JP2022098590A - Ai model generation method and inspection device - Google Patents

Abstract

To provide a generation method of an AI (artificial intelligence) model and an inspection device which corrects an image by the AI model where the accuracy of inspection by correcting noise removal and resolution enhancement may be improved by the AI model generated by learning using AI.SOLUTION: In a generation method of an AI model in an inspection device 1 which corrects an image of an inspection object using the AI model generated by machine learning, a high-quality image of the inspection object picked up by the inspection device 1 is used as a teacher image, a low-quality image of the inspection object picked up by the inspection device 1 and sharing the same visual field as the teacher image is used as a learning image, the learning image is used as an input, an image corrected by the AI model is used as an output, and the output is compared with the teacher image to generate an AI model.SELECTED DRAWING: Figure 6

Description

The present invention relates to a method for creating an AI model and an inspection device for correcting an image by the AI model.

The inspection device for measuring the solder shape on the front surface and the back surface of the substrate is inspected using an image of the object to be inspected (see, for example, Patent Document 1 or 2). For example, in an automatic optical inspection (AOI) type inspection device, an object to be inspected is irradiated with illumination light and imaged by a camera. At this time, in order to improve the throughput of the production line, the imaging conditions may be changed (the imaging range (field of view) may be widened or the exposure time may be shortened). Further, in the tomosynthesis type X-ray inspection device, a plurality of transmitted images are captured by changing the relative positions of the radiation source (radiation generator) and the inspected object and the detector, and a reconstructed image is obtained from these transmitted images. It is configured to generate. Therefore, it takes time to capture a plurality of transparent images and generate a reconstructed image using the transmitted images, the throughput of the production line decreases, and in order to increase the throughput, only a part of the inspected object can be inspected. There is. Further, although it depends on the composition such as the material and shape of the object to be inspected, there is also a request to reduce the irradiation dose (exposure dose) of radiation at the time of inspection as much as possible.

Japanese Unexamined Patent Publication No. 2012-053015 Japanese Unexamined Patent Publication No. 2008-026334

When the imaging conditions are changed in the inspection device as described above, there is a problem that the noise of the image captured is increased and a problem that the resolution is lowered. For example, the radiation emitted by the X-ray inspection apparatus is weakened, or the exposure time (radiation irradiation time) is shortened, and the field of view (the imaging region on the object to be inspected, hereinafter referred to as "FOV (Field Of View)"". By widening the (called) or reducing the number of transmitted images used to generate the reconstructed image, the imaging time can be shortened, thereby reducing the exposure dose at the time of inspection and shortening the inspection time. As a result, the noise of the cross-sectional image obtained from the transparent image or the reconstructed image increases, or the resolution decreases.

The present invention has been made in view of such a problem, and is inspected by correcting the noise and the decrease in resolution that increase due to the change of the imaging condition by the AI model created by learning using AI (artificial intelligence). It is an object of the present invention to provide a method for creating an AI model and an inspection device for correcting an image by the AI model, which improves the accuracy of the AI model.

In order to solve the above-mentioned problems, the method of creating an AI model according to the present invention is a method of creating an AI model in an inspection device that corrects an image of an inspected object by using an AI model created by machine learning. The high-quality image of the inspected object captured by the inspection device is used as a teacher image, and the low-quality image of the inspected object captured by the inspection device is the same field as the teacher's image. The AI model is created by using the image of the above as a learning image, the learning image as an input, the image corrected by the AI model as an output, and learning by comparing the output with the teacher's image. do.

Further, in the method for creating an AI model according to the present invention, when an image of the subject to be inspected is divided into a plurality of imaging regions and acquired, in at least two imaging regions of the plurality of imaging regions, It is desirable to create the AI model specialized for each of the imaging regions.

Further, in the method for creating an AI model according to the present invention, when an image of the subject to be inspected is divided into a plurality of regions and acquired, the AI model is said to be in at least two of the plurality of imaging regions. It is desirable to create the AI model common to the imaging region.

Further, in the method for creating an AI model according to the present invention, the image of the object to be inspected is divided into a plurality of partial regions, and at least two of the plurality of partial regions are characterized by each of the partial regions. It is desirable to create the above-mentioned AI model.

Further, in the method for creating an AI model according to the present invention, it is desirable to create an AI model specialized for an image of a specific region of the inspected body among the images of the inspected body.

Further, in the method for creating an AI model according to the present invention, it is desirable that the image quality is a noise amount or a resolution.

Further, the inspection device according to the present invention detects a light source, a holding portion for holding the inspected object, and light emitted from the light source and reflected by the inspected object or transmitted through the inspected object. It has a detector for acquiring an image of the inspected object and a control unit, and the control unit irradiates the inspected object with light from the light source and uses the detector to obtain an image of the inspected object. The teacher image and the learning image are acquired, and the transmission image is corrected by the method for creating an AI model according to any one of claims 1 to 6 using the teacher image and the learning image. Create the AI model for this purpose.

Further, in the inspection device according to the present invention, the light source is a radiation source that emits radiation, and the detector detects radiation from the radiation source that has passed through the inspected object and detects the radiation of the inspected object. The control unit acquires an image, changes the relative positions of the radiation source, the holding unit, and the detector, and acquires the light emitted from the radiation source and transmitted through the inspected object by the detector. By doing so, the teacher image and the learning image are acquired as the transmission image of the inspected object, and the transmission image is corrected by the AI model creation method described above using the teacher image and the learning image. It is desirable to create the AI model for this purpose.

Further, in the inspection apparatus according to the present invention, the light source is a radiation source that emits radiation, and the detector detects radiation from the radiation source that has passed through the inspected body and images of the inspection body. The control unit changes the relative positions of the radiation source, the holding unit, and the detector, and acquires the light emitted from the radiation source and transmitted through the inspected body by the detector. Thereby, the transmission image of the inspected object is acquired, the transmission image is reconstructed to generate the teacher image and the learning image which are cross-sectional images of the inspected object, and the teacher image and the said. It is desirable to create the AI model for correcting the cross-sectional image by the method for creating the AI model according to any one of claims 1 to 6 using the learning image.

Further, in the inspection device according to the present invention, the control unit changes the relative positions of the radiation source, the holding unit, and the detector to emit light emitted from the radiation source and transmitted through the inspected object. It is desirable to acquire a transmission image of the inspected object by acquiring it with the detector, and to correct the acquired transmission image or the cross-sectional image generated from the transmission image with the AI model.

Further, in the inspection device according to the present invention, it is desirable that the correction is performed on the image of the specified region of the inspected object.

