary projection data obtained when the inspection object is irradiated with radiation based on the attenuation information and shape information of the inspection object, and specifies defects based on the temporary projection data.

[0015] An information processing apparatus according to an eighth aspect of the present invention is the information processing apparatus according to the seventh aspect, wherein the processor may identify the defect by distinguishing data having a different magnitude relationship from the temporary projection data in the projection data.

[0016] An information processing device according to a ninth aspect of the present invention is the information processing device according to the seventh aspect, wherein the processor may identify the defect by distinguishing data having a difference from the temporary projection data in the projection data.

[0017] An information processing device according to a tenth aspect of the present invention is the information processing device according to the first aspect, wherein the processor may identify the defect based on information on scattering of the inspection object.

[0018] In addition, the eleventh embodiment of the present invention is an information processing method, in which a processor performs the following processing: obtaining projection data corresponding to at least two photographic positions that are different in the irradiation direction of radiation on the inspection object, and identifying defects of the inspection object based on the projection data, attenuation information of the inspection object, and shape information of the inspection object.

[0019] In addition, the twelfth embodiment of the present invention is an information processing program that enables a processor to perform the following processing: obtaining projection data corresponding to at least two photographic positions that are different in the irradiation direction of radiation on the inspection object, and identifying defects of the inspection object based on the projection data, attenuation information of the inspection object, and shape information of the inspection object.

[0020] Effects of the Invention

[0021] According to the above method, the information processing device, information processing method and information processing program of the present invention can identify defects with high precision together with the three-dimensional position based on the projection data, regardless of the state of the defect, even if the projection data is small in number or has a limited projection direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a structural diagram showing an example of the overall structure of a defect identification system according to an exemplary embodiment.

[0023] Figure 2 This is a perspective view showing an example of an inspection object and defects.

[0024] Figure 3 This is a diagram for explaining shape information.

[0025] Figure 4 This is a block diagram showing an example of the configuration of an information processing device.

[0026] Figure 5 This is a flowchart showing an example of a process for identifying a defect of an inspection object executed by an information processing device.

[0027] Figure 6 This is a functional block diagram showing an example of a configuration related to photographing position identification processing of an information processing device.

[0028] Figure 7 This is a flowchart showing an example of imaging position identification processing executed by the information processing device.

[0029] Figure 8A This is a diagram for explaining parameters required to specify an imaging position relative to an inspection object.

[0030] Figure 8B This is a diagram for explaining the coordinate system of the inspection object.

[0031] Figure 8C This is a diagram for explaining the coordinate system for photography.

[0032] Figure 9This is a diagram for explaining an example of line attenuation information.

[0033] Figure 10A This is a diagram for explaining the setting of the path of radiation irradiated from a radiation source.

[0034] Figure 10B This is a diagram for explaining the path of radiation passing through a voxel.

[0035] Figure 11 This is a functional block diagram showing an example of a configuration related to defect identification processing of an information processing device.

[0036] Figure 12 This is a flowchart showing an example of defect identification processing executed by the information processing apparatus according to the first exemplary embodiment.

[0037] Figure 13 This is a diagram for explaining the setting of subset numbers and projection data numbers.

[0038] Figure 14 This is a flowchart showing an example of defect identification processing executed by the information processing apparatus according to the second exemplary embodiment.

[0039] Figure 15 It is a diagram for explaining the scattering angle. DETAILED DESCRIPTION

[0040] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, these exemplary embodiments do not limit the present invention.

[0041] First, an example of the overall structure of the defect identification system 1 according to this exemplary embodiment will be described. Figure 1 1 is a block diagram showing an example of the overall structure of the defect specific system 1 according to this exemplary embodiment. Figure 1 As shown in FIG. 1 , the defect identification system 1 of this exemplary embodiment includes an information processing device 10, an inspection object 60, a radiation source 64, and a detector 68. Figure 1 Although one detector 68 is shown in FIG, the number of detectors 68 included in the defect identification system 1 is not limited. For example, a detector 68 may be provided in each irradiation direction of the radiation R. Alternatively, a plurality of detectors 68 may be arranged in a row to obtain a single projection data set from the plurality of detectors 68.

[0042] The information processing apparatus 10 of this exemplary embodiment uses at least two projection data obtained by the detector 68 by irradiating the inspection object 60 with radiation R from different directions by the radiation source 64 to identify the defect 61 of the inspection object 60. Figure 2As shown in FIG, the information processing apparatus 10 of this exemplary embodiment specifies that the defect 61 exists inside the inspection object 60. The information processing apparatus 10 of this exemplary embodiment specifies the three-dimensional position of the defect 61 inside the inspection object 60. Figure 2 In FIG. 6 , a cylindrical inspection object is shown as the inspection object 60 . However, the shape of the inspection object 60 is not limited to the cylindrical shape, and may be any shape.

[0043] Specifically, the information processing device 10 generates virtual temporary projection data corresponding to the projection data obtained by the detector 68 when the inspection object 60 is irradiated with radiation R from at least two directions, based on the attenuation information and shape information of the inspection object 60. The projection data is then matched with the actual measurement data, i.e., the projection data, to identify the three-dimensional position of the defect 61 in the inspection object 60. Hereinafter, the term "projection data" refers to the actual measurement data obtained by the detector 68.

[0044] Here, attenuation information refers to information about the linear attenuation coefficient of components included in the inspection object 60 when the inspection object 60 is free of defects. The linear attenuation coefficient refers to the probability of radiation R interacting with the inspection object 60 per unit distance traveled, and is the ratio of decrease in the intensity or photon count of the radiation R. The linear attenuation coefficient varies depending on the energy of the radiation. The radiation R irradiated from the radiation source 64 has an energy distribution. Therefore, the attenuation information preferably includes information about the linear attenuation coefficients in at least two energy bands of the radiation R irradiated onto the inspection object 60.

[0045] Projection data is the sum of the energies of radiation R in multiple energy bands incident on each detector 68. The information processing device 10 of this exemplary embodiment effectively utilizes information on the linear attenuation coefficients of the inspection object 60 in the multiple energy bands of radiation R irradiated onto the inspection object 60, enabling high-precision identification of the imaging position, described in detail below. Furthermore, the information processing device 10 can generate a defect ratio tomographic image, described in detail below, with high precision. Consequently, the information processing device 10 of this exemplary embodiment enables high-precision identification of defects 61, including their three-dimensional positions.

[0046] When the radiation R passes through the inspection object 60, it is attenuated due to interactions with the inspection object 60 such as the photoelectric effect, interferential scattering (Thomson scattering), non-interferential scattering (Compton scattering) and electron pair generation, and a portion of the attenuated radiation R is generated as new radiation (scattered radiation).

[0047] The detector 68 detects the radiation R (hereinafter referred to as direct radiation) irradiated from the radiation source 64 and passing through the inspection object 60, and also detects the newly generated (scattered) radiation due to the above-mentioned interaction. That is, in addition to the data based on the direct radiation, the projection data obtained by the detector 68 also includes data based on the scattered radiation. In other words, the projection data obtained by the detector 68 is affected by the scattered radiation. Therefore, in order to reduce the influence of the scattered radiation included in the projection data, it is preferred that the attenuation information include the scattered radiation information. As for the probability and angle of occurrence of scattering, it can be calculated in theory. For example, for Compton scattering, it can be calculated by the Klein-Nishina formula (Klein-Nishina model). However, the probability and angle of occurrence of the scattering vary depending on the type of components included in the inspection object 60. Therefore, the attenuation information preferably includes the scattering information of each component included in the inspection object 60.

[0048] In addition, the probability and angle of occurrence of scattering also vary depending on the energy of the irradiated radiation R. Therefore, the attenuation information preferably includes scattering information in at least two energy bands. In the case of Compton scattering, the energy after scattering varies according to the scattering angle. In order to be able to calculate multiple scatterings (second scattering, third scattering, ...) that are further scattered after such energy changes, the attenuation information preferably includes scattering information in at least two energy bands. By effectively utilizing the scattering information of the inspection object 60, the information processing device 10 can accurately identify the photographic position described in detail later. In addition, the information processing device 10 can accurately generate a defect ratio tomographic image described in detail later. As a result, according to the information processing device 10 of this exemplary embodiment, the defect 61 can be identified with high precision, including the three-dimensional position.

[0049] Shape information refers to information on the distribution of each component included in the inspection object 60 in the three-dimensional space when there is no defect 61 in the inspection object 60. In addition, an arbitrary area that includes the entire inspection object 60 can also be set, and the area can be used as the inspection object 60. In this case, inside the area set in a manner to include the original inspection object 60, there is a part that is not included in the original inspection object 60. In this part, there is a substance corresponding to the environment in which the original inspection object 60 exists. Therefore, when setting the area that includes the original inspection object 60 and using the area as the inspection object 60, it is sufficient to include the substance corresponding to the environment in which the original inspection object 60 exists as a component of the inspection object 60, and also include information on the distribution of the substance in the three-dimensional space in the shape information. Since there is usually air around the original inspection object 60, it is sufficient to include air as a component of the inspection object 60, and also include information on the distribution of air in the three-dimensional space in the shape information. For example, as Figure 3 As shown, an arbitrary rectangular parallelepiped 62 including the entire inspection object 60 can be set and the set rectangular parallelepiped 62 can be used as the inspection object 60. In this case, air can be included in the components of the inspection object 60, and information on the distribution of air in three-dimensional space can also be included in the shape information. In this exemplary embodiment, as Figure 3 As shown, a rectangular parallelepiped 62 including the inspection object 60 is set, and defects are identified using the set rectangular parallelepiped 62 as the inspection object 60 .

[0050] By combining attenuation information and shape information, it is possible to obtain information on the three-dimensional distribution of the linear attenuation coefficient (in at least two energy bands) (and the distribution of the scattering probability in at least two energy bands) of the inspection object 60. Furthermore, in this exemplary embodiment, information on the three-dimensional distribution of the linear attenuation coefficient (and the scattering probability) of the inspection object 60 is used to specify the imaging position and generate a defect ratio tomographic image, as described in detail later. However, the attenuation information and shape information may be any form of information as long as information on the three-dimensional distribution of the linear attenuation coefficient (and the scattering probability) can be obtained. For example, information on the three-dimensional distribution of the linear attenuation coefficient (and the scattering probability) may be used.

[0051] In this exemplary embodiment, when a defect 61 is present in the inspection object 60, information on the linear attenuation coefficient of the defect 61 can also be effectively utilized. In this case, the information on the linear attenuation coefficient of the defect 61 preferably includes information on the linear attenuation coefficients in at least two energy bands of the radiation R irradiating the inspection object 60. By effectively utilizing this information on the linear attenuation coefficient of the defect (in at least two energy bands), a defect ratio tomographic image, described in detail later, can be generated with high precision. As a result, the information processing device 10 of this exemplary embodiment can accurately identify the defect 61, including its three-dimensional position.

[0052] Hereinafter, the identification of the defect 61 of the inspection object 60 by the information processing device 10 of the defect identification system 1 according to this exemplary embodiment will be described in detail for each exemplary embodiment.

[0053] [First exemplary embodiment]

[0054] (Structure of Information Processing Device)

[0055] First, the hardware structure of the information processing device 10 of this exemplary embodiment will be described. Figure 4 1 is a block diagram showing an example of the hardware configuration of the information processing device 10 according to this exemplary embodiment. Figure 4 As shown, the information processing device 10 includes a processor 20 such as a CPU (Central Processing Unit), a memory 22, an interface (Interface) unit 23, a storage unit 24, a display 26, and an input device 28. The processor 20, the memory 22, the interface unit 23, the storage unit 24, the display 26, and the input device 28 are connected via a bus 29 such as a system bus or a control bus so that they can exchange various information with each other.

[0056] The processor 20 reads various programs stored in the storage unit 24, including the imaging position identification program 30 and the defect identification program 32, into the memory 22, and executes processing according to the read programs. Thus, the processor 20 performs control related to the identification of the defect 61. The memory 22 is a working memory used by the processor 20 to execute processing.

[0057] The imaging position identifying program 30 and the defect identifying program 32 executed by the processor 20 are stored in the storage unit 24. Specific examples of the storage unit 24 include a HDD (Hard Disk Drive) and an SSD (Solid State Drive).

[0058] The I / F unit 23 communicates various information with the detector 68 through wireless communication or wired communication. The display 26 and the input device 28 function as a user interface. The display 26 provides the user with various information related to the analysis of the sample. The display 26 is not particularly limited, and examples thereof include a liquid crystal display and an LED (Light Emitting Diode) display. In addition, the input device 28 is operated by the user to input various instructions related to the specific aspects of the defect 61. The input device 28 is not particularly limited, and examples thereof include a keyboard, a stylus pen, and a mouse. In addition, in the information processing device 10, a touch panel display that integrates the display 26 and the input device 28 is used.

[0059] (Overview of overall processing)

[0060] Next, the overall flow of the process for identifying the defect 61 of the inspection target 60 by the information processing apparatus 10 of this exemplary embodiment will be described. Figure 5 , a flowchart showing an example of a process for specifying a defect 61 of an inspection object 60 executed by the information processing device 10 is shown. Figure 5 The processing shown is executed when a defect detection instruction is received from the user via the input device 28, for example.

[0061] exist Figure 5 In step S10, the processor 20 performs the imaging position identification process described in detail later to identify the imaging position. In this exemplary embodiment, the imaging position refers to the position and posture of the imaging coordinate system. Specifically, the imaging position refers to the position (three-dimensional position) of the radiation source 64 relative to the inspection object 60 and the irradiation direction of the radiation R. In this exemplary embodiment, since the imaging position is unknown, the imaging position is first identified in step S10. If the imaging position is already known, the process of step S10 can be omitted.

[0062] In the next step S12 , the processor 20 executes a defect identification process, which will be described in detail later, to identify the defect 61 .

[0063] In this manner, the information processing apparatus 10 of this exemplary embodiment specifies the defect 61 through two steps: the imaging position specifying process and the defect specifying process. Detailed descriptions of each process will be given below.

[0064] (Photography location specific processing)

[0065] The above-mentioned photographing position specifying process will be described.

[0066] First, the structure of the information processing device 10 related to the photographing position identification process will be described. Figure 6FIG. 1 is a functional block diagram showing an example of the structure of the information processing device 10 related to the photographing position identification process. Figure 6 As shown, the information processing device 10 includes a first acquisition unit 40 and a photographing position specifying unit 42. When the processor 20 executes the photographing position specifying program 30, the processor 20 functions as the first acquisition unit 40 and the photographing position specifying unit 42.

[0067] The first acquisition unit 40 has a function of acquiring a plurality of projection data obtained by the detector 68 when the inspection object 60 is irradiated with radiation R from different directions by the radiation source 64. For example, in the information processing apparatus 10 of this exemplary embodiment, after imaging the inspection object 60, the first acquisition unit 40 acquires a plurality of projection data corresponding to different irradiation directions of the radiation R from the detector 68 at arbitrary timings and stores the data in the storage unit 24. Furthermore, when performing imaging position identification processing, the first acquisition unit 40 acquires the plurality of projection data corresponding to different irradiation directions of the radiation R stored in the storage unit 24 and outputs the data to the imaging position identification unit 42.

[0068] like Figure 6 As shown, the imaging position specifying unit 42 includes a first temporary projection data generating unit 44. The first temporary projection data generating unit 44 has the function of generating temporary projection data obtained when the inspection object 60 is irradiated with radiation R, based on the attenuation information and shape information of the inspection object 60. In other words, the first temporary projection data generating unit 44 generates virtual temporary projection data of the inspection object 60 corresponding to each of the plurality of projection data acquired by the first acquiring unit 40 through simulation based on the attenuation information and shape information of the inspection object 60. Furthermore, the first temporary projection data generating unit 44 generates temporary projection data for each of the imaging positions used in the simulation (hereinafter referred to as temporary imaging positions) by varying the temporary imaging positions.

[0069] The imaging position specifying unit 42 has the function of specifying the imaging position of each projection data item based on the plurality of projection data items acquired by the first acquisition unit 40 and the plurality of temporary projection data items generated by the first temporary projection data generating unit 44 corresponding to each of the plurality of projection data items. Specifically, the imaging position specifying unit 42 specifies, as the actual imaging position, the temporary imaging position corresponding to the temporary projection data item for which the similarity between the temporary projection data item and the projection data item exceeds a threshold value for each of the plurality of projection data items.

[0070] Next, the operation related to the photographing position identification process performed by the information processing device 10 of this exemplary embodiment will be described in detail. Figure 7 2 is a flowchart showing an example of the imaging position specifying process executed by the information processing device 10 according to this exemplary embodiment.

[0071] exist Figure 7 In step S100 , as described above, the first acquisition unit 40 acquires a plurality of projection data from the storage unit 24 and outputs the acquired data to the imaging position identification unit 42 .

[0072] In the next step S102 , the photographing position specifying unit 42 sets an initial temporary photographing position.

