Radiation image acquisition device, radiation image acquisition system and radiation image acquisition method

Abstract

The image acquisition apparatus (1) of the present invention comprises: a camera (10) that scans and captures radiation passing through an object in one direction to acquire an X-ray image; a scintillator (11) disposed on the camera (10) to convert X-rays into light; and a control device (20) that inputs the X-ray image into a learned model pre-constructed using image data through machine learning and performs noise removal processing to remove noise from the X-ray image. The camera (10) comprises: a scanning camera (12) which is composed of pixel lines (74) having M (M is an integer greater than 2) pixels (72) arranged in one direction arranged in N columns (N is an integer greater than 2) in a direction orthogonal to one direction, and outputs a light-related detection signal for each pixel (72); and a readout circuit (73) that adds the detection signals output from at least 2 of the M pixels (72) for each pixel line (74) in the N columns of the scanning camera (12), and sequentially outputs the added N detection signals to output an X-ray image.

Description

Technical Field

[0001] One aspect of the implementation relates to a radiation image acquisition apparatus, a radiation image acquisition system, and a radiation image acquisition method. Background Technology

[0002] Currently, an apparatus is used that obtains image data by arranging multiple rows of line sensors orthogonal to the transport direction of the object, and summing the detection data output from the multiple rows of line sensors. This apparatus allows the acquisition of an integrated exposure effect from the image data of the electromagnetic waves transmitted through the object.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Re-publication Patent No. WO2019 / 082276

[0006] Patent Document 2: Japanese Patent Application Publication No. 2019-158663 Summary of the Invention

[0007] The technical problem that the invention aims to solve

[0008] In existing devices as described above, by adding detection data obtained from multi-line sensors, the noise level tends to increase as the signal value increases in the addition result. Therefore, the signal / noise ratio (S / N) in image data is sometimes not sufficiently improved.

[0009] Therefore, one aspect of the implementation is made in view of this problem, and the objective is to provide a radiation image acquisition apparatus, radiation image acquisition system and radiation image acquisition method that can effectively improve the S / N ratio of radiation images.

[0010] Technical means to solve the problem

[0011] One aspect of the implementation of a radiation image acquisition apparatus includes: an imaging device that scans and captures radiation passing through an object in one direction to acquire a radiation image; a scintillator disposed on the imaging device to convert radiation into light; and an image processing module that inputs the radiation image into a learned model pre-constructed using image data through machine learning and performs noise removal processing to remove noise from the radiation image. The imaging device includes: a detection element composed of pixel lines having M pixels arranged in one direction arranged in N columns (N being an integer of 2 or more) arranged in a direction orthogonal to the one direction, and outputting a light-related detection signal for each pixel; and a readout circuit that adds the detection signals output from at least two pixels out of the M pixels for each pixel line in the N columns of the detection element, and sequentially outputs the added N detection signals to output a radiation image.

[0012] Alternatively, other aspects of the implementation of the radiation image acquisition system include: the radiation image acquisition device described above; a source that irradiates a target object with radiation; and a conveying device that conveys the target object relative to the imaging device in one direction.

[0013] Alternatively, another aspect of the implementation method for acquiring a radiation image includes: a step of scanning and capturing scintillation light corresponding to radiation passing through an object in one direction to acquire a radiation image; and a step of inputting the radiation image into a learned model pre-constructed using image data through machine learning, and performing noise removal processing to remove noise from the radiation image. In the acquisition step, a detection element is used to add detection signals output from at least two pixels out of M pixels for each pixel line of N columns of the detection element, and sequentially outputting the added N detection signals to output a radiation image. The detection element is composed of pixel lines having M pixels (M is an integer of 2 or more) arranged in one direction arranged in N columns (N is an integer of 2 or more) in a direction orthogonal to one direction, and outputting a detection signal related to scintillation light for each pixel.

[0014] According to one or more of the above aspects, scintillation light corresponding to radiation transmitted through an object is detected by a detection element arranged in N columns of pixel lines (each pixel line has M pixels arranged along the scanning direction of the object). For each pixel line, the detection signals of at least two pixels from the output M pixel detection signals are added together, and the N added detection signals are output sequentially, thereby generating a radiation image. Furthermore, noise removal processing is performed on the output radiation image by inputting it into a pre-learned model constructed using image data through machine learning. This increases the signal component and removes noise components in the radiation image, effectively improving the signal-to-noise ratio (S / N ratio) of the radiation image.

[0015] The effects of the invention

[0016] According to the implementation method, the signal-to-noise ratio of radiographic images can be effectively improved. Attached Figure Description

[0017] Figure 1 This is a schematic structural diagram of the image acquisition device 1 according to the first embodiment.

[0018] Figure 2 It is shown Figure 1 A top view of the structure of the scanning camera 12.

[0019] Figure 3 It is shown Figure 1 A block diagram illustrating an example of the hardware structure of the control device 20.

[0020] Figure 4 It is shown Figure 1 A block diagram of the functional structure of the control device 20.

[0021] Figure 5 It is shown Figure 4 An example of an X-ray image acquired by the image acquisition unit 203.

[0022] Figure 6 It is shown by Figure 4 The noise standard deviation diagram generated by the noise diagram generation unit 204 is an example of a noise diagram generation.

[0023] Figure 7 It is shown Figure 4 A diagram showing an example of the input and output data of Model 207 after learning is completed.

[0024] Figure 8 This is a diagram showing an example of a training image used to construct the learned model 207.

[0025] Figure 9 This is a flowchart showing the production sequence of the teaching data (training data), i.e., the image data, used by the construction unit 206 to construct the learned model 207.

[0026] Figure 10 This is a flowchart illustrating the sequence of observation processing performed by the image acquisition device 1.

[0027] Figure 11 This is a block diagram illustrating the functional structure of the control device 20A, a modified example of this disclosure.

[0028] Figure 12 This is a flowchart illustrating the sequence of observation processing performed by the image acquisition device 1 of a variant of this disclosure.

[0029] Figure 13 It is shown by Figure 11 A graph showing an example of the simulation results of the transmitted X-ray energy spectrum performed by the calculation unit 202A.

[0030] Figure 14 It is shown by Figure 11 A graph showing an example of the simulation results derived from the calculation unit 202A, showing the relationship between the thickness of the object and the average energy and transmittance.

[0031] Figure 15 It is shown by Figure 11 A graph showing an example of the simulation results of the relationship between the thickness of an object and the transmittance of X-rays, derived from the calculation unit 202A.

[0032] Figure 16 It is shown by Figure 11 A graph illustrating an example of the simulation results derived from the calculation unit 202A, showing the relationship between the thickness of an object and the average energy of transmitted X-rays.

[0033] Figure 17 It is shown by Figure 11 A chart illustrating an example of the simulation results of the relationship between pixel values ​​and average energy of an X-ray image, derived from the calculation unit 202A.

[0034] Figure 18 This is a graph illustrating an example of the simulation results showing the relationship between the pixel values ​​and the standard deviation of the noise values ​​in an X-ray image.

[0035] Figure 19 It is shown in Figure 11 A chart illustrating the relationship between the standard deviation of pixel values ​​and noise values ​​when the material of an object changes, derived from the calculation unit 202A.

[0036] Figure 20 This is a block diagram illustrating the functional structure of the control device 20B, another variation of this disclosure.

[0037] Figure 21 This is a flowchart illustrating the sequence of observation processing performed by the image acquisition device 1, another variation of this disclosure.

[0038] Figure 22 It is shown by Figure 20 The diagram shows an example of generating a noise standard deviation diagram for the noise diagram generation unit 204B.

[0039] Figure 23 This is a perspective view showing an example of the structure of the jig used for taking pictures in the image acquisition apparatus 1, another variation of this disclosure.

[0040] Figure 24 It is shown Figure 23An example of a photograph of a jig.

[0041] Figure 25 This is a block diagram showing the functional structure of the control device 20C according to the second embodiment.

[0042] Figure 26 It shows the construction Figure 25 The image is an example of the teaching data, i.e., the image data, of the completed model 206C.

[0043] Figure 27 It is shown Figure 25 An example of an X-ray transmission image of a selected object, 204C.

[0044] Figure 28 It is shown Figure 25 An example of a thickness-brightness characteristic curve obtained from the selection section 204C.

[0045] Figure 29 It is shown Figure 25 The figure shows an example of the brightness-SNR characteristic curve obtained by the selection unit 204C.

[0046] Figure 30 It is shown by Figure 25 The selection function of the learned model based on image characteristics is executed by the selection unit 204C.

[0047] Figure 31 It is shown that it is used by Figure 25 An example of an X-ray transmission image used to evaluate the resolution of the selected section 204C.

[0048] Figure 32 It is shown that it is used by Figure 25 A perspective view of an example of the structure of a fixture used for evaluating the luminance-noise ratio in the 204C selection section.

[0049] Figure 33 It is shown that Figure 32 The image shows a noise-removed X-ray transmitted image obtained using a fixture as the object.

[0050] Figure 34 This is a flowchart illustrating the sequence of observation processing using the image acquisition device 1 of the second embodiment.

[0051] Figure 35 This is a block diagram showing the functional structure of the control device 20D, a modified example of the second embodiment.

[0052] Figure 36 This is a flowchart showing the sequence of observation processing of the image acquisition apparatus 1 using a modified example of the second embodiment.

[0053] Figure 37 This is a block diagram showing the functional structure of the control device 20E, a modified example of the third embodiment.

[0054] Figure 38 It shows the construction Figure 37 The image is an example of the teaching data, i.e., the image data, of the completed model 206E.

[0055] Figure 39 It is shown Figure 37 An example of an X-ray transmission image of the object being analyzed by the determination unit 202E.

[0056] Figure 40 It is shown Figure 37 A figure showing an example of the thickness-brightness characteristic curve obtained by the determination unit 202E.

[0057] Figure 41 It is shown Figure 37 A diagram showing an example of the luminance-SNR characteristic curve obtained by the determination unit 202E.

[0058] Figure 42 It is shown that it is used by Figure 37 An example of an X-ray transmission image used to evaluate the resolution of the determination unit 202E.

[0059] Figure 43 Used to illustrate by Figure 37 The selection function of the learned model based on image characteristics is executed by the selection unit 204E.

[0060] Figure 44 It is shown that it is used by Figure 37 A perspective view of an example of the structure of a fixture used for evaluating the luminance-noise ratio in the selection section 204E.

[0061] Figure 45 It is shown that Figure 44 The image shows a noise-removed X-ray transmitted image obtained using a fixture as the object.

[0062] Figure 46 This is a flowchart illustrating the observation processing sequence using the image acquisition apparatus 1 of the third embodiment. Detailed Implementation

[0063] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Furthermore, in the description, the same reference numerals are used for the same elements or elements having the same function, and repeated descriptions are omitted.

[0064] Figure 1 This is a structural diagram of the radiation image acquisition apparatus and radiation image acquisition system of this embodiment, namely, image acquisition apparatus 1. Figure 1 As shown, the image acquisition device 1 is an apparatus that irradiates an object F being transported along the transport direction TD with X-rays (radiation) and acquires an X-ray transmission image (radiation image) of the object F based on the X-rays passing through it. The image acquisition device 1 uses the X-ray transmission image to perform foreign object inspection, weight inspection, and inspection of the object F. Examples of its applications include food inspection, carry-on baggage inspection, substrate inspection, battery inspection, and material inspection. The image acquisition device 1 is configured to include: a belt conveyor (transportation device) 60, an X-ray irradiator (radiation source) 50, an X-ray inspection camera (image capturing device) 10, a control device (image processing module) 20, a display device 30, and an input device 40 for performing various inputs. Furthermore, the radiation images in the embodiments of this disclosure are not limited to X-ray images, but also include images generated by electromagnetic radiation other than X-rays such as gamma rays.

[0065] The belt conveyor 60 has a belt section carrying an object F. By moving this belt section along a conveying direction (one direction) TD, the object F is conveyed at a predetermined conveying speed along the conveying direction TD. The conveying speed of the object F is, for example, 48 m / min. The belt conveyor 60 can change the conveying speed as needed, for example, to 24 m / min or 96 m / min. In addition, the belt conveyor 60 can appropriately change the height position of the belt section to change the distance between the X-ray irradiator 50 and the object F. Furthermore, the object F conveyed by the belt conveyor 60 can include, for example, various items such as edible meat, seafood, crops, snacks and other food products, rubber products such as tires, resin products, metal products, mineral resources and other materials, waste, and electronic components or electronic boards. The X-ray irradiator 50 is a device that irradiates (outputs) the object F with X-rays as an X-ray source. The X-ray irradiator 50 is a point source that diffuses X-rays within a predetermined angle range along a certain irradiation direction. An X-ray irradiator 50 is positioned above the belt conveyor 60, with the X-ray irradiation direction facing the belt conveyor 60 and the diffused X-rays reaching the entire width direction (the direction intersecting the transport direction TD) of the object F. Furthermore, the X-ray irradiator 50 defines a predetermined segmentation range along the length direction of the object F (parallel to the transport direction TD) as the irradiation range, and the object F is transported by the belt conveyor 60 in the transport direction TD, thereby irradiating the entire length direction of the object F with X-rays. The X-ray irradiator 50 uses a control device 20 to set the tube voltage and tube current, irradiating the belt conveyor 60 with X-rays of a predetermined energy and radiation dose corresponding to the set tube voltage and tube current. Additionally, a filter 51 is provided near the belt conveyor 60 side of the X-ray irradiator 50 to allow X-rays to pass through a predetermined wavelength band.

