CN115170461A

CN115170461A - Training data creation support device and training data creation support method

Info

Publication number: CN115170461A
Application number: CN202210304428.XA
Authority: CN
Inventors: 野口威; 冈山敏之
Original assignee: Screen Holdings Co Ltd
Current assignee: Screen Holdings Co Ltd
Priority date: 2021-03-18
Filing date: 2022-03-16
Publication date: 2022-10-11
Also published as: TW202238455A; JP2022144121A; KR20220130577A; TWI837586B

Abstract

The invention provides a training data creation assistance device and a training data creation assistance method capable of easily creating training data for performing examination with high accuracy. The abnormal data acquisition unit acquires abnormal data indicating an image of the inspection object determined to be defective in advance. The normal data acquisition unit acquires normal data indicating an image of a normal inspection object so as to correspond to the abnormal data acquired by the abnormal data acquisition unit. A difference data generation unit generates difference data indicating a difference between the abnormal data acquired by the abnormal data acquisition unit and the normal data acquired by the normal data acquisition unit so as to correspond to the abnormal data. Based on the difference data generated by the difference data generation unit, an image of a portion of the inspection target object is presented by the presentation unit.

Description

Training data creation support device and training data creation support method

Technical Field

The present invention relates to a training data creation assistance device and a training data creation assistance method for assisting creation of training data.

Background

In the case of products such as food and the like, inspection for determining whether or not the products are acceptable is appropriately performed in each step from the middle of production to before distribution. For example, japanese patent application laid-open No. 2011-119471 discloses a defect inspection apparatus for inspecting various defects generated in a manufacturing process of a semiconductor wafer.

In this defect inspection apparatus, a plurality of SEM (scanning electron microscope) images representing a plurality of wafers to be inspected are generated. The user designates, as a template, an SEM image of a wafer indicating a non-defective product from among the plurality of SEM images generated. Defects on the circuit pattern of the wafer shown in the inspection image are detected by calculating the difference between each of a plurality of SEM images (inspection images) other than the template and the template.

Disclosure of Invention

In a defect inspection apparatus, even if an inspection object has no defect, it is sometimes determined that a defect exists. In this case, although there is no defect, the inspection object is discarded because an erroneous inspection result is obtained, and the yield is lowered. Therefore, it is desirable to perform inspection with higher accuracy.

An object of the present invention is to provide a training data creation assistance device and a training data creation assistance method that can easily create training data for performing an inspection with high accuracy.

(1) A training data creation support device according to an aspect of the present invention supports creation of training data used for reexamination of an object to be examined, and includes: an abnormal data acquisition unit that acquires abnormal data representing an image of an inspection object previously determined to be defective; a normal data acquisition unit that acquires normal data indicating an image of a normal inspection target object so as to correspond to the abnormal data acquired by the abnormal data acquisition unit; a difference data generation unit that generates difference data indicating a difference between the abnormal data acquired by the abnormal data acquisition unit and the normal data acquired by the normal data acquisition unit so as to correspond to the abnormal data; and a presentation unit that presents an image of a portion of the inspection target object based on the difference data generated by the difference data generation unit.

In the training data creation support device, an image of a portion of the inspection target is presented based on difference data indicating a difference between abnormal data indicating an image of the inspection target determined to be defective in advance and normal data indicating an image of a normal inspection target. The portion of the inspection target object in the image shown based on the difference data is highly likely to need to be inspected again. Therefore, the user can create training data for rechecking the part by confirming the presented image. This makes it possible to easily create training data for examination with high accuracy.

(2) The training data creation assistance device further has: an accepting unit that accepts selection of the difference data presented by the presenting unit; and a registration unit that registers the difference data selected by the receiving unit as training data. In this case, the difference data selected by the user is registered as training data. Thus, training data can be created more easily.

(3) The receiving unit also receives the correction of the difference data presented by the presenting unit, and the registering unit registers the corrected difference data as training data. In this case, the portion of the inspection target object that needs to be reexamined in the image based on the difference data can be more appropriately corrected. Thus, training data for performing inspection with higher accuracy can be created.

(4) The normal data acquisition unit acquires a plurality of normal data so as to correspond to the abnormal data, and the differential data generation unit generates a plurality of differential data based on the abnormal data and a plurality of normal data corresponding to the abnormal data. In this case, a plurality of differential data are generated based on one abnormal data. This can improve the work efficiency of training data creation.