[0073] Here, the parameters required to specify the imaging position relative to the inspection object 60 are considered (see Figure 8A ).like Figure 8B As shown, the coordinate system of the inspection object 60 is set as Figure 8C As shown in FIG, the coordinate system of the photographing is set. In addition, in this exemplary embodiment, the projection data is as follows Figures 8A to 8C Although the case of projection data captured by the cone beam method as shown is described, projection data captured by any other scanning method may also be used, for example, projection data captured by a pencil beam method or a fan beam method.

[0074] First, by making Figure 8B The coordinate system of the inspection object 60 is moved parallel to the x-axis, y-axis, and z-axis by Δx_p, Δy_p, and Δz_p, respectively, so that the origin of the coordinate system is aligned with Figure 8C The position of the radiation source 64 shown is consistent. That is, any position of the imaging system ( Figure 8C Any position in the coordinate system of the photograph) can be obtained by using Figure 8B The coordinate system of the inspection object 60 is expressed by parallel displacement of Δx_p, Δy_p, and Δz_p.

[0075] Next, any pose of the camera system ( Figure 8C Any pose of the coordinate system of the photograph) can be obtained by parallel translation Figure 8B The coordinate system of the inspection object 60 is represented by rotating the coordinate system around the x-axis, y-axis, and z-axis by angles θ_x_p, θ_y_p, and θ_z_p, respectively. Specifically, considering the order of rotation around the three axes, any posture of the imaging system can be represented by first rotating around the z-axis by angle θ_z_p, then around the y-axis (of the coordinate system rotated around the z-axis) by angle θ_y_p, and finally around the x-axis (of the coordinate system rotated around the y-axis) by angle θ_x_p.

[0076] In this way, the three-dimensional position and posture of the imaging system relative to the inspection object 60 can be determined by the six parameters of the parallel movement Δx_p, Δy_p, Δz_p and the rotation angles θ_x_p, θ_y_p, θ_z_p. These six parameters representing the imaging position are set as unknown numbers, and the initial temporary imaging position is set. For example, the rotation angles θ_x_p, θ_y_p, and θ_z_p can all be set to "0" and the parallel movement Δx_p and Δy_p can also be set to "0", and only the parallel movement Δz_p is set as the reference value. In addition, here, the distance D_sd_p from the radiation source 64 to the detector 68 (refer to Figure 8C ) is known, but the seven parameters including the distance D_sd_p can also be set as unknowns. In this case, the initial value of the distance D_sd_p can also be set as the reference value.

[0077] In this exemplary embodiment, for ease of handling, the coordinate system of the inspection object 60 is fixed, and the positions and postures of the coordinate systems of the images of each of the plurality of projection data are expressed by parallel translation or rotation.

[0078] Next, in step S104, the first temporary projection data generator 44 generates temporary projection data. Specifically, the first temporary projection data generator 44 generates temporary projection data by simulation based on attenuation information and shape information of the inspection object 60 under the conditions of the set temporary imaging position.

[0079] First, the first provisional projection data generator 44 obtains the three-dimensional linear attenuation coefficient distribution of the inspection object 60 based on the attenuation information and shape information of the inspection object 60. As described above, the attenuation information and shape information may be any type of information as long as the three-dimensional linear attenuation coefficient distribution can be obtained. For example, the three-dimensional linear attenuation coefficient distribution itself may be used. In this exemplary embodiment, the attenuation information is denoted as f(no_f, no_e), and the shape information is denoted as no_f(no_pix). The three-dimensional linear attenuation coefficient distribution of the inspection object 60 is represented by the combination of the attenuation information f(no_f, no_e) and the shape information no_f(no_pix). Here, "no_f" represents the identification number of a component included in the inspection object 60, and a component with the identification number no_f is labeled "component no_f." Furthermore, "no_e" represents the identification number of the energy (equivalent to the wavelength) of the radiation R, and the energy with the identification number no_e is labeled "energy no_e." In addition, the attenuation information f (no_f, no_e) indicates the linear attenuation coefficient of the energy no_e of the radiation R with respect to the component no_f. On the other hand, the shape information no_f (no_pix) indicates the component no_f in the voxel no_pix. The total number of voxels of the inspection object 60 is num_pix_t, and the voxel no_pix is ​​an arbitrary value between 1 and num_pix_t. In this exemplary embodiment, as described above, the inspection object 60 is a rectangular parallelepiped (see Figure 3 Therefore, within the inspection object 60, a rectangular region formed by fine grids (voxels) is assigned numbers 1 to num_pix_t. Furthermore, the position of each voxel is originally assigned three-dimensionally using x, y, and z coordinates. However, as the number of dimensions increases, reference becomes complicated. Therefore, in this exemplary embodiment, each voxel is assigned a one-dimensional number.

[0080] exist Figure 9 An example of the linear attenuation coefficient f (no_f, no_e) is shown in FIG. Figure 9 The document contains three graphs showing the linear attenuation coefficients of air, aluminum, and iron, from smallest to smallest. The linear attenuation coefficient values ​​are based on "Tables of X-Ray Mass Attenuation Coefficients and Mass Energy—Absorption Coefficients from 1keV to 20MeV for Elements Z=1 to 92 and 48 Additional Substances of Dosimetric Interest, NIST, J.H. Hubbell and SMSeltzer."

[0081] Assume that the component numbers no_f for air, aluminum, and iron are 1, 2, and 3, respectively. The component of each voxel of inspection object 60 is any of these three. Alternatively, the linear attenuation coefficient value for air can be approximately set to 0.0 in the total energy. In other words, a component no_f with a linear attenuation coefficient value of 0.0 in the total energy is considered to be air.

[0082] Then, if Figure 10A As shown in FIG. 7 , the path 70 of the radiation R is set. Figure 10A As shown, the radiation source 64 is not a point but is spatially extended. That is, in order to ensure the dose of the radiation R, that is, to ensure the brightness or shading difference of the projection data, the radiation source 64 has a certain size. Moreover, due to the extension of the radiation source 64, blurring occurs in the projection data. In order to accurately simulate the projection data including such blurring and generate temporary projection data, it is preferable to set the path 70 of the radiation R based on the spatial extension of the radiation source 64. Figure 10A In the embodiment, a plurality of radiation sources (radiation sources 64) are set at predetermined three-dimensional intervals within a predetermined region (radiation source 64) extending in a three-dimensional space. Figure 10A There are 28 radiation sources 64S in the example, and a path 70 of the radiation R starting from each radiation source 64S is set.

[0083] like Figure 10A As shown, the paths 70 of the radiation R are set at equal intervals of small angles along two orthogonal directions from the radiation source 64. Figure 10B As shown, the first temporary projection data generating unit 44 specifies the voxels 74 that each path 70 passes through. Figure 10B In the example shown, the path 70 passes through four voxels 74. For each voxel 74, the intensity of the radiation R emitted from the radiation source 64S is attenuated and incident on the detector 68 based on the linear attenuation coefficient and the path length L passed through. Then, the first temporary projection data generating unit 44 accumulates the intensity of the radiation R incident on the detector 68 for each path 70 for each detector 68 to generate temporary projection data. Figure 10B In FIG, the path 70 of the radiation R in the three-dimensional space and the three-dimensional voxel 74 are simply represented two-dimensionally. Figure 10A In the figure, the angles in the two directions are set to θ and φ respectively, and their combination (θ, φ) represents an arbitrary direction. When each small angle is set to Δθ and Δφ, each path 70 is set to the following direction.

[0084] ..., (-Δθ, -2×Δφ / cos(Δθ)), (-Δθ, -Δφ / cos(Δθ)), (-Δθ, 0), (-Δθ, +Δφ / cos(Δθ)), (-Δθ, +2×Δφ / cos(Δθ)), ……

[0085] ……, (0,-2×Δφ), (0,-Δφ), (0,0), (0,+Δφ), (0,+2×Δφ),…

[0086] ..., (+Δθ, -2×Δφ / cos(Δθ)), (+Δθ, -Δφ / cos(Δθ)), (+Δθ, 0), (+Δθ, +Δφ / cos(Δθ)), (+Δθ, +2×Δφ / cos(Δθ)), ……

[0087] In addition, in order to make the range of the small solid angle centered on each path 70 equal, the small angle Δφ in the angle θ in the above formula is corrected to Δφ / cos(θ) (the range of θ is set to -90 degrees to +90 degrees, and the correction is made in such a way that Δφ increases as the absolute value of θ increases). Figure 10A In the example, setting more paths 70, in other words, setting the angles Δθ and Δφ smaller, allows for more accurate simulation of the provisional projection data, but this increases computational time. The intensity I of the radiation R incident on the detector 68 for each path 70 is obtained using the following equation (1).

[0088] I=(e(1)×I_e(1))+(e(2)×I_e(2))+……+(e(num_e)×I_e(num_e))…(1)

[0089] In the above formula (1), e(1), e(2), ..., e(num_e) represent energies, which are known. As shown in formula (1), the intensity I is the sum of the energies of the num_e types of photons incident on the detector.

[0090] In the above formula (1), the intensity I_e(no_e) of the radiation R at the energy e(no_e) is obtained by the following formula (2).

[0091] I_e(no_e)=I0_e(no_e)×exp(-{f(no_f(no_pix(1)), no_e)×plen_pix(1))+(f(no_f(no_pix( 2)), no_e)×plen_pix(2))+…+(f(no_f(no_pix(num_pix)), no_e)×plen_pix(num_pix)})…(2)

[0092] The above formula (2) represents the intensity I_e(no_e) of radiation R of intensity I0_e(no_e) among the energy e(no_e) emitted from the radiation source 64S, which passes through each voxel 74 while attenuating and enters the detector 68 .

[0093] In the above formula (2), f(no_f(no_pix(no_pix_path)), no_e) represents the linear attenuation coefficient f in the energy no_e of the component no_f in the no_pix_path voxel no_pix through which the path 70 passes, and plen_pix(no_pix_path) represents the path length plen_pix in the no_pix_path voxel no_pix through which the path 70 passes.

[0094] The intensity I0_e(no_e) in the above formula (2) represents the Figure 10A The total number of photons (of energy e(no_e)) emitted per unit time in each path 70 within a small angular range of -Δθ / 2 to +Δθ / 2 and -Δφ / cos(θ) / 2 to +Δφ / cos(θ) / 2 in two directions centered on the path 70 (a range where the solid angles of each path 70 are equal). Figure 10A The narrower the interval of the angle of the path 70 is set, in other words, the more paths 70 are set, the smaller the value of the intensity I0_e(no_e) is set. I0_e(0), I0_e(1), ..., I0_e(num_e) are determined by the radiation source 64 and are known.

[0095] The intensity of the radiation R incident on the detector 68 of each path 70 is calculated by the above formula (1), and the intensity is accumulated for each detector 68. Thus, the first temporary projection data generating unit 44 can generate temporary projection data. Figure 8C The voxel 74 through which each path 70 passes and the path length L in the voxel 74 can be simply obtained by calculating the path length L in the coordinate system of the photography shown in FIG. Figure 8B The direction of the path 70 and the position of the radiation source 64S as the starting point in the coordinate system of the inspection object 60 are simply obtained. Figure 8B The direction of the path 70 in the coordinate system of the inspection object 60 can be determined by making Figure 8C The direction of the path 70 in the coordinate system of the photograph is obtained by reverse rotation based on the rotation angles θ_x_p, θ_y_p, and θ_z_p. Figure 8B The position of the radiation source 64S as the starting point of the path 70 in the coordinate system of the inspection object 60 can be adjusted so that Figure 8CThe position of the radiation source 64S in the imaging coordinate system is obtained by reverse rotation based on the rotation angles θ_x_p, θ_y_p, and θ_z_p, followed by parallel translation by Δx_p, Δy_p, and Δz_p.

[0096] Thus, the first temporary projection data generator 44 generates temporary projection data. The temporary projection data is data obtained by simulating the projection data of the inspection object 60 without the defect 61. Furthermore, in the case of the cone beam method, the detector 68 is arranged two-dimensionally. Therefore, the temporary projection data generated by the first temporary projection data generator 44 is two-dimensional data.

[0097] In the next step S106, the photographic position specifying unit 42 calculates the similarity between the measured projection data and the temporary projection data. As described above, the temporary projection data is data obtained by simulating the projection data of the inspection object 60 without the defect 61. On the other hand, sometimes the inspection object 60 in the measured projection data includes a defect 61. In the case where the defect 61 is included, the projection data and the temporary projection data are not completely consistent. However, even in the case where the defect 61 is included, the area occupied by it is only a small part of the inspection object 60, and the effect on the overall similarity of the projection data of the two is very small. In comparison, the difference between the photographic positions of the actual measurement and the simulation has a greater impact on the similarity of the projection data of the two. Therefore, by making the similarity between the projection data and the temporary projection data greater than a threshold value, the photographic positions of the actual measurement and the simulation can be made consistent. As an index of the similarity of two data (an index indicating the degree of similarity), various well-known indices are known. For example, known similarity indices include the sum of squared differences (SSD), the sum of absolute differences (SAD), normalized cross-correlation (NCC), zero-mean normalized cross-correlation (ZNCC), image uniformity ratio, and mutual information. Any known similarity indices may also be used.

[0098] Furthermore, particularly in the case of a cone-beam method, the projection data includes not only direct radiation emitted from the radiation source 64 and attenuated by passing through the inspection object 60, but also scattered radiation generated as new radiation from a portion of the attenuated direct radiation as the direct radiation attenuates. Therefore, it is preferable to generate provisional projection data that also simulates scattered radiation in step S104, and to calculate the similarity between the projection data and the provisional projection data that account for scattered radiation. However, since the calculation of scattered radiation is computationally intensive and time-consuming, the imaging position identification process can also omit the scattered radiation simulation and simply remove the low-frequency components from the spatial frequency components included in the measured projection data. The low-frequency components are also removed from the provisional projection data, and the similarity between the two is calculated. Since the image based on scattered radiation is added to the projection data as an image that varies spatially at a low frequency, removing the low-frequency components from the spatial frequency components included in the projection data allows for the appropriate removal of the image based on scattered radiation. Furthermore, even when low-frequency components are removed, edges that significantly contribute to similarity (portions where projection data values ​​vary dramatically spatially, such as the boundary between the inspection object 60 and the background, or boundaries within the inspection object 60) can be retained. Removal of low-frequency components from projection data can be achieved by converting the projection data into data in the frequency domain using a Fourier transform, cutting off low-frequency components that are significantly affected by scattered radiation, and restoring the original projection data using an inverse Fourier transform. Alternatively, even without performing a Fourier transform, in other words, without converting data into the frequency domain, removal can be achieved by convolving the projection data with a transmission filter that applies a mid- to high-frequency component.

[0099] If the derived similarity is less than or equal to the threshold (similarity ≤ threshold), the determination in step S108 becomes negative, and the process proceeds to step S110. This is because the set temporary photography position is not suitable as the actual photography position.

[0100] In step S110, the imaging position specifying unit 42 updates the set temporary imaging position. Specifically, the imaging position specifying unit 42 updates the six parameters representing the temporary imaging position: the translation amounts Δx_p, Δy_p, and Δz_p, and the rotation angles θ_x_p, θ_y_p, and θ_z_p. Furthermore, if the distance D_sd_p from the radiation source 64S to the detector 68 is also unknown, the seven parameters including this distance are updated. Upon completion of step S110, the process returns to step S104, and steps S104 through S108 are repeated.

[0101] On the other hand, in step S108 , when the similarity is greater than the threshold value, the determination is affirmative, and the process proceeds to step S112 .

[0102] In step S112, the photographing position specifying unit 42 specifies the set temporary photographing position as the actual photographing position. Figure 7 The photographing position specifying process shown is completed.

[0103] In this way, the operation of repeatedly calculating the evaluation value (similarity in this exemplary embodiment) while updating the unknown parameters and the method of exploring the parameter whose evaluation value becomes the minimum or maximum (maximum in this exemplary embodiment) is called an optimization algorithm or an exploration algorithm, and various known methods are known. The six (or seven) parameters can also be updated using any known method. For example, in the case of updating by the gradient method, first, the six (or seven) parameters are increased by a small amount, the similarity between the projection data and the temporary projection data is calculated, and the difference (change) between the similarity before the increase is obtained. The ratio of the change in similarity obtained for each parameter to the small increase in the parameter is equivalent to the slope (partial differential). Then, based on the slope (partial differential) of the similarity for each parameter, each parameter is updated in the direction of increasing similarity. In addition, since the units are different between the parameters of parallel translation and distance and the parameters of rotation, the step width of the update can also be changed. In the case of updating by the gradient method, there is a possibility of converging to a local solution. To eliminate this possibility, the following procedure can be used: First, for each parameter, a predetermined number of candidate values ​​are set within a possible range. Similarities are calculated for all combinations of candidate values ​​for each parameter to find the combination with the highest similarity. This combination is used as the initial value, and a gradient method is used to find parameters with a similarity greater than a threshold. Alternatively, a genetic algorithm can be used to find parameters with a similarity greater than a threshold. Alternatively, any other optimization algorithm can be used to find parameters with a similarity greater than a threshold.