[0066] The X-ray inspection camera 10 detects X-rays that pass through the object F from the X-rays irradiated by the X-ray irradiator 50, acquires and outputs a signal based on the X-rays. The image acquisition device 1 of this embodiment sequentially outputs detection signals based on the X-rays that pass through the object F being transported by the belt conveyor 60, scans the X-ray transmission image along the transport direction TD, and outputs the captured X-ray transmission image.

[0067] The X-ray inspection camera 10 includes: a filter 19, a scintillator 11, a scanning camera 12 (detection element), a sensor control unit 13, an amplifier 14, an AD converter 15, a correction circuit 16, an output interface 17, and an amplifier control unit 18. The scintillator 11, the scanning camera 12, the amplifier 14, the AD converter 15, the correction circuit 16, and the output interface 17 are electrically connected.

[0068] The scintillator 11 is fixed to the scanning camera 12 by adhesive or the like, and converts the X-rays passing through the object F into scintillation light. The scintillator 11 outputs the scintillation light to the scanning camera 12. The filter 19 allows the X-rays of a specified wavelength band to pass through towards the scintillator 11.

[0069] The scanning camera 12 detects the flashing light from the scintillator 11 and converts it into electrical charge, which is then output to the amplifier 14 as a detection signal (electrical signal). Figure 2 This is a top view showing the structure of the scanning camera 12. (See attached image.) Figure 2 As shown, the scanning camera 12 includes: a photodiode (photoelectric conversion element), i.e., a plurality of pixels 72, which are arranged in two dimensions on a substrate 71; a readout circuit 73, which outputs a detection signal to the outside by photoelectric conversion of the flashing light by the plurality of pixels 72; and a wiring section W, which electrically connects the readout circuit 73 to each of the plurality of pixels 72.

[0070] In detail, the scanning camera 12 has the following structure: on the substrate 71, pixel lines (pixel groups) 74, consisting of M pixels 72 arranged along the transport direction TD, are arranged in N columns (N is an integer of 2 or more) in a direction substantially orthogonal to the transport direction TD. For example, the number of pixels M is 4, and the number of pixel lines N is any integer between 200 and 30,000.

[0071] Under the control of the sensor control unit 13, the readout circuit 73 sequentially receives detection signals output from M pixels 72 at predetermined detection intervals (details will be described later) for each pixel line 74, and performs a process of adding (totaling) the detection signals from at least two pixels 72 among the detection signals from the M pixels 72. It then combines the detection signals from each pixel line 74 that have undergone addition processing and outputs them externally as the detection signal for one row of the object F orthogonal to the transport direction TD. In this embodiment, the readout circuit 73 performs addition processing on all M detection signals. Furthermore, by staggering the predetermined detection intervals, the readout circuit 73 performs addition processing on the detection signals sequentially output from the M pixels 72, thereby outputting the detection signal for the next row of the object F orthogonal to the transport direction TD. Similarly, the readout circuit 73 sequentially outputs multiple rows of detection signals for the object F orthogonal to the transport direction TD.

[0072] The sensor control unit 13 controls the scanning camera 12 to repeatedly capture images at a predetermined detection cycle, such that all pixels 72 within the pixel line 74 of the scanning camera 12 can capture X-rays passing through the same area of ​​the object F. The predetermined detection cycle can also be set based on the pixel width of the pixels 72 within the pixel line 74 of the scanning camera 12. The predetermined detection cycle can also be determined, for example, based on the distance between the pixels 72 within the pixel line 74 of the scanning camera 12, the speed of the belt conveyor 60, the distance between the X-ray irradiator 50 and the object F on the belt conveyor 60 (FOD (Focus Object Distance)), and the distance between the X-ray irradiator 50 and the scanning camera 12 (FDD (Focus Detector Distance)), to determine the deviation (delay time) in the detection timing of the pixels 72 within the pixel line 74 of the scanning camera 12, and set the predetermined detection cycle based on this deviation.

[0073] Amplifier 14 amplifies the detection signal at a predetermined set amplification rate and generates an amplified signal, which is then output to AD converter 15. The set amplification rate is set by amplifier control unit 18. Amplifier control unit 18 sets the set amplification rate of amplifier 14 based on predetermined shooting conditions.

[0074] The AD converter 15 converts the amplified signal (voltage signal) output from the amplifier 14 into a digital signal and outputs it to the correction circuit 16. The correction circuit 16 performs prescribed corrections on the digital signal, such as signal amplification, and outputs the corrected digital signal to the output interface 17. The output interface 17 outputs the digital signal to the outside of the X-ray inspection camera 10.

[0075] The control device 20 is, for example, a computer such as a PC (Personal Computer). The control device 20 generates an X-ray transmission image based on digital signals (amplified signals) corresponding to multiple lines of detection signals sequentially output from the X-ray inspection camera 10 (more specifically, output interface 17). In this embodiment, the control device 20 generates one X-ray transmission image based on 128 lines of digital signals output from the output interface 17. The generated X-ray transmission image is output to the display device 30 after noise removal processing described later, and is displayed by the display device 30. Furthermore, the control device 20 controls the X-ray irradiator 50, the amplifier control unit 18, and the sensor control unit 13. In this embodiment, the control device 20 is a device independently installed outside the X-ray inspection camera 10, but it can also be integrated inside the X-ray inspection camera 10.

[0076] Figure 3 The hardware structure of the control device 20 is shown. For example... Figure 3 As shown, the control device 20 physically includes a computer, such as a CPU (Central Processing Unit) 101 and a GPU 105 (Graphics Processing Unit) as processors, RAM (Random Access Memory) 102 and ROM (Read Only Memory) 103 as recording media, a communication module 104, and an input / output module 106, all electrically connected. Furthermore, the control device 20 may also include a display, keyboard, mouse, touch panel display, etc., as input devices 40 and display devices 30, and may also include data recording devices such as hard disk drives and semiconductor memory. Additionally, the control device 20 may be composed of multiple computers.

[0077] Figure 4 This is a block diagram showing the functional structure of the control device 20. The control device 20 includes: an input unit 201, a calculation unit 202, an image acquisition unit 203, a noise map generation unit 204, a processing unit 205, and a construction unit 206. Figure 4 The functional units of the control device 20 shown are implemented by loading a program (the radiation image processing program of the first embodiment) onto hardware such as the CPU 101, GPU 105, and RAM 102. Under the control of the CPU 101 and GPU 105, the communication module 104 and input / output module 106 are operated, and data is read from and written to the RAM 102. The CPU 101 and GPU 105 of the control device 20 execute this computer program to make the control device 20 function as... Figure 4Each functional unit performs its function, sequentially executing the processing corresponding to the radiographic image acquisition and processing method described later. Furthermore, the CPU 101 and GPU 105 can be standalone hardware units, or either one can be used alone. Additionally, the CPU 101 and GPU 105 can be installed in a programmable logic device such as an FPGA, like a software processor. Regarding RAM or ROM, it can also be a standalone hardware unit, or it can be built into a programmable logic device such as an FPGA. All data required to execute the computer program, and all data generated by executing the computer program, are stored in built-in memory such as ROM 103, RAM 102, or a memory medium such as a hard disk drive. Furthermore, the built-in memory or memory medium within the control device 20 pre-stores a learned model 207 (described later), which, when read by the CPU 101 and GPU 105, enables the CPU 101 and GPU 105 to perform noise removal processing on the X-ray image (X-ray transmitted image) (described later).

[0078] The functions of each functional unit of the control device 20 will be explained in detail below.

[0079] The input unit 201 receives input information such as the conditions of the radiation source when irradiating the object F, or arbitrary imaging conditions. Specifically, the input unit 201 receives input from the user of the image acquisition device 1 regarding the operating conditions of the X-ray irradiator (radiation source) 50 when imaging the object F, or imaging conditions of the X-ray detection camera 10. Operating conditions may include all or some of the following: tube voltage, target angle, target material, etc. Examples of conditional information indicating shooting conditions include: the material and thickness of filters 51 and 19 disposed between the X-ray irradiator 50 and the X-ray inspection camera 10; the distance (FDD) between the X-ray irradiator 50 and the X-ray inspection camera 10; the type of window material of the X-ray inspection camera 10; information related to the material and thickness of the scintillator 11 of the X-ray inspection camera 10; X-ray inspection camera information (e.g., gain setting value, circuit noise value, saturation charge, conversion factor value (e- / count), camera line rate (Hz) or line speed (m / min)); and information about the object F. The input unit 201 may receive conditional information as direct input of numerical values ​​or other information, or as selective input of numerical values ​​or other information preset in the internal memory. The input unit 201 receives the aforementioned conditional information from the user, but may also obtain a portion of the conditional information (such as tube voltage) based on the detection results of the control state executed by the control device 20.

[0080] The calculation unit 202 calculates the average energy value related to the X-rays (radiation) transmitted through the object F based on the condition information. The condition information includes at least one of the following: the tube voltage of the generating source, information related to the object F, information about the filters of the camera used to capture the object F, information about the scintillator of the camera, and information about the filters of the X-ray generating source. Specifically, the calculation unit 202 calculates the average energy value of the X-rays transmitted through the object F and detected by the X-ray detection camera 10 using the image acquisition device 1, based on the condition information received by the input unit 201. For example, the calculation unit 202 calculates the spectrum of the X-rays detected by the X-ray detection camera 10 using an approximation formula, such as that of Tucker et al., based on information included in the condition information, such as tube voltage, target angle, target material, material and thickness of filters 51 and 19 and their presence or absence, type of window material of the X-ray detection camera 10 and its presence or absence, and material and thickness of the scintillator 11 of the X-ray detection camera 10. Furthermore, the calculation unit 202 calculates the integral value of the spectral intensity and the integral value of the photon number based on the spectrum of the X-rays, and divides the integral value of the spectral intensity by the integral value of the photon number to calculate the average energy of the X-rays.

[0081] The calculation method using the well-known Tucker approximation is described below. For example, when the target is determined to be tungsten and the target angle is determined to be 25°, the calculation unit 202 can determine: Em: kinetic energy of the electron target collision, T: kinetic energy of the electron in the target, A: proportionality constant determined by the atomic number of the target material, ρ: density of the target, μ(E): linear attenuation coefficient of the target material, B: a gradually changing function of Z and T, C: Thomson-Whiddington constant, θ: target angle, c: speed of light in vacuum. Based on this, the calculation unit 202 can calculate the X-ray spectrum of the irradiation by calculating the following formula (1).

[0082]

[0083] In addition, Em can be determined based on the tube voltage information, A, ρ, and μ(E) can be determined based on the material information of the object F, and θ can be determined based on the angle information of the object F.

[0084] Next, the calculation unit 202 can use the X-ray attenuation formula of the following formula (2) to calculate the X-ray energy spectrum that passes through the filter and the object F and is absorbed by the scintillator.

[0085] I = I0e -μx (2)

[0086] Here, μ is the attenuation number of the object F, filter, scintillator, etc., and x is the thickness of the object F, filter, scintillator, etc. μ can be determined based on the material information of the object F, filter, and scintillator, and x can be determined based on the thickness information of the object F, filter, and scintillator. The X-ray photon number spectrum is obtained by dividing the X-ray energy spectrum by the energy of each X-ray. The calculation unit 202 calculates the average energy of the X-ray by dividing the integral value of the energy intensity by the integral value of the photon number using the following formula (3).

[0087] Average energy E = integral value of spectral intensity / integral value of photon number…(3)

[0088] Through the above calculation process, the calculation unit 202 calculates the average energy of the X-rays. In addition, the X-ray spectrum can also be calculated using well-known approximations such as those by Kramers or Birch.

[0089] The image acquisition unit 203 irradiates the object F with radiation and acquires a radiation image containing the radiation that has passed through the object F. Specifically, the image acquisition unit 203 generates an X-ray image based on the digital signal (amplified signal) output from the X-ray detection camera 10 (more specifically, the output interface 17). The image acquisition unit 203 generates one X-ray image based on the multi-line digital signal output from the output interface 17. Figure 5 This is a diagram showing an example of an X-ray image acquired by the image acquisition unit 203.

[0090] The noise map generation unit 204 derives evaluation values ​​from the pixel values ​​of each pixel in the radiation image based on relational data representing the relationship between pixel values ​​and evaluation values ​​that represent the amplification of evaluation noise values. It then generates a noise map by establishing a correspondence between the derived evaluation values ​​and each pixel in the radiation image. Specifically, the noise map generation unit 204 derives evaluation values ​​from the average energy related to the radiation transmitted through the object F and the pixel values ​​of each image in the radiation image. More specifically, the noise map generation unit 204 uses a formula (relational data) relating pixel values ​​to the standard deviation of noise values ​​(evaluation values ​​that represent the amplification of evaluation noise values) to derive the standard deviation of noise values ​​from the average energy of X-rays calculated by the calculation unit 202 and the pixel values ​​of each pixel in the X-ray image (radiation image) acquired by the image acquisition unit 203. The noise map generation unit 204 generates a noise standard deviation map (noise map) by establishing a correspondence between the derived standard deviation of noise values ​​and each pixel in the X-ray image.

[0091] The relationship between the pixel value and the standard deviation of the average energy and the noise value used by the noise map generation unit 204 is expressed by the following formula (4).