Further, it is desirable that the inspection device according to the present invention has a user interface, and the control unit accepts the learning conditions by the user interface.

According to the AI model creation method and the inspection apparatus according to the present invention, even when the imaging conditions are changed (for example, the radiation is weakened or the imaging time is shortened in the X-ray inspection apparatus), there is little noise and the resolution is high. High images (transparent images and cross-sectional images) can be obtained, and inspection accuracy can be improved.

It is explanatory drawing for demonstrating the structure of the inspection apparatus which concerns on embodiment. It is explanatory drawing for demonstrating each functional block processed by the control part of the said inspection apparatus. It is a flowchart for demonstrating the flow of inspection. It is a flowchart for demonstrating the flow of the process of taking a transmission image and generating a reconstructed image. It is explanatory drawing for demonstrating the movement of a substrate holding part and a detector, the radiation of X-rays from a radiation generator, and the timing of an image pickup by a detector, (a) shows a timing chart, (b) is The timing of exposure is shown. It is a flowchart of the learning process of AI model. It is explanatory drawing for demonstrating the division of an image in learning and correction. It is a graph which shows the relationship between the number of epochs and loss in learning. It is explanatory drawing for demonstrating the relationship between the input image (learning image), the output image and the teacher image in learning. It is a flowchart of a transparent image shooting / reconstruction image generation process executed at the time of inspection.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. Although the description will be given here based on the X-ray inspection device, the same applies to other types of inspection devices such as the AOI method, as will be described later. As shown in FIG. 1, the inspection device 1 according to the present embodiment includes a control unit 10, a monitor 12, and an image pickup unit 32, which are composed of a processing device such as a personal computer (PC). .. Further, the imaging unit 32 further includes a radiation quality changing unit 14, a radiation generator driving unit 16, a substrate holding unit driving unit 18, a detector driving unit 20, a radiation generator 22, a substrate holding unit 24, and a detector 26. have.

The radiation generator 22 is a device (radioactive source) that generates radiation such as X-rays, and generates radiation by colliding accelerated electrons with a target such as tungsten or diamond. The radiation in the present embodiment will be described with respect to the case of X-rays, but the radiation is not limited thereto. For example, the radiation may be alpha rays, beta rays, gamma rays, ultraviolet rays, visible light, or infrared rays. Further, the radiation may be a microwave or a terahertz wave. The radiation generator 22 has a function as a light source (radioactive source) for irradiating the object to be inspected with light (radiation).

The substrate holding portion 24 holds the substrate to be inspected. The substrate held by the substrate holding portion 24 is irradiated with the radiation generated by the radiation generator 22, and the radiation transmitted through the substrate is captured as an image by the detector 26. Hereinafter, the radiation transmission image of the substrate captured by the detector 26 is referred to as a “transmission image”. As will be described later, in the present embodiment, the substrate holding portion 24 holding the substrate and the detector 26 are relatively moved with respect to the radiation generator 22 to acquire a plurality of transmitted images, and the reconstructed image is obtained. To generate.

The transmitted image captured by the detector 26 is sent to the control unit 10, and for example, using a known technique such as a filter-corrected back projection method (Filtered-Back Projection method (FBP method)), the solder of the joint portion is soldered. It is reconstructed into an image containing a three-dimensional shape. Then, the reconstructed image and the transparent image are stored in the storage in the control unit 10 and the external storage (not shown). Hereinafter, an image reconstructed into a three-dimensional image including the three-dimensional shape of the solder of the joint portion based on the transparent image is referred to as a “reconstructed image”. Further, an image obtained by cutting out an arbitrary cross section from the reconstructed image is called a "cross-section image". Such a reconstructed image and a cross-sectional image are output to the monitor 12. The monitor 12 displays not only a reconstructed image and a cross-sectional image, but also an inspection result of a solder joint state, which will be described later. Further, the reconstructed image in the present embodiment is also referred to as "planar CT" because it is reconstructed from the planar image captured by the detector 26 as described above.

The radiation quality changing unit 14 changes the radiation quality of the radiation generated by the radiation generator 22. The radiation quality is determined by the voltage applied to accelerate the electrons that collide with the target (hereinafter referred to as "tube voltage") and the current that determines the number of electrons (hereinafter referred to as "tube current"). The line quality changing unit 14 is a device that controls these tube voltages and tube currents. This radiation quality changing unit 14 can be realized by using a known technique such as a transformer or a rectifier.

Here, the radiation quality is determined by the brightness and hardness (spectral distribution of radiation) of the radiation. Increasing the tube current increases the number of electrons that collide with the target and the number of photons of the generated radiation. As a result, the brightness of the radiation increases. For example, some parts such as capacitors are thicker than other parts, and it is necessary to irradiate them with high-intensity radiation in order to capture a transmitted image of these parts. In such a case, the brightness of the radiation is adjusted by adjusting the tube current. Further, when the tube voltage is increased, the energy of the electrons colliding with the target becomes large, and the energy (spectral) of the generated radiation becomes large. In general, the greater the energy of radiation, the greater the penetrating force of a substance and the less likely it is to be absorbed by the substance. A transmitted image captured using such radiation has a low contrast. Therefore, the tube voltage can be used to adjust the contrast of the transmitted image.

The radiation generator drive unit 16 has a drive mechanism such as a motor (not shown), and moves the radiation generator 22 up and down along an axis passing through its focal point (the direction of this axis is defined as the "Z-axis direction"). Can be moved. As a result, the irradiation field can be changed by changing the distance between the radiation generator 22 and the object to be inspected (board) held by the substrate holding portion 24, and the magnification of the transmitted image captured by the detector 26 can be changed. It will be possible. The position of the radiation generator 22 in the Z-axis direction is detected by the generator position detection unit 23 and output to the control unit 10.

The detector drive unit 20 also has a drive mechanism such as a motor (not shown), and rotates and moves the detector 26 along the detector rotation trajectory 30. Further, the board holding section drive unit 18 also has a drive mechanism such as a motor (not shown), and the board holding section 24 is translated on a plane provided with the board rotation track 28. Further, the substrate holding portion 24 is configured to rotate and move on the substrate rotation track 28 in conjunction with the rotational movement of the detector 26. This makes it possible to capture a plurality of transmitted images having different projection directions and angles while changing the relative positional relationship between the substrate held by the substrate holding unit 24 and the radiation generator 22.