[0104] By photographing position specific processing (refer to Figure 7) The photographic position of each projection data corresponding to at least two photographic positions that are different from the irradiation direction of the radiation R is specified. When the relationship between the photographic positions of the projection data is known, that is, when the relationship between the position and posture of the coordinate system of the photographic images of the projection data is known, it is sufficient to specify the photographic position of only one projection data relative to the coordinate system of the inspection object 60 through the photographic position specifying process. Specifically, in the photographic position specifying process, by setting an initial value of the photographic position of one projection data, repeatedly calculating the similarity between the projection data and the temporary projection data and updating the temporary photographic position, it is sufficient to obtain a temporary photographic position whose similarity is greater than a threshold. In this case, as long as the photographic position of one projection data (six parameters of parallel movement amounts Δx_p, Δy_p, Δz_p and rotation angles θ_x_p, θ_y_p, θ_z_p) is set relative to the inspection object 60, the photographic positions (six parameters) of all other projection data are determined because the relationship between the photographic position of one projection data and the photographic positions of all other projection data is known. Therefore, in step S104 of the imaging position identification process, provisional projection data corresponding to the projection data in all directions is generated. Furthermore, in step S106, the sum of the similarities between the projection data in all directions and the provisional projection data is calculated. Furthermore, in step S110, the provisional imaging position (parameter) is updated in a direction where the sum of the similarities between the projection data in all directions and the provisional projection data increases. As a result, the imaging position identification process determines a parameter such that the sum of the similarities between the projection data in all directions and the provisional projection data exceeds a threshold value.

[0105] On the other hand, when the relationship between the imaging positions of the projection data is unknown, the imaging position identification process (see Figure 7 ), for each projection data, the parameters (six or seven) whose similarity between the projection data and the temporary projection data is greater than a threshold value are obtained.

[0106] By the imaging position identification processing described above, the imaging position of each projection data corresponding to at least two imaging positions having different irradiation directions of the radiation R can be identified.

[0107] (Defect-specific handling)

[0108] The above-mentioned defect identification process will be described. In the following description, variable names used in the imaging position identification process may be redefined as other variables as needed.

[0109] First, the structure of the information processing device 10 related to the defect identification process will be described. Figure 11FIG. 1 is a functional block diagram showing an example of the structure of the information processing device 10 related to the defect identification process. Figure 11 As shown, the information processing device 10 includes a second acquisition unit 50 and a defect identification unit 52. When the processor 20 executes the defect identification program 32, the processor 20 functions as the second acquisition unit 50 and the defect identification unit 52.

[0110] The second acquisition unit 50 has a function of acquiring a plurality of projection data items obtained by the detector 68 by irradiating the inspection object 60 with radiation R from different directions by the radiation source 64. In this exemplary embodiment, as described above, the first acquisition unit 40 of the information processing apparatus 10 during the imaging position identification process causes the storage unit 24 to store a plurality of projection data items having different irradiation directions of the radiation R. Therefore, the second acquisition unit 50 acquires a plurality of projection data items having different irradiation directions of the radiation R from the storage unit 24.

[0111] like Figure 11 As shown, the defect identifying unit 52 includes a second temporary projection data generating unit 54. The second temporary projection data generating unit 54 has the function of generating temporary projection data obtained when irradiating the inspection object 60 with radiation R based on the attenuation information and shape information of the inspection object 60. Furthermore, the temporary projection data generated by the second temporary projection data generating unit 54 is different from the temporary projection data generated by the first temporary projection data generating unit 44 during the imaging position identifying process. As described above, the temporary projection data generated by the first temporary projection data generating unit 44 is temporary projection data of the inspection object 60 that does not include the defect 61. On the other hand, the temporary projection data generated by the defect identifying unit 52 is temporary projection data of the inspection object 60 that assumes the defect 61.

[0112] The defect identifying unit 52 has a function of identifying the defect 61 based on a plurality of provisional projection data having different irradiation directions of the radiation R generated by the second provisional projection data generating unit 54 and a plurality of projection data having different irradiation directions of the radiation R acquired by the second acquiring unit 50. Specifically, the defect identifying unit 52 generates a defect ratio tomographic image based on a comparison result between the provisional projection data and the projection data, and identifies the three-dimensional position, size, and shape of the defect 61 based on the defect ratio tomographic image.

[0113] Furthermore, in this exemplary embodiment, in contrast to the "projection data" acquired by irradiating the inspection object 60 with radiation R, the image of the interior of the inspection object 60 is referred to as a "tomographic image," including both two-dimensional and three-dimensional images. A "defect ratio tomographic image" is a tomographic image that indicates the ratio of defects 61. In this exemplary embodiment, the maximum value is set to 1.0, with a higher value indicating a higher probability that the tomographic image represents a defect 61. Furthermore, in this exemplary embodiment, a three-dimensional image of the defect ratio is generated as a defect ratio tomographic image. Therefore, the defect ratio tomographic image will sometimes be referred to as a three-dimensional defect ratio image below.

[0114] Next, the operation related to the defect identification process performed by the information processing device 10 of this exemplary embodiment will be described in detail. Since the imaging position identification process described above specifies the imaging position of each projection data relative to the inspection object 60, various tables used in the defect identification process are generated in advance based on the geometrical arrangement of the radiation source 64, the inspection object 60, and the detector 68. Figure 10A As shown, the paths 70 of the radiation R are set at equal intervals of a small angle along two orthogonal directions from the radiation source 64S, and the detector 68 to which each path 70 is incident is specified. As described above, the more paths 70 are set, in other words, the smaller the angles Δθ and Δφ are set, the more accurately the three-dimensional image of the defect ratio can be generated, but it takes calculation time. For each detector 68, the paths 70 incident on the detector 68 are extracted to tabulate the number of paths 70. Then, for each path 70, the voxels 74 passed through are extracted to tabulate the number of extracted voxels 74 and the number of each voxel 74, and the path length in each voxel 74 is calculated and tabulated. In addition, there is no need to Figure 10A Each of the plurality of radiation sources 64S shown is tabulated separately, and the paths 70 of all the radiation sources 64S may be tabulated together.

[0115] For example, the following table is generated.

[0116] Table num_path(no_s, no_d): A table showing the number of paths 70 incident on detector no_d for scan no_s. "no_s" represents the scan identification number, and a scan with identification number no_s is labeled "scan no_s." In this exemplary embodiment, imaging in each direction is referred to as a scan. "no_d" represents the identification number of detector 68, and a detector 68 with identification number no_d is labeled "detector no_d." As described above, in this exemplary embodiment, detectors 68 are arranged two-dimensionally, so "no_d" represents the identification number of detector 68 in the two-dimensional arrangement.

[0117] Table num_pix(no_s, no_d, no_path): a table showing the number of voxels 74 passed by path no_path of scan no_s incident on detector no_d. "no_path" indicates the identification number of path 70, and a path 70 with identification number no_path is labeled "path no_path".

[0118] Table no_pix(no_s, no_d, no_path, no_pix_path): A table of identification numbers for voxels 74 in the no_pix_path th path that the radiation from scan no_s passes through on the incident path no_path to detector no_d. "no_pix_path" represents the identification numbers of voxels 74 in the no_path path that the radiation from scan no_s passes through on the incident path to detector no_d, with numbers assigned from smallest to largest starting from the direction closest to radiation source 64.

[0119] Table plen_pix(no_s, no_d, no_path, no_pix_path): table of path lengths in the no_pix_path-th voxel 74 traversed by the path no_pat of the scan no_s incident to the detector no_d.

[0120] Furthermore, for each subset (described in detail later) obtained by dividing all scans into several groups (subsets), the sum of the path lengths of all paths 70 incident on all detectors 68 and passing through each voxel 74 of all scans belonging to the subset is calculated and tabulated.

[0121] For example, the following table is generated.

[0122] Table plen_pix_sub(no_sub, no_pix): a table showing the sum of the path lengths of all paths 70 incident on all detectors 68 passing through voxel no_pix for all scans belonging to subset no_sub. "no_sub" represents the identification number of the subset, and the subset with identification number no_sub is labeled "subset no sub".

[0123] In this exemplary embodiment, the linear attenuation coefficient of the defect 61 is set in advance as f_d_l(no_e) and f_d_h(no_e). Here, "no_e" represents an identification number indicating the energy band of the radiation R, and "f_d_l(no_e)" represents the linear attenuation coefficient of the radiation R at energy no_e for a defect with a small linear attenuation coefficient. The linear attenuation coefficient f_d_l of the defect with a small linear attenuation coefficient is set to be smaller than the linear attenuation coefficient f of all components included in the inspection object 60 except air at all energies no_e. Furthermore, "f_d_h(no_e)" represents the linear attenuation coefficient of the radiation R at energy no_e for a defect with a large linear attenuation coefficient. The linear attenuation coefficient f_d_h of the defect with a large linear attenuation coefficient is set to be larger than the linear attenuation coefficient f of all components included in the inspection object 60 except air at all energies no_e.

[0124] As defects 61 included in the inspection object 60, there are, for example, bubbles, lack of penetration, poor fusion, cracks and foreign matter, and most of these defects 61 can be regarded as air. Therefore, it is preferable to set air as a defect with a small linear attenuation coefficient. The linear attenuation coefficient f_d_l of air is set to a value smaller than the linear attenuation coefficient f of all parts included in the inspection object 60 except air in all energies no_e. On the other hand, foreign matter can be set as a defect with a large linear attenuation coefficient. The linear attenuation coefficient f_d_h of the foreign matter is set to a value larger than the linear attenuation coefficient f of all parts included in the inspection object 60 except air in all energies no_e. In addition, when the linear attenuation coefficient of the foreign matter is unknown, in other words, when the type of the foreign matter is unknown, a virtual value can be set as the linear attenuation coefficient f_d_h of the foreign matter.

[0125] exist Figure 12 , a flowchart showing an example of a defect identification process executed by the information processing apparatus 10 of this exemplary embodiment is shown. Figure 12 In step S200 shown, the second acquisition unit 50 acquires a plurality of projection data from the storage unit 24 and outputs the acquired data to the defect identifying unit 52 .

[0126] In the next step S202 , the defect identifying unit 52 divides the projection data in all directions (all scans) into a plurality of groups (subsets).

[0127] For example, when the number of projection data (number of scans, number of projection directions) is 8, Figure 13 The subset number and projection data number can be set as shown in the example. Figure 13As shown in the example, it is preferable to include projection data with projection directions as close to opposite directions as possible in the same subset. In addition, it is preferable to include projection data with projection directions as close to orthogonal as possible in the same subset. In addition, it is preferable to set the subset numbers in the order of the subsets with projection directions as close to orthogonal as possible (updating the defect ratio tomographic image (detailed description later)). In addition, Figure 13 In the example, the projection directions of all projection data are contained in one plane, and the projection directions (angles) of the projection data are set to 45 degrees apart. However, the projection directions of all projection data are not limited to being contained in one plane, and the projection angles of the projection data are not limited to being equally spaced. Figure 13 In the example, in order to clearly illustrate the preferred method of dividing the subsets and the method of determining the order, it is assumed that the projection directions of all projection data are contained within a plane, and the projection angles of the projection data within the plane are equally spaced. In this exemplary embodiment, the position and posture of the imaging system of each projection data relative to the inspection object 60 can be set to be arbitrary. That is, when the coordinate system of the inspection object 60 is fixed, the imaging system between each projection data, Figure 8C The yz plane or xz plane shown may not be consistent. In addition, in the case of a normal cone beam method, the yz plane or xz plane between each projection data is consistent.

[0128] exist Figure 12 In the defect identification process shown, the defect identification unit 52 updates the defect ratio tomographic image using only the projection data belonging to the subset. This process is repeated for each subset, and updating all subsets is considered to be a single overall update. After the defect identification unit 52 has performed a predetermined number of overall updates, it completes the generation of the defect ratio tomographic image.

[0129] Therefore, in step S204, the defect identifying unit 52 generates an initial defect ratio tomographic image (described in detail later). In the next step S206, the defect identifying unit 52 sets the total number of updates to "1".

[0130] In the next step S208 , the defect identifying unit 52 initializes the subset number j to “1”. The subset number j is initialized every update count i. In the next step S210 , the defect identifying unit 52 initializes the projection data number k to “1”. The projection data number k is initialized for each subset j.

[0131] In the next step S212 , the defect identifying unit 52 calculates the defect ratio (described in detail later) of each voxel 74 of the defect ratio tomographic image based on the projection data and temporary projection data corresponding to the scan number of the combination of subset number j and projection data number k.

[0132] In the next step S214, the defect identification unit 52 increments the projection data number k (k=k+1). In the next step S216, the defect identification unit 52 determines whether the projection data number k exceeds the total number Nk of projection data belonging to subset j (k>Nk). If the projection data number k exceeds the total number Nk, the determination in step S216 becomes negative, and the process returns to step S212, repeating the calculation of the defect ratio for each voxel 74 in the defect ratio tomographic image. On the other hand, if the projection data number k exceeds the total number Nk, the determination in step S216 becomes positive, and the process proceeds to step S218.

[0133] In step S218 , the defect identifying unit 52 updates the defect ratio tomographic image based on the defect ratio calculated in step S212 .

[0134] In the next step S220, the defect identification unit 52 increments the subset number j (j = j + 1). In the next step S222, the defect identification unit 52 determines whether the subset number j exceeds the total number Nj of subsets (j > Nj). If the subset number j exceeds the total number Nj, ​​the determination in step S222 becomes negative, and the process returns to step S210, repeating the updating of the defect ratio tomographic image. On the other hand, if the subset number j exceeds the total number Nj, ​​the determination in step S222 becomes positive, and the process proceeds to step S224.

[0135] In step S224, the defect identification unit 52 increments the update count i (i = i + 1). In the next step S226, the defect identification unit 52 determines whether the update count i exceeds the preset count Ni (i > Ni). If the update count i exceeds the preset count Ni, the determination in step S226 becomes negative, and the process returns to step S208, repeating the updating of the defect ratio tomographic image. On the other hand, if the update count i exceeds the preset count Ni, the determination in step S226 becomes positive, and the process proceeds to step S228.

[0136] In the next step S228, the defect identifying unit 52 identifies the three-dimensional position, size, and shape of the defect 61 based on the defect ratio tomographic image generated by the above-mentioned processing (described in detail later). In addition, the defect identifying unit 52 displays the identification result on the display 26. When the processing of step S228 is completed, Figure 12 The defect specific processing shown ends.

[0137] The details of steps S204, S212, and S218 are described below. Furthermore, defect ratio tomographic images are generated for each of the defect with a small linear attenuation coefficient and the defect with a large linear attenuation coefficient. Specifically, the temporary projection data (temporary projection data excluding the defect 61 generated based on the attenuation information f (no_f, no_e) and the shape information no_f (no_pix)) whose imaging position is determined are compared with the measured projection data. Then, a defect ratio tomographic image of the defect 61 with a small linear attenuation coefficient is generated based on the larger projection data of the measured side, and a defect ratio tomographic image of the defect 61 with a large linear attenuation coefficient is generated based on the smaller projection data of the measured side. That is, the above-described defect ratio tomographic image generation is performed for each of the defect 61 with a small linear attenuation coefficient and the defect 61 with a large linear attenuation coefficient. Furthermore, in generating the defect ratio tomographic image of the defect 61 with a small linear attenuation coefficient, for the measured projection data that is smaller than the temporary projection data, the difference between the measured projection data and the temporary projection data is replaced with "0." Specifically, the value of the measured projection data is replaced with the value of the temporary projection data. Furthermore, when generating a defect ratio tomographic image for a defect 61 with a large linear attenuation coefficient, conversely, for measured projection data with a value greater than the provisional projection data, the measured projection data values ​​are replaced with the provisional projection data values. In other words, the difference between the measured projection data and the provisional projection data is replaced with "0" (described in detail later). The following describes in detail the method for generating a defect ratio tomographic image for a defect 61 with a small linear attenuation coefficient. The method for generating a defect ratio tomographic image for a defect 61 with a large linear attenuation coefficient is supplementally described only where necessary.

[0138] Step S204 will be described in detail. In step S204, as described above, an initial defect ratio tomographic image r_d(no_pix) is generated. r_d(no_pix) represents the ratio of defects included in voxel no_pix and takes a value between 0.0 and 1.0. When r_d(no_pix) is 0.0, voxel no_pix is ​​deemed to contain no defects, while when it is 1.0, voxel no_pix is ​​deemed to be occupied by defect 61. The total number of voxels 74 in the defect ratio tomographic image is set to num_pix_t, and no_pix is ​​an arbitrary value between 1 and num_pix_t. The linear attenuation coefficient for each energy no_e of the defect is assigned based on the previously generated linear attenuation coefficient table f_d_l(no_e) for the defect 61. As the initial value of the defect ratio tomographic image r_d(no_pix), the value of all voxels 74 is set to 0.0, indicating that no defect 61 is included in any voxel 74.