[0092]

[0093] In the above equation (4), the variable Noise represents the standard deviation of the noise value, the variable Signal represents the signal value (pixel value) of the pixel, the constant F represents the noise factor, the constant M represents the magnification of the scintillator, the constant C represents the coupling efficiency between the scanning camera 12 and the scintillator 11 in the X-ray detection camera 10, the constant Q represents the quantum efficiency of the scanning camera 12, the constant cf represents the conversion coefficient in the scanning camera 12 that converts the signal value of the pixel into charge, the variable Em represents the average energy of the X-ray, the constant D represents the dark current noise generated by thermal noise in the image sensor, and the constant R represents the readout noise in the scanning camera 12. When using the above equation (4), the noise map generation unit 204 substitutes the pixel value of each pixel of the X-ray image obtained by the image acquisition unit 203 into the variable Signal, and substitutes the value of the average energy calculated by the calculation unit 202 into the variable Em. Furthermore, the noise graph generation unit 204 obtains the variable Noise, calculated using the above formula (4), as the standard deviation of the noise value. In addition, other parameters, including average energy, can also be obtained by receiving input from the input unit 201, or can be preset.

[0094] Figure 6 This diagram illustrates an example of generating a noise standard deviation diagram by the noise diagram generation unit 204. The noise diagram generation unit 204 uses the relationship formula (4) between pixel values ​​and the standard deviation of noise values, substituting various pixel values ​​into the variable Signal to obtain the correspondence between pixel values ​​and the variable Noise, thereby deriving a relationship chart G3 representing the correspondence between pixel values ​​and the standard deviation of noise values. Furthermore, the noise diagram generation unit 204 derives relationship data G2 representing the correspondence between each pixel position and pixel value from the X-ray image G1 acquired by the image acquisition unit 203. Additionally, the noise diagram generation unit 204 applies the correspondence shown in the relationship chart G3 to each pixel value in the relationship data G2, deriving the standard deviation of the noise value corresponding to each pixel position in the X-ray image. As a result, the noise diagram generation unit 204 establishes a correspondence between the derived noise standard deviation and each pixel position, deriving relationship data G4 showing the correspondence between each pixel position and the standard deviation of noise. Furthermore, the noise diagram generation unit 204 generates a noise standard deviation diagram G5 based on the derived relationship data G4.

[0095] The processing unit 205 inputs the radiation image and noise image into the pre-learned model 207, which has been constructed through machine learning, and performs image processing to remove noise from the radiation image. That is, as... Figure 7As shown, the processing unit 205 retrieves the learned model 207 (described later) constructed by the construction unit 206 from the built-in memory or memory medium within the control device 20. The processing unit 205 inputs the X-ray image G1 acquired by the image acquisition unit 203 and the noise standard deviation map G5 generated by the noise map generation unit 204 into the learned model 207. Therefore, the processing unit 205 performs image processing to remove noise from the X-ray image G1 using the learned model 207, thereby generating an output image G6. Furthermore, the processing unit 205 outputs the generated output image G6 to the display device 30, etc.

[0096] The construction unit 206 uses a radiation image (training image), a noise map generated from the training image based on the relationship between pixel values ​​and the standard deviation of noise values, and noise-removed image data (data after removing noise from the training image) as training data. It then constructs a learned model 207 based on the training image and the noise map, outputting noise-removed image data, through machine learning. The construction unit 206 stores the constructed learned model 207 in the built-in memory or memory medium within the control device 20. Machine learning includes taught learning, untaught learning, and reinforcement learning; among these, deep learning and neural network learning are examples. In the first embodiment, as an example of a deep learning algorithm, a 2D convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is used. Furthermore, the learned model 207 can be generated and downloaded to the control device 20 via an external computer or the like, in addition to being constructed by the construction unit 206. Moreover, the radiation image used for machine learning includes a radiation image of a known structure or an image reproducing that radiation image.

[0097] Figure 8 An example of training images, one of the training data used to construct the model 207, is shown. As training images, X-ray transmission images of patterns with various thicknesses, materials, and resolutions can be used. Figure 8 The example shown is a training image G7 generated using chicken as the subject. This training image G7 can also be an X-ray image generated using image acquisition device 1 with various known structures as the subject, or it can be image data generated through simulation calculations. Regarding the X-ray image, it can also be an image acquired using a device different from image acquisition device 1.

[0098] The construction unit 206, as a preprocessing step for machine learning, derives evaluation values ​​from the pixel values ​​of each pixel in the radiographic image based on relational data representing the relationship between pixel values ​​and the expanded evaluation noise values. It then generates a noise map, which establishes a correspondence between the derived evaluation values ​​and each pixel in the radiographic image. Specifically, when the learning model 207 is constructed, the construction unit 206 acquires training images generated through actual imaging or simulation calculations from the image acquisition unit 203, etc. Furthermore, the construction unit 206 sets, for example, the operating conditions of the X-ray irradiator 50 of the image acquisition device 1 or the imaging conditions of the image acquisition device 1. Alternatively, the construction unit 206 sets the operating conditions or imaging conditions of the X-ray irradiator 50 during simulation calculations. The construction unit 206 calculates the average energy of the X-rays based on the aforementioned operating conditions or imaging conditions using the same method as the calculation unit 202. Furthermore, the construction unit 206 uses... Figure 6 The noise map generation unit 204 shown uses the same method to generate a noise standard deviation map based on the average energy of the X-rays and the training image. That is, the preprocessing method of the machine learning method includes a noise map generation step, wherein, based on relational data representing the relationship between pixel values ​​and the expanded evaluation values ​​of the evaluation noise values, evaluation values ​​are derived from the pixel values ​​of each pixel in the radiation image, and data corresponding to the derived evaluation values ​​and each pixel of the radiation image, i.e., a noise map, is generated.

[0099] The construction unit 206 uses the training image, the noise map generated from the training image, and noise-removed image data (pre-removed noise data from the training image) as training data to construct a learned model 207 through machine learning. Specifically, the construction unit 206 pre-obtains noise-removed image data (noise removed from the training image). When the training image is an X-ray image generated through simulation calculation, the construction unit 206 sets the image before noise was added during the generation process as the noise-removed image data. On the other hand, when the training image is an X-ray image generated by the image acquisition device 1 and targeting various known structures, the construction unit 206 sets the image after noise removal from the X-ray image using image processing techniques such as average filter, median filter, bilateral filter, and NLM filter as the noise-removed image data. The construction unit 206 performs machine learning training to construct a learned model 207 that outputs noise-removed image data based on the training image and the noise standard deviation map.

[0100] Figure 9 A flowchart showing the production sequence of the teaching data (training data), i.e., the image data, used by the construction unit 206 to construct the learned model 207.

[0101] The teaching data, i.e., image data (also called teaching image data), is created by the computer in the following order. First, an image of a structure with a specified structure (structure image) is created (step S301). For example, an image of a structure with a specified structure can also be created through simulation calculation. Alternatively, an X-ray image of a structure such as a chart with a specified structure can be obtained to create a structure image. Next, for a pixel selected from the multiple pixels constituting the structure image, the standard deviation of the pixel value, i.e., σ (sigma), is calculated (step S302). Then, a normal distribution (Poisson distribution) representing the noise distribution is set based on the σ value obtained in step S302 (step S303). In this way, teaching data for various noise conditions can be generated by setting a normal distribution based on the σ value. Next, a randomly set noise value is calculated according to the normal distribution set based on the σ value in step S303 (step S304). Furthermore, by appending the noise value obtained in step S304 to the pixel value of a pixel, pixel values ​​constituting the teaching data, i.e., image data, are generated (step S305). Steps S302 to S305 are then processed on multiple pixels constituting the structure image (step S306) to generate teaching image data that becomes teaching data (step S307). Additionally, if teaching image data is still needed, steps S301 to S307 are processed on other structure images (step S308) to generate other teaching image data that becomes teaching data. Furthermore, the other structure images can be images of structures with the same structure or images of structures with different structures.

[0102] Furthermore, multiple teaching data, i.e., image data, are needed for constructing the learned model 207. Additionally, images with low noise are preferred among the structural images; ideally, noise-free images are best. Therefore, if structural images are generated through simulation calculations, multiple noise-free images can be produced, making simulation calculations an effective method for generating structural images.

[0103] Next, the sequence of observation and processing of the X-ray transmitted image of the object F using the image acquisition apparatus 1 of the first embodiment, i.e., the flow of the radiation image acquisition method of the first embodiment, will be described. Figure 10 This is a flowchart illustrating the sequence of observation processes performed by the image acquisition device 1.

[0104] First, the construction unit 206 uses training images, a noise standard deviation map generated from the training images based on a relation, and noise-removed image data as training data to construct a learned model 207 that outputs noise-removed image data based on the training images and the noise standard deviation map through machine learning (step S100). Next, the input unit 201 receives input from the operator (user) of the image acquisition device 1, indicating the operating conditions of the X-ray irradiator 50 or the shooting conditions of the X-ray detection camera 10 (step S101). Then, the calculation unit 202 calculates the average energy value of the X-rays detected by the X-ray detection camera 10 based on the condition information (step S102).

[0105] Next, an object F is set in the image acquisition device 1, and an image of the object F is captured. An X-ray image of the object F is acquired by the control device 20 (step S103). Then, the control device 20, based on the relationship between the standard deviation of pixel values ​​and noise values, derives the standard deviation of noise values ​​from the average energy of X-rays and the pixel values ​​of each pixel in the X-ray image. The derived standard deviation of noise is then correlated with each pixel value to generate a noise standard deviation map (step S104).

[0106] Next, the processing unit 205 processes the X-ray image and noise standard deviation diagram of the input object F, which is based on the pre-built and memorized learning model 207, and performs noise removal processing on the X-ray image (step S105). Then, the X-ray image that has undergone noise removal processing by the processing unit 205, i.e., the output image, is output to the display device 30 (step S106).

[0107] According to the image acquisition apparatus 1 described above, scintillation light corresponding to the X-rays passing through the object F is detected by a scanning camera 12 with pixel lines 74 arranged in N columns. Each pixel line 74 has M pixels 72 arranged along the scanning direction TD of the object F. The detection signals of at least two pixels 72 from the detection signals output to each pixel line 74 are added together, and the N added detection signals are output sequentially to generate an X-ray image. Furthermore, the output X-ray image is input into a pre-learned model 207 constructed using machine learning of image data to perform noise removal processing on the X-ray image. As a result, the signal components in the X-ray image can be increased and noise components can be removed, effectively improving the signal-to-noise ratio (S / N) of the X-ray image. Specifically, it is known that when noise removal processing using the learned model 207 is implemented, the CNR (Contrast to Noise Ratio) is improved by approximately 6.4 times compared to the case without noise removal processing, and the improvement effect is also greater than the approximately 1.9 times improvement effect of noise removal processing using a bilateral filter.

[0108] Furthermore, in the image acquisition device 1, the learned model 207 is constructed through machine learning by using image data obtained by adding normally distributed noise values ​​to an X-ray image of a specified structure as teaching data. Therefore, the teaching data, i.e., the image data, for constructing the learned model 207 is readily available, and the learned model 207 can be constructed effectively.

[0109] Furthermore, according to the image acquisition device 1, using the relationship between the standard deviation of pixel values ​​and noise values, the standard deviation of noise values ​​is derived from the pixel values ​​of each image in the X-ray image, generating a noise standard deviation map that establishes a correspondence between the derived standard deviation of noise values ​​and each pixel of the X-ray image. Then, the X-ray image and the noise standard deviation map are input into a pre-learned model 207 constructed through machine learning, and image processing to remove noise from the X-ray image is performed. Based on this structure, considering the standard deviation of noise values ​​derived from the pixel values ​​of each pixel in the X-ray image, noise in each pixel of the X-ray image is removed through machine learning. Thus, noise removal corresponding to the relationship between the pixel values ​​and the standard deviation of noise values ​​in the X-ray image can be achieved using the learned model 207. As a result, noise in the X-ray image can be effectively removed.

[0110] In particular, the noise patterns in X-ray images vary depending on factors such as tube voltage, filters, scintillators, X-ray camera conditions (gain setting, circuit noise level, saturation charge, conversion factor (e- / count), camera line frequency), and the object being inspected. Therefore, when aiming to remove noise through machine learning, it is advisable to prepare learning models that can learn under various conditions. Specifically, as a comparative example, the following method can be used: construct multiple learning models based on the conditions during X-ray image measurement, select a learning model for each condition, and perform noise removal processing. In this comparative example, for instance, it may be necessary to construct a learning model for each noise condition, such as the average energy of the X-rays, the gain of the X-ray camera, and the type of X-ray camera, requiring a large number of learning models and significant time investment. For example, with 10 different average X-ray energies, 8 different X-ray camera gains, and 3 different object types, 240 fully learned models are needed. Assuming each model takes one day to build, the machine learning process would take 240 days. In this embodiment, a noise map is generated from the X-ray image and set as input data for machine learning. This reduces the noise conditions required to generate the learned model and significantly reduces the learning time for constructing the learned model 207.

[0111] [Modifications of the control device 20 in the first embodiment]

[0112] Figure 11 This is a block diagram showing the functional structure of the control device 20A, a modified example of the first embodiment. The control device 20A differs from the first embodiment in that the calculation unit 202A has the function of deriving the average energy of X-rays from the pixel values ​​of the X-ray image, and the noise map generation unit 204A has the function of deriving a noise standard deviation map based on the pixel values ​​of the X-ray image and the average energy of the X-rays derived from the X-ray image. Figure 12 It is shown that it is composed of Figure 11 A flowchart illustrating the sequence of observation processing performed by the image acquisition device 1 in the control device 20A. (See flowchart for example.) Figure 12 As shown, in the control device 20A, Figure 10 The process shown in step S103 of the control device 20 of the first embodiment is performed immediately after step S100. Furthermore, in the control device 20A, the processes shown in steps S102A and S104A are performed instead of the processes in steps S102 and S104 of the control device 20.