Here, the radius of gyration between the substrate rotation track 28 and the detector rotation track 30 is not fixed, but can be freely changed. This makes it possible to arbitrarily change the irradiation angle of the radiation irradiating the components arranged on the substrate. The orbital planes of the substrate rotating orbital 28 and the detector rotating orbital 30 are orthogonal to the Z-axis direction described above, and if the directions orthogonal to the orbital planes are the X-axis direction and the Y-axis direction, the substrate holding portion 24 The positions in the X-axis direction and the Y-axis direction are detected by the board position detection unit 29 and output to the control unit 10, and the positions of the detector 26 in the X-axis direction and the Y-axis direction are detected by the detector position detection unit 31. It is detected and output to the control unit 10.

The control unit 10 controls all the operations of the inspection device 1 described above. Hereinafter, various functions of the control unit 10 will be described with reference to FIG. Although not shown, an input device such as a keyboard and a mouse is connected to the control unit 10 in addition to the monitor 12 described above, and these monitors, the keyboard, the mouse, and the like are the users of the inspection device 1. It constitutes an interface.

The control unit 10 includes a storage unit 34, an image pickup processing unit 35, a cross-section image generation unit 36, a learning processing unit 37, a substrate inspection surface detection unit 38, a pseudo cross-section image generation unit 40, and an inspection unit 42. Although not shown, the control unit 10 also includes an image pickup control unit that controls the operation of the radiation quality change unit 14, the radiation generator drive unit 16, the substrate holding unit drive unit 18, and the detector drive unit 20. Further, each of these functional blocks is realized by the cooperation of a CPU that executes various arithmetic processes, hardware such as RAM used as a work area for storing data and executing a program, and software. Therefore, these functional blocks can be realized in various forms by a combination of hardware and software.

The storage unit 34 stores information such as imaging conditions for capturing a transmitted image of the substrate and the design of the substrate to be inspected. The storage unit 34 also stores a transparent image of the substrate, a reconstructed image (cross-sectional image, pseudo cross-sectional image), an inspection result of the inspection unit 42 described later, and the like. Further, although the details will be described later, the storage unit 34 stores the AI model created or updated by the learning processing unit 37. The storage unit 34 further has a speed at which the radiation generator driving unit 16 drives the radiation generator 22, a speed at which the substrate holding unit driving unit 18 drives the substrate holding unit 24, and a detector driving unit 20 driving the detector 26. The speed is also stored.

The image pickup processing unit 35 drives the radiation generator 22, the substrate holding unit 24, and the detector 26 by the radiation generator driving unit 16, the substrate holding unit driving unit 18, and the detector driving unit 20, and is driven by the substrate holding unit 24. A transmitted image of the held object to be inspected is imaged, and a reconstructed image is generated from the transmitted image. The method of capturing a transmission image and generating a reconstructed image by the imaging processing unit 35 will be described later.

The cross-sectional image generation unit 36 generates a cross-sectional image based on a plurality of transparent images acquired from the storage unit 34. This can be achieved by using known techniques such as the FBP method and the maximum likelihood estimation method. Different reconstruction algorithms have different properties of the resulting reconstructed image and the time required for reconstruction. Therefore, a plurality of reconstruction algorithms and parameters used for the algorithms may be prepared in advance so that the user can select them. As a result, it is possible to provide the user with a degree of freedom in selection, such as giving priority to shortening the time required for reconstruction, or giving priority to good image quality even if it takes time. The generated cross-sectional image is output to the storage unit 34 and recorded in the storage unit 34. The image pickup processing unit 35 described above can also correct a transmission image or a cross-sectional image by using the AI model stored in the storage unit 34. The method of correcting the transparent image or the cross-sectional image will be described later.

The substrate inspection surface detection unit 38 determines a position (cross-sectional image) on the substrate on which the surface to be inspected (for example, the surface of the substrate) is projected from among the plurality of cross-sectional images generated by the cross-sectional image generation unit 36. Identify. Hereinafter, the cross-sectional image showing the inspection surface of the substrate is referred to as an "inspection surface image". Details of the inspection surface image detection method will be described later.

The pseudo cross-sectional image generation unit 40 images a region of the substrate thicker than the cross-sectional image by stacking a predetermined number of continuous cross-sectional images with respect to the cross-sectional image generated by the cross-sectional image generation unit 36. The number of cross-sectional images to be stacked is determined by the thickness of the area of the substrate on which the cross-sectional images are projected (hereinafter referred to as "slice thickness") and the slice thickness of the pseudo cross-sectional images. For example, if the slice thickness of the cross-sectional image is 50 μm and the height (for example, 500 μm) of the BGA solder ball (hereinafter simply referred to as “solder”) as the pseudo cross-sectional image is to be the slice thickness, 500/50 = 10 All you have to do is stack the cross-sectional images. At this time, in order to specify the position of the solder, the inspection surface image specified by the substrate inspection surface detection unit 38 is used.

The inspection unit 42 is based on the cross-sectional image generated by the cross-sectional image generation unit 36, the inspection surface image specified by the substrate inspection surface detection unit 38, and the pseudo cross-sectional image generated by the pseudo cross-sectional image generation unit 40. To inspect. Since the solder that joins the board and the components is near the board inspection surface, the solder can be attached to the board by inspecting the inspection surface image and the cross-sectional image that reflects the area on the radiation generator 22 side with respect to the inspection surface image. It can be determined whether or not the parts are properly joined.

Here, the "soldering state" refers to whether or not the substrate and the component are joined by soldering to generate an appropriate conductive path. Inspection of the solder joint condition includes bridge inspection, melt condition inspection, and void inspection. "Bridge" refers to an unfavorable conductive path between conductors created by joining solder. Further, the "melted state" refers to a state of whether or not the bonding between the substrate and the component is insufficient due to insufficient melting of the solder, that is, a state of whether or not it is "floating". "Void" refers to a defect in the solder joint due to air bubbles in the solder joint. Therefore, the inspection unit 42 includes a bridge inspection unit 44, a melting state inspection unit 46, and a void inspection unit 48.

The details of the operations of the bridge inspection unit 44, the melting state inspection unit 46, and the void inspection unit 48 will be described later, but the bridge inspection unit 44 and the void inspection unit 48 are based on the pseudo-section image generated by the pseudo-section image generation unit 40. The bridge and voids are inspected, respectively, and the melt state inspection unit 46 inspects the melt state of the solder based on the inspection surface image specified by the substrate inspection surface detection unit 38. The inspection results of the bridge inspection unit 44, the melting state inspection unit 46, and the void inspection unit 48 are recorded in the storage unit 34.

FIG. 3 is a flowchart showing a flow from capturing a transmission image, generating a reconstructed image, specifying an inspection surface image, and inspecting the solder joint state. Further, FIG. 4 is a flowchart showing the flow of the processing portion of the imaging of the transparent image and the generation of the reconstructed image. The process in this flowchart starts, for example, when the control unit 10 receives an instruction to start inspection from an input device (not shown).