[0139] The details of step S212 will be described. The defect identifying unit 52 calculates the defect ratio of the defect ratio tomographic image based on the projection data g(no_s, 1) to g(no_s, num_d). The scan number no_s is determined in accordance with the combination of the subset number j and the projection data number k (see Figure 13 ). “num_d” represents the number of detectors 68.

[0140] The calculation method of the projection data g(no_s, no_d) based on the detector no_d is described. First, the provisional projection data g_v of the scan no_s and the detector no_d is obtained based on the current tomographic image using the following formula (3).

[0141] g_v=I(1)+I(2)+…+I(num_path(no_s, no_d))…(3)

[0142] In addition, num_path(no_s, no_d) in the above formula (3) is the number of paths 70 that enter the detector no_d in the scan no_s, which is obtained in advance.

[0143] In the above formula (3), I(no_path) represents the intensity of the radiation incident on the path no_path of the detector, and is obtained by the following formula (4).

[0144] I(no_path)=(e(1)×I_e(1))+(e(2)×I_e(2))+……+(e(num_e)×I_e(num_e))…(4)

[0145] In the above formula (4), as described above, e(1), e(2), ..., e(num_e) represent energy and are known.

[0146] In the above formula (4), the intensity I_e(no_e) of the radiation R at the energy e(no_e) is obtained by the following formula (5).

[0147] I_e(no_e)=I0_e(no_e)×exp(-{(f_ext(1,no_e)×plen_pix(no_s,no_d,no_path,1))+(f_ext(2,no_e)×plen_pix(no_s,no_d,no _path, 2))+……+(f_ext(num_pix(no_s, no_d, no_path), no_e)×plen_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)))})

[0148] in,

[0149] f_ext(no_pix_path, no_e)=f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e)+{r_d(no_pix(no_s , no_d, no_path, no_pix_path))×(f_d(no_e)-f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e))}

[0150] In the generation of defect ratio tomographic images of defects with small linear attenuation coefficients, f_d(no_e)=f_d_l(no_e)

[0151] In the generation of the defect ratio tomographic image of a defect with a large linear attenuation coefficient, f_d(no_e)=f_d_h(no_e)…(5)

[0152] The above formula (5) represents the intensity I_e(no_e) of the radiation R of intensity I0_e(no_e) among the energy e(no_e) irradiated from the radiation source 64 , which passes through each voxel 74 while attenuating and enters the detector 68 .

[0153] In the above formula (5), f_ext(no_pix_path, no_e) represents the linear attenuation coefficient of energy no_e in the no_pix_path voxel no_pix through which the path no_path of the scan no_s incident on the detector no_d passes. The linear attenuation coefficient f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e) based on the original attenuation information f(no_f, no_e) and shape information no_f(no_pix) is added at a ratio of 1-r_d(no_pix(no_s, no_d, no_path, no_pix_path)) and the linear attenuation coefficient f_d_l(no_e) of the defect is added at a ratio of r_d(no_pix(no_s, no_d, no_path, no_pix_path)) to obtain the linear attenuation coefficient f_d_l(no_e) of the defect.

[0154] As mentioned above, the intensity I0_e(no_e) in the above formula (5) is expressed in Figure 10A The sum of the number of photons (of energy e(no_e)) irradiated per unit time in a small angle range of Figure 10AThe narrower the interval of the angle of the path 70 is set, in other words, the more the paths 70 are set, the smaller the value of I0_e(no_e) is set. I0_e(0), I0_e(1), ..., I0_e(num_e) are determined by the radiation source 64 and are known.

[0155] The provisional projection data g_v obtained using equation (3) is compared with the measured projection data g(no_s, no_d). If the provisional projection data g_v is larger, the defect ratio of all voxels (excluding those in the air) passing through the path of the scan no_s incident on the detector no_d is subtracted by Δr. The provisional projection data g_v is then recalculated and compared with the measured projection data g(no_s, no_d). This process of subtracting Δr from the defect ratio, recalculating the provisional projection data g_v, and comparing it with the projection data g(no_s, no_d) is repeated until the provisional projection data g_v is equal to or smaller than the measured projection data g(no_s, no_d). Next, the following data is used: between the provisional projection data g_v that is about to fall below the measured projection data g(no_s, no_d) and the provisional projection data g_v that falls below the measured projection data g(no_s, no_d), the provisional projection data g_v that is closer to the measured projection data g(no_s, no_d) is used. The decrement in the defect ratio when calculating this provisional projection data g_v is stored as dr_d(no_s, no_d). In this case, dr_d(no_s, no_d) is a negative value. When recalculating the provisional projection data g_v by reducing the defect ratio by α, in equation (5) above, r_d(no_pix(no_s, no_d, no_path, no_pix_path)) in f_ext(no_pix_path, no_e) is replaced with r_d(no_pix(no_s, no_d, no_path, no_pix_path))-α. In this way, the intensity I_e(no_e) is recalculated by the above formula (5), and the intensity I(no_path) is recalculated by the above formula (4) based on the recalculated I_e(1), I_e(2), ..., Ie(num_e). Based on the recalculated I(1), I(2), ..., I(num_path(no_s, no_d)), the temporary projection data g_v is recalculated by the above formula (3).

[0156] Furthermore, when calculating equation (5) above, if there is a voxel 74 among voxels no_pix(no_s, no_d, no_path, 1) to no_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)) where r_d(no_pix(no_s, no_d, no_path, no_pix_path))-α is less than 0.0, the defect ratio r_d is set to 0.0 for that voxel 74. In this way, the linear attenuation coefficient f_ext is not set to a value greater than the original linear attenuation coefficient f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e) that does not include the defect 61.

[0157] If the provisional projection data g_v is smaller than the measured projection data g(no_s, no_d), conversely, for all paths 70 incident on detector no_d during scan no_s, the defect ratio in all voxels 74 passed through is repeatedly increased by Δr, and the provisional projection data g_v is recalculated until the provisional projection data g_v reaches or exceeds the measured projection data g(no_s, no_d). The provisional projection data g_v that is closer to the measured projection data g(no_s, no_d) is then used: the provisional projection data g_v just before reaching or exceeding the measured projection data g(no_s, no_d) and the provisional projection data g_v that reaches or exceeds the measured projection data g(no_s, no_d). The increment in defect ratio when calculating this provisional projection data g_v is stored as dr_d(no_s, no_d). At this point, dr_d(no_s, no_d) is a positive value. When the defect ratio is increased, if there is a voxel 74 whose defect ratio exceeds the maximum value of 1.0 as a result of the increase, the defect ratio of this voxel 74 is set to 1.0.

[0158] Through the above calculation, the defect ratio increments dr_d(no_s, 1) to dr_d(no_s, num_d) can be obtained based on the projection data g(no_s, 1) to g(no_s, num_d).

[0159] Furthermore, for data in which the measured projection data g (no_s, 1) to g (no_s, num_d) are smaller than the provisional projection data g_v when no defect is present, the value of the measured projection data g is replaced in advance with the value of the provisional projection data g_v when no defect is present. The provisional projection data g_v when no defect is present is the provisional projection data g_v obtained from the original attenuation information f (no_f, no_e) and shape information no_f (no_pix) using the above-mentioned equations (3) and (4), and the equation (5) in which the defect ratio r_d is set to 0. For example, if the value of the measured projection data g (no_s, no_d) of the detector no_d is smaller than the provisional projection data g_v obtained from the original attenuation information f (no_f, no_e) and shape information no_f (no_pix) using the above-mentioned equations (3), (4), and (5), it is considered that the defect 61 having a small linear attenuation coefficient is not included in each voxel 74 through which the path 70 incident on the detector 68 passes. Furthermore, in order to set the defect ratio r_d (of a defect with a small linear attenuation coefficient) of each voxel 74 to 0.0, the value of the measured projection data g(no_s, no_d) is replaced in advance with the value of the temporary projection data g_v. By this replacement, when generating a defect ratio tomographic image of a defect 61 with a small linear attenuation coefficient, the values ​​of the measured projection data g(no_s, 1) to g(no_s, num_d) are guaranteed to be equal to or greater than the value of the temporary projection data obtained based on the original attenuation information f(no_f, no_e) and shape information no_f (no_pix) when the defect 61 is not included. The replacement is performed in step S204 (refer to Figure 12 ) before. In addition, since the defect ratio tomographic images of the defect 61 with a small linear attenuation coefficient and the defect 61 with a large linear attenuation coefficient are replaced in the generation of the defect ratio tomographic images, the measured projection data g(no_s, 1) to g(no_s, num_d) before the replacement are separately saved.

[0160] In generating a defect ratio tomographic image of a defect with a large linear attenuation coefficient, for data in which the measured projection data g(no_s, no_d) is larger than the temporary projection data g_v when no defect is included among the measured projection data g(no_s, 1) to g(no_s, num_d), the value of the measured projection data g(no_s, no_d) is replaced in advance with the value of the temporary projection data g_v when no defect is included. The temporary projection data g_v when no defect is included is the temporary projection data g_v obtained based on the original attenuation information f(no_f, no_e) and shape information no_f(no_pix) using the above-mentioned equations (3) and (4) and equation (5) in which the defect ratio r_d is set to 0. For example, if the value of the measured projection data g(no_s, no_d) for detector no_d is greater than the provisional projection data g_v calculated using equations (3), (4), and (5) based on the original attenuation information f(no_f, no_e) and shape information no_f(no_pix), it is assumed that no defect 61 with a large linear attenuation coefficient is present in each voxel 74 through which path 70 incident on detector 68 passes. Furthermore, to set the defect ratio r_d (of defects 61 with large linear attenuation coefficients) for each voxel 74 to 0.0, the value of the measured projection data g(no_s, no_d) is replaced with the value of the provisional projection data g_v. This replacement ensures that the values ​​of the measured projection data g(no_s, 1) to g(no_s, num_d) are equal to or less than the value of the provisional projection data obtained based on the original attenuation information f(no_f, no_e) and shape information no_f(no_pix) when the defect 61 is not present.

[0161] In generating a defect ratio tomographic image of a defect 61 having a large linear attenuation coefficient, for detector no_d scanning no_s, provisional projection data g_v is first calculated based on the current tomographic image using equations (3), (4), and (5). If the provisional projection data g_v is small, the process of calculating provisional projection data g_v by subtracting Δr from the defect ratio of all voxels 74 (excluding air) passing through all paths 70 incident on the detector 68 is repeated until the provisional projection data g_v reaches or exceeds the measured projection data g(no_s, no_d). The provisional projection data g_v that is closer to the measured projection data g(no_s, no_d) is used, between the provisional projection data g_v that is about to reach or exceeds the measured projection data g(no_s, no_d) and the provisional projection data g_v that reaches or exceeds the measured projection data g(no_s, no_d). The decrement in the defect ratio during this calculation is stored as dr_d(no_s, no_d). In this case, dr_d(no_s, no_d) is a negative value. In this case, for voxels 74 where the result of subtracting the defect ratio is less than the minimum value of 0.0, the defect ratio is set to 0.0. That is, in equation (5), the linear attenuation coefficient f_ext is not set to a value smaller than the original linear attenuation coefficient f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e) that does not include the defect 61. If the provisional projection data g_v is larger than the measured projection data g(no_s, no_d), the defect ratio is increased by Δr to obtain the provisional projection data g_v. This process is repeated until the provisional projection data g_v falls below the measured projection data g(no_s, no_d). The provisional projection data g_v that is closer to the measured projection data g(no_s, no_d) is used, as between the provisional projection data g_v immediately before falling below the measured projection data g(no_s, no_d) and the provisional projection data g_v that has fallen below the measured projection data g(no_s, no_d). The increment in defect ratio during this calculation is stored as dr_d(no_s, no_d). In this case, dr_d(no_s, no_d) is a positive value. For voxels 74 where the result of increasing the defect ratio exceeds the maximum value of 1.0, the defect ratio is set to 1.0.

[0162] Furthermore, in both the generation of the defect ratio tomographic image for the defect 61 with a small linear attenuation coefficient and the generation of the defect ratio tomographic image for the defect 61 with a large linear attenuation coefficient, in the defect ratio tomographic image r_d generated in step S204, the defect ratio for voxels 74 occupied by air among voxels 1 to num_pix_t is fixed at 0.0, since no defect will occur in these voxels 74. Furthermore, when the defect ratio increments dr_d(no_s, 1) to dr_d(no_s, num_d) are calculated in step S212, the defect ratio for voxels 74 occupied by air is also fixed at 0.0.

[0163] Step S218 will be described in detail. After calculating the defect ratio increments dr_d(no_s, 1) to dr_d(no_s, num_d) for all projection data belonging to subset j, the defect identification unit 52 updates the defect ratio tomographic image based on these increments. Focusing on a certain voxel no_pix, assume that a path 70 incident on detector no_d for scan no_s passes through that voxel 74. In this case, dr_d(no_s, no_d) is an estimated value of the defect ratio increment for the path 70 incident on detector no_d for scan no_s that passes through voxel no_pix. Therefore, it is necessary to average the estimated defect ratio increments for all paths 70 that pass through voxel no_pix, weighted by the length of each path. First, the defect ratio tomographic image representing the average of these increments is set as dr_d_ave(no_pix), and dr_d_ave(no_pix) is initialized by setting the values ​​of all voxels 74 to "0." Here, dr_d_ave(no_pix) represents the average of the defect ratio increments in voxel no_pix, where no_pix is ​​any value between 1 and the total number of voxels, num_pix_t. Then, for all projection data belonging to subset j, the defect ratio increments dr_d(no_s, 1) to dr_d(no_s, num_d) are weighted by the path length and added to each voxel 74 in dr_d_ave. Finally, the average of these increments is calculated for each voxel 74 by dividing the dr_d_ave value for each voxel 74 by the sum of the path lengths for each voxel 74, as previously listed in the table, plen_pix_sub. This average is then added to the defect ratio tomographic image before the update to update the defect ratio tomographic image.

[0164] The addition process is specifically described using the projection data of scan no_s belonging to subset j as an example. The number of voxels traversed by path no_path from scan no_s to detector no_d is num_pix(no_s, no_d, no_path). In the incremental average tomographic image dr_d_ave, the value of the no_pix_path-th voxel no_pix(no_s, no_d, no_path, no_pix_path) traversed by path no_path from scan nos to detector no_d is multiplied by the increment multiplied by the path length: dr_d(no_s, no_d) × plen_pix(no_s, no_d, no_path, no_pix_path). Similarly, in the incremental average tomographic image dr_d_ave, the values ​​of all num_pix(no_s, no_d, no_path) voxels 74 passed by the path no_path are incremented by dr_d(no_s, no_d)×path length (plen_pix(no_s, no_d, no_path, 1) to plen_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path))). Furthermore, if there is a voxel in which the result obtained by adding the increment dr_d(no_s, no_d) to the defect ratio before updating exceeds the maximum defect ratio value of 1.0 among voxels no_pix(no_s, no_d, no_path, 1) to no_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)), the increment is corrected so that the result of addition reaches the maximum defect ratio value of 1.0, and then added to dr_d_ave. Similarly, if there is a voxel in which the result obtained by adding the increment dr_d(no_s, no_d) to the defect ratio before updating is less than the minimum defect ratio value of 0.0, the increment is corrected so that the result of addition reaches 0.0, and then added to dr_d_ave. This addition process is performed for all paths 1 to num_path(no_s, no_d) incident on detector no_d during scan no_s. Then, the above addition process is performed on all detectors 1 to num_d of scan no_s. The above is the addition process for the projection data of scan no_s belonging to subset j. This addition process is performed on all projection data belonging to subset j.Finally, dr_d_ave(1) to dr_d_ave(num_pix_t) generated by the above addition operation are divided by the sum of the path lengths of each voxel 74, plen_pix_sub(j, 1) to plenpix_sub(j, num_pix_t), to obtain the tomographic images dr_d_ave(1) to dr_d_ave(num_pix_t) of the average value of the defect ratio increment, and the defect ratio tomographic image is updated by adding it to the tomographic images r_d(1) to r_d(num_pix_t) before the update.

[0165] Furthermore, as described above, in the defect ratio tomographic images r_d(1) to r_d(num_pix_t), the value of the defect ratio is fixed to 0.0 for the voxel 74 occupied by air.