[0113] The calculation unit 202A calculates the average energy from the pixel values ​​of each pixel in the X-ray image (step S102A). Specifically, the calculation unit 202A pre-derives the relationship between pixel values ​​and average energy for each piece of conditional information based on simulation calculations of the X-ray spectrum, etc. The calculation unit 202A acquires conditional information that includes at least the tube voltage obtained through the input unit 201 and information about the scintillator provided by the X-ray detection camera 10. Furthermore, based on this conditional information, the calculation unit 202A selects a relationship corresponding to the pre-derived relationship between pixel values ​​and average energy. Then, based on the selected relationship, the calculation unit 202A derives the average energy of each pixel from the pixel values ​​of each pixel in the X-ray image obtained by the image acquisition unit 203.

[0114] The following section describes the derivation of the relationship between the pixel value and the average energy for each piece of conditional information, performed by the calculation unit 202A. Figures 13-17 Please provide an explanation.

[0115] First, calculation unit 202A, based on conditional information, derives graph G18 showing the relationship between the thickness of object F and X-ray transmittance, and graph G19 showing the relationship between the thickness of object F and average X-ray energy. Specifically, as follows... Figure 13 As shown in parts (a) to (d), the calculation unit 202A calculates the energy spectrum G14 to G17 of the transmitted X-rays under various changes in the thickness of the object F by simulation calculation based on condition information including at least tube voltage and information of the scintillator of the X-ray inspection camera 10. Figure 13This is a graph showing an example of the simulation calculation results of the energy spectrum of X-rays transmitted through an object F, performed by the calculation unit 202A. Here, the energy spectra of transmitted X-rays G14 to G17 are shown when the thickness of the object F, which is composed of water, is gradually increased in stages and the simulation calculation is performed. Furthermore, based on the calculated energy spectra G14 to G17, the calculation unit 202A calculates the average energy of the transmitted X-rays under various changes in the thickness of the object F. In addition to simulation calculations, the calculation unit 202A can also obtain the relationship between the thickness of the object F and the average energy based on X-ray images obtained by taking pictures of a structure with a known thickness.

[0116] Furthermore, the calculation unit 202A also derives the relationship between the thickness of the object F and the X-ray transmittance based on the above simulation results. Figure 14 This is a graph illustrating an example of the relationship between the thickness of the object F and its average energy and transmittance, derived from the calculation unit 202A. For example... Figure 14 As shown, the average energy and X-ray transmittance of transmitted X-rays are derived by corresponding to the energy spectra G14 to G17 calculated for each thickness of the object F.

[0117] Next, the X-ray transmittance derived from the object F of various thicknesses by the calculation unit 202A is used to derive a graph G18 showing the relationship between the thickness of the object F and the X-ray transmittance. Figure 15 This is a graph showing the relationship between the thickness of the object F and the X-ray transmittance of the object F, derived by the calculation unit 202A. Furthermore, the calculation unit 202A derives a graph G19 showing the relationship between the thickness of the object F and the average energy of the X-rays from the average energy of the X-rays derived for object F of various thicknesses. Figure 16 This is a graph showing an example of the relationship between the thickness of the object F and the average energy of the X-rays passing through the object F, derived by the calculation unit 202A.

[0118] Furthermore, the calculation unit 202A calculates two charts G18 and G19 derived from each of the various conditional information, such as... Figure 17 The graph G20 shown represents the relationship between pixel values ​​and average energy in an X-ray image. Figure 17This is a graph showing the relationship between the pixel values ​​and average energy of the X-ray image derived by the calculation unit 202A. Specifically, the calculation unit 202A derives the pixel value I0 of the X-ray transmitted image when the object F is absent, based on conditional information. Furthermore, the calculation unit 202A sets the pixel value I of the X-ray image when the object F is present and calculates the X-ray transmittance, i.e., I / I0. Next, based on the graph G18 showing the thickness of the object F and the X-ray transmittance through the object F, the calculation unit 202A derives the thickness of the object F from the calculated X-ray transmittance, i.e., I / I0. Finally, based on the derived thickness of the object F and the graph G19 showing the thickness of the object F and the average energy of the transmitted X-rays, the calculation unit 202A derives the average energy of the transmitted X-rays corresponding to that thickness. Next, the calculation unit 202A performs the above-described derivation on each of the various condition information by changing the pixel value I of the X-ray image in various ways, thereby deriving a graph G20 for each condition information that represents the relationship between the pixel value of the X-ray image and the average energy of the transmitted X-ray.

[0119] Here, we will describe an example of deriving the average energy based on pixel values ​​from the calculation unit 202A. For example, assume the following situation: the calculation unit 202A, based on condition information, derives the pixel value of the X-ray transmitted image when the object F does not exist as I0 = 5000, and sets the pixel value of the X-ray image when the object F exists as I = 500. In this case, the calculation unit 202A calculates the X-ray transmittance as I / I0 = 0.1. Next, based on graph G18 showing the relationship between the thickness of the object F and the X-ray transmittance of the object F, the calculation unit 202A derives a thickness of 30 mm corresponding to an X-ray transmittance of 0.1. Furthermore, based on graph G19 showing the relationship between the thickness of the object F and the average energy of the transmitted X-rays, the calculation unit 202A derives an average energy of 27 keV corresponding to a pixel value of 500. Finally, section 202A calculates the average energy of the X-rays for each repetition of each pixel value, and derives a graph G20 showing the relationship between the pixel values ​​and the average energy of the X-ray image.

[0120] Then, the calculation unit 202A selects the chart G20 corresponding to the condition information obtained by the input unit 201 from the multiple charts G20 that have been pre-exported in the above order. Based on the selected chart G20, the calculation unit 202A outputs the average energy of transmitted X-rays corresponding to the pixel values ​​of each pixel of the X-ray image obtained by the image acquisition unit 203.

[0121] Furthermore, the calculation unit 202A can also derive the average energy of X-rays from the conditional information obtained by the input unit 201 and the pixel values ​​of each pixel in the X-ray image, without pre-deriving the relationship between pixel values ​​and average energy of X-rays for each piece of conditional information. Specifically, the calculation unit 202A derives the pixel value I0 of the X-ray image when the object is absent, based on the conditional information. Furthermore, the calculation unit 202A calculates the transmittance by obtaining the ratio of each pixel value I0 relative to the pixel value I0 for each pixel in the X-ray image obtained by the image acquisition unit 203. Next, the calculation unit 202A derives the thickness based on the graph G18 showing the relationship between thickness and X-ray transmittance and the calculated transmittance. Furthermore, the calculation unit 202A derives the average energy based on the graph G19 showing the relationship between thickness and average energy and the derived thickness, thereby deriving the average energy for each pixel value of the X-ray image.

[0122] The noise map generation unit 204A generates a noise standard deviation map (step S104A) from the X-ray image acquired by the image acquisition unit 203 and the average energy of the X-rays corresponding to each pixel of the X-ray image derived by the calculation unit 202A. Specifically, the noise map generation unit 204A substitutes the pixel values ​​of each pixel of the X-ray image acquired by the image acquisition unit 203 and the average energy derived for each pixel by the calculation unit 202A into the relation (4) to derive the standard deviation of the noise value of each pixel considering the thickness of the object. The noise map generation unit 204A generates the standard deviation of the noise value corresponding to each pixel of the X-ray image as a noise standard deviation map.

[0123] Figure 18 This is a graph illustrating an example of the relationship between the standard deviation of pixel values ​​and noise values. The graph shows the relationship between the standard deviation of noise values ​​derived from the pixel values ​​of the X-ray image and the pixel values ​​of the X-ray image, obtained through the calculation unit 202A and the noise image generation unit 204A of this modified example. In this modified example, since the standard deviation of the noise values ​​is derived considering the thickness of the object, the more the pixel value increases, the smaller the thickness of the object, and the lower the average energy of the pixel. Therefore, as can be deduced from equation (4), the change in the standard deviation of the noise values ​​as the pixel values ​​increase differs between the first embodiment and this modified example. Figure 18 In the example shown, the standard deviation of the noise value increases less when the pixel value increases in graph G22 of this modified example compared to graph G21 of the first embodiment.

[0124] In the control device 20A of the modified embodiment of the first embodiment, the average energy is calculated from the pixel value of each pixel in the X-ray image. Here, for example, if multiple objects with different thicknesses or materials exist in the X-ray image, the average energy for each object is significantly different, making it impossible to adequately remove noise from the X-ray image. According to this structure, since the average energy of the X-rays passing through the object F is calculated for each pixel value in the X-ray image, differences in thickness or material can be taken into account, and noise removal corresponding to the relationship between the pixel value of each pixel in the X-ray image and the noise can be achieved. As a result, noise in the X-ray image can be effectively removed.

[0125] Furthermore, in this modified example, the control device 20A uses a chart G20 derived from each of the various conditional information to derive the average energy from the pixel values ​​of the X-ray image. At this time, the material differences of the object F can also be ignored, and the average energy can be derived from the pixel values. Figure 19 This is a graph showing the relationship between the pixel values ​​and the standard deviation of the noise value in an X-ray image, derived by the calculation unit 202A. Here, the material variation of the object F is also taken into consideration as conditional information, and the relationship is derived as follows: Graph G24 shows an example when the material is aluminum, Graph G23 shows an example when the material is PET (Polyethylene terephthalate), and Graph G25 shows an example when the material is copper. Thus, even when the material of the object F changes, if the tube voltage of the X-ray irradiator 50 and the information of the scintillator provided by the X-ray detection camera 10 used to image the object F are the same, the relationship between the pixel value and the average energy of the transmitted X-ray does not change significantly, and therefore, the relationship between the pixel value and the standard deviation of the noise value does not change significantly. Considering this property, the control device 20A can ignore the material difference of the object F as conditional information and derive the average energy from the pixel values ​​of the X-ray image. Even in this case, the control device 20A according to this modified example can achieve noise removal corresponding to the relationship between the pixel value and the standard deviation of the noise. As a result, noise in the X-ray image can be effectively removed.

[0126] [Another variation of the control device 20 of the first embodiment]

[0127] Figure 20 This is a block diagram illustrating the functional structure of the control device 20B, which is a variation of the first embodiment. The control device 20B differs from the first embodiment in that the image acquisition unit 203B has the function of acquiring an X-ray image of the fixture, and the noise map generation unit 204B has the function of deriving a graph representing the relationship between the standard deviation of pixel values ​​and noise values ​​from the X-ray image of the fixture. Figure 21 It is shown that it is composed of Figure 20A flowchart illustrating the observation processing sequence performed by the image acquisition device 1 of the control device 20B. (See flowchart for example.) Figure 21 As shown, in the control device 20B of this modified example, the displacement... Figure 10 The first embodiment shown executes the processes shown in steps S101, S102 and S104 executed by the control device 20, and then executes the processes shown in steps S201 and S202.

[0128] The image acquisition unit 203B irradiates the fixture with radiation and acquires a radiographic image of the fixture that captures the radiation passing through it (step S201). Specifically, the image acquisition unit 203B uses the image acquisition device 1 to irradiate the fixture and the object F with X-rays and acquires a captured X-ray image. The fixture is a flat plate-shaped member with known thickness and material. That is, before observing and processing the object F, the image acquisition unit 203B uses the image acquisition device 1 to acquire a captured X-ray image of the fixture. Furthermore, the image acquisition unit 203B uses the image acquisition device 1 to acquire a captured X-ray image of the object F. However, the timing of acquiring the X-ray images of the fixture and the object F is not limited to the above; they can be acquired simultaneously or in reverse order (step S103). In addition, the image acquisition unit 203B, like the image acquisition unit 203, irradiates the object F with X-rays and acquires a captured X-ray image that captures the X-rays passing through the object F.

[0129] The image acquisition device 1 sets up a fixture and takes a picture of the fixture. The noise map generation unit 204B derives relational data (step S202) from the X-ray image of the fixture obtained as a result, which shows the relationship between the pixel value and the evaluation value of the expansion of the evaluation noise value. Specifically, the noise map generation unit 204B derives a noise standard deviation map (Noise Standard Deviation Map) from the X-ray image of the fixture, which shows the relationship between the standard deviation of the pixel value and the noise value.

[0130] Figure 22This diagram illustrates an example of generating a noise standard deviation map by the noise map generation unit 204B. The noise map generation unit 204B derives a relationship chart G27 from the X-ray image G26 of the fixture, showing the correspondence between pixel values ​​and the standard deviations of noise values. Furthermore, similar to the first embodiment, the noise map generation unit 204B derives relationship data G2 from the X-ray image G1 acquired by the image acquisition unit 203B, showing the correspondence between each pixel position and pixel value. Then, the noise map generation unit 204B applies the correspondence shown in the relationship chart G27 to each pixel in the relationship data G2, thereby deriving the standard deviation of the noise value corresponding to each pixel position in the X-ray image. As a result, the noise map generation unit 204B establishes a correspondence between the derived noise standard deviation and each pixel position, deriving relationship data G4 showing the correspondence between each pixel position and the standard deviation of noise. Based on the derived relationship data G4, the noise map generation unit 204B generates a noise standard deviation map G5.

[0131] The derivation of the relationship chart G27, which represents the relationship between the standard deviation of pixel values ​​and noise values, from the X-ray image G26 of the fixture, by the noise image generation unit 204B will be explained. Figure 23 The image shows an example of the construction of a jig used for photographing in this modified example. The jig uses, for example, a member P1 whose thickness varies in a stepped manner along one direction. Figure 24 It is shown Figure 23 An example of an X-ray image of a fixture. First, the noise image generation unit 204B, in the X-ray image G26 of the fixture, derives the pixel values ​​(hereinafter referred to as true pixel values) of the fixture when there is no noise for each step, and derives the standard deviation of the noise values ​​based on the true pixel values. Specifically, the noise image generation unit 204B derives the average value of the pixel values ​​of a certain step of the fixture. Furthermore, the noise image generation unit 204B sets the average value of the derived pixel values ​​as the true pixel value of that step. The noise image generation unit 204B, for that step, derives the difference between each pixel value and the true pixel value as the noise value. The noise image generation unit 204B derives the standard deviation of the noise values ​​from the noise values ​​of each derived pixel value.