(5) The normal data acquisition unit acquires a plurality of normal data so as to correspond to the abnormal data, and the differential data generation unit generates differential data based on an average of the abnormal data and the plurality of normal data corresponding to the abnormal data. According to this configuration, even when a noise component unrelated to a defect is accidentally mixed in any of the plurality of normal data, since the plurality of normal data are averaged, the noise component hardly affects the pixel value of the averaged normal data. Therefore, training data for inspection with higher accuracy can be created.

(6) The normal data acquisition unit acquires normal data so as to correspond to a plurality of abnormal data, and the differential data generation unit generates differential data based on each of the abnormal data and the normal data corresponding to the abnormal data. In this case, the training data can be created at high speed using the common normal data.

(7) The normal data acquired by the normal data acquisition unit includes image data representing an image of the inspection object that has not been previously determined to be defective. In this case, it is possible to easily acquire normal data indicating an image of a normal inspection target.

(8) The normal data acquired by the normal data acquisition unit includes master data (master data) representing a design drawing of the inspection object. In this case, it is possible to easily acquire normal data indicating an image of a normal inspection target.

(9) The normal data acquisition unit acquires main data corrected based on the machining accuracy of the inspection object as normal data. With this configuration, even when the inspection target area is small, normal data indicating an image of a normal inspection target can be easily acquired.

(10) A non-inspection target region is set in the abnormal data acquired by the abnormal data acquisition unit and the normal data acquired by the normal data acquisition unit, and a differential data generation unit generates differential data excluding the set non-inspection target region. In this case, it is prevented that a portion other than the inspection target region is included in the image shown by the differential data. Thus, training data for performing inspection with higher accuracy can be created.

(11) The abnormal data acquiring unit further acquires defect information indicating a type of a defect in the inspection object corresponding to the acquired abnormal data, and the difference data generating unit adds the defect information acquired by the abnormal data acquiring unit to the generated difference data. In this case, the user does not need to perform the operation of giving the defect information to the training data. This reduces the burden on the user and improves the work efficiency of creating the training data. In addition, since errors accompanying the user's work do not occur, more accurate training data can be created.

(12) An abnormal data acquisition unit acquires abnormal data after binarization processing, and a normal data acquisition unit acquires normal data after binarization processing. In this case, since the data amount of the abnormal data and the normal data is reduced, the training data can be created at high speed.

(13) A training data creation assistance method according to another aspect of the present invention assists creation of training data used for reexamination of an object to be examined, the training data creation assistance method including: acquiring abnormal data indicating an image of an inspection object previously determined to be defective; acquiring normal data indicating a normal image of the inspection object so as to correspond to the acquired abnormal data; generating difference data indicating a difference between the acquired abnormal data and normal data acquired in a manner corresponding to the abnormal data; and presenting an image of a portion of the inspection target object based on the generated difference data.

According to the training data creation support method, the image of the portion of the inspection target is presented based on the difference data indicating the difference between the abnormal data indicating the image of the inspection target determined to be defective in advance and the normal data indicating the image of the normal inspection target. The portion of the inspection target object in the image shown based on the difference data is highly likely to need to be inspected again. Therefore, the user can create training data for rechecking the part by confirming the prompted image. This makes it possible to easily create training data for examination with high accuracy.

Drawings

Fig. 1 is a diagram showing a configuration of a processing system including an auxiliary device according to a first embodiment of the present invention.

Fig. 2 is a diagram showing a configuration of the auxiliary device of fig. 1.

Fig. 3 is a diagram showing various data used to create training data.

Fig. 4 is a diagram showing an example of a display screen of the display device in creating the training data.

Fig. 5 is a flowchart showing an assist process of the assist apparatus of fig. 2.

Fig. 6 is a diagram showing various data for creating training data in the first modification.

Fig. 7 is a diagram showing various data for creating training data in the second modification.

Fig. 8 is a diagram showing various data used for creating training data in the third modification.

Fig. 9 is a diagram showing various data used for creating training data in the second embodiment.

Fig. 10 is a diagram showing various data used for creating training data in the third embodiment.

Fig. 11 is a diagram showing various data used for creating training data in the fourth embodiment.

Detailed Description

[1] First embodiment

(1) Processing system

Hereinafter, a training data creation support device and a training data creation support method according to an embodiment of the present invention will be described with reference to the drawings. In the following description, the training data creation assistance device will be referred to simply as an assistance device. Fig. 1 is a diagram showing a configuration of a processing system including an auxiliary device according to a first embodiment of the present invention. As shown in fig. 1, the processing system 100 includes a processing apparatus 10, an inspection apparatus 20, and a database storage apparatus 30.