[0166] Step S228 will be described in detail. The defect identifying unit 52 identifies the three-dimensional position, size, and shape of the defect 61 based on the defect ratio tomographic image r_d. The defects 61 included in the inspection object 60 are very sparse. Therefore, in the detector no_d of most scans no_s, the difference between the temporary projection data virtually generated by simulation based on the attenuation information and shape information of the inspection object 60 without the defect 61 and the actual measured projection data is close to 0. Figure 12 The repetitive processing shown generates an image of the defect ratios of each detector 68, whose size and shape match the differences between simulation and actual measurements, in areas of the defect ratio three-dimensional image that intersect paths 70 corresponding to projection data from detector no_d for the few scans no_s where the differences are non-zero. Regions in the three-dimensional defect ratio image with a defect ratio exceeding a predetermined value (e.g., 0.8) can be identified as defects 61. Alternatively, after extracting regions with a defect ratio exceeding a predetermined value (e.g., 0.8) from the three-dimensional defect ratio image, connected regions can be identified. In each connected region, the sum of the defect ratios within the region exceeds a predetermined value. In other words, regions with large defect ratios and large volumes can be identified as defects 61. Alternatively, after applying a three-dimensional smoothing filter to the three-dimensional defect ratio image, regions with a defect ratio exceeding a predetermined value (e.g., 0.6) can be extracted. Then, connected regions can be identified. Within each connected region, the maximum defect ratio within the region exceeds a predetermined value and is identified as a defect. Applying the smoothing filter increases the defect ratio and the defect ratio of regions with large volumes. After identifying the defect, the defect identifying unit 52 displays the three-dimensional position, size, and shape of the defect 61 identified inside the inspection object 60 on the display 26 so that the user can know the defect.

[0167] As described above, according to this exemplary embodiment, the three-dimensional position, size, and shape of the defect 61 can be specified with high precision.

[0168] [Second exemplary embodiment]

[0169] This exemplary embodiment describes defect identification processing that differs from the first exemplary embodiment. Furthermore, in this exemplary embodiment, as in the first exemplary embodiment, a table num_path (no_s, no_d) for the number of paths 70, a table num_pix (no_s, no_d, no_path) for the number of voxels 74, a table no_pix (no_s, no_d, no_path, no_pixpath) for the numbers of voxels 74, and a table plen_pix (no_s, no_d, no_path, no_pix_path) for path lengths are also generated. Furthermore, tables f_d_l (no_e) and f_d_h (no_e) for the linear attenuation coefficients of defects are also generated. Furthermore, a table plen_pix_sub (no_sub, no_pix) for the total path lengths of the subset no_sub is not necessary.

[0170] exist Figure 14 A flowchart showing an example of defect identification processing performed by the information processing apparatus 10 of this exemplary embodiment is shown in FIG. Figure 14 As shown, the defect specific processing of this exemplary embodiment replaces the defect specific processing of the first exemplary embodiment (see Figure 12 ) steps S210 to S216 and includes the processing of step S209. In addition, the specific content of the processing in step S218 after step S209 is different from the first exemplary embodiment.

[0171] Furthermore, in this exemplary embodiment, defect ratio tomographic images are generated for each of the defects 61 with a small linear attenuation coefficient and the defects 61 with a large linear attenuation coefficient. In this case, when generating a defect ratio tomographic image for the defect 61 with a small linear attenuation coefficient, for any of the measured projection data g(no_s, no_d) to g(no_s, num_d) where the measured projection data g(no_s, no_d) is smaller than the temporary projection data g_v when the defect is not present, the values ​​of the measured projection data g(no_s, no_d) are replaced in advance with the values ​​of the temporary projection data g_v obtained from the original attenuation information f(no_f, no_e) and shape information no_f(no_pix) using the aforementioned equations (3) and (4), and equation (5) with the defect ratio r_d set to 0. In addition, in the generation of the defect ratio tomographic image of the defect 61 with a large linear attenuation coefficient, conversely, for data where the measured projection data g(no_s, no_d) is larger than the temporary projection data g_v, the value of the measured projection data g(no_s, no_d) is replaced in advance with the value of the temporary projection data g_v.

[0172] In this exemplary embodiment, an evaluation function is determined that evaluates the difference between the provisional projection data g_v obtained from the generated defect ratio tomographic image r_d and the measured projection data g(no_s, no_d). The defect ratio tomographic image is then updated so that the evaluation function decreases. Specifically, the change in the evaluation function when the value of the defect ratio tomographic image (defect ratio) is slightly changed in each voxel 1 to num_pix_t of the defect ratio tomographic image r_d is calculated, and this value is used as the gradient. The defect ratio tomographic image is then updated based on the gradient calculated for each voxel 1 to num_pix_t. The gradient is calculated for each subset j, and the defect ratio tomographic image is updated. Steps S209 and S218 are described in detail below.

[0173] In step S209, the defect identifying unit 52 calculates the gradient based on the projection data belonging to the subset j. First, the evaluation function ef is determined in advance as follows.

[0174] ef=∑∑(g_v(no_s, no_d)-g(no_s, no_d)) 2

[0175] in,

[0176] The first ∑ is the sum associated with the scan no_s belonging to subset j.

[0177] The second Σ is the sum associated with detector no_d (the sum of 1 to num_d).

[0178] g_v(no_s, no_d) represents temporary projection data of scan no_s and detector no_d obtained from the defect ratio tomographic image.

[0179] g(no_s, no_d) represents the measured projection data of the scanner no_s and the detector no_d.

[0180] …(6)

[0181] First, based on the current defect ratio tomographic image r_d, each provisional projection data g_v(no_s, no_d) is calculated using equations (3), (4), and (5). Then, the evaluation function ef is calculated using equation (6). Next, r_d(1), i.e., the value (defect ratio) of voxel 1 in the defect ratio tomographic image r_d, is temporarily increased by a predetermined amount α. Each provisional projection data g_v(no_s, no_d) is again calculated using equations (3), (4), and (5), and the evaluation function ef is calculated using equation (6). Furthermore, when r_d(1) is temporarily increased by α, if voxel no_pix(no_s, no_d, no_path, no_pix_path) in equation (5) is equal to 1, the intensity I_e(no_e) can be calculated by replacing r_d(no_pix(no_s, no_d, no_path, no_pix_path)) in f_ext(no_pix_path, no_e) with r_d(no_pix(no_s, no_d, no_path, no_pix_path))+α. Thus, the evaluation function ef before the increase of the value of voxel 1 in the defect ratio tomographic image r_d by α is subtracted from the evaluation function ef before the increase, and the result is the gradient Δef(1) in voxel 1.

[0182] Furthermore, in the calculation of the gradient, it is ideal to be able to calculate the gradient even when r_d(1)+α exceeds the maximum defect ratio of 1.0. Therefore, when r_d(1)+α exceeds the maximum defect ratio of 1.0, when voxel no_pix is ​​equal to 1 in equation (5), r_d(1) in f_ext is replaced by 1.0-α and 1.0, respectively, and the intensity I_e(no_e) is calculated to calculate the evaluation function ef. Then, the evaluation function ef calculated by replacing r_d(1) by 1.0-α is subtracted from the evaluation function ef calculated by replacing r_d(1) by 1.0, and the result is the gradient Δef(1) in voxel 1.

[0183] Similarly, the gradients Δef(2), Δef(3), ..., Δef(num_pix_t) are calculated for voxels 2, 3, ..., num_pix_t, respectively. Furthermore, to quickly calculate the evaluation function ef after adding α, if none of the voxels no_pix(no_s, no_d, no_path, 1) to no_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)) through which the path no_path of the scan no_s incident on the detector no_d passes when calculating the above equation (5) does not correspond to the voxel for which α has been added, the calculation of equations (4) and (5) above may be omitted for that path 70, and the intensity I(no_path) of the radiation R in the path no_path may be set to the same value as the intensity I(no_path) before adding α. Furthermore, for voxels occupied by air, the gradient Δef is set to 0 to fix the defect ratio to 0.0. In this case, the calculations of the above-mentioned equations (3), (4), and (5) are not performed.

[0184] The evaluation function ef may be any function as long as it can evaluate the difference between the provisional projection data g_v obtained from the defect ratio tomographic image r_d and the measured projection data g(no_s, no_d). For example, the following function may be used.

[0185] ef=∑∑abs(g_v(no_s, no_d)-g(no_s, no_d))

[0186] in,

[0187] The first ∑ is the sum associated with the scan no_s belonging to subset j.

[0188] The second Σ is the sum associated with detector no_d (the sum of 1 to num_d).

[0189] abs(x) represents the absolute value of x.

[0190] g_v(no_s, no_d) represents temporary projection data of scan no_s and detector no_d obtained from the defect ratio tomographic image.

[0191] g(no_s, no_d) represents the measured projection data of the scanner no_s and the detector no_d.

[0192] …(7)

[0193] The process of step S218 will be described. The defect identifying unit 52 updates the defect ratio tomographic image r_d using the following equation.

[0194] r_d(1)←r_d(1)-β×Δef(1)

[0195] r_d(2)←r_d(2)-β×Δef(2)

[0196] …

[0197] r_d(num_pix_t)←r_d(num_pix_t)-β×Δef(num_pix_t)

[0198] Among them, β represents the update step size.

[0199] …(8)

[0200] Furthermore, if there is a voxel 74 for which the defect ratio r_d value is less than the minimum defect ratio value of 0.0 as a result of the update, the defect ratio r_d value is set to 0.0 for that voxel 74. Furthermore, if there is a voxel 74 for which the defect ratio r_d value exceeds the maximum defect ratio value of 1.0 as a result of the update, the defect ratio r_d value is set to the maximum value of 1.0 for that voxel 74.

[0201] Through the above-described processing, also in this exemplary embodiment, it is possible to generate a defect ratio tomographic image r_d for each of the defect 61 having a small attenuation coefficient and the defect 61 having a large linear attenuation coefficient.

[0202] The technology of the present invention is not limited to the above-mentioned embodiments, and various modifications are possible. For example, the following modifications are possible.

[0203] While the exemplary embodiments described above illustrate an example in which the scanning method is a cone beam method, there are no particular limitations on the scanning method, and any scanning method may be applied. For example, in the case of a pencil beam (parallel beam) method, two-dimensional data obtained by performing a series of parallel scans in the x and y directions by the radiation source 64 and the detector 68 is used as a single projection data set. Furthermore, by using data at a different projection angle as another projection data set, imaging position identification processing and defect identification processing can be performed in the same manner as in the case of the cone beam method. Figure 8C This is a diagram assuming an imaging system of a cone beam method. However, in the case of a pencil beam (parallel beam) method, the radiation source 64 and each detector 68 exist in correspondence. Figure 8C In the xy plane, each radiation source 64 exists at the same x and y coordinates as each detector 68. Of course, in the case of the pencil beam method, the position and posture of the imaging system of each projection data relative to the inspection object 60 can also be arbitrary. When the coordinate system of the inspection object 60 is fixed, the imaging system of each projection data Figure 8CThe yz plane or xz plane shown may not be consistent. In the case of the pencil beam method, in the imaging position specific processing, Figure 7 The method shown can specify the imaging position in such a way that the similarity between the measured projection data and the temporary projection data is greater than a threshold value. In the defect specific processing, each projection data can be divided into subsets. Figure 12 or Figure 14 The defect ratio tomographic image is generated by the defect identification process shown in FIG. 1 , and the three-dimensional position, shape, and size of the defect 61 are identified based on the defect ratio tomographic image. In addition, in the case of the pencil beam (parallel beam) method, since one detector 68 corresponds to the radiation source 64, Figure 10A The method shown only requires setting the path 70 incident on one detector 68 .

[0204] In the case of the fan beam method, after the two-dimensional data obtained by connecting the one-dimensional data detected by the one-dimensional detector 68 in the x-direction or the y-direction is set as one projection data and the data of different projection angles is set as another projection data, the imaging position identification process and the defect identification process can be performed in the same manner as in the case of the cone beam method. In the case of the fan beam method, since the radiation source 64 exists in correspondence with each (one-dimensional) detector 68, Figure 8C In the xy plane, each radiation source 64 is located along the x-axis or y-axis at the same x-coordinate or y-coordinate as each (one-dimensional) detector 68. In the case of the fan beam method, the position and posture of the imaging system of each projection data relative to the inspection object 60 can also be arbitrary. In the case of the fan beam method, the above-mentioned Figure 7 The photographic position can be specified by the photographic position specifying process shown. Figure 12 or Figure 14 The defect ratio tomographic image is generated by the defect identification process shown in FIG. 1 , and the three-dimensional position, shape, and size of the defect 61 are identified based on the defect ratio tomographic image. In the case of the fan beam method, since the detector 68 is arranged in one dimension and corresponds to the radiation source 64, Figure 10A The method shown only requires setting a path 70 incident on a detector 68 configured in one dimension.

[0205] Furthermore, in the exemplary embodiments described above, the paths 70 of the radiation R are set at equal intervals. However, by varying the intervals for each path 70, it is possible to express differences in the intensity of the radiation R along each path 70. For example, if it is desired to increase the intensity of the irradiated radiation R toward the center of the radiation source 64 and decrease it toward the ends, the paths 70 can be set with narrower intervals (dense) toward the center of the radiation source 64 and wider intervals (sparser) toward the ends. Alternatively, the intensity I0_e (no_e) of radiation irradiated from the radiation source 64 can be varied for each path 70. In this case, a table I0_e(no_s, no_d, no_path, no_e) of the intensity of the path no_path of the scan no_s incident to the detector no_d is generated in advance, and in the above formula (2) or (5), the intensity I0_e(no_e) in the energy e(no_e) of the radiation R irradiated from the radiation source 64 is set to I0_e(no_s, no_d, no_path, no_e), and it can be changed according to the path 70.

[0206] In addition, each of the above exemplary embodiments can also be implemented when the information of the linear attenuation coefficient of the inspection object 60 is not a value in two or more energy bands but a single value. When the information of the linear attenuation coefficient is a single value, Figure 9 The linear attenuation coefficient table f(no_f, no_e) shown in FIG. 1 shows a value f(no_f) for each component number no_f. In addition, the above formula (2) is expressed as the following formula (2_2).

[0207] I=I0×exp(-{(f(no_f(no_pix(1)))×plen_pix(1))+(f(no_f(no_pix(2)))×plen_pix(2))+……+(f(no_f(no_pix(num_pix)))×plen_pix(num_pix)})…(2_2)

[0208] Furthermore, by omitting the above-mentioned formula (1), it is possible to accumulate the intensities I obtained for each path 70 by the above-mentioned formula (2-2) for each detector 68 to generate temporary projection data.

[0209] Furthermore, the linear attenuation coefficients of the defect, f_d_l(no_e) and f_d_h(no_e), also become the same values, f_d_l and f_d_h. Furthermore, the above equation (5) becomes the following equation (5_2).

[0210] I=I0×exp(-{(f_ext(1)×plen_pix(no_s, no_d, no_path, 1))+(f_ext(2)×plen_pix(no_s, no_d, no_path, 2))+… …+(f_ext(num_pix(no_s, no_d, no_path))×plen_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)))})

[0211] in,

[0212] f_ext(no_pix_path)=f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)))+{r_d(no_pix(no_s, no_d, no_path, no_pix_path))×(f_d-f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)))}

[0213] In the generation of defect ratio tomographic images of defects with small linear attenuation coefficients, f_d = f_d_l

[0214] In the generation of the defect ratio tomographic image of a defect with a large linear attenuation coefficient, f_d=f_d_h...(5_2)

[0215] Then, by omitting the above formula (4) and setting the intensities I obtained for the respective paths 70 by the above formula (5_2) as I(1), I(2), ..., the provisional projection data g_v can be obtained by the above formula (3).

[0216] In each of the exemplary embodiments described above, the paths 70 incident on each detector 68 can be approximately combined into a single path. In this case, the imaging position identification process also requires the pre-production of a table similar to that used in the defect identification process. When combining the paths 70 into a single path, the pre-production table is as follows. Furthermore, during the imaging position identification process, the following table is newly generated each time the imaging position is updated.

[0217] Table num_path(no_s, no_d): indicates the number of paths 70 incident on detector no_d for scan no_s. Even when the paths 70 are aggregated into one, the first exemplary embodiment requires a table showing the number of paths 70 before aggregation.

[0218] Table num_pix(no_s, no_d): This table represents the number of voxels 74 passed through by at least one of the paths 1 to num_path(no_s, no_d) incident on detector no_d during scan no_s, before the paths 70 are combined into one. If paths 70 are not combined into one, table num_pix(no_s, no_d, no_path) is generated. However, if paths 70 are combined into one, table no_path can be omitted.

[0219] Table no_pix(no_s, no_d, no_pix_path): This table shows the number of voxel 74 in the no_pix_pathth order, passed by at least one of the paths 1 to num_path(no_s, no_d) incident on detector no_d during scan no_s, before paths 70 are consolidated into one. When paths 70 are not consolidated into one, table no_pix(no_s, no_d, no_path, no_pix_path) is generated. However, when paths 70 are consolidated into one, table no_path can be omitted.