[0132] Furthermore, the noise map generation unit 204B derives the relationship between the standard deviation of the true pixel value and the noise value, as a graph G27 showing the relationship between the standard deviation of the pixel value and the noise value. Specifically, the noise map generation unit 204B derives the standard deviation of the true pixel value and the noise value for each step of the fixture. By plotting the relationship between the derived standard deviation of the true pixel value and the noise value into a graph, the noise map generation unit 204B depicts an approximate curve, thereby deriving the graph G27 representing the relationship between the standard deviation of the pixel value and the noise value. Moreover, for the approximate curve, exponential approximation, linear approximation, logarithmic approximation, polynomial approximation, power approximation, etc., are used.

[0133] In the control device 20B of this modified example, relational data is generated based on the radiographic image obtained by photographing the actual fixture. This yields relational data for noise removal of the radiographic image most suitable for the object F. As a result, noise in the radiographic image can be removed more effectively.

[0134] Furthermore, the noise map generation unit 204B can also derive the relationship between the standard deviation of pixel values ​​and noise values ​​from images captured with varying tube currents or exposure times in an object-free state, without using a fixture. Based on this structure, since the noise map is generated by producing relational data from radiographic images obtained through actual imaging, noise removal corresponding to the relationship between pixel values ​​and noise amplification can be achieved. As a result, noise in radiographic images can be removed more effectively.

[0135] Specifically, the image acquisition unit 203B may acquire multiple X-ray images taken in a state without an object (step S201), and the noise map generation unit 204B may derive the relationship between the standard deviation of pixel values ​​and noise values ​​from the X-ray images acquired by the image acquisition unit 203B (step S202). The multiple X-ray images are multiple images in which at least one of the conditions of the radiation source or the imaging conditions is different. As an example, the image acquisition unit 203B changes the tube current or exposure time and acquires multiple X-ray images taken using the image acquisition device 1 in a state without an object F before the observation and processing of the object F. Furthermore, the noise map generation unit 204B derives the true pixel value for each X-ray image, and similarly to this modified example, derives the standard deviation of noise based on the true pixel value. Furthermore, similar to this modified example, the noise map generation unit 204B derives a relationship chart representing the relationship between pixel values ​​and the standard deviation of noise by plotting the relationship between the true pixel value and the standard deviation of noise into a graph and drawing an approximate curve. Finally, similarly to the first embodiment, the noise map generation unit 204B generates a noise standard deviation map from the X-ray image acquired by the image acquisition unit 203B based on the derived relational chart.

[0136] [Second Implementation]

[0137] Figure 25 This is a block diagram showing the functional structure of the control device 20C according to the second embodiment. The control device 20C includes: an input unit 201C, a calculation unit 202C, a filtering unit 203C, a selection unit 204C, and a processing unit 205C.

[0138] Furthermore, the control device 20C pre-stores multiple learned models 206C that perform noise removal processing on X-ray transmitted images. These multiple learned models 206C are learning models generated by machine learning, pre-constructed using image data as teaching data. Machine learning methods include teaching-based learning, deep learning, reinforcement learning, and neural network learning. In this embodiment, as an example of a deep learning algorithm, a two-dimensional convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is used. The multiple learned models 206C can also be generated by an external computer and downloaded to the control device 20C, or they can be generated within the control device 20C.

[0139] exist Figure 26 The image data shown is an example of the teaching data used to construct the model 206C after learning. As teaching data, X-ray transmission images of patterns of various thicknesses, materials, and resolutions can be used. Figure 26The example shown is an X-ray transmission image of chicken. This image data can also be X-ray transmission images actually generated using image acquisition device 1 with various objects, or image data generated through simulation calculations. Regarding the X-ray transmission image, it can also be an image obtained using a device different from image acquisition device 1. Furthermore, X-ray transmission images can be combined with image data generated through simulation calculations. Multiple learned models 206C are pre-constructed using image data obtained with transmitted X-rays of different average energies and with known noise distributions. The average energy of the X-rays in the image data is preset to different values ​​by setting the operating conditions of the X-ray irradiator (radiation source) 50 of image acquisition device 1, or the imaging conditions of image acquisition device 1, or by setting the operating conditions or imaging conditions of the X-ray irradiator 50 during simulation calculations (the method for setting the average energy through operating conditions or imaging conditions will be described later). That is, multiple learned models 206C are constructed using training images as training data through machine learning (construction step). The training images are X-ray images corresponding to the average energy of X-rays transmitted through the object F, calculated based on conditional information such as the operating conditions of the X-ray irradiator (radiation source) 50 representing the X-ray transmission image of the object F, or the imaging conditions of the X-ray detection camera 10. For example, in this embodiment, the multiple learned models 206C are constructed using various image data with average energies set to 10keV, 20keV, 30keV, ... and 10keV scale values ​​for multiple frames (e.g., 20,000 frames).

[0140] The teaching data, i.e., the image data, used to construct the learned model 206C is generated in the same production order as in the first embodiment described above.

[0141] Below, return Figure 25 The functions of each functional unit of the control device 20C will be explained in detail.

[0142] The input unit 201C receives from the user of the image acquisition device 1 all inputs, including the operating conditions of the X-ray irradiator (radiation source) 50 when X-rays of the object F pass through the image, and the imaging conditions of the X-ray inspection camera 10. Operating conditions may include all or some of the following: tube voltage, target angle, target material, etc. Imaging conditions may include: the material and thickness of the filters 51 and 19 (filters provided by the camera for imaging the object or filters provided by the source) disposed between the X-ray irradiator 50 and the X-ray inspection camera 10; the distance (FDD) between the X-ray irradiator 50 and the X-ray inspection camera 10; the type of window material of the X-ray inspection camera 10; information related to the material and thickness of the scintillator 11 of the X-ray inspection camera 10; X-ray inspection camera information (e.g., gain setting value, circuit noise value, saturation charge, conversion factor value (electron count), camera line rate (Hz) or line speed (m / min)); and object information, etc. The input unit 201C can receive condition information input as direct input of numerical information, or as selection input of numerical information preset in the internal memory. The input unit 201C receives the aforementioned condition information input from the user, but it can also obtain some condition information (such as tube voltage) based on the detection results of the control state executed by the control device 20C.

[0143] The calculation unit 202C calculates the average energy of the X-rays (radiation) transmitted through the object F by the image acquisition device 1 and detected by the X-ray detection camera 10 based on the conditional information received by the input unit 201C. For example, the calculation unit 202C calculates the spectrum of the X-rays detected by the X-ray detection camera 10 using an approximation formula, such as that of Tucker et al., based on information included in the conditional information, such as tube voltage, target angle, target material, filter material and thickness and presence or absence, window material type and presence or absence, and scintillator 11 material and thickness of the X-ray detection camera 10. Furthermore, the calculation unit 202C calculates the spectral intensity integral value and the photon number integral value based on the X-ray spectrum, and divides the spectral intensity integral value by the photon number integral value to calculate the average energy of the X-rays.

[0144] The calculation method using the well-known Tucker approximation is described below. For example, when the target is determined to be tungsten and the target angle is determined to be 25°, the following can be determined by the calculation unit 202C: Em: kinetic energy of the electron target collision, T: kinetic energy of the electron in the target, A: proportionality constant determined by the atomic number of the target material, ρ: density of the target, μ(E): linear attenuation coefficient of the target material, B: a gradually changing function of Z and T, C: Thomson-Whiddington constant, θ: target angle, c: speed of light in vacuum. Based on this, the calculation unit 202C can calculate the X-ray spectrum of the irradiation by calculating the above equation (1).

[0145] Next, the calculation unit 202C can use the X-ray attenuation formula of equation (2) above to calculate the X-ray energy spectrum that passes through the filter and the object F and is absorbed by the scintillator. The X-ray photon number spectrum is obtained by dividing the X-ray energy spectrum by the energy of each X-ray. The calculation unit 202C calculates the average energy of the X-ray by dividing the integral value of the energy intensity by the integral value of the photon number using the formula (3) above. Through the above calculation process, the calculation unit 202C calculates the average energy of the X-ray. In addition, the known approximation formulas of Kramers, Birch, etc., can also be used for the calculation of the X-ray spectrum.

[0146] The screening unit 203C, based on the average energy value calculated by the calculation unit 202C, filters candidates for learned models from a plurality of pre-constructed learned models 206C. Specifically, the screening unit 203C compares the calculated average energy value with the average energy value of the X-rays in the image data used to construct the plurality of learned models 206C, and filters the plurality of learned models 206C constructed from image data with similar average energy values ​​as candidates. More specifically, if the average energy value calculated by the calculation unit 202C is 53 keV, the screening unit 203C sets the learned models 206C constructed from image data with average energy values ​​of 40 keV, 50 keV, and 60 keV, where the difference from this value is less than a predetermined threshold (e.g., 15 keV), as candidates for learned models.

[0147] The selection unit 204C selects the learned model 206C from the candidates filtered by the screening unit 203C for noise removal processing of the X-ray transmission image of the object F. Specifically, the selection unit 204C acquires an X-ray transmission image taken by the image acquisition device 1 with a fixture as the object, and selects the learned model 206C to be used based on the image characteristics of the X-ray transmission image. At this time, the selection unit 204C analyzes energy characteristics, noise characteristics, or resolution characteristics as image characteristics of the X-ray transmission image, and selects the learned model 206C based on the analysis results.

[0148] More specifically, the selection unit 204C acquires an X-ray transmission image of a flat plate-shaped component, which is a fixture, has a known thickness and material, and whose relationship between average X-ray energy and X-ray transmittance is known. It compares the brightness of the X-ray image transmitted through the fixture with the brightness of the X-ray image transmitted through air, and calculates the X-ray transmittance of the fixture at one point (or an average of multiple points). For example, if the brightness of the X-ray image transmitted through the fixture is 5550 and the brightness of the X-ray image transmitted through air is 15000, the transmittance is calculated to be 37%. Furthermore, the selection unit 204C determines the average energy of the transmitted X-rays (e.g., 50 keV) estimated based on the 37% transmittance as the energy characteristic of the fixture's X-ray transmission image. The selection unit 204C selects an image data whose average energy is closest to the determined average energy value and constructs a learned model 206C.

[0149] In addition, the selection unit 204C can also analyze the multi-point characteristics of the fixture with varying thickness or material, as well as the energy characteristics of the X-ray transmission image of the fixture. Figure 27 This is a diagram showing an example of an X-ray transmission image of the object being analyzed by the selection unit 204C. Figure 27 This is an X-ray transmission image of a fixture with a stepped thickness. The selection unit 204C selects multiple measurement areas (ROIs) of varying thicknesses based on this X-ray transmission image, analyzes the average brightness of each of the multiple measurement areas, and obtains a thickness-brightness characteristic curve as an energy characteristic. Figure 28 The image shows an example of a thickness-brightness characteristic curve obtained by the selection unit 204C.

[0150] Furthermore, selection unit 204C similarly uses the image data used to construct the learned model 206C selected by screening unit 203C as the object, and obtains a thickness-brightness characteristic curve. The learned model 206C constructed from image data with characteristics closest to the characteristic curve obtained with the fixture as the object is selected as the final learned model 206C. However, the image characteristics of the image data used to construct this learned model 206C can also refer to image characteristics calculated in advance outside the control device 20C. In this way, by setting multiple measurement areas, the learned model with the most suitable noise of the X-ray transmission image for the object F can be selected. In particular, the differences in the X-ray spectrum or the effect of the filter during the measurement of the X-ray transmission image can be estimated with high precision.

[0151] Furthermore, the selection unit 204C can analyze the luminance value and noise of each of multiple measurement areas as the noise characteristics of the X-ray transmission image of the fixture, and obtain a luminance-noise ratio characteristic curve as the noise characteristic. That is, the selection unit 204C selects multiple measurement areas (ROIs) with different thicknesses or materials based on the X-ray transmission image, analyzes the standard deviation and average luminance value of the multiple measurement areas (ROIs), and obtains a luminance-SNR (SN ratio) characteristic curve as the noise characteristic. At this time, the selection unit 204C calculates the SNR of each measurement area (ROI) using SNR = (average luminance value) ÷ (standard deviation of luminance value). Figure 29 The image shows an example of the luminance-SNR characteristic curve obtained by the selection unit 204C. Furthermore, the selection unit 204C selects the learned model 206C, which is constructed from image data having noise characteristics that are closest to the obtained characteristic curve, as the final learned model 206C.

[0152] Here, the selection unit 204C can also obtain a noise characteristic curve with the vertical axis set to the standard deviation of the brightness value, instead of the brightness-SNR characteristic curve mentioned above, as the noise characteristic. By using such a brightness-noise characteristic, the dominant noise factor (shot noise, readout noise, etc.) can be determined for each signal quantity detected by the X-ray inspection camera 10 based on the slope of the curve of the region of each signal quantity, and the learned model 206C is selected based on the determination result.