As shown in FIG. 3, the control unit 10 emits radiation emitted by the radiation generator 22 by the radiation generator drive unit 16 when the inspection object (inspected object) is carried into the inspection device (step S100). The irradiation field is set, the substrate holding unit 24 is moved by the substrate holding unit driving unit 18, and the detector 26 is moved by the detector driving unit 20 to change the imaging position, while the radiation quality changing unit 14 emits radiation. The radiation quality of the generator 22 is set, radiation is applied to the substrate to capture a transmitted image, and further, from the plurality of transmitted images captured in this way, the cross-sectional image generation unit 36 and the pseudo-cross-sectional image generation unit 36 and the pseudo-cross-sectional image generation unit. A reconstructed image (cross-sectional image / pseudo-cross-sectional image) is generated by 40 (step S120). The moving path of the substrate holding unit 24 by the substrate holding unit driving unit 18 and the moving path of the detector 26 by the detector driving unit 20 when capturing a transmitted image are information stored in the storage unit 34. It is assumed that the board holding drive unit 18 and the detector drive unit 20 are set in advance by the method of reading or the method of inputting from the input device. Further, it is assumed that the position of the radiation generator 22 in the Z-axis direction is also preset by the same method.

The details of the process of this step S120 will be described with reference to FIGS. 4 and 5. As shown in FIG. 4, when the step S120 is started, the image pickup processing unit 35 of the control unit 10 turns on the operation signal output to the substrate holding unit drive unit 18 and the detector drive unit 20 (step S1000). ). It corresponds to the time t0 in FIG. 5 (a). When this operation signal is turned on, the board holding unit drive 18 starts the movement of the board holding unit 24 (step S1002), and the detector drive unit 20 starts the movement of the detector 26 (step S1004). The substrate holding unit 24 and the detector 26 are moved along a preset movement path as described above.

When the image pickup processing unit 35 determines whether or not it is the image pickup timing (step S1006) and determines that it is not the image pickup timing (“N” in step S1006), the image pickup processing unit 35 repeats this step again after a predetermined time, and at the image pickup timing. When it is determined that there is (“Y” in step S1006), the image pickup start signal (trigger) is transmitted to the detector 26 (step S1008). For example, in the example of FIG. 5A, the trigger for the detector 26 is turned on at time t1.

The detector 26 that has detected that the trigger has been turned on by the image pickup processing unit 35 starts imaging of the transmitted image and transmits a response signal indicating that the imaging has started to the image pickup processing unit 35 (step S1010). Further, the detector 26 transmits an exposure signal to the radiation generator driving unit 16 (step S1012). For example, in the example of FIG. 5A, the exposure signal output to the radiation generator driving unit 16 is turned on from the time t2 to the time T. In this way, if the detector 26 is configured to transmit the exposure signal to the radiation generator driving unit 16, the delay from the start of imaging to the start of exposure can be reduced as much as possible.

The radiation generator driving unit 16 that has received the exposure signal from the detector 26 generates radiation from the radiation generator 22 while the exposure signal is on, and this radiation is applied to the object to be inspected (step S1014). Here, when the detector 26 adopts the rolling shutter method, the X-ray information (intensity, etc.) detected by the light receiving element of the detector 26 is along a plurality of scanning lines arranged in a predetermined direction. It is acquired, but the start time is shifted for each scanning line. For example, as shown in FIG. 5B, when the detector 26 is composed of n scanning lines extending in the left-right direction, L1, L2, L3, ..., Ln-1, Ln from above. The information detected with the start time shifted in the order of is acquired. Therefore, the information obtained from each scanning line is obtained by generating X-rays from the radiation generator 22 during the time when all the scanning lines are acquiring data (during time T in the case of FIG. 5B). Since the information is obtained by X-rays irradiated at the same time, it is possible to prevent distortion of the acquired transmitted image.

Further, the image pickup processing unit 35 that has received the response signal transmitted from the detector 26 acquires the position information of the board holding unit 24 from the board position detection unit 29, and determines the position of the detector 26 from the detector position detection unit 31. Acquire and store (step S1016). Since the movement of the board holding unit 24 by the board holding drive unit 18 and the movement of the detector 26 by the detector drive unit 20 are controlled along a predetermined movement path as described above, the board holding unit If either the position of 24 or the position of the detector 26 is known, the position of the other can be known. Therefore, both the positions of the substrate holding portion 24 and the detector 26 may be stored, or one of the positions may be stored. You may. Further, the positions of the substrate holding portion 24 and the detector 26 may be stored in the above-mentioned XY Cartesian coordinate system (in the form of positions (x, y) in the X-axis direction and the Y-axis direction), or the substrate rotation trajectory 28. And may be stored in a polar coordinate system (a form of a position (r, θ) specified by a distance r from the origin and an angle θ) with the center of the orbital surface of the detector rotation orbit 30 as the origin.

When the imaging of the transmitted image is completed as described above, the detector 26 transmits the captured transmitted image to the imaging processing unit 35 (step S1018). Then, the image pickup processing unit 35 that has acquired this transparent image stores the position information of the substrate holding unit 24 acquired in step S1016 and the position information of the detector 26 in association with the acquired transparent image in the storage unit 34 (. Step S1020).

Further, when the image pickup processing unit 35 determines whether or not there is a next image pickup position (step S1022) and determines that there is a next image pickup position (“Y” in step S1022), the process returns to step S1006 and described above. The processing (steps S1006 to S1020) is repeated. On the other hand, when the image pickup processing unit 35 determines that there is no next image pickup position (“N” in step S1022), the image pickup processing unit 35 turns off the operation signals output to the substrate holding unit drive unit 18 and the detector drive unit 20 (step). S1024), the board holding unit drive unit 18 that has detected that the operation signal has been turned off stops the movement of the board holding unit 24 (step S1026), and the detector drive unit 20 stops the movement of the detector 26 (step S1024). Step S1028). For example, it corresponds to the time t3 in FIG. 5 (a).

Finally, the image pickup processing unit 35 generates a reconstructed image from the transmitted image stored in the storage unit 34 by the cross-section image generation unit 36 and the pseudo cross-section image generation unit 40 (step S1030). The generated reconstructed image may be stored in the storage unit 34.