[0220] Table plen_pix(no_s, no_d, no_pix_path): This table represents the path lengths of at least one path 70 from among paths 1 to num_path(no_s, no_d) incident on detector no_d during scan no_s before the paths 70 are combined into one, i.e., the average path lengths. plen_pix(no_s, no_d, no_pix_path) is calculated by extracting all voxels 74 passed by at least one path 70 from among paths 1 to num_path(no_s, no_d) before the paths 70 are combined into one, calculating the sum of the path lengths of paths 1 to num_path(no_s, no_d) in each voxel 74, and dividing this sum by num_path(no_s, no_d). Furthermore, the voxel number no_pix_path in the table no_pix(no_s, no_d, no_pix_path) is made common to the table plen_pix(no_s, no_d, no_pix_path). When the paths 70 are not combined into one, the table plen_pix(no_s, no_d, no_path, no_pix_path) is generated. However, when the paths 70 are combined into one, no_path can be omitted.

[0221] Table I0_e (no_s, no_d, no_e): represents the sum of the intensities of the energy e(no_e) of the radiation R incident on detector no_d for all paths 70 of scan no_s before the paths 70 are combined into one. When the paths incident on each detector are combined into one, the number of paths 70 incident on each detector 68 before the combination differs for each detector 68. That is, in equations (2) and (5), the intensity I0_e(no_e) of the energy e(no_e) emitted from the radiation source 64 differs for each detector 68. Therefore, instead of the intensity I0_e(no_e), table I0_e(no_s, no_d, no_e) is generated in advance. Furthermore, I0_e(no_s, no_d, no_e) does not depend on the imaging position, so it only needs to be generated once in advance.

[0222] Table plen_pix_sub(no_sub, no_pix): This table represents the sum of the path lengths of all paths 70 incident on all detectors 68 of all scans, belonging to subset no_sub, passing through voxel no_pix. Even when paths 70 are combined into one, in the first exemplary embodiment, a table summing the previous path lengths is still required.

[0223] When the paths 70 incident on each detector 68 are approximately combined into one, the above formula (2) becomes the following formula (2_3).

[0224] I_e(no_e)=I0_e(no_s,no_d,no_e)×exp(-{(f(no_f(no_pix(no_s,no_d,1)),no_e)×plen_pix(no_s,no_d,1))+(f(no_f(no_pix(no_s,no_d,2)),no_ e)×plen_pix(no_s, no_d, 2))+……+(f(no_f(no_pix(no_s, no_d, num_pix(no_s, no_d))), no_e)×plen_pix(no_s, no_d, num_pix(no_s, no_d)))})…(2_3)

[0225] Then, the intensity I incident on the detector no_d can be obtained by the above-mentioned formula (1) to generate temporary projection data.

[0226] In addition, the above formula (5) becomes the following formula (5_3).

[0227] I_e(no_e)=I0e(no_s,no_d,no_e)×exp(-{(f_ext(1,no_e)×plen_pix(no_s,no_d,1))+(f_ext(2,no_e)×plen_ pix(no_s, no_d, 2))+……+(f_ext(num_pix(no_s, no_d), no_e)×plen_pix(no_s, no_d, num_pix(no_s, no_d)))})

[0228] in,

[0229] f_ext(no_pix_path, no_e)=f(no_f(no_pix(no_s, no_d, no_pix_path)), no_e)+{r_d(no_pix( no_s, no_d, no_pix_path))×(f_d(no_e)-f(no_f(no_pix(no_s, no_d, no_pix_path)), no_e))}

[0230] In the generation of the defect ratio tomographic image of a defect with a small linear attenuation coefficient, f_d(no_e)=f_d_1(no_e)

[0231] In the generation of the defect ratio tomographic image of a defect with a large linear attenuation coefficient, f_d(no_e)=f_d_h(no_e)…(5_3)

[0232] Furthermore, the provisional projection data g_v can be obtained by replacing the above equation (4) with the following equation (4_3). In this case, the above equation (3) is unnecessary.

[0233] g_v=(e(1)×I_e(1))+(e(2)×I_e(2))+……+(e(num_e)×I_e(num_e))…(4_3)

[0234] In addition, Figure 12In step S218 of the defect-specific processing of the first exemplary embodiment shown, the estimated value dr_d(no_s, no_d) of the increment of the defect ratio in the path 70 of the scan no_s incident to the detector no_d is weighted by the path length and added to the no_pix_path voxel no_pix(no_s, no_d, no_pix_path) of the tomographic image dr_d_ave of the average value of the increment, and the formula of the increment obtained by weighting by the path length becomes dr_d(no_s, no_d)×plen_pix(no_s, no_d, no_pix_path)×num_path(no_s, no_d).

[0235] When the paths 70 incident on each detector 68 are approximately combined into one, and when the information on the linear attenuation coefficient of the inspection object 60 is not a single value in two or more energy bands, the intensity I incident on the detector no_d can be calculated by replacing the above equation (2) with the following equation (2_4) to generate provisional projection data. In this case, the above equation (1) is unnecessary.

[0236] I=I0(no_s,no_d)×exp(-{(f(no_f(no_pix(no_s,no_d,1)))×plen_pix(no_s,no_d,1))+(f(no_f(no_pix(no_s,no_d,2)))×plen_pi x(no_s,no_d,2))+……+(f(no_f(no_pix(no_s,no_d,num_pix(no_s,no_d))))×plen_pix(no_s,no_d,num_pix(no_s,no_d)))})…(2_4)

[0237] Alternatively, the provisional projection data g_v can be obtained by replacing the above equation (5) with the following equation (5_4). In this case, the above equations (3) and (4) are unnecessary.

[0238] g_v=I0_e(no_s, no_d)×exp(-{(f_ext(1)×plen_pix(no_s, no_d, 1))+(f_ext(2)×plen_pix(no_ s, no_d, 2))+……+(f_ext(num_pix(no_s, no_d))×plen_pix(no_s, no_d, num_pix(no_s, no_d)))})

[0239] in,

[0240] f_ext(no_pix_path)=f(no_f(no_pix(no_s,no_d,no_pix_path)))+{r_d(no_pix(no_s,no_d,no_pix_path))×(f_d-f(no_f(no_pix(no_s,no_d,no_pix_path))))}

[0241] In the generation of defect ratio tomographic images of defects with small linear attenuation coefficients, f_d = f_d_l

[0242] In the generation of the defect ratio tomographic image of a defect with a large linear attenuation coefficient, f_d = f_d_h...(5_4)

[0243] In addition, in each of the above-mentioned exemplary embodiments, the detection efficiency of the detector 68 at each energy can also be set to de(1), de(2)...de(num_e), and the above-mentioned equations (1) and (4) can be set to the following equations (1_5) and (4_5), respectively, so that the detection efficiency of the detector 68 is reflected in the intensity I of the radiation R.

[0244] I=(e(1)×I_e(1)×de(1))+(e(2)×I_e(2)×de(2))+……+(e(num_e)×I_e(num_e)×de(num_e))…(1_5)

[0245] I(no_path)=(e(1)×I_e(1)×de(1))+(e(2)×I_e(2)×de(2))+……+(e(num_e)×I_e(num_e)×de(num_e))…(45)

[0246] In addition, in the defect specific processing of the first exemplary embodiment (see Figure 12), multiple types may be set for at least one of the defect 61 with a small linear attenuation coefficient and the defect 61 with a large linear attenuation coefficient. For example, as the defect 61 with a small linear attenuation coefficient, in addition to air, a foreign object with a small linear attenuation coefficient may be set. In addition, as the defect 61 with a large linear attenuation coefficient, multiple different types of foreign objects with large linear attenuation coefficients may be set. When multiple types of defects 61 are set, as the linear attenuation coefficients of the defect 61, f_d_l (no_f_d_l, no_e) and f_d_h (no_f_d_h, no_e) are set in advance instead of f_d_l (no_e) and f_d_h (no_e). Here, no_f_d_l and no_f_d_h represent the number of the defect 61 with a small linear attenuation coefficient and the number of the defect 61 with a large linear attenuation coefficient, respectively. Furthermore, f_d_l (no_f_d_l, no_e) and f_d_h (no_f_d_h, no_e) represent the linear attenuation coefficients at energy no_e for defects no_f_d_l and no_f_d_h, respectively. Furthermore, for all types of defects 61 with small linear attenuation coefficients, the linear attenuation coefficient f_d_l is set to be smaller than the linear attenuation coefficient f of all components of the inspection object 60 except air at all energies no_e. For all types of defects 61 with large linear attenuation coefficients, the linear attenuation coefficient f_d_h is set to be larger than the linear attenuation coefficient f of all components of the inspection object except air at all energies no_e. In addition, the linear attenuation coefficient f_d_l of the defect numbered no_f_d_l is smaller than that of the defect numbered no_f_d_l-1 among all energies no_e, and the linear attenuation coefficient f_d_h of the defect numbered no_f_d_h is larger than that of the defect numbered no_f_d_h-1 among all energies no_e.

[0247] Next, the defect specific processing when multiple types of defects 61 are set for the defect 61 with a small linear attenuation coefficient (see Figure 12 ) is described. In addition, the same processing as in the case of setting one defect 61 is omitted. In addition, the description of the defect 61 with a large linear attenuation coefficient is omitted.

[0248] First, in step S204, the defect identification unit 52 generates a defect ratio tomographic image r_d(no_pix) and a defect type tomographic image no_f_d(no_pix). Here, the number of defect types is num_f_d. The defect numbered num_f_d is the defect with the smallest linear attenuation coefficient. Furthermore, as the initial value of no_f_d(no_pix), the value of all voxels is set to 1, the value when the defect has the largest linear attenuation coefficient. Furthermore, as the initial value of r_d(no_pix), the value of all voxels is set to 0.0, the value when no defect exists.

[0249] In addition, in step S212, the above formula (5) is set to the following formula (5_5).

[0250] I_e(no_e)=I0_e(no_e)×exp(-{(f_ext(1,no_e)×plen_pix(no_s,no_d,no_path,1))+(f_ext(2,no_e)×plen_pix(no_s,no_d,no _path, 2))+……+(f_ext(num_pix(no_s, no_d, no_path), no_e)×plen_pix(no_s, no_d, no_path, num_pix(no_s, no_d, no_path)))})

[0251] in,

[0252] When the defect number no_f_d(no_pix(no_s, no_d, no_path, no_pix_path)) is 1,

[0253] f_ext(no_pix_path, no_e)=f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e)+{r_d(no_pix(no_s, no_d, no_path, no_pix_path))×(f_d(1, no_e)-f(no_f(no_pix(no_s, no_d, no_path, no_pix_path)), no_e))}

[0254] When the defect number no_f_d(no_pix(no_s, no_d, no_path, no_pix_path)) is 2 or more,

[0255] f_ext(no_pix_path, no_e)=f_d(no_f_d(no_pix(no_s, no_d, no_path, no_pix_path))-1, no_e)+{r_d(no_pix(no_s, no_d, no_path, no_pix_pa th))×(f_d(no_f_d(no_pix(no_s, no_d, no_path, no_pix_path)), no_e)-f_d(no_f_d(no_pix(no_s, no_d, no_path, no_pix_path))-1, no_e))}

[0256] In the generation of the defect ratio tomographic image of a defect with a small linear attenuation coefficient, f_d(no_f_d, no_e)=f_d_1(no_f_d, no_e)

[0257] In the generation of the defect ratio tomographic image of a defect with a large linear attenuation coefficient, f_d(no_f_d, no_e)=f_d_h(no_f_d, no_e)…(5_5)

[0258] The above formula (5_5) means: the linear attenuation coefficient of the energy no_e in the no_pix_path voxel no_pix through which the path no_path of the scan no_s incident to the detector no_d passes is set to f_ext, and when the defect number no_f_d of the voxel is 1, the linear attenuation coefficient f(no_f(no_pix), no_e) based on the original attenuation information f and shape information no_f is added at a ratio of 1-r_d and the linear attenuation coefficient f_d_l(1, no_e) of the defect number 1 is added at a ratio of r_d is set to f_ext. In addition, it means that when the defect number no_f_d of the voxel is 2 or greater, the linear attenuation coefficient f_dl (no_f_d-1, no_e) of the defect number no_f_d-1 is added at a ratio of 1-r_d and the linear attenuation coefficient f_d_l (no_f_d, no_e) of the defect number no_f_d is added at a ratio of r_d and is set to f_ext.

[0259] In step S212, when multiple types of defects 61 are set, the same as when one type of defect 61 is set. For each detector no_d of scan no_s, the defect ratio increment is calculated so that the temporary projection data g_v is consistent with the measured projection data g(no_s, no_d), and the increment is stored as dr_d(no_s, no_d). Here, when multiple types of defects 61 are set, the defect ratio increment is changed in the range of 0.0 to 1.0 for the defect number 1 (the linear attenuation coefficient is the range of the original linear attenuation coefficient f without defects to the linear attenuation coefficient f_d_l(1) of the defect number 1), the range of 0.0 to 1.0 for the defect number 2 (the linear attenuation coefficient is the range of the linear attenuation coefficient f_d_l(1) of the defect number 1 to the linear attenuation coefficient f_d_l(2) of the defect number 2), and so on. For example, let the increment in the defect ratio of defect number 1 between 0.0 and 1.0 be α, the increment in the defect ratio of defect number 2 between 0.0 and 1.0 corresponding to this increment be γ2×α, and the increment in the defect ratio of defect number 3 between 0.0 and 1.0 be γ3×α, etc. γ2, γ3, etc. are the conversion ratios for the increments in the defect ratio in each interval, determined so that the sum (or average) of the changes in the linear attenuation coefficient at each energy level based on the increments α, γ2×α, γ3×α, etc. in the defect ratio is the same in each interval. Furthermore, when the defect ratio in all voxels (excluding air) passing through all paths incident on detector no_d in each scan no_s is varied (increased or decreased) by α, the defect ratio in each voxel changes according to the following rule: If the defect ratio r_d, obtained by varying the current defect ratio r_d by γ×α, is within the defect ratio range of 0.0 to 1.0, the defect ratio remains unchanged. Furthermore, γ represents the conversion ratio for the defect ratio of the defect number no_f_d in the voxel between 0.0 and 1.0. If the defect ratio is less than 0.0, when the defect number no_f_d is 1, the defect ratio is set to 0.0, and the linear attenuation coefficient f_ext is not set to a value smaller than the original linear attenuation coefficient f without the defect. Furthermore, when the defect number no_f_d is 2 or greater, the defect ratio is first changed to 0.0 within the interval of the defect number no_f_d. Then, 1 is subtracted from no_f_d, and the defect ratio is changed from 1.0 within the interval of the defect number from which 1 was subtracted (by multiplying the conversion ratio within that interval).In addition, in the case of a defect ratio greater than 1.0, when the defect number no_f_d is num_f_d (maximum), the defect ratio is set to 1.0. In addition, first, the defect ratio is changed to 1.0 in the interval of the defect number no_f_d, and then no_f_d is increased by 1, and the defect ratio is changed from 0.0 in the interval of the defect number increased by 1 by the remaining amount (multiplied by the transformation ratio in the interval).

[0260] Furthermore, in step S218, when multiple defect types are set, similarly to setting a single defect, the tomographic image is updated based on the defect ratio increments dr_d(no_s, 1) to dr_d(no_s, num_d) calculated for all projection data belonging to subset j. In the tomographic image dr_d_ave, which represents the average of these increments, the increments weighted by the path length are added to each voxel passed through by all paths 70 of all detectors 68 across all scans belonging to subset j. Finally, the average of these increments is calculated for each voxel by dividing the dr_d_ave value for each voxel by the sum of the path lengths for each voxel, plen_pix_sub. This average is then added to the tomographic image r_d before the update to update the tomographic image. Here, if there is a voxel where the result of adding the defect ratio dr_d before the update is a defect number no_f_d of num_f_d (maximum) and the defect ratio exceeds the maximum value of 1.0, the increment for this voxel is corrected so that the result of the addition is a defect number no_f_d of num_f_d (maximum) and the defect ratio is the maximum value of 1.0, and the increment is added to dr_d_ave. Similarly, if there is a voxel where the result of adding the defect ratio dr_d before the update is a defect number no_f_d of 1 and the defect ratio is less than the minimum value of 0.0, the increment for this voxel is corrected so that the result of the addition is a defect number no_f_d of 1 and the defect ratio is the minimum value of 0.0, and the increment is added to dr_d_ave. When multiple types of defects are set, the defect ratio tomographic image r_d(no_pix) is updated along with the defect type tomographic image no_f_d(no_pix).