The processing device 10 is composed of a CPU (central processing unit) 11, a RAM (random access memory) 12, a ROM (read only memory) 13, a storage device 14, an operation unit 15, a display device 16, and an input/output I/F (interface) 17. The CPU11, RAM12, ROM13, storage device 14, operation unit 15, display device 16, and input/output I/F17 are connected to a bus 18.

The RAM12 is used as a work area of the CPU 11. The ROM13 stores a system program. The storage device 14 includes a storage medium such as a hard disk or a semiconductor memory, and stores a training data creation support program (hereinafter, referred to simply as a support program). The auxiliary programs may also be stored in the ROM13 or other external storage device. The CPU11, the RAM12, and the ROM13 constitute an assisting apparatus 40 for executing training data creation assisting processing (hereinafter, referred to simply as assisting processing). In the auxiliary processing, the creation of training data is assisted.

The operation unit 15 is an input device such as a keyboard, a mouse, or a touch panel. The user can give a predetermined instruction to the assist device 40 by operating the operation unit 15. The display device 16 is a display device such as a liquid crystal display device, and displays a GUI (Graphical User Interface) for receiving an instruction from a User. The input/output I/F17 is connected to the inspection device 20.

The inspection apparatus 20 is, for example, an AOI (automatic optical inspection) apparatus, sequentially images an inspection target to generate a plurality of image data each representing an image of the plurality of inspection targets, and stores each generated image data. A unique identification number is assigned to each image data stored.

The inspection apparatus 20 will be described below with reference to a substrate as an example of an object to be inspected, but the object to be inspected is not limited to a substrate. The substrate refers to a semiconductor substrate, a substrate for FPD (Flat Panel Display) such as a liquid crystal Display device or an organic EL (Electro Luminescence) Display device, a substrate for optical disk, a substrate for magnetic disk, a substrate for optical disk, a substrate for photomask, a ceramic substrate, a substrate for solar cell, or the like.

The inspection device 20 performs processing based on a predetermined algorithm on each of the stored image data, thereby inspecting the substrate corresponding to each image data. The inspection apparatus 20 may inspect the substrate corresponding to each image data based on the deep learning. In the inspection, it is determined whether there is a defect on the substrate. In addition, the type of the defect is determined for the substrate determined to be defective.

In the inspection apparatus 20, even a substrate having no defect may be determined to have a defect. Although there is no defect, if the substrate is discarded due to an erroneous determination, the yield is lowered. Here, re-inspection based on training learning is performed for the substrate determined to be defective. The assisting apparatus 40 assists the creation of training data used in the re-examination. The database storage device 30 includes a large-capacity storage device such as a server. The created training data is registered in the database storage device 30. The assist device 40 will be described in detail below.

(2) Auxiliary device

Fig. 2 is a diagram showing a structure of the auxiliary device 40 of fig. 1. Fig. 3 is a diagram representing various data used to create training data. Fig. 4 shows an example of a display screen of the display device 16 during creation of training data. As shown in fig. 2, the support device 40 includes, as functional units, an abnormal data acquisition unit 41, a normal data acquisition unit 42, a differential data generation unit 43, a presentation unit 44, a reception unit 45, and a registration unit 46. The CPU11 in fig. 1 realizes a functional section of the support device 40 by executing a support program stored in the ROM13, the storage device 14, or the like. A part or all of the functional units of the support device 40 may be realized by hardware such as an electronic circuit.

The abnormal data acquisition unit 41 acquires, from the inspection apparatus 20, image data (hereinafter referred to as abnormal data) indicating an image of each substrate determined to be defective in advance by the inspection apparatus 20. The image data may show an image of the entire substrate, or may show a partial image of the substrate in the same area. A part of the image of the substrate based on the abnormality data acquired by the abnormality data acquisition unit 41 is shown in the upper left part of fig. 3.

The normal data acquisition unit 42 acquires image data (hereinafter, referred to as normal data) indicating an image of a normal substrate so as to correspond to each of the abnormal data acquired by the abnormal data acquisition unit 41. In this example, the normal data is predetermined image data indicating an image of a substrate that has not been determined to be defective by the inspection apparatus 20 in advance, and is acquired from the inspection apparatus 20. A part of the image of the substrate based on the normal data acquired by the normal data acquisition section 42 is shown in the lower left part of fig. 3.