Next, returning to FIG. 3, the control unit 10 automatically inspects the inspected object based on the preset information (step S140). Specifically, the substrate inspection surface detection unit 38 of the control unit 10 receives a transmission image or a reconstructed image (cross-section image) from the cross-section image generation unit 36, and identifies the inspection surface image from the transmitted image (cross-section image) (step S141). The bridge inspection unit 44 acquires a pseudo cross-section image having a slice thickness similar to that of the solder ball projecting the solder ball from the pseudo cross-section image generation unit 40, and inspects the presence or absence of the bridge (step S142). When the bridge is not detected (“N” in step S143), the melting state inspection unit 46 acquires an inspection surface image from the substrate inspection surface detection unit 38 and inspects whether or not the solder is melted (step S144). ). When the solder is melted (“Y” in step S145), the void inspection unit 48 acquires a pseudo cross-section image partially projecting the solder ball from the pseudo cross-section image generation unit 40, and the void is present. It is inspected whether or not (step S146). If the void is not found (“N” in step S147), the void inspection unit 48 determines that the solder bonding state is normal (step S148), and outputs that fact to the storage unit 34. When a bridge is detected (“Y” in step S143), the solder is not melted (“N” in step S145), or a void is present (“Y” in step S147), respectively. The bridge inspection unit 44, the melting state inspection unit 46, and the void inspection unit 48 determine that the solder joint state is abnormal (step S149), and output that fact to the storage unit 34. When the solder state is output to the storage unit 34, the automatic inspection process ends.

Finally, the control unit 10 outputs the inspection result to the monitor 12 or the like (step S160), carries out the inspection target (inspected object) (step S180), and ends the inspection according to this flowchart (or next). The inspection by the above-mentioned treatment of the inspected object is started).

According to the above method, the position where the transmitted image is captured is not the information of the time when the imaging processing unit 35 transmits the trigger to the detector 20, but the time when the detector 26 starts acquiring the image (from the detector 26). It is the information of the time when the response signal was received). In the case where the relative positions of the radiation generator 22 and the substrate holding unit 24 and the detector 26 are changed (the substrate holding unit 24 and the detector 26 continue to move), the image pickup processing unit 35 triggers. Since there is a delay from the transmission to the start of image acquisition by the detector 26, the positions of the board holding unit 24 and the detector 26 at the time when the trigger is transmitted are the positions where the transmitted image is actually captured. It may be out of alignment. Therefore, as described above, when the detector 26 starts acquiring an image and the image pickup processing unit 35 receives the response signal transmitted from the detector 26 at that time, the substrate holding unit 24 and the detector 26 By acquiring the position, accurate position information can be acquired, and thereby the accuracy of the reconstructed image can be improved. Further, the moving path of the board holding portion 24 by the board holding unit driving unit 18 and the moving path of the detector 26 by the detector driving unit 20 may deviate from the positions specified in advance depending on the characteristics of the driving unit and the like. As described above, since these positions are the positions detected by the board position detection unit 29 and the detector position detection unit 31, accurate position information can be obtained, and the accuracy of the reconstructed image is further configured. Can be made to.

The position information and the transmitted image of the substrate holding unit 24 and the detector 26 are stored by cyclically using a predetermined area of the storage area (memory, hard disk, etc.) of the control unit 10 (predetermined area). The storage area can be used efficiently by adopting a method of storing information sequentially from the beginning of the above, and when the information is stored at the end of the predetermined area, returning to the beginning of the predetermined area and storing the information). ..

Further, when the detector 26 is capturing a transmission image by the rolling shutter method, if the transmission image is acquired while the relative positions of the radiation generator 22 and the substrate holding portion 24 and the detector 26 are changed, the image is obtained. Although it may be distorted, as described above, the on / off of the X-ray emitted from the radiation generator 22 (on / off of the exposure signal) is synchronized with the signal (response signal) of the rolling shutter of the detector 26. This makes it possible to acquire a transmission image without distortion.

As described above, in the inspection device 1 having the above configuration, in order to shorten the inspection time and reduce the amount of radiation irradiated to the inspected object, the imaging time of the inspected object at the time of inspection is shortened. Or, it is necessary to weaken the emitted radiation. In order to shorten the imaging time, it is conceivable to shorten the exposure time (irradiation time of radiation), widen the FOV, or reduce the number of transmitted images used to generate the reconstructed image. However, if the imaging time is shortened or the emitted radiation is weakened by such a method, the noise of the transmitted image or the cross-sectional image cut out from the reconstructed image generated from the transmitted image increases or becomes. The resolution may be low and, as a result, the accuracy of inspections using these images may be reduced.

The inspection device 1 according to the present embodiment is configured to correct using an AI model in order to remove noise from a transmission image or a cross-sectional image and to increase the resolution of these images. That is, an image captured for inspection (transmission image or cross-sectional image) is used as an input image, and by applying an AI model to this input image, an image corrected by AI can be obtained as an output image. The correction method using the AI model will be described below.

The correction method according to the present embodiment is a machine such as deep learning using an image (transmission image or cross-sectional image) of the object to be inspected captured by the inspection device 1 as teacher data and learning data in advance. It is configured to perform learning (referred to as "AI learning" in the following description) to generate an AI model, and in the inspection, the generated AI model is used to correct a transmission image or a cross-sectional image. First, the process of AI learning (deep learning) will be described with reference to FIGS. 6 to 9. In the following description, the "teacher image" is an image of an object to be inspected with a sufficient exposure time and radiation quality (brightness and hardness of radiation) with little noise, or a sufficient resolution. It shows an image of the object to be inspected (resolution that can obtain a predetermined inspection accuracy in the inspection), and the "learning image" is the exposure time, radiation quality, or radiation quality set in the actual inspection. It shall indicate an image of the subject to be inspected, captured at resolution, containing noise, or at low resolution.

Further, in the inspection device 1 according to the present embodiment, the object of correction using the AI model may be a transmission image or a cross-sectional image. Further, there are cases where the correction target reduces noise and cases where the resolution is increased. Here, a case where noise is reduced will be described.

Further, in the inspection device 1 according to the present embodiment, the entire area Ra (for example, an area of about 2000 × 2000 pixels) of the transmission image or the cross-sectional image is divided into a plurality of partial areas Rs (for example, an area of 256 × 256 pixels). However, each partial region Rs is configured to perform learning or correction (using the partial region Rs at the same position of the teacher image and the learning image). Further, in the AI learning process, as shown in FIG. 7, 3/4 of the entire image area Ra is used for learning, and the remaining 1/4 area is verified (performance evaluation). Used for. The arrangement of the verification area and the learning area shown in FIG. 7 is an example, and any partial area Rs can be assigned for verification or learning. Further, the ratio of the verification area and the learning area in one image is not limited to the combination of 3/4 and 1/4, and can be appropriately set.