[0261] When multiple types of defects are set, the defects are identified along with their types and three-dimensional positions based on the defect ratio tomographic image and the defect type tomographic image. For example, in the defect ratio tomographic image, it is possible to identify the defects along with the defect types as follows: an area with a defect number of 1 and a defect ratio below a preset value (e.g., 0.8) is identified as having no defects; an area with a defect number of 1 and a defect ratio of a preset value (e.g., 0.8) or higher, and an area with a defect number of 2 and a defect ratio of 0.5 or less is identified as a defect number of 1; an area with a defect ratio of 2 and a defect ratio greater than 0.5, and an area with a defect ratio of 3 and a defect ratio of 0.5 or less is identified as a defect number of 2, and so on. Alternatively, identification can be performed after temporarily converting to a defect ratio tomographic image that also includes the defect types. For example, identification can be performed along with the defect type as follows: in a defect ratio tomographic image, the defect ratio of a voxel numbered 2 is added to 1.0, the defect ratio of a voxel numbered 3 is added to 2.0, and so on. After these additions are made, regions with a defect ratio lower than a predetermined value (e.g., 0.8) are identified as having no defects; regions with a defect ratio greater than or equal to a predetermined value (e.g., 0.8) and less than or equal to 1.5 are identified as having defect number 1; regions with a defect ratio greater than or equal to 1.5 and less than or equal to 2.5 are identified as having defect number 2, etc. Alternatively, after conversion to a defect ratio tomographic image that also includes the defect type, regions within a predetermined range of values ​​centered around the defect ratio value (0.0, 1.0, 2.0, etc.) are extracted from the defect ratio tomographic image for each defect number, and then connected regions (connected regions) are identified. Only regions in each connected region where the sum of the defect ratios in the region is greater than or equal to the predetermined value corresponding to the defect number (regions with large defect ratio values ​​and volumes) are identified as defects of each number. After the defect is identified, it is sufficient to display it in a manner that allows the three-dimensional position, size, and shape of the defect identified inside the inspection object to be known and in a manner that allows the type of the defect to be known.

[0262] In the defect specific processing of the second exemplary embodiment (refer to Figure 14), a plurality of types may be set for at least one of defects with a small linear attenuation coefficient and defects with a large linear attenuation coefficient. In this case, in the second exemplary embodiment, a defect type tomographic image no_f_d(no_pix) is also generated while generating a defect ratio tomographic image r_d(no_pix). In step S209 of the defect specific processing of the second exemplary embodiment, when the gradient is obtained by temporarily increasing the defect ratio by a predetermined amount α for each voxel of the tomographic image, the defect ratio r_d in each voxel is temporarily increased by γ×α obtained by multiplying the defect ratio γ of the defect number no_f_d of the voxel. Then, if the result of the increase by γ×α is greater than the defect ratio 1.0, after the defect ratio is increased to 1.0 in the interval of the defect number no_f_d, no_f_d is increased by 1, and the remaining amount (multiplied by the conversion ratio in the interval) is increased in the interval of the defect number increased by 1. Thus, after temporarily increasing the defect ratio (and, if necessary, increasing the defect number), each g_v(no_s, no_d) is calculated using equations (3), (4), and (5-5). Then, the evaluation function ef is calculated using equations (6) or (7). The gradient is calculated by subtracting the evaluation function ef calculated using equations (3), (4), (5-5), and (6) (or (7)) based on the current tomographic image r_d and no_f_d from this evaluation function ef. In addition, for a voxel in which the defect number no_f_d is num_f_d (maximum) and the defect ratio exceeds the maximum value 1.0 as a result of temporarily increasing the defect ratio by a specified amount α, the gradient can be obtained by subtracting the evaluation function ef calculated by setting the defect number no_f_d to num_f_d and the defect ratio to 1.0 in the voxel from the evaluation function ef calculated by setting the defect number no_f_d to num_f_d and the defect ratio to 1.0.

[0263] In the defect specific processing of the second exemplary embodiment (refer to Figure 14In step S218, the tomographic image r_d is updated according to the following rules. If the updated defect ratio r_d(no_pix) - γ × β × Δef(no_pix) (γ is the transformation ratio for the interval of defect number no_f_d in that voxel) for each voxel no_pix is ​​within the defect ratio range of 0.0 to 1.0, the image is updated as is. If the updated defect ratio r_d(no_pix) - γ × β × Δef(no_pix) is less than 0.0, the defect ratio is set to 0.0 when the defect number no_f_d is 1. If the defect number no_f_d is 2 or greater, the defect ratio is first changed to 0.0 for the interval of the defect number no_f_d. Then, no_f_d is decremented by 1, and the defect ratio is updated by the remainder of the change from 1.0 (multiplied by the transformation ratio for that interval) for the interval of the defect number that was decremented by 1. Furthermore, if the updated defect ratio r_d(no_pix) - γ × β × Δef(no_pix) exceeds the defect ratio of 1.0, and if the defect number no_f_d reaches num_f_d (maximum), the defect ratio is set to 1.0. Furthermore, first, the defect ratio is changed to 1.0 in the interval corresponding to the defect number no_f_d. Then, no_f_d is incremented by 1, and the defect ratio is updated by the remainder of the interval (multiplied by the conversion ratio in that interval) in the interval corresponding to the defect number incremented by 1. Following the above rule, the defect ratio tomographic image r_d(no_pix) is updated simultaneously with the defect type tomographic image no_f_d(no_pix).

[0264] In addition, in the defect specific processing of the first exemplary embodiment (see Figure 12), in step S212, instead of calculating the defect ratio difference dr_d(no_s, no_d), rr_d(no_s, no_d) may be calculated as the defect ratio ratio. Specifically, when generating a defect ratio tomographic image for defects with small linear attenuation coefficients, the provisional projection data g_v is compared with the measured projection data g(no_s, no_d). If the provisional projection data g_v is larger or smaller, the defect ratio of all voxels passing through all paths incident on the detector no_d during scan no_s is reduced or increased by a predetermined amount. The provisional projection data g_v is then calculated again and compared with the measured projection data g(no_s, no_d). This process is repeated until g_v is equal to or greater than g(no_s, no_d). Then, the following data is used: the temporary projection data g_v that is closer to the measured projection data g(no_s, no_d) than the temporary projection data g_v that is closer to the measured projection data g(no_s, no_d) or the temporary projection data g_v that is closer to the measured projection data g(no_s, no_d). The defect ratio ratio calculated during this calculation may also be stored as rr_d(no_s, no_d). Here, "subtracting or adding a predetermined amount to the ratio" means determining a predetermined amount increment in the ratio and multiplying the ratio by the ratio while subtracting or adding the increment. For example, determining a predetermined amount increment in the ratio as 0.01 means multiplying the ratio by 0.99, 0.98, and so on, or by 1.01, 1.02, and so on. To calculate the provisional projection data g_v by subtracting or adding the ratio α to the predetermined defect ratio, in equation (5) above, r_d(no_pix(no_s, no_d, no_path, no_pix_path)) of f_ext(no_pix_path, no_e) can be replaced with α×r_d(no_pix(no_s, no_d, no_path, no_pix_path)). If there is a voxel where α×r_d(no_pix(no_s, no_d, no_path, no_pix_path)) exceeds the maximum defect ratio value of 1.0, the defect ratio for that voxel is fixed at the maximum value of 1.0.

[0265] When calculating the defect ratio rr_d(no_s,no_d), in the defect specific processing (refer to Figure 12In step S218 of the present invention, the tomographic image of the average value of the ratios is first set to rr_d_ave(no_pix), and all voxels are set to zero to initialize rr_d_ave(no_pix). Then, for all projection data belonging to subset j, the defect ratio ratios rr_d(no_s, 1) to rr_d(no_s, num_d) are weighted by the path length and added to each voxel in the tomographic image of the average value of the ratios rr_d_ave. Finally, the average value of the ratios is calculated for each voxel by dividing the value of each voxel in rr_d_ave by the sum of the path lengths of each voxel, plen_pix_sub. The tomographic image is updated by multiplying the tomographic image r_d before the update by this ratio. When the defect ratio ratio rr_d(no_s, no_d) is added to each voxel of the tomographic image rr_d_ave of the average value, if there is a voxel in which the result of multiplying the tomographic image r_d before the update by the ratio exceeds the maximum value of the defect ratio, for this voxel, the ratio is corrected so that the multiplication result reaches the maximum value of 1.0 and then added to rr_d_ave.

[0266] In addition, in the defect specific processing of each exemplary embodiment described above (see Figure 12 、 Figure 14 ), it may also be that after the defect ratio tomographic image r_d is generated, in other words, after the overall update number i exceeds the prescribed number Ni and the generation of the defect ratio tomographic image is completed, the regions (connected regions) whose defect ratios are greater than a predetermined value (for example, 0.1) and are connected are respectively identified, and the region in each connected region where at least one of the average value and the total value of the defect ratio in the region is greater than a predetermined value and the total value is the largest (the region with the largest defect ratio value and volume) is determined (specifically) to be a defect, and after the defect ratios of all voxels in the region are fixed at 1.0, the process returns to step S206, the overall update number i is set to 1 again, and the generation of the defect ratio tomographic image is repeated again. Alternatively, the determination (specification) of defects and the generation of defect ratio tomographic images may be repeated in this manner (determining the region with the largest sum of defect ratios as a defect → fixing the defect ratio of the region where defects are determined to be 1.0 and generating a defect ratio tomographic image → determining the region with the largest sum of defect ratios as a defect → fixing the defect ratio of the region where defects are determined to be 1.0 and generating a defect ratio tomographic image → ... and repeating this), and when no connected region is observed in which at least one of the average value and the sum of the final defect ratios is greater than a predetermined value, the generation of the defect ratio tomographic image is completed. In this way, by determining the most likely defect, the remaining defect ratio tomographic images can be generated more accurately. That is, by repeating the determination of the most likely defect and the generation of the remaining defect ratio tomographic images, the defect ratio tomographic image can be generated more accurately, and the defect can be specified more accurately.

[0267] In the defect specific processing of each exemplary embodiment described above (see Figure 12 and Figure 14 ), the defect ratio tomographic image may be generated by subtracting noise from the measured projection data g(no_s, 1) to g(no_s, num_d). Specifically, in generating a defect ratio tomographic image of a defect 61 having a small linear attenuation coefficient, for data in which the measured projection data g minus the provisional projection data g_v is less than a predetermined value among the measured projection data g(no_s, 1) to g(no_s, num_d), the value of the measured projection data g is replaced in advance with the value of the provisional projection data g_v, along with data in which the measured projection data g is smaller than the provisional projection data g_v when no defect is present. The data in which the measured projection data g minus the provisional projection data g_v is less than a predetermined value means that even if the measured projection data g is larger, the difference therebetween is less than the predetermined value, and the provisional projection data g_v when no defect is present refers to the provisional projection data g_v obtained from the original attenuation information f(no_f, no_e) and shape information no_f(no_pix) by using the above-mentioned equations (3) and (4), and equation (5) in which the defect ratio r_d is set to 0. In addition, in the generation of defect ratio tomographic images of defects with large linear attenuation coefficients, conversely, for data where the temporary projection data g_v minus the measured projection data g is less than a preset value, together with data where the measured projection data g is larger than the temporary projection data g_v when the defect is not included, the value of the measured projection data g is replaced in advance with the value of the temporary projection data g_v, wherein the data where the temporary projection data g_v minus the measured projection data g is less than a preset value means that even if the measured projection data g is smaller, the difference is less than the preset value. Thus, among the measured projection data g(no_s, 1) to g(no_s, num_d), only the data with a higher reliability whose difference from the temporary projection data when the defect is not included is greater than the preset value is retained, and a defect ratio tomographic image with less noise can be generated based only on the highly reliable projection data. However, when the preset value is set to a large value, in other words, when the difference between the measured projection data and the temporary projection data is excessively set to 0, it is sometimes difficult to detect. Figure 2 The thin planar defect 61 shown or the small non-planar defect 61 requires attention.

[0268] The three-dimensional position, shape, and size of the defect can also be specified by the following method: among the measured projection data g(no_s, 1) to g(no_s, num_d), only the data whose difference with the temporary projection data when the defect is not included is larger than a predetermined value is retained, and then the defect is simply regarded as existing in the area where the path of the radiation R corresponding to each projection data whose difference with the retained temporary projection data is not 0 (in each scan no_s, the path of the radiation R emitted from the radiation source 64 to the detector no_d corresponding to each projection data whose difference is not 0) intersects or is densely packed. However, in the case of this method, if the noise is not reliably eliminated in each measured projection data, specifically, if the difference with the temporary projection data of the noise is not reliably set to 0, many intersection areas and dense areas of the path corresponding to the noise may sometimes be specified as many erroneous defects. However, when the predetermined value is set to a large value that can reliably eliminate the noise in each projection data, Figure 2 Differences due to thin planar defects 61, as shown, or differences due to small, non-planar defects 61 (the difference between the measured projection data and the provisional projection data excluding the defect) also disappear, potentially making defect detection difficult. Specifically, with this method, if the differences with noise (with the provisional projection data) and the differences with defects in each measured projection data are not clearly distinguished in magnitude, and if the differences with noise are not reliably eliminated to a predetermined value, leaving only the differences with defects, there is a risk of erroneous detection of defects or failure to detect them.

[0269] As described in the exemplary embodiments above, the defect identification method (including the method described above for subtracting noise based on a preset value) generates a defect ratio tomographic image by differentially matching the measured projection data in each projection direction with the temporary projection data when the defect is not present. Even defects that are difficult to distinguish from noise based solely on the projection data (thin planar defects or small defects) can be identified by visualizing them as regions where the defect ratio and size are clearly different from those of the noise in the defect ratio tomographic image generated by matching these projection data. Furthermore, the three-dimensional shape of the defect can be visualized. Furthermore, compared to conventional techniques for detecting defects by irradiating an inspection object with radiation from multiple different directions to acquire projection data and generate tomographic images of the inspection object, the defect identification method, as described in the exemplary embodiments above, by differentially matching the measured projection data in each projection direction with the temporary projection data allows for high-precision detection of defects based on a relatively small amount of projection data or projection data with limited projection directions. In particular, even thin planar defects or small defects can be detected with high precision based on the differentials in the projection data.

[0270] As described above, projection data includes not only direct radiation emitted from the radiation source and attenuated by passing through the inspection object, but also scattered radiation generated as new radiation from a portion of the attenuated direct radiation as the direct radiation attenuates. In particular, the cone beam method includes a large number of scattered radiation. Therefore, in each of the exemplary embodiments described above, it is preferable to also consider scattered radiation when generating a defect ratio tomographic image. The probability of occurrence of scattered radiation at each scattering angle is represented by p_s(no_e_in, no_e_out, no_th). no_e_in represents the energy of the incident radiation, no_e_out represents the energy of the scattered radiation, and no_th represents the angle (hereinafter referred to as the scattering angle) of the direction of the scattered radiation relative to the direction of the incident radiation (hereinafter referred to as the incident direction). The occurrence probability p_s represents the probability that, when radiation with energy no_e_in travels a unit distance, scattered radiation with energy no_e_out will be generated at an angle no_th relative to the direction of the radiation. In addition, in this exemplary embodiment, since the energy no_e_in of the incident radiation, the energy no_e_out of the scattered radiation, and the scattering angle no_th are handled discretely, they are represented by numbers. Figure 15 In the figure, the direction of the incident light is indicated by a thick solid arrow, the direction of the scattered light is indicated by a thin solid arrow, the scattering angle is indicated by θ, the direction of the scattered light projected on the plane (xy plane) at right angles to the incident light is indicated by a thin wavy arrow, and the angle between the direction of the scattered light projected on the xy plane at right angles to the incident light and the x-axis (positive direction) is indicated by φ. Figure 15 As shown, the scattering angle θ is the polar angle (zenith angle) in three-dimensional polar coordinates with the incident direction being the z-axis (positive direction), and the angle φ is the azimuth angle (declination angle). The occurrence probability p_s is equiprobable with respect to the angle φ.

[0271] When radiation passes through an object under inspection, the probability of each interaction, such as the photoelectric effect, interferometric scattering (Thomson scattering), non-interferometric scattering (Compton scattering), and electron pair formation, varies depending on the energy of the radiation. Furthermore, the angular distribution of scattered radiation, that is, the probability of scattering at each angle, varies depending on the type of interaction. Therefore, in this exemplary embodiment, the sum of the probabilities of scattered radiation at scattering angle no_th due to each interaction, which varies depending on the energy no_e_in, is expressed as p_s(no_e_in, no_e_out, no_th). In other words, the sum of the probabilities of scattered radiation due to each interaction at each energy no_e_in is expressed as a function p_s(no_e_in, no_e_out, no_th). Furthermore, depending on the type of interaction, the energy before and after scattering varies. For example, in the case of interferometric scattering, the energy of the incident radiation is the same as the energy of the scattered radiation, but in the case of non-interferometric scattering, the energy of the incident radiation is different from the energy of the scattered radiation, and the energy of the scattered radiation varies depending on the scattering angle. Therefore, in this exemplary embodiment, the probability of scattered radiation occurrence p_s is determined as a function of the energy no_e_in of the incident radiation and the energy no_e_out of the scattered radiation. Furthermore, since the probability of scattered radiation occurrence p_s varies depending on the component included in the inspection object, in this exemplary embodiment, each component has a specific probability p_s. The probability of scattered radiation occurrence p_s can be determined by actually measuring the intensity of scattered radiation at each angle no_th with each energy no_e_out for each component, or it can be determined theoretically. For example, in the case of non-interferometric scattering, the probability of scattering at each angle can be calculated using the Klein-Nishina equation. The probability of scattered radiation occurrence p_s can also be assigned using any method. For example, it can be assigned using a three-dimensional table (no_e_in, no_e_out, no_th) or a function with the three variables no_e_in, no_e_out, and no_th as arguments. In this exemplary embodiment, the sum of the probabilities of occurrence of scattered rays due to various interactions is expressed as a function p_s. However, it may be given by a combination of the probability of occurrence of each interaction at each energy no_e_in and the probability of scattering at each energy no_e_out and each angle no_th when each interaction occurs.