The abnormal data and the normal data correspond to each other. For example, when the abnormal data is acquired by the abnormal data acquiring unit 41, the normal data acquiring unit 42 may acquire image data representing an image of a normal substrate having an identification number immediately preceding the identification number of the abnormal data as normal data. Alternatively, when the abnormal data is acquired by the abnormal data acquiring unit 41, the normal data acquiring unit 42 may acquire image data representing an image of a normal substrate having an identification number subsequent to the identification number of the abnormal data as normal data.

The difference data generation unit 43 calculates a difference between each pixel value of the abnormal data acquired by the abnormal data acquisition unit 41 and each pixel value of the normal data acquired by the normal data acquisition unit 42 so as to correspond to the abnormal data, thereby generating new image data. The image data generated by the difference data generation unit 43 is referred to as difference data. An image of the substrate based on the difference data generated by the difference data generation section 43 is shown in the right part of fig. 3. The difference data represents an image of a portion of the substrate that is highly likely to require re-inspection. Therefore, the differential data can be a label indicating a portion on the substrate that needs to be re-inspected.

The presentation unit 44 presents each difference data to the user by displaying a GUI50 (fig. 4) including an image based on each difference data generated by the difference data generation unit 43 on the display device 16. As shown in fig. 4, the GUI50 includes an image display area 51, a login button 52, and a correction button 53. A plurality of images of the object to be measured are displayed in the image display area 51. In the present example, the image based on the difference data and the image based on the abnormal data are displayed in the image display area 51 so as to overlap each other, but only the image based on the difference data may be displayed in the image display area 51.

The receiving unit 45 receives an instruction to register difference data. Specifically, the receiving unit 45 receives, from the operation unit 15, selection of difference data registered as training data in the GUI50 displayed by the presentation unit 44. The user can select an arbitrary image using the operation unit 15 while visually checking the image displayed in the image display area 51, and can give an instruction to the receiving unit 45 to select the difference data indicating the image as training data by operating the registration button 52. The registration unit 46 registers the difference data selected by the receiving unit 45 in the database storage device 30 as training data.

The receiving unit 45 can receive the correction of the difference data. The user can select an arbitrary image in the image display area 51 using the operation unit 15 and instruct the receiving unit 45 to correct the difference data corresponding to the selected image by operating the correction button 53. The user can designate a portion of the selected image to be reexamined by, for example, painting the selected image using the operation unit 15.

When the designation is received by the receiving unit 45, the difference data generating unit 43 assigns a label indicating the portion to which the designation is received to the selected difference data. Thereby, the differential data is corrected. After the difference data is corrected, if the registration button 52 is operated, the registration unit 46 registers the corrected difference data as training data in the database storage device 30.

(3) Auxiliary treatment

Fig. 5 is a flowchart showing an assist process of the assist device 40 of fig. 2. The assist processing of fig. 5 is performed by the CPU11 of fig. 1 executing an assist program stored in the ROM13, the storage device 14, or the like on the RAM 12. The assist processing will be described below with reference to the assist device 40 of fig. 2 and the flowchart of fig. 5.

First, the abnormal data acquiring unit 41 acquires various abnormal data from the inspection device 20 (step S1). Next, the normal data acquisition unit 42 acquires normal data corresponding to the abnormal data acquired in step S1 from the inspection device 20 (step S2). Steps S1 and S2 may be performed simultaneously.

Next, the differential data generating unit 43 generates each differential data based on the corresponding abnormal data and normal data acquired in steps S1 and S2, respectively (step S3). Then, the presentation unit 44 presents each difference data to the user by displaying an image based on each difference data generated in step S3 on the display device 16 (step S4).

Next, the receiving unit 45 determines whether or not the correction of any one of the difference data presented in step S4 is received (step S5). If the corrected difference data is not received, the receiving unit 45 proceeds to step S7. When the correction of any one of the differential data is received, the differential data generating unit 43 adds a tag indicating the designated portion to which the correction is received to the differential data, thereby correcting the differential data (step S6), and the process proceeds to step S7.

In step S7, the receiving unit 45 determines whether or not an instruction to register any one of the difference data presented in step S4 or the difference data corrected in step S6 is given (step S7). If the registration of the difference data is not instructed, the receiving unit 45 proceeds to step S9. When the instruction to register any one of the difference data is given, the registration unit 46 registers the instructed difference data as training data in the database storage device 30 (step S8), and the process proceeds to step S9.