Further, as already described, in the inspection of the inspected object, a plurality of transmitted images are obtained by changing the relative positions of the radiation generator 22 and the substrate holding portion 24 and the detector 26 on which the inspected object is placed. Since tens to hundreds of transparent images (a set of a teacher's image and a learning image) can be obtained for one FOV, an AI model for correcting the transparent image is generated. In the case, learning is performed using these a plurality of transparent images. In addition, a reconstructed image is generated from these transparent images (a reconstructed image of the teacher image is generated from the transparent image of the teacher image, and a reconstructed image of the learning image is generated from the transparent image of the learning image. ), Hundreds of cross-sectional images (a set of teacher image and learning image) can be obtained from this reconstructed image. Therefore, when generating an AI model for correcting the cross-sectional image, these multiple cross-sectional images are obtained. Learning is done using.

Before starting learning, the object to be inspected is placed on the substrate holding portion 24 of the inspection device 1. In addition, an input device or the like is used to input which part of the object to be learned, that is, the FOV (image area) to be learned. In addition, the image pickup conditions (radio quality, exposure time, magnification, etc.) of the teacher image and the image pickup conditions of the learning image are input by an input device or the like. Further, either a transparent image or a cross-sectional image is input as the target of the output AI model.

Then, when the start of the learning process is instructed by the input device or the like, as shown in FIG. 6, the learning process unit 37 of the control unit 10 arranges the FOV via the image pickup process unit 35, that is, the object to be inspected. The upper imaging area is set (step S200).

When the FOV is arranged in step S200, the learning processing unit 37 captures a teacher's image (transmission image) with the exposure time, radiation quality, magnification, etc. input in advance via the image pickup processing unit 35. It is stored in the storage unit 34 (step S202), and similarly, a learning image (transmission image) is imaged with the exposure time, radiation quality, magnification, etc. input in advance and stored in the storage unit 34 (step). S204). As described above, for one FOV, a plurality of transmitted images are obtained by changing the relative positions of the radiation generator 22 and the substrate holding portion 24 and the detector 26 on which the inspected object is placed. Since the image is taken, the substrate holding unit 24 and the detector 26 are moved, and steps S202 and S204 are executed at a plurality of positions. Further, when performing AI learning for correcting the cross-sectional image, in step S202, a reconstructed image for the teacher is generated from the captured image for the teacher (transparent image), and further, the reconstructed image for the teacher is generated from the reconstructed image. A cross-sectional image is generated and stored in the storage unit 34. Similarly, in step S204, a reconstructed image for learning is generated from the captured learning image (transparent image), and a cross-sectional image for learning is generated from the reconstructed image and stored in the storage unit 34.

Next, the learning processing unit 37 specifies an image set for AI learning and learning parameters (step S206). Here, the learning parameter is information related to the number of times of learning and the network structure. This learning parameter may be stored in the storage unit 34 in advance, or may be input from an input device or the like.

When the AI learning image set or the like is performed in step S206, the learning processing unit 37 determines whether or not to use the existing AI model (step S208). The AI model used for correction may be an AI model specialized for each FOV, or may be the same AI model for a plurality of FOVs.

Here, an AI model specialized for each specific object in the FOV (in the image), for example, a BGA, a chip resistor, an IC lead portion, or the like, or an AI specialized only for the solder joint portion may be used. It may be a model. By specializing only in solder bonding, it is possible to make a specialized high-performance model, and the inspection surface image is specified by the above-mentioned processing, and the radiation generator 22 side with respect to the inspection surface image and the inspection surface image. By inspecting the cross-sectional image showing the area of, the processing time can be shortened by limiting the correction by the AI model to only the area to be inspected. As shown in FIG. 7, since it is divided into a plurality of partial regions Rs, it may be configured to create an AI model specialized for each divided region (partial region Rs). On the other hand, by using the same AI model for a plurality of FOVs, it is possible to obtain a general-purpose AI model that straddles the FOVs.

Therefore, when building an AI model specialized for FOV, create a new AI model without using the existing AI model, and when building a common AI model with multiple FOVs, use the existing AI model. Will be used. Further, when the image of the object to be inspected is divided into a plurality of imaging regions (FOVs) and imaged, an AI model specialized for the FOV is created for some FOVs, which is common to some FOVs. It may be configured to create the AI model. Whether or not the existing AI model can be used may be stored in the storage unit 34 in advance, or may be input from the input device. When the existing AI model is not used (“N” in step S208), the learning processing unit 37 creates a new AI model (step S210), and when the existing AI model is used (step S208). “Y”), the learning processing unit 37 reads an existing AI model from the storage unit 34 (step S212). When the AI model is stored in the storage unit 34, the FOV information corresponding to each AI model is also managed.

Further, the learning processing unit 37 performs the AI learning described above using the teacher image and the learning image (step S214). As described with reference to FIG. 7, AI learning is performed for each of the subregions Rs for learning among the divided subregions Rs. In AI learning, a learning image (which may be a transparent image or a cross-sectional image) is input by a method such as deep learning, and a teacher's image (similar to the learning image, a transparent image and a cross-sectional image). Train the AI model so that the image is close to (may be). Then, the learning processing unit 37 updates the AI model based on the result of AI learning (step S216). Further, the learning processing unit 37 may be configured to display the learning progress status on the monitor 12, and if the learning progress status is displayed, the display is updated (step S218). Progress includes performance evaluation of AI models.

The performance evaluation of the AI model is performed by calculating the loss. The AI model trained by AI learning is applied to each image (input image) for learning and verification, and a corrected image (high-quality output image) is generated. The loss is calculated by taking the difference between the evaluation values of the generated output image and the teacher's image (for example, the mean square error of the luminance value). The loss here is the quality at which the difference between the teacher image and the image corrected by the AI model is lost or impaired (or irrecoverable). Therefore, in expressing the loss, the luminance values may be simply compared as described above, or other than that, the edge portion may be emphasized and compared, or the frequency component may be compared. Therefore, the loss defines a loss function determined according to the above policy and is calculated based on this loss function.

FIG. 8 shows the loss of the image to which the AI model is applied to the learning image (the output image, which is shown as “learning” in FIG. 8) and the loss of the performance evaluation (as “evaluation” in FIG. 8). (Shown) is a graph showing the number of epochs (number of repetitions of learning). If the loss of performance evaluation is large even though the loss of the output image is small (dashed line in FIG. 8), overfitting (a state in which an AI model that is excessively specialized in training data and is not versatile is generated) occurs. It is thought that it is. On the other hand, when the loss of both the output image and the performance evaluation becomes small, it is considered that the learning is proceeding normally. In FIG. 9, (a) is an input image (learning image), and (b) is an output image corrected by applying an AI model to this input image. When the learning is normally performed, the output image becomes an image close to the teacher image of (c), but in the overfitting state, it becomes an image of an abnormal state as shown in (d). The abnormal state is a state in which there is a portion where the luminance value is significantly different, or a state in which the shape of the imaged object is significantly different from the original shape.