[0153] Figure 30 This diagram illustrates the selection function of the learned model based on image characteristics, performed by the selection unit 204C. Figure 30 Part (a) shows the brightness-SNR characteristic curves G1, G2, and G3 of the various image data used to construct multiple learned models 206C. Part (b) shows, in addition to these characteristic curves G1, G2, and G3, the brightness-SNR characteristic curve G of the X-ray transmitted image of the fixture. T In the characteristic curves G1, G2, G3, G... T In the case of an object, select section 204C to select the one closest to the characteristic curve G. T The learned model 206C is constructed using the image data of the characteristic curve G2, which represents the characteristics of the model, and performs its function. During selection, the selection unit 204C selects the characteristic curves G1, G2, G3, and characteristic curve G... TIn between, the SNR error of the brightness values ​​at regular intervals is calculated, and the root mean square error (RMSE) of these errors is calculated. The learned model 206C corresponding to the characteristic curves G1, G2, and G3 with the smallest RMSE is selected. Alternatively, when the selection unit 204C uses energy characteristics for selection, the learned model 206C can be selected in the same way.

[0154] The X-ray transmission image of the fixture can also be selected as the object. Based on the characteristics of the image after noise removal processing by applying multiple learned models, the learned model 206C is selected.

[0155] For example, the selection unit 204C uses X-ray transmission images of a fixture capturing graphs at various resolutions, applies multiple learned models 206C to these images, and evaluates the resulting noise-removed image. Furthermore, the selection unit 204C selects the learned model 206C that produces the smallest change in resolution before and after the noise removal process. Figure 31 The image shown is an example of an X-ray transmission image used for resolution evaluation. In this X-ray transmission image, a graph showing the resolution varying stepwise along one direction is used as the subject of the photograph. The resolution of an X-ray transmission image can be determined using MTF (Modulation Transfer Function) or CTF (Contrast Transfer Function).

[0156] In addition to evaluating the resolution change mentioned above, the selection unit 204C can also evaluate the brightness-to-noise ratio characteristics of the noise-removed image and select the learned model 206C used to generate the image with the highest brightness-to-noise ratio. Figure 32 The image shows an example of the structure of a fixture used for evaluating the luminance-noise ratio. For example, a fixture could be used in which foreign objects P2 of various materials and sizes are scattered in a component P1 whose thickness varies in a stepped manner along one direction. Figure 33 Showing Figure 32 The fixture obtains a noise-removed X-ray transmitted image of the object. The selection unit 204C selects an image region R1 containing the foreign object P2 and an image region R2 near R1 that does not contain the foreign object P2 from the X-ray transmitted image. The minimum brightness L of image region R1 is then calculated. MIN The average brightness L of image region R2 AVE The standard deviation L of the brightness of image region R2. SD Furthermore, the selection unit 204C uses the following formula:

[0157] CNR=(L AVE-L MIN ) / L SD

[0158] The brightness-to-noise ratio (CNR) is calculated. Then, the selection unit 204C calculates the brightness-to-noise ratio (CNR) for each of the X-ray transmission images after applying multiple learned models 206C, and selects the learned model 206C that generates the X-ray transmission image with the highest brightness-to-noise ratio (CNR).

[0159] Alternatively, in selection 204C, the brightness can also be based on the average value L of the brightness of image region R1. AVE_R1 The average brightness L of image region R2 AVE_R2 The standard deviation L of the brightness of image region R2. SD The following formula is used for calculation.

[0160] CNR=(L AVE_R1 -L MIN_R2 ) / L SD

[0161] The processing unit 205C applies the learned model 206C selected by the selection unit 204C to the X-ray transmission image obtained with the object F as the target, and performs noise removal image processing to generate an output image. The processing unit 205C then outputs the generated output image to a display device 30 or the like.

[0162] Next, the sequence of observation and processing of the X-ray transmitted image of the object F using the image acquisition apparatus 1 of the second embodiment, i.e., the flow of the radiographic image acquisition method of the second embodiment, will be described. Figure 34 This is a flowchart illustrating the sequence of observation processes performed by the image acquisition device 1.

[0163] First, the operator (user) of the image acquisition device 1 receives input from the control device 20C, including the operating conditions of the X-ray irradiator 50 and the imaging conditions of the X-ray detection camera 10 (step S1). Next, the control device 20C calculates the average energy value of the X-rays detected by the X-ray detection camera 10 based on the condition information (step S2).

[0164] Then, the average energy value of the X-rays used to construct the image data of the learned model 206C stored in the control device 20C is determined by the control device 20C (step S3). Thereafter, the determination of the average energy value of the X-rays is repeated for all the learned models 206C stored in the control device 20C (step S4).

[0165] Next, multiple candidates for the learned model 206C are selected by comparing the average energy value of the X-rays calculated by the control device 20C (step S5). Then, an X-ray transmission image of the fixture is obtained by setting the fixture in the image acquisition device 1 and taking a picture of the fixture (step S6).

[0166] Subsequently, the control device 20C acquires the image characteristics of the X-ray transmitted image of the fixture (average energy value of X-rays, thickness-brightness characteristics, brightness-noise ratio characteristics, brightness-noise characteristics, resolution variation characteristics, etc.) (step S7). Then, based on the acquired image characteristics, the control device 20C selects the final learned model 206C (step S8).

[0167] Next, by setting the object F in the image acquisition device 1 and taking a picture of the object F, an X-ray transmission image of the object F is obtained (step S9). Then, by applying the finally selected learned model 206C to the X-ray transmission image of the object F by the control device 20C, noise removal processing is performed on the X-ray transmission image as the object (step S10). Finally, by the control device 20C, the noise-removed X-ray transmission image, i.e., the output image, is output to the display device 30 (step S11).

[0168] According to the image acquisition apparatus 1 described above, the signal component in the X-ray transmitted image can be increased and the noise component removed, effectively improving the signal-to-noise ratio (S / N ratio) of the X-ray transmitted image. Furthermore, based on the operating conditions of the X-ray source or the imaging conditions of the X-ray transmitted image when acquiring the X-ray transmitted image of the object F, the average energy of the X-rays transmitted through the object F is calculated. Based on this average energy, candidates for the learned model 206C used for noise removal are selected from the pre-built learned model 206C. Therefore, since the learned model 206C corresponding to the average energy of the X-rays of the object is used for noise removal, noise removal corresponding to the relationship between the brightness and noise of the X-ray transmitted image can be achieved. As a result, noise in the X-ray transmitted image can be effectively removed, for example, improving foreign object detection performance. In particular, the manner of noise in the X-ray transmitted image varies depending on factors such as tube voltage, filter, scintillator, X-ray detection camera conditions (gain setting, circuit noise value, saturation charge, conversion factor value (e- / count), camera line frequency), and the object. Therefore, when aiming to remove noise through machine learning, it is necessary to prepare multiple learning models that can learn under various conditions. Currently, it is not possible to select a learning model that conforms to the noise level from multiple learning models based on the conditions of X-ray transmission image measurement. According to this embodiment, by selecting the learned model 206C corresponding to the average energy of the X-rays of the object being photographed, the selection of a learning model that always conforms to the noise level is achieved.

[0169] Generally, X-ray transmission images contain noise generated by X-rays. Increasing the X-ray dose to improve the signal-to-noise ratio (SN ratio) of the X-ray transmission image is also considered. However, in this case, increasing the X-ray dose increases the irradiation of the sensor, shortening its lifespan, and also shortens the lifespan of the X-ray source, making it difficult to simultaneously improve the SN ratio and extend its lifespan. In this embodiment, since it is not necessary to increase the X-ray dose, both improving the SN ratio and extending the lifespan can be achieved.

[0170] Furthermore, the control device 20C in this embodiment has the following function: using the selected learned model 206C, it performs image processing to remove noise from the X-ray transmitted image of the object F. This function enables noise removal that corresponds to the relationship between the brightness and noise of the X-ray transmitted image, and effectively removes noise from the X-ray transmitted image.

[0171] Furthermore, the control device 20C of this embodiment has the following function: by comparing the average energy value of the X-rays calculated based on the selection information with the average energy value determined based on the image data used to construct the learned model 206C, candidate models for the learned model are selected. This function reliably achieves noise removal corresponding to the relationship between the brightness and noise of the X-ray transmitted image.

[0172] Furthermore, the control device 20C of this embodiment has the following function: selecting a learned model 206C from candidates based on the image characteristics of the X-ray transmission image of the fixture. Through this function, the learned model 206C most suitable for noise removal of the X-ray transmission image of the object F can be selected. As a result, noise removal corresponding to the relationship between the brightness and noise of the X-ray transmission image can be achieved more reliably.

[0173] [Modifications of the Second Embodiment]

[0174] In addition, the control device 20C of the second embodiment described above selects candidates for the learned model 206C based on the average energy value of the X-ray calculated according to the condition information, but it may also have functions to correspond to the performance degradation of the X-ray detection camera 10, the output change of the X-ray irradiator 50, or the performance degradation.

[0175] Figure 35 This is a block diagram showing the functional structure of the control device 20D, a modified example of the second embodiment. The control device 20D differs from the control device 20C of the second embodiment in that it has a measuring unit 207C, and in the functions of the calculation unit 202D and the screening unit 203D.

[0176] In control device 20C, by setting the performance degradation of X-ray inspection camera 10 and output variation or performance degradation of X-ray irradiator 50 to none, and by estimating the relationship between brightness and noise of X-ray transmitted image based on average X-ray energy, a learned model 206C is selected. In contrast, control device 20D of this variant has the following function: considering the performance degradation of X-ray inspection camera 10, output variation or performance degradation of X-ray irradiator 50, calculating the X-ray conversion coefficient, and selecting the learned model 206C based on the X-ray conversion coefficient. The X-ray conversion coefficient is a parameter representing the efficiency of X-ray conversion from visible light by the scintillator to electrons (electrical signals) by the camera's sensor.

[0177] Generally speaking, the X-ray conversion factor F TWhen the average energy of the X-ray is set to E[keV], the scintillator emission is set to EM[photon / keV], the coupling efficiency of the sensor is set to C, and the quantum efficiency of the sensor is set to QE, the following formula can be used for calculation.

[0178] F T =E×EM×C×QE

[0179] Additionally, the signal-to-noise ratio (SNR) of the X-ray transmitted image is calculated using the X-ray conversion factor F. T X-ray photon number N P The camera readout noise Nr is obtained using the following formula.

[0180] SNR = F T N P / {(F T N P +Nr 2 ) 1 / 2}

[0181] Therefore, it can be based on the X-ray conversion factor F T It is assumed that the relationship between the brightness and noise of the X-ray transmitted image is taken into account due to the degradation of camera performance.

[0182] The measurement unit 207C of the control device 20D has the following functions: measuring the decrease in luminous intensity EM, which represents performance degradation of the scintillator 11; the decrease in quantum efficiency QE, which represents performance degradation of the sensor of the scanning camera 12; and the change in average energy E, which represents output variation and performance degradation of the X-ray irradiator 50. For example, the measurement unit 207C measures the decrease in luminous intensity between the scintillator 11 in its undegraded state (when it was new) and the current state of the scintillator 11, and estimates the current luminous intensity EM based on this decrease. Additionally, the measurement unit 207C measures the decrease in brightness between the scanning camera 12 in its undegraded state (when it was new) and the current state of the scanning camera 12, and estimates the current quantum efficiency QE based on this decrease. Furthermore, the measurement unit 207C estimates the current average energy E based on the change in average energy between the X-ray irradiator 50 in its undegraded state (when it was new) and the current state of the X-ray irradiator 50. The average energy E can be obtained from the imaging data of a flat plate component where the thickness and material are known, and the relationship between the average energy of X-rays and the X-ray transmittance is known. Alternatively, it can be obtained from the imaging data of multiple points of a fixture with varying thickness or material.

[0183] The calculation unit 202D of the control device 20D uses the calculated average energy E of the X-rays and the luminous intensity EM and quantum efficiency QE estimated by the measurement unit 207C to calculate the X-ray conversion coefficient F. TThe screening unit 203D of the control device 20D has the following function: It filters the calculated X-ray conversion coefficient F... T X-ray conversion coefficients F of the image data used to construct the model 206C after learning. T Comparisons are made to filter candidates for the learned model 206C.

[0184] In addition, the control device 20D of the above-mentioned modified example selects the learned model based on the image characteristics obtained by the shooting fixture after screening the candidates of the learned model. However, it is also possible to perform noise removal processing on the X-ray transmitted image of the object without shooting the fixture. Figure 36 This is a flowchart illustrating the sequence of observation processing performed by the image acquisition device 1 in another variation. Thus, it can also be omitted. Figure 34 In steps S6 to S8, noise removal is performed using a learned model filtered based on average energy.

[0185] [Third Implementation]

[0186] Figure 37 This is a block diagram showing the functional structure of the control device 20E according to the third embodiment. The control device 20E includes: an acquisition unit 201E, a determination unit 202E, a selection unit 204E, and a processing unit 205E.

[0187] Furthermore, the control device 20E contains pre-stored multiple learned models 206E that perform noise removal processing on X-ray transmitted images. These multiple learned models 206E are learning models pre-constructed using image data as teaching data and generated by machine learning. Machine learning methods include teaching-based learning, deep learning, reinforcement learning, and neural network learning. In this embodiment, as an example of a deep learning algorithm, a two-dimensional convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is used. The multiple learned models 206E can also be generated by an external computer and downloaded to the control device 20E, or they can be generated within the control device 20E.