[0272] Scatter calculations are computationally intensive and time-consuming. In this exemplary embodiment, after the imaging position identification step is completed and the imaging position of each projection data item relative to the inspection object is determined, various tables are generated before defect identification processing, and then scatter calculations are performed. Because the component of each voxel is known based on the inspection object's shape information no_f (no_pix), the scatter ray occurrence probability p_s for the corresponding component can be assigned to each voxel to perform scatter calculations. In the scatter calculation, scattered ray projection data are generated as g_s(1, 1) to g_s(num_s, num_d) based on the inspection object's shape information no_f (no_pix), the scattered ray occurrence probability p_s for each component, and the imaging position of each projection data item relative to the inspection object. Then, the calculated scattered ray projection data g_s(1,1) to g_s(num_s, num_d) are subtracted from the measured projection data g(1,1) to g(hum_s, num_d), and only the measured direct ray projection data are obtained as g_p(1,1) to g_p(num_s, num_d). In addition, num_s represents the number of scans (the number of projection directions). Figure 12 or Figure 14 In the defect identification process shown, a defect ratio tomographic image is generated using only the direct line projection data g_p(1,1) to g_p(num_s, num_d) instead of the measured projection data g(no_s, 1) to g(no_s, num_d). Internal defects within the inspection object are very small, and their influence on the scattered radiation generated by the inspection object is minimal. Therefore, scatter calculations can be performed only once, after the imaging position of each projection data relative to the inspection object has been determined.

[0273] The following describes a method for calculating scattering based on the probability of scattered ray occurrence p_s(no_e_in, no_e_out, no_th) for each voxel and generating scattered ray projection data g_s(1, 1) to g_s(num_s, num_d). When calculating scattering by only calculating the first scattering (hereinafter referred to as primary scattering) and omitting multiple scattering (second, third, and so on), the scattered ray projection data g_s can be generated as follows. Focusing on the no_pix_path-th voxel no_pix(no_s, no_d, no_path, no_pix_path) through which the path no_path incident on the detector no_d of the scan no_s passes, the intensity of the scattered ray generated when radiation of energy no_e on the path no_path is attenuated in the voxel no_pix(no_s, no_d, no_path, no_pix_path) is defined as I_pix_out(no_e_out, no_th) and assigned using the following equation (9).

[0274] I_pix_out(no_e_out,no_th)=I_pix(no_e)×plen_pix(no_s,no_d,no_path,no_pix_path)×p_s(no_e,no_e_out,no_th)…(9)

[0275] In equation (9), I_pix(no_e) represents the intensity of radiation with energy no_e on path no_path when it enters voxel no_pix(no_s, no_d, no_path, no_pix_path). I_pix_out(no_e_out, no_th) in equation (9) represents the intensity of radiation with incident intensity I_pix(no_e) scattered at each scattering angle no_th relative to the incident direction of path no_path and at each energy no_e_out, according to the probability p_s(no_e, no_e_out, no_th), as radiation with incident intensity I_pix(no_e) decays along path no_path over path length plen_pix(no_s, no_d, no_path, no_pix_path). In equation (9), it should be noted that the scattering angle no_th is the polar angle (zenith angle) relative to the incident direction of path no_path. In equation (9), the scattering probability p_s(no_e, no_e_out, no_th) is equiprobable with respect to azimuth (declination). As a result, the scattering intensity I_pix_out(no_e_out, no_th) is uniform with respect to azimuth (declination). Furthermore, when calculating the scattering intensity I_pix_out(no_e_out, no_th) for incident radiation of energy no_e using equation (9), calculations may be omitted for combinations of no_e_out and no_th for which the scattering probability p_s(no_e, no_e_out, no_th) is zero. After the intensity I_pix_out (no_e_out, no_th) of the scattered rays scattered at each energy no_e_out and each scattering angle no_th is calculated by formula (9), the detector into which each scattered ray enters can be identified among detectors 1 to num_d through geometric calculation based on the positional relationship between the path no_path in the voxel no_pix (no_s, no_d, no_path, no_pix_path) and the detectors 1 to num_d that scan no_s (assuming that each scattered ray does not undergo multiple scattering (scattering after the second time)). Similarly, geometric calculations can be used to determine the voxels through which each scattered ray passes after being generated along the path no_path in voxel no_pix (no_s, no_d, no_path, no_pix_path) at the scattering angle no_th and reaching the detector, as well as the path lengths in these voxels. Therefore, based on the attenuation coefficient and path length at each energy no_e_out of each voxel passed through, the intensity of the scattered ray generated at each energy no_e_out and each scattering angle no_th when attenuated and incident on the detector can be determined.Furthermore, scattered radiation generated along the path of a direct ray incident on detector no_d can be detected in all detectors 1 to num_d in the cone beam method, but can only be detected in detector no_d in the pencil beam method. In the fan beam method, it can only be detected in the one-dimensionally arranged detectors to which detector no_d belongs. Taking into account the fact that the detectors capable of detection are limited by the scanning method, the detector on which scattered radiation generated along the path of a direct ray incident on detector no_d is identified after the detector capable of detecting scattered radiation is limited based on detector number no_d. The calculations described above allow for the determination of the intensity of scattered radiation incident on each detector in voxel no_pix(no_s, no_d, no_path, no_pix_path) no_pix_path through which path no_path incident on detector no_d passes, resulting from the attenuation of radiation of energy no_e along path no_path, in scan no_s. Similarly, calculations can be performed for all energies 1 to num_e, all voxels 1 to num_pix(no_s, no_d, no_path), all paths 1 to num_path(no_s, no_d), and all detectors 1 to num_d. Through the above calculations, the sum of the intensities of scattered rays incident on each detector 1 to num_d during the scan no_s at each energy 1 to num_e can be calculated. Furthermore, for each detector 1 to num_d scanning no_s, the intensities of scattered rays at each energy are multiplied by the energy and then added together to generate the detected intensities, i.e., projection data g_s(no_s, 1) to g_s(no_s, num_d) (see equations (1) and (4) above). Furthermore, when summing the intensities of scattered rays at each energy, the detection efficiency of the detector can also be reflected (see equations (1-5) or (4-5) above). Similarly, projection data g_s(1, 1) to g_s(num_s, num_d) for all scans can be generated. Furthermore, the detectors that each scattered ray generated along the path of the direct ray enters, the voxels that each scattered ray passes through before reaching the detector, and the path lengths in these voxels can be calculated in advance and tabulated.The method described above can generate projection data g_s(1,1) to g_s(num_s, num_d) of only single-scattered rays. However, projection data g_s(1,1) to g_s(1,num_d), g_s(2,1) to g_s(2,num_d), ..., of each scan may be blurred to virtually generate projection data of multiple scattered rays (secondary, tertiary, ... scattered rays), and then added to the original projection data. In this way, g_s(1,1) to g_s(num_s, num_d) may be converted into projection data of single-scattered rays plus multiple scattered rays.

[0276] Furthermore, the degree of blurring of projection data varies depending on the scanning method. For example, in the fan-beam method, the two-dimensional projection data from detectors 1 to num_d are blurred more strongly in the one-dimensional direction corresponding to the same beam (radiation source) than in the direction orthogonal to it in each scan. Here, each one-dimensional data point represents primary scattered rays generated by different beams, but data points with very close beams are treated as if they were generated by the same beam, and are also blurred in the direction orthogonal to the one-dimensional direction (blurring in two dimensions). Furthermore, in the pencil-beam method, the projection data from detectors 1 to num_d are blurred less strongly in two dimensions in each scan than in the cone-beam or fan-beam methods. Here, each data point represents primary scattered rays generated by different beams, but data points with very close beams are treated as if they were generated by the same beam, and are blurred in two dimensions.

[0277] The following method will be described. Based on the scattered radiation occurrence probability p_s (no_e_in, no_e_out, no_th) for each voxel, calculations are performed to include not only primary scattering but also multiple scattering to generate projection data g_s(1, 1) to g_s(num_s, num_d). When multiple scattering is also included, projection data can be generated using Monte Carlo simulation. Specifically, a large number of photons of radiation of various energies no_e are emitted in various directions from a radiation source 64. For each photon, based on the linear attenuation coefficient and the scattered radiation occurrence probability p_s (no_e_in, no_e_out, no_th) for each voxel 74, a determination is made as to whether the photon travels in a straight line, is absorbed, or is scattered. If the photon is scattered, the determination of the direction of scattering is repeated until the photon reaches the detector 68. This allows the generation of projection data g_s(1, 1) to g_s(num_s, num_d). Note that the settings of the radiation source 64 and the detector 68 capable of detecting scattered radiation differ depending on the scanning method, such as the pencil beam method, the fan beam method, and the cone beam method.

[0278] In each of the exemplary embodiments described above, neither the imaging position specific processing nor the defect specific processing takes into account the attenuation of the radiation R in the area outside the inspection object 60 (the area outside the inspection object 60). The area outside the inspection object 60 is filled with a material corresponding to the environment in which the inspection object 60 exists, and its linear attenuation coefficient is known. In addition, in each imaging position, the path length from the radiation source 64 to the area of ​​the inspection object 60 and the path length from the area of ​​the inspection object 60 to the detector 68 can be obtained in each path 70 where the radiation emitted from the radiation source 64 is incident on each detector 68. Therefore, temporary projection data can be obtained more accurately based on this information. For example, if the area outside the inspection object 60 is filled with air, the attenuation in the area outside the inspection object 60 can be calculated based on the linear attenuation coefficient of air, thereby more accurately obtaining temporary projection data. As a result, the imaging position can be specified more accurately, and a defect ratio tomographic image can be generated more accurately to specify the defect. However, the presence or absence of calculation of attenuation in the region other than the inspection object 60 is considered to have little influence on the accuracy of identification of the imaging position and identification of the defect 61 .

[0279] Furthermore, in each of the above-described embodiments, for example, various processors described below can be used as the hardware configuration of processing units that execute various processes, such as the first acquisition unit 40, the imaging position identification unit 42, the first provisional projection data generation unit 44, the second acquisition unit 50, the defect identification unit 52, and the second provisional projection data generation unit 54. These various processors include, in addition to general-purpose processors (CPUs) that execute software (programs) to function as various processing units, as described above, processors whose circuit configuration can be modified after manufacture, such as FPGAs (Field Programmable Gate Arrays), programmable logic devices (PLDs), and processors (specific circuits) with circuit configurations specifically designed to execute specific processes, such as ASICs (Application Specific Integrated Circuits).

[0280] A processing unit may be composed of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, a combination of a CPU and an FPGA). Alternatively, multiple processing units may be composed of one processor.

[0281] As examples of multiple processing units composed of a single processor, there is a first approach, typified by computers such as clients and servers, where a single processor is composed of a combination of one or more CPUs and software, with this processor functioning as multiple processing units. A second approach, typified by systems on a chip (SoC), uses a processor that implements the functions of the entire system, including multiple processing units, on a single integrated circuit (IC). In this way, various processing units are constructed using one or more of the aforementioned processors as hardware structures.

[0282] Furthermore, as the hardware configuration of these various processors, more specifically, a circuit (circuitry) formed by combining circuit elements such as semiconductor elements can be used.

[0283] In addition, in the above exemplary embodiments, the imaging position specifying program 30 and the defect specifying program 32 are pre-stored (installed) in the storage unit 24 of the information processing device 10, but the present invention is not limited to this. The imaging position specifying program 30 and the defect specifying program 32 can be provided by recording them on a recording medium such as a CD-ROM (Compact Disc Read Only Memory), a DVD-ROM (Digital Versatile Disc Read Only Memory), or a USB (Universal Serial Bus) memory device. Alternatively, the imaging position specifying program 30 and the defect specifying program 32 can be downloaded from an external device via a network.

[0284] Based on the above description, the invention described in the following supplementary notes can be grasped.

[0285] [Note 1]

[0286] An information processing device, wherein:

[0287] having at least one processor,

[0288] The processor

[0289] Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object,

[0290] A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

[0291] [Note 2]

[0292] The information processing device according to Supplementary Note 1, wherein:

[0293] The attenuation information includes attenuation information corresponding to at least two components included in the inspection object.

[0294] [Note 3]

[0295] The information processing device according to Supplement 1 or 2, wherein:

[0296] The attenuation information includes air attenuation information.

[0297] [Note 4]

[0298] The information processing device according to any one of Supplementary Notes 1 to 3, wherein:

[0299] The attenuation information includes attenuation information in at least two energy bands.

[0300] [Note 5]

[0301] The information processing device according to any one of Supplementary Notes 1 to 4, wherein:

[0302] The processor

[0303] obtaining attenuation information of the defect,

[0304] The defect is identified based on the attenuation information of the defect.

[0305] [Note 6]

[0306] The information processing device according to any one of Supplementary Notes 1 to 5, wherein:

[0307] The processor

[0308] The defect is identified by distinguishing between a region where the inspection object does not exist and a region where the inspection object exists in the distribution of the inspection object in the three-dimensional space based on the shape information.

[0309] [Note 7]

[0310] The information processing device according to any one of Supplementary Notes 1 to 6, wherein:

[0311] The processor

[0312] Based on the attenuation information and shape information of the inspection object, temporary projection data obtained when the inspection object is irradiated with radiation is generated.

[0313] The defect is identified based on the temporary projection data.

[0314] [Note 8]

[0315] The information processing device according to Supplementary Note 7, wherein:

[0316] The processor

[0317] The defect is identified by distinguishing data in the projection data that is different in magnitude from the temporary projection data.

[0318] [Note 9]

[0319] The information processing device according to Supplementary Note 7, wherein:

[0320] The processor

[0321] The defect is identified by distinguishing data having a difference from the provisional projection data in the projection data.

[0322] [Note 10]

[0323] The information processing device according to any one of Supplementary Notes 1 to 9, wherein:

[0324] The processor identifies the defect based on information on scattering of the inspection object.

[0325] [Note 11]

[0326] An information processing method, wherein a processor performs the following processing:

[0327] Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object,

[0328] A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

[0329] [Note 12]

[0330] An information processing program configured to cause a processor to perform the following processing:

[0331] Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object,

[0332] A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

[0333] Application filed in Japan on January 23, 2023: The disclosure of Japanese Patent Application No. 2023-008366 is incorporated herein by reference in its entirety.

[0334] All documents, patent applications, and technical specifications described in this specification are incorporated herein by reference to the same extent as if each individual document, patent application, or technical specification was specifically and individually described as being incorporated by reference.

Claims

1. An information processing device, wherein: having at least one processor, The processor Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object, A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

2. The information processing device according to claim 1, wherein The attenuation information includes attenuation information corresponding to at least two components included in the inspection object.

3. The information processing device according to claim 1, wherein The attenuation information includes air attenuation information.

4. The information processing device according to claim 1, wherein: The attenuation information includes attenuation information in at least two energy bands. The information processing device according to claim 1 , wherein: The processor obtaining attenuation information of the defect, The defect is identified based on the attenuation information of the defect. The information processing apparatus according to claim 1 , wherein: The processor The defect is identified by distinguishing between a region where the inspection object does not exist and a region where the inspection object exists in the distribution of the inspection object in the three-dimensional space based on the shape information.

7. The information processing apparatus according to claim 1, wherein: The processor Based on the attenuation information and shape information of the inspection object, temporary projection data obtained when the inspection object is irradiated with radiation is generated. The defect is identified based on the temporary projection data.

8. The information processing apparatus according to claim 7, wherein: The processor The defect is identified by distinguishing data in the projection data that is different in magnitude from the temporary projection data.

9. The information processing apparatus according to claim 7, wherein: The processor The defect is identified by distinguishing data having a difference from the provisional projection data in the projection data.

10. The information processing apparatus according to claim 1, wherein: The processor The defect is identified based on information on scattering of the inspection object.

11. An information processing method, wherein: The processor performs the following processing: Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object, A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

12. An information processing program configured to cause a processor to execute the following processing: Acquiring projection data corresponding to at least two imaging positions having different irradiation directions of the inspection object, A defect of the inspection object is identified based on the projection data, the attenuation information of the inspection object, and the shape information of the inspection object.

Images