From the above, in step S218, when the progress of learning is displayed on the monitor 12 or the like, the loss graph shown in FIG. 8 and the input image, output image, and teacher image shown in FIG. 9 are displayed side by side. , The learning situation can be visually confirmed. As described above, in the loss graph, when the loss of performance evaluation is larger than the loss of the output image, or when the output image and the teacher image are significantly different, it is abnormal learning (over-learning). It can be determined that abnormal learning (wasted time) can be reduced by interrupting learning. Further, the learning time can be shortened by ending the learning when the loss becomes small or the output image becomes close to the teacher's image. In addition, in the display of the progress status of learning, the elapsed time from the start of learning and the estimated remaining time until the end of learning may be displayed.

Returning to FIG. 6, the learning processing unit 37 determines whether or not to perform the next learning (step S220). If there is a transparent image or a cross-sectional image that has not been learned yet, or if the learning of a predetermined number of repetitions has not been completed (“Y” in step S220), the learning processing unit 37 returns to step S216 and performs the above-mentioned processing. repeat. On the other hand, when there is no transparent image or cross-sectional image that has not been learned yet, or when an instruction to end learning is given from an input device or the like (“N” in step S220), the learning processing unit 37 stores the AI model. It is output to 34 (“Y” in step S220), and the AI learning process is terminated. As described above, when it is determined to be overfitting and interrupted (“interruption” in step S220), the learning processing unit 37 ends the AI learning process without outputting the AI model.

The above explanation has described the learning of the AI model for correction to remove noise, but in the case of learning the AI model for correction to increase the resolution, the resolutions of the teacher image and the learning image are changed. It is necessary to obtain it. For example, in step S202 of FIG. 6, when acquiring a teacher's image (transparent image), the FOV at the time of the actual inspection is divided into four, the magnification (magnification) is increased, and the four images are imaged. Is used as a teacher image, and the learning image (transmission image) captured in step S204 is captured by the FOV at the time of the actual inspection. This makes it possible to acquire a high-resolution teacher image and a low-resolution learning image.

Further, reducing the number of transmitted images for generating the reconstructed image is also effective in shortening the imaging time and reducing the exposure dose. If the number of transparent images is reduced, the resolution of the reconstructed image (cross-sectional image) is lowered, which may affect the accuracy of the inspection. Therefore, the reconstructed image (cross-sectional image) can be made high in resolution by generating a pseudo-transparent image from the captured transparent image and increasing the number of transparent images. Here, the pseudo-transparent image can be obtained by generating a transparent image between two adjacent transparent images among the captured transparent images. Such interpolation is called synogram interpolation.

The AI learning process described above can also be applied to interpolation of synograms. When generating an AI model for interpolation of a synogram, a teacher image (transparent image) is acquired in step S202 of FIG. At this time, the number of transparent images for one FOV is set so that the resolution of the reconstructed image (cross-sectional image) generated from those transparent images is the resolution at which a predetermined inspection accuracy can be obtained. Then, in step S204, instead of capturing the learning image (transparent image), a predetermined number of transparent images are thinned out from the teacher image acquired in step S202 to be a learning image. For example, when 150 transparent images are acquired as teacher images in step S202, 75 images are thinned out from the teacher image (transparent image) in step S204, and the remaining 75 images are used as learning images (transparent images). do.

Further, in the AI learning process of steps S214 to S220, an AI model for generating a transparent image between two adjacent transparent images of the learning image is constructed by learning. For example, when interpolating the Nth transparent image, the N-1st transparent image and the N + 1th transparent image are used as input images, and the AI model is applied to the input images to apply the Nth transparent image (pseudo transparent image). To learn to obtain as an output image.

In the above description, whether the correction target of the AI model is a transparent image or a cross-sectional image, noise is removed from the transparent image or the cross-sectional image or the resolution is increased, the synogram is interpolated, or the subject is inspected. Whether to learn the image of a specific area (board surface or specific component), create an AI model for each FOV, create an AI model common to FOV, or create a teacher image and a learning image. The learning conditions such as the image quality and the learning end condition can be set for the control unit 10 from the user interface including the monitor 12 and the keyboard / mouse. In addition, the learning status can be confirmed and the learning process can be completed via this user interface.

When the AI model obtained as described above is applied to an inspection, the AI model described above is applied to the acquired transmission image or reconstructed image (cross-sectional image) in step S120 shown in FIG. 3 for correction. (Correct noise, increase resolution, interpolate synogram, etc.). Hereinafter, the process of correcting the transmission image or the reconstructed image (cross-sectional image) by the AI model will be described with reference to FIG.

FIG. 10A shows a process when the correction by the AI model is applied to the transparent image. The control unit 10 captures a transmitted image of the object to be inspected under the imaging conditions for inspection (step S121a), and corrects the captured transmitted image using the AI model (step S122a). Then, the control unit 10 generates a reconstructed image (cross-sectional image) from the corrected transparent image (step S123a), and ends the transparent image shooting / reconstructed image generation process S120.

Further, FIG. 10B shows a process when the correction by the AI model is applied to the reconstructed image (cross-sectional image). The control unit 10 captures a transmitted image of the object to be inspected under the imaging conditions for inspection (step S121b), and generates a reconstructed image (cross-sectional image) from the captured transmitted image (step S122b). Then, the control unit 10 corrects the generated reconstructed image (cross-sectional image) using the AI model (step S123b), and ends the transmission image capturing / reconstructed image generation process S120.

Further, FIG. 10C shows a process in which the AI model is applied only to the reconstructed image (cross-sectional image) of the specified region in the reconstructed image (cross-sectional image) to correct the image. The control unit 10 captures a transmitted image of the object to be inspected under the imaging conditions for inspection (step S121c), and generates a reconstructed image (cross-sectional image) from the captured transmitted image (step S122c). Then, the control unit 10 identifies the substrate surface (inspection surface image) from the generated reconstructed image (cross-sectional image) by the method described in step S141 (step S123c), and the specified inspection surface image and inspection. Correction using the AI model is performed on the cross-sectional image showing the region on the radiation generator 22 side with respect to the surface image (step S124c), and the transmission image shooting / reconstruction image generation process S120 is completed.