[0188] exist Figure 38 The image data shown is an example of the teaching data used to construct the model 206E after learning. As teaching data, X-ray transmission images of patterns of various thicknesses, materials, and resolutions can be used. Figure 38The example shown is an X-ray transmission image of chicken. This image data can also be X-ray transmission images actually generated using image acquisition device 1 with various objects, or image data generated through simulation calculations. Regarding the X-ray transmission image, it can also be an image obtained using a device different from image acquisition device 1. Furthermore, X-ray transmission images can be combined with image data generated through simulation calculations. Multiple learned models 206E are pre-constructed using image data obtained with transmitted X-rays of different average energies and with known noise distributions. The average energy of the X-rays in the image data is preset to different values ​​by setting the operating conditions of the X-ray irradiator (radiation source) 50 of image acquisition device 1, or the imaging conditions of image acquisition device 1, or by setting the operating conditions or imaging conditions of the X-ray irradiator 50 during simulation calculations. That is, multiple learned models 206E are constructed using training images as training data through machine learning (construction step). The training images are X-ray images corresponding to the average energy of X-rays transmitted through the object F, calculated based on conditional information such as the operating conditions of the X-ray irradiator (radiation source) 50 representing the X-ray transmission image of the object F, or the imaging conditions of the X-ray detection camera 10. For example, in this embodiment, the multiple learned models 206E are constructed using various image data with average energies set to 10keV, 20keV, 30keV, ... and 10keV scale values ​​for multiple frames (e.g., 20,000 frames).

[0189] The teaching data, i.e., the image data, used to construct the learned model 206E is generated in the same production order as in the first embodiment described above.

[0190] Below, return Figure 37 The functions of each functional unit of the control device 20E will be explained in detail.

[0191] The acquisition unit 201E acquires X-ray transmission images of a fixture and an object F irradiated with X-rays using the image acquisition device 1. The fixture is a flat plate with known thickness and material, and a known relationship between the average X-ray energy and X-ray transmittance, or a fixture with charts taken at various resolutions. That is, the acquisition unit 201E acquires the X-ray transmission image of the fixture taken by the image acquisition device 1 before observing the object F. Furthermore, the acquisition unit 201E acquires the X-ray transmission image of the object F taken by the image acquisition device 1 at a time after the model 206E has been learned, based on the X-ray transmission image of the fixture. However, the acquisition time of the X-ray transmission images of the fixture and the object F is not limited to the above; they can be acquired simultaneously or at opposite times.

[0192] The determination unit 202E determines the image characteristics of the X-ray transmission image of the fixture acquired by the acquisition unit 201E. Specifically, the selection unit 204E determines the image characteristics of the X-ray transmission image, such as energy characteristics, noise characteristics, resolution characteristics, or frequency characteristics.

[0193] For example, when using a flat plate member of known thickness and material as a fixture, the determination unit 202E compares the brightness of the X-ray image transmitted through the fixture with the brightness of the X-ray image transmitted through air, and calculates the X-ray transmittance of one point (or an average of multiple points) of the fixture. For example, if the brightness of the X-ray image transmitted through the fixture is 5550 and the brightness of the X-ray image transmitted through air is 15000, the transmittance is calculated to be 37%. Furthermore, the determination unit 202E determines the average energy of the transmitted X-rays (e.g., 50 keV) estimated based on the transmittance of 37% as the energy characteristic of the X-ray transmitted image of the fixture.

[0194] In addition, the determination unit 202E can also analyze the characteristics of multiple points of the fixture with changes in thickness or material, as well as the energy characteristics of the X-ray transmission image of the fixture. Figure 39 This is a diagram showing an example of an X-ray transmission image of the object being analyzed by the determination unit 202E. Figure 39 This is an X-ray transmission image of a fixture with a stepped thickness. The determination unit 202E selects multiple measurement areas (ROIs) of varying thicknesses based on this X-ray transmission image, analyzes the average brightness of each of the multiple measurement areas, and obtains a thickness-brightness characteristic curve as an energy characteristic. Figure 40 The image shows an example of a thickness-brightness characteristic curve obtained by the determination unit 202E.

[0195] Furthermore, the determination unit 202E can analyze the luminance value and noise of each of multiple measurement areas as the noise characteristics of the X-ray transmission image of the fixture, and obtain a luminance-noise ratio characteristic curve as the noise characteristic. That is, the determination unit 202E selects multiple measurement areas (ROIs) with different thicknesses or materials based on the X-ray transmission image, analyzes the standard deviation and average luminance value of the multiple measurement areas (ROIs), and obtains a luminance-SNR (SN ratio) characteristic curve as the noise characteristic. At this time, the determination unit 202E calculates the SNR of each measurement area (ROI) using SNR = (average luminance value) ÷ (standard deviation of luminance value). Figure 41 The diagram shows an example of a luminance-SNR characteristic curve obtained by the determination unit 202E. Alternatively, the determination unit 202E may obtain a noise characteristic curve with the vertical axis set to the standard deviation of the luminance value, instead of the aforementioned luminance-SNR characteristic curve, as the noise characteristic.

[0196] Furthermore, when using a fixture with a graph, the determination unit 202E can also obtain the resolution distribution of the X-ray transmission image of the fixture as a resolution characteristic. Moreover, the determination unit 202E has the function of obtaining resolution characteristics for the image after applying multiple learned models 206E to the X-ray transmission image of the fixture and performing noise removal processing. Figure 42 The image shown is an example of an X-ray transmission image used for resolution evaluation. In this X-ray transmission image, a graph showing the resolution varying stepwise along one direction is used as the subject of the photograph. The resolution of an X-ray transmission image can be determined using MTF (Modulation Transfer Function) or CTF (Contrast Transfer Function).

[0197] Refer again Figure 37 The selection unit 204E selects, from among the multiple learned models 206E stored in the control device 20E, a learned model 206E that will ultimately be used for noise removal processing of the X-ray transmission image of the object F, based on the image characteristics obtained by the determination unit 202E. That is, the selection unit 204E compares the image characteristics determined by the determination unit 202E with the image characteristics determined based on the image data used to construct the multiple learned models 206E, and selects a learned model 206E that is similar in both.

[0198] For example, the selection unit 204E selects a learned model 206E, which is constructed using image data whose average energy is closest to the value of the average energy of transmitted X-rays determined by the determination unit 202E.

[0199] Furthermore, similarly to the determination method performed by the determination unit 202E, the selection unit 204E obtains a thickness-brightness characteristic curve based on the image data used to construct multiple learned models 206E. The learned model 206E constructed using image data with characteristics closest to the thickness-brightness characteristic curve obtained with a fixture as the object is selected as the final learned model 206E. However, the image characteristics of the image data used to construct the learned model 206E can also be referenced to image characteristics calculated in advance outside the control device 20E. Thus, by using the image characteristics obtained by setting multiple measurement areas, the most suitable learned model for noise removal of the X-ray transmission image of the object F can be selected. In particular, differences in the X-ray spectrum or the effect of the filter during the measurement of the X-ray transmission image can be estimated with high precision.

[0200] Alternatively, the selection unit 204E may select the learned model 206E constructed from image data having a luminance-noise ratio characteristic that is closest to the luminance-noise ratio characteristic obtained by the determination unit 202E as the final learned model 206E. However, the image characteristics of the image data used to construct the learned model 206E may also be obtained by the selection unit 204E based on the image data, or may refer to image characteristics calculated in advance outside the control device 20E. Here, the selection unit 204E may also use the luminance-noise characteristic instead of the luminance-noise ratio characteristic as the noise characteristic to select the learned model 206E. By using such a luminance-noise characteristic, the dominant noise factor (shot noise, readout noise, etc.) can be determined for each signal quantity detected by the X-ray detection camera 10 based on the slope of the curve of the region of each signal quantity, and the learned model 206E can be selected based on the determination result.

[0201] Figure 43 This diagram illustrates the selection function of the learned model based on image characteristics, performed by the selection unit 204E. Figure 43 In the diagram, section (a) shows the brightness-SNR characteristic curves G1, G2, and G3 of the various image data used to construct multiple learned models 206E. In section (b), in addition to these characteristic curves G1, G2, and G3, the brightness-SNR characteristic curve G of the X-ray transmitted image of the fixture is also shown. T In the characteristic curves G1, G2, G3, G... T In the case of an object, select section 204E to select the one closest to the characteristic curve G. TThe learned model 206E is constructed using the image data of the characteristic curve G2, which represents the characteristics of the model, and then functions accordingly. During selection, the selection unit 204E selects the characteristic curves G1, G2, G3, and characteristic curve G... T In between, the SNR error of the brightness values ​​at regular intervals is calculated, and the root mean square error (RMSE) of these errors is calculated. The learned model 206E corresponding to the characteristic curves G1, G2, and G3 with the smallest RMSE is selected. Alternatively, when the selection unit 204E uses energy characteristics for selection, the learned model 206E can also be selected in the same way.

[0202] The selection unit 204E can also use the X-ray transmission image of the fixture as the object. Based on the characteristics of the image after noise removal processing is performed using multiple learned models, the learned model 206E that produces the image with relatively superior characteristics is selected.

[0203] For example, the selection unit 204E uses X-ray transmission images of a fixture that captures charts with various resolutions, applies multiple learned models 206E to these images, and evaluates the resolution characteristics of the resulting noise-removed image. Furthermore, the selection unit 204E selects the learned model 206E from the images whose resolution changes least before and after noise removal processing.

[0204] In addition to evaluating the resolution change mentioned above, the selection unit 204E can also evaluate the brightness-to-noise ratio characteristics of the noise-removed image and select the learned model 206E that generates the image with the highest brightness-to-noise ratio. Figure 44 The image shows an example of the structure of a fixture used for evaluating the luminance-noise ratio. For example, a fixture could be used in which foreign objects P2 of various materials and sizes are scattered in a component P1 whose thickness varies in a stepped manner along one direction. Figure 45 Showing Figure 44 The fixture obtains a noise-removed X-ray transmitted image of the object. The selection unit 204E selects an image region R1 containing the foreign object P2 and an image region R2 near R1 that does not contain the foreign object P2 from the X-ray transmitted image. The minimum brightness L of image region R1 is calculated. MIN The average brightness L of image region R2 AVE The standard deviation L of the brightness of image region R2. SD Furthermore, the selection unit 204E uses the following formula:

[0205] CNR=(L AVE -L MIN ) / L SD

[0206] The brightness-to-noise ratio (CNR) is calculated. Then, the selection unit 204E calculates the brightness-to-noise ratio (CNR) for each of the X-ray transmission images after applying multiple learned models 206E, and selects the learned model 206E that generates the X-ray transmission image with the highest brightness-to-noise ratio (CNR).

[0207] Alternatively, selection unit 204E can also be based on the average brightness L of image region R1. AVE_R1 The average brightness L of image region R2 AVE_R2 The standard deviation L of the brightness of image region R2. SD The following formula is used for calculation.

[0208] CNR=(L AVE_R1 -L MIN_R2 ) / L SD

[0209] The processing unit 205E applies the learned model 206E selected by the selection unit 204E to the X-ray transmission image obtained with the object F as the target, and performs noise removal image processing to generate an output image. The processing unit 205E then outputs the generated output image to a display device 30 or the like.

[0210] Next, the sequence of observation and processing of the X-ray transmitted image of the object F using the image acquisition apparatus 1 of the third embodiment, i.e., the flow of the radiographic image acquisition method of the third embodiment, will be described. Figure 46 This is a flowchart illustrating the sequence of observation processes performed by the image acquisition device 1.

[0211] First, the operator (user) of the image acquisition device 1 sets the imaging conditions of the image acquisition device 1, such as the tube voltage of the X-ray irradiator 50 or the gain of the X-ray detection camera 10 (step S1E). Next, a fixture is set in the image acquisition device 1, and an X-ray transmission image is acquired with the fixture as the object by the control device 20E (step S2E). At this time, X-ray transmission images of various fixtures can also be acquired sequentially.

[0212] Correspondingly, the control device 20E determines the image characteristics (energy characteristics, noise characteristics, and resolution characteristics) of the X-ray transmission image of the fixture (step S3E). Furthermore, the control device 20E applies multiple learned models 206E to the X-ray transmission image of the fixture, and determines the image characteristics (resolution characteristics or brightness-to-noise ratio, etc.) of each X-ray transmission image after applying the multiple learned models 206E (step S4E).

[0213] Next, via control device 20E, the learned model 206E is selected based on a comparison of the energy characteristics of the X-ray transmitted image of the fixture with the energy characteristics of the image data used to construct the learned model 206E, and the degree of change in the resolution characteristics of the X-ray transmitted image of the fixture before and after the application of the learned model (step S5E). Alternatively, the learned model 206E can also be selected based on a comparison of the noise characteristics of the X-ray transmitted image of the fixture with the noise characteristics of the image data used to construct the learned model 206E, and the degree of change in the resolution characteristics of the X-ray transmitted image of the fixture before and after the application of the learned model. Furthermore, in step S5E, instead of the above processing, the learned model 206E with the highest brightness-to-noise ratio (CNR) after applying the learned model can be selected.

[0214] Next, by setting the object F in the image acquisition device 1 and taking a picture of the object F, an X-ray transmission image of the object F is obtained (step S7E). Then, by applying the finally selected learned model 206E to the X-ray transmission image of the object F by the control device 20E, noise removal processing is performed on the X-ray transmission image as the object (step S8E). Finally, the X-ray transmission image with noise removal processing, i.e., the output image, is output to the display device 30 by the control device 20E (step S9E).

[0215] According to the image acquisition apparatus 1 described above, the signal component in the X-ray transmitted image can be increased and the noise component removed, effectively improving the signal-to-noise ratio (S / N ratio) of the X-ray transmitted image. Furthermore, the image characteristics of the X-ray image of the fixture are determined, and based on these image characteristics, a learned model for noise removal is selected from a pre-built learned model. Thus, the characteristics of the X-ray transmitted image, which vary due to the operating conditions of the X-ray irradiator 50 of the image acquisition apparatus 1, can be estimated. The learned model 206E selected based on this estimation result is then used for noise removal. Therefore, noise removal corresponding to the relationship between the brightness and noise of the X-ray transmitted image can be achieved. As a result, noise in the X-ray transmitted image can be effectively removed.