In step S9, the registration unit 46 determines whether or not the end is instructed (step S9). The user can instruct the end or the continuation by performing a predetermined operation using the operation unit 15. If the end is not instructed, the registration unit 46 returns to step S5. When the difference data is further registered, the user instructs to continue without instructing the end. When the termination is instructed, the registration unit 46 terminates the assist processing.

(4) Effect

In the support device 40 of the present embodiment, the abnormal data acquiring unit 41 acquires abnormal data indicating an image of the inspection target determined to be defective in advance. The normal data acquiring unit 42 acquires normal data indicating an image of a normal inspection target so as to correspond to the abnormal data acquired by the abnormal data acquiring unit 41. The difference data generating unit 43 generates difference data indicating a difference between the abnormal data acquired by the abnormal data acquiring unit 41 and the normal data acquired by the normal data acquiring unit 42 so that the abnormal data corresponds to the normal data. The portion of the inspection target object in the image shown based on the difference data is highly likely to need to be inspected again.

Therefore, the presentation unit 44 presents an image of a portion of the inspection target based on the difference data generated by the difference data generation unit 43. The selection of the difference data presented by the presentation unit 44 is received by the reception unit 45. The selected difference data is received by the receiving unit 45 as training data, and is registered in the database storage device 30 by the registration unit 46. In this case, the user can register the selected difference data as the training data by selecting the difference data corresponding to the desired image while visually recognizing the presented image. Thus, training data can be created more easily.

The receiving unit 45 also receives the correction of the difference data presented by the presenting unit 44. The registration unit 46 registers the corrected difference data as training data. In this case, the user can more appropriately correct the portion of the inspection target object that needs to be reexamined in the image based on the difference data. Thus, training data for performing inspection with higher accuracy can be created.

(5) Modification examples

Fig. 6 is a diagram showing various data for creating training data in the first modification. In the first modification, the user can register a non-inspection target region indicating the outside of the inspection target region in the inspection apparatus 20 in advance. When the non-inspection target region is registered, the inspection device 20 sets the non-inspection target region in the generated image data.

Therefore, as shown in the upper left part of fig. 6, a non-inspection target region is set in the abnormal data acquired by the abnormal data acquiring unit 41. Similarly, as shown in the lower left part of fig. 6, a non-inspection target region is set in the normal data acquired by the normal data acquisition unit 42. In this case, as shown in the right part of fig. 6, the difference data generating unit 43 calculates the difference between the pixel values of the abnormal data and the normal data corresponding to the abnormal data in a state where the set non-inspection target region is excluded, thereby generating difference data. The pixel value of the non-inspection target region in the difference data may be 0.

In this case, a label indicating a portion that needs to be re-inspected is prevented from being given to a portion outside the inspection target region. Thus, training data for performing inspection with higher accuracy can be created. Further, since the portion outside the inspection target region is not to be re-inspected, the re-inspection of the substrate can be performed at high speed by using the created training data.

Fig. 7 is a diagram showing various data for creating training data in the second modification. As shown in the upper left part of fig. 7, the abnormal data acquiring unit 41 acquires the abnormal data together with defect information indicating the type of the defect of the substrate in the image indicated by the abnormal data. As shown in the right part of fig. 7, the difference data generation unit 43 adds the defect information acquired by the abnormal data acquisition unit 41 to the generated difference data. In the image display area 51 of the GUI50, an image based on the difference data may also be shown in a manner (e.g., color) indicating the type of defect.

In this case, the user does not need to perform the operation of giving the defect information to the training data. This reduces the burden on the user and improves the work efficiency of creating the training data. In addition, since errors accompanying the user's work do not occur, more accurate training data can be created.

Fig. 8 is a diagram showing various data used for creating training data in the third modification. In the third modification, the user can set in advance in the inspection apparatus 20 to perform binarization processing on the image data. When it is set to perform the binarization processing, the inspection device 20 generates binarized image data.

Therefore, as shown in the upper left part of fig. 8, the abnormal data acquiring section 41 acquires the abnormal data after the binarization processing. Similarly, as shown in the lower left part of fig. 8, the normal data acquisition section 42 acquires the normal data after the binarization processing. As shown in the right part of fig. 8, the difference data generation unit 43 calculates the difference between the pixel values of the binarized abnormal data and the binarized normal data corresponding to the abnormal data, thereby generating difference data.

In this case, since the data amount of the image data is reduced, the training data can be created at high speed. In addition, by using the created training data, re-inspection of the substrate can be performed at high speed.

[2] Second embodiment

In the first embodiment, the normal data acquisition unit 42 acquires normal data so that one normal data corresponds to one abnormal data, but the embodiment is not limited thereto. The following description deals with the point of difference between the assist processing in the second to fourth embodiments and the assist processing in the first embodiment.

Fig. 9 is a diagram showing various data used for creating training data in the second embodiment. In the present embodiment, as shown in the lower left part of fig. 9, a plurality of normal data are acquired so that one abnormal data corresponds to the plurality of normal data. The difference data generation unit 43 calculates the difference between the pixel values of one abnormal data and the respective normal data corresponding to the abnormal data, thereby generating difference data. Therefore, as shown in the right part of fig. 9, a plurality of difference data are generated corresponding to one abnormal data.

According to this configuration, a plurality of difference data are generated from one abnormal data. This can improve the work efficiency of training data creation.

[3] Third embodiment

Fig. 10 is a diagram showing various data used for creating training data in the third embodiment. In the present embodiment, as shown in the lower left part of fig. 10, a plurality of normal data are acquired so that one abnormal data corresponds to the plurality of normal data. The difference data generation unit 43 calculates the difference between the average pixel value of one abnormal data and the average pixel value of a plurality of normal data corresponding to the abnormal data, thereby generating difference data. Therefore, as shown in the right part of fig. 10, one piece of difference data is generated corresponding to one piece of abnormal data.

According to this configuration, even when a noise component unrelated to a defect is accidentally mixed in any of the plurality of normal data, since the plurality of normal data are averaged, the noise component hardly affects the pixel value of the averaged normal data. Therefore, by using the averaged normal data, it is possible to create training data for examination with higher accuracy.

[4] Fourth embodiment

Fig. 11 is a diagram showing various data used for creating training data in the fourth embodiment. In the present embodiment, as shown in the upper left part of fig. 11, normal data is acquired such that one normal data corresponds to a plurality of abnormal data. The differential data generating unit 43 generates differential data based on each abnormal data and normal data corresponding to the abnormal data. Therefore, as shown in the right part of fig. 11, a plurality of difference data corresponding to the plurality of abnormal data are generated.

According to this configuration, since it is not necessary to acquire normal data every time abnormal data is acquired, it is possible to create training data at high speed using common normal data. In the present embodiment, the normal data acquiring unit 42 may acquire image data representing an image of a normal substrate, which is prepared in advance, as normal data. In this case, step S2 in the auxiliary processing may be performed before step S1.

The image data prepared in advance may be any of the image data generated by the inspection apparatus 20, or may be master data such as CAD data representing a design drawing of a substrate. Here, the pattern width of the substrate in the image, the radius of curvature of the corner of the pattern, and the like differ between the image data generated by the inspection apparatus 20 and the master data depending on the processing accuracy of the substrate by etching or the like. Therefore, the main data may be corrected based on the processing accuracy of the substrate to change the pattern width in the image, the radius of curvature of the corner portion of the pattern, or the like.

In the first to third embodiments, image data representing an image of a substrate that has not been determined to be defective by the inspection apparatus 20 in advance is also acquired from the inspection apparatus 20 as normal data, but the embodiments are not limited to this. At least one of the normal data may be previously prepared main data or corrected main data.

[5] Other embodiments

In the above embodiment, the assisting apparatus 40 includes the receiving unit 45 and the registration unit 46, but the embodiment is not limited thereto. The support device 40 may not include the receiving unit 45 and the registration unit 46. Even in this case, the user can create training data for rechecking the part by confirming the image presented on the GUI 50. This makes it possible to easily create training data for examination with high accuracy.

[6] Correspondence between each structural element of the technical solution and each part of the embodiment

Hereinafter, examples of correspondence between each constituent element of the embodiments and each element of the embodiments will be described, but the present invention is not limited to the following examples. As each structural element of the claims, other kinds of elements having the structures or functions described in the claims can be used.

In the above embodiment, the support device 40 is an example of a training data creation support device, the abnormal data acquisition unit 41 is an example of an abnormal data acquisition unit, and the normal data acquisition unit 42 is an example of a normal data acquisition unit. The difference data generating unit 43 is an example of a difference data generating unit, the presenting unit 44 is an example of a presenting unit, the receiving unit 45 is an example of a receiving unit, and the registering unit 46 is an example of a registering unit.

Claims