At this time, if an AI model specialized for the FOV is created, the AI model corresponding to the FOV is applied to each FOV and correction is performed. As described above, among the plurality of FOVs, an AI model specialized for the FOV may be applied to some FOVs, and a common AI model may be applied to some FOVs.

As described above, even if the transmitted radiation or the reconstructed image (cross-sectional image) is corrected using the AI model at the time of inspection, the irradiated radiation may be weakened or the imaging time may be shortened. In the subsequent processing of FIG. 3, the inspection can be performed by the reconstructed image (cross-sectional image / pseudo-cross-sectional image) from which noise has been removed or the resolution has been increased, and the inspection accuracy can be improved.

As described above, in the inspection apparatus according to the present embodiment, noise is removed, the resolution is increased, or the synogram is interpolated by the correction using the AI model, so that the radiation dose to be irradiated is weakened or the image is taken. Even if the time is shortened, a high-quality image (transmission image or tomographic image) of the inspected object can be obtained, so that the inspection accuracy can be improved. Further, since the time required for inspection can be shortened, even if the entire area of the inspected object is inspected, the overall throughput is not reduced.

Here, in the learning of the AI model, in the inspection device on which the AI model is mounted, a teacher's image and a learning image are used as an image of an image to be inspected to be inspected. For example, when the object to be inspected is an electronic substrate, various types of IC packages are mounted (SOJ (Small Outline J-leaded), BGA (Ball grid array), etc.). Therefore, according to the above-mentioned method, it is possible to construct an AI model specialized for each inspected object by learning using an image of the inspected object itself to be inspected as a teacher image and a learning image. This can improve the image quality of images (transparent images and cross-sectional images) at the time of inspection, and can be expected to improve inspection accuracy. An AI model is applied to such an image for learning (teacher image and learning image), and the image is imaged by an inspection device in which the image at the time of inspection is corrected using this AI model. It can be expected that the effect will be further improved.

The teacher image and the learning image are acquired by the inspection device 1, learning is performed using the acquired images on a computer different from the control unit 10, and the result (AI model) obtained by learning is used as the control unit. It may be configured to be mounted on the 10 to perform the inspection, or as described above, the control unit 10 may perform the learning and the result (AI model) may be used to perform the inspection.

Further, the above-mentioned AI model learning method and image correction by the generated AI model are performed not only by the X-ray inspection device but also by the AOI type inspection device, specifically, the light source and the detection above the object to be inspected. It can also be applied to an inspection device in which a camera is arranged as a vessel and light from a light source (illumination light having a striped pattern or the like) is applied to an object to be inspected to acquire an image by the camera. In the case of an inspection device having such a configuration, images having different resolutions can be acquired as a teacher image and a learning image by changing the magnification of the zoom lens of the camera or using a pixel shifting method. Further, by averaging the acquired plurality of images and generating one image, images having different noise amounts can be used as a teacher image and a learning image.

1 Inspection device 10 Control unit 22 Radiation generator (light source, radiation source)
24 Board holding part (holding part)
26 detector

Claims (12)

It is a method of creating an AI model in an inspection device that corrects an image of an inspected object using an AI model created by machine learning.
A high-quality image of the subject to be inspected taken by the inspection device is used as a teacher's image.
An image having a low image quality of the subject to be inspected taken by the inspection device and having the same field of view as the teacher's image is used as a learning image.
A method of creating an AI model for creating an AI model by inputting the learning image, using an image corrected by the AI model as an output, and learning by comparing the output with the teacher's image. When the image of the object to be inspected is divided into a plurality of imaging regions and acquired, in at least two imaging regions of the plurality of imaging regions, the AI model specialized for each of the imaging regions is used. The method for creating an AI model according to claim 1. A claim that when an image of an object to be inspected is divided into a plurality of regions and acquired, the AI model common to the imaging regions is created in at least two imaging regions of the plurality of imaging regions. The method for creating an AI model according to 1 or 2. Claims 1 to 3 divide the image of the object to be inspected into a plurality of subregions, and in at least two of the plurality of subregions, create the AI model specialized for each of the subregions. The method for creating an AI model according to any one of the above. The method for creating an AI model according to any one of claims 1 to 4, wherein an AI model specialized for an image of a specific region of the inspected object is created among the images of the inspected object. The method for creating an AI model according to any one of claims 1 to 5, wherein the image quality is a noise amount or a resolution. Light source and
A holding part that holds the object to be inspected,
A detector that acquires an image of the inspected object by detecting light emitted from the light source and reflected by the inspected object or transmitted through the inspected object.
Has a control unit,
The control unit
The light from the light source is applied to the inspected object, and the teacher image and the learning image are acquired as images of the inspected object by the detector.
An inspection device for creating the AI model for correcting the transmitted image by the method for creating an AI model according to any one of claims 1 to 6 using the teacher image and the learning image. The light source is a radiation source that emits radiation.
The detector detects radiation from the radiation source that has passed through the inspected object and acquires an image of the inspected object.
The control unit
By changing the relative positions of the radiation source, the holding portion, and the detector, and acquiring the light emitted from the radiation source and transmitted through the inspected object by the detector, the transmission of the inspected object is obtained. The teacher image and the learning image are acquired as images, and the image is obtained.
An inspection device for creating the AI model for correcting the transmitted image by the method for creating an AI model according to any one of claims 1 to 6 using the teacher image and the learning image. The light source is a radiation source that emits radiation.
The detector detects radiation from the radiation source that has passed through the inspected object and acquires an image of the inspected object.
The control unit
By changing the relative positions of the radiation source, the holding portion, and the detector, and acquiring the light emitted from the radiation source and transmitted through the object to be inspected by the detector, the transmission of the object to be inspected is obtained. Get the image,
The transparent image is reconstructed to generate the teacher image and the learning image which are cross-sectional images of the object to be inspected.
An inspection device for creating an AI model for correcting a cross-sectional image by the method for creating an AI model according to any one of claims 1 to 6 using the teacher image and the learning image. The control unit
By changing the relative positions of the radiation source, the holding portion, and the detector, and acquiring the light emitted from the radiation source and transmitted through the object to be inspected by the detector, the transmission of the object to be inspected is obtained. Get the image,
The inspection device according to claim 8 or 9, wherein the acquired transparent image or the cross-sectional image generated from the transparent image is corrected by the AI model. The inspection device according to any one of claims 7 to 10, wherein the correction is performed on the image of the specified region of the inspected object. Has a user interface
The inspection device according to any one of claims 7 to 11, wherein the control unit receives the learning condition by the user interface.

Images