[0216] Generally, X-ray transmission images contain noise generated by X-rays. Increasing the X-ray dose to improve the signal-to-noise ratio (SN ratio) of the X-ray transmission image is also considered. However, in this case, increasing the X-ray dose increases the irradiation of the sensor, shortening its lifespan, and also shortens the lifespan of the X-ray source, making it difficult to simultaneously improve the SN ratio and extend its lifespan. In this embodiment, since it is not necessary to increase the X-ray dose, both improving the SN ratio and extending the lifespan can be achieved.

[0217] In this embodiment, during the selection of the learned model, the image characteristics of the X-ray transmission image of the fixture are compared with the image characteristics of the image data used to construct the learned model. Therefore, the learned model 206E constructed using image data corresponding to the image characteristics of the X-ray transmission image of the fixture is selected, thus effectively removing noise from the X-ray transmission image of the object F.

[0218] Furthermore, in this embodiment, the learned model is selected by applying image characteristics of the X-ray transmission image of the fixture to multiple learned models 206E. In this case, by actually applying the image characteristics of the X-ray transmission image of the fixture to multiple learned models 206E, the learned model 206E is selected, thus effectively removing noise from the X-ray transmission image of the object F.

[0219] In particular, in this embodiment, energy characteristics or noise characteristics are used as image characteristics. In this case, a learned model 206E is selected by constructing an image whose characteristics are similar to the energy characteristics or noise characteristics of the X-ray transmission image of the fixture, which varies due to the imaging conditions of the image acquisition device 1. As a result, noise removal of the X-ray transmission image of the object F corresponding to the changes in the conditions of the image acquisition device 1 becomes possible.

[0220] In this embodiment, resolution characteristics or brightness-to-noise ratio are also used as image characteristics. Based on this structure, by applying the selected learned model 206E, an X-ray transmission image with good resolution characteristics or brightness-to-noise ratio can be obtained. As a result, noise removal of the X-ray transmission image of the object becomes possible, corresponding to changes in the conditions of the image acquisition device 1.

[0221] In the above embodiments, it is preferable to construct a fully learned model through machine learning, whereby image data obtained by adding normally distributed noise values ​​to a radial image of a specified structure is used as teaching data. Therefore, the teaching data, i.e., the image data, for constructing the fully learned model is readily available, and the fully learned model can be constructed effectively.

[0222] Furthermore, in the above embodiments, it is also preferable that the image processing module includes: a noise map generation unit that, based on relational data representing the relationship between pixel values ​​and the expanded evaluation values ​​of the evaluation noise values, derives evaluation values ​​from the pixel values ​​of each pixel in the radiation image, and generates data corresponding to the derived evaluation values ​​and each pixel of the radiation image, i.e., a noise map; and a processing unit that inputs the radiation image and the noise map into the learned model and performs noise removal processing to remove noise from the radiation image. Furthermore, it is also preferable that, in the execution step, based on the relational data representing the relationship between pixel values ​​and the expanded evaluation values ​​of the evaluation noise values, evaluation values ​​are derived from the pixel values ​​of each pixel in the radiation image, and data corresponding to the derived evaluation values ​​and each pixel of the radiation image, i.e., a noise map, is generated; the radiation image and the noise map are input into the learned model, and noise removal processing to remove noise from the radiation image is performed. In this case, based on the relational data representing the relationship between pixel values ​​and the expanded evaluation values ​​of the evaluation noise values, evaluation values ​​are derived from the pixel values ​​of each pixel in the radiation image, and data corresponding to the derived evaluation values ​​and each pixel of the radiation image, i.e., a noise map, is generated. Furthermore, the radiation image and noise map are input into a pre-learned model constructed using machine learning, and noise removal processing is performed to remove noise from the radiation image. Thus, the amplification of noise values ​​evaluated based on the pixel values ​​of each pixel in the radiation image can be considered, and noise in each pixel of the radiation image can be removed using machine learning. The learned model achieves noise removal corresponding to the relationship between the pixel values ​​of the radiation image and noise amplification. As a result, noise in the radiation image can be effectively removed.

[0223] Furthermore, in the above embodiment, it is preferable that the image processing module includes: an input unit that receives input of conditions indicating the radiation source when irradiating and photographing an object, or condition information representing arbitrary photographing conditions; a calculation unit that calculates the average energy related to the radiation passing through the object based on the condition information; and a filtering unit that filters, based on the average energy, a learned model for noise removal processing from a plurality of learned models pre-constructed using image data through machine learning. Moreover, it is preferable that, in the executed steps, the input of conditions indicating the radiation source when irradiating and photographing an object, or condition information representing arbitrary photographing conditions, is received; the average energy related to the radiation passing through the object is calculated based on the condition information; and the learned model for noise removal processing is filtered from a plurality of learned models pre-constructed using image data through machine learning based on the average energy. In this case, the average energy of the radiation passing through the object is calculated based on the conditions of the radiation source or photographing conditions when acquiring the radiation image of the object. Furthermore, based on this average energy, candidates for learned models for noise removal are filtered from the pre-constructed learned models. Therefore, by using a learned model corresponding to the average energy of the radiation from the photographed object for noise removal, noise removal corresponding to the relationship between brightness and noise in the radiation image can be achieved. As a result, noise in radiation images can be effectively removed.

[0224] Furthermore, in the above embodiment, it is also preferable that the image processing module includes: a determination unit that determines the image characteristics of a radiation image acquired by an imaging device with the fixture as the object; a selection unit that selects a learned model from a plurality of learned models pre-constructed using image data through machine learning based on the image characteristics; and a processing unit that performs noise removal processing using the selected learned model. Moreover, it is also preferable that, in the execution step, the image characteristics of the radiation image acquired with the fixture as the object are determined, a learned model is selected from a plurality of learned models pre-constructed using image data through machine learning based on the image characteristics, and noise removal processing is performed using the selected learned model. According to this structure, the image characteristics of the radiation image of the fixture are determined, and a learned model for noise removal is selected from the pre-constructed learned models based on these image characteristics. Thus, the characteristics of the radiation image, which vary according to the conditions of the radiation source of the system, can be estimated, and the learned model selected based on this estimation result can be used for noise removal. Therefore, noise removal corresponding to the relationship between the brightness and noise of the radiation image can be achieved. As a result, noise in the radiation image can be effectively removed.

[0225] [Industry availability]

[0226] The embodiments, which are used for acquiring radiographic images, radiographic images systems, and radiographic images, can effectively improve the signal-to-noise ratio (S / N) of radiographic images.

[0227] Symbol Explanation

[0228] 1…Image acquisition device (radiographic image acquisition device, radiographic image acquisition system); 10…X-ray inspection camera (imaging device); 11…Scintillator; 12…Scanning camera (detection element); 20, 20A~20E…Control device (image processing module); 50…X-ray irradiator (radiation source); 60…Belt conveyor (transfer device); 72…Pixel; 74…Pixel line (pixel group); 73…Readout circuit; 2 01, 201C…Input Unit; 202, 202A, 202C, 202D…Calculation Unit; 202E…Determination Unit; 203C, 203D…Filtering Unit; 204, 204A, 204B…Noise Map Generation Unit; 204C, 204E…Selection Unit; 205, 205C, 205E…Processing Unit; 206C, 206E, 207…Learned Model; F…Object; TD…Transfer Direction (one direction).

Claims

1. A radiation image taking apparatus, wherein provided with: a photographing device that scans and photographs a radiation that has passed through an object in one direction to take a radiation image; a scintillator that is provided on the photographing device to convert the radiation into light; and an image processing module that inputs the radiation image into a learned model that is constructed using machine learning with image data in advance, and performs noise removal processing that removes noise from the radiation image, the learned model is constructed using machine learning with image data that is obtained by adding a noise value to a pixel value of a radiation image of a structure that is obtained through simulation calculation, the noise value being a noise value along a normal distribution based on a σ value of a standard deviation of a pixel value of one pixel of the radiation image of the structure, the photographing device includes: a detection element that is configured by pixel lines having M pixels arranged in the one direction, and N columns arranged in a direction orthogonal to the one direction, and outputs a detection signal related to the light for each of the pixels, where M is an integer of 2 or more, and N is an integer of 2 or more; and a readout circuit that adds the detection signals output from at least two of the M pixels for each of the pixel lines of the N columns of the detection element, and sequentially outputs the added N detection signals to thereby output the radiation image.

2. The radiation image taking apparatus according to claim 1, wherein the image processing module has: a noise map generation section that derives an evaluation value that represents a pixel value and an evaluation noise value from a pixel value of each pixel of the radiation image based on relationship data that represents a relationship of the evaluation value and the evaluation noise value, and generates a noise map that is data in which the derived evaluation value and each pixel of the radiation image are made to correspond to each other; and a processing section that inputs the radiation image and the noise map into the learned model, and performs noise removal processing that removes noise from the radiation image.

3. The radiation image taking apparatus according to claim 1, wherein the image processing module has: an input section that receives input of a condition of a source of the radiation that irradiates an object and photographs the object, or condition information that represents an arbitrary photographing condition; a calculation section that calculates an average energy related to the radiation that has passed through the object based on the condition information; and a selection section that selects a learned model for the noise removal processing from a plurality of learned models that are respectively constructed using machine learning with image data in advance, based on the average energy.

4. The radiation image taking apparatus according to claim 1, wherein the image processing module has: a determination section that determines an image characteristic of a radiation image that is taken by the photographing device with a jig as an object; a selection section that selects a learned model from a plurality of learned models that are respectively constructed using machine learning with image data in advance, based on the image characteristic; and a processing section that performs the noise removal processing using the selected learned model. ​ 5. A radiation image taking system comprising: the radiation image taking apparatus according to any one of claims 1 to 4; a radiation generating source that irradiates radiation toward the object; and a conveyance apparatus that conveys the object toward the one direction with respect to the taking apparatus.

6. A radiation image taking method comprising: a step of taking a radiation image by scanning and taking scintillating light corresponding to radiation that has passed through an object in one direction; and a step of inputting the radiation image into a learned model constructed by machine learning using image data in advance, and performing noise removal processing for removing noise from the radiation image, the learned model being constructed by machine learning using image data obtained by adding a noise value to a pixel value of each pixel of a radiation image of a structure obtained by simulation calculation, the noise value being a noise value following a normal distribution along a σ value based on a standard deviation of a pixel value of one pixel of the radiation image of the structure, in the taking step, The detection elements are used to add detection signals output from at least two of the M pixels for each pixel line of N columns of the detection elements, sequentially output the added N detection signals, and thereby output the radiation image, wherein the detection element is configured by arranging the pixel lines having the M pixels arranged in the one direction in N columns in a direction orthogonal to the one direction, and outputs the detection signal related to the scintillating light for each of the pixels, M being an integer of 2 or more, and N being an integer of 2 or more.

7. The radiation image taking method according to claim 6, wherein the learned model is constructed by machine learning using image data obtained by adding a noise value following a normal distribution along a σ value based on a standard deviation of a pixel value of one pixel of the radiation image of the structure as teaching data.

8. The radiation image taking method according to claim 6, wherein in the performing step, an evaluation value representing a relationship between a pixel value and an evaluation noise value is derived from a pixel value of each pixel of the radiation image based on relationship data representing a relationship between a pixel value and an evaluation noise value, a noise map is generated, and the noise removal processing for removing noise from the radiation image is performed by inputting the radiation image and the noise map into the learned model, the noise map being data in which the derived evaluation value is associated with each pixel of the radiation image.

9. The radiation image taking method according to claim 7, wherein in the performing step, an evaluation value representing a relationship between a pixel value and an evaluation noise value is derived from a pixel value of each pixel of the radiation image based on relationship data representing a relationship between a pixel value and an evaluation noise value, a noise map is generated, and the noise removal processing for removing noise from the radiation image is performed by inputting the radiation image and the noise map into the learned model, the noise map being data in which the derived evaluation value is associated with each pixel of the radiation image.

10. The radiation image taking method according to claim 6, wherein In the execution step, input of a condition of a generation source of a radiation at the time of irradiation of the radiation and imaging of an object, or condition information indicating an arbitrary imaging condition is received, based on the condition information, an average energy related to the radiation transmitted through the object is calculated, and based on the average energy, a learning completed model used for the noise removal processing is selected from among a plurality of learning completed models respectively constructed in advance using image data through machine learning.

11. The radiation image taking method according to claim 7, wherein, In the execution step, input of a condition of a generation source of a radiation at the time of irradiation of the radiation and imaging of an object, or condition information indicating an arbitrary imaging condition is received, based on the condition information, an average energy related to the radiation transmitted through the object is calculated, and based on the average energy, a learning completed model used for the noise removal processing is selected from among a plurality of learning completed models respectively constructed in advance using image data through machine learning.

12. The radiation image taking method according to claim 6, wherein, In the execution step, an image characteristic of a radiation image taken with the jig as an object is determined, based on the image characteristic, a learning completed model is selected from among a plurality of learning completed models respectively constructed in advance using image data through machine learning, and the noise removal processing is executed using the selected learning completed model.

13. The radiation image taking method according to claim 7, wherein, In the execution step, an image characteristic of a radiation image taken with the jig as an object is determined, based on the image characteristic, a learning completed model is selected from among a plurality of learning completed models respectively constructed in advance using image data through machine learning, and the noise removal processing is executed using the selected learning completed model.

14. The radiation image capturing method according to any one of claims 6 to 13, wherein Further provided are: a step of irradiating a radiation toward the object; and a step of conveying the object toward the one direction with respect to the detection element. a step of irradiating a radiation toward the object; and a step of conveying the object toward the one direction with respect to the detection element.

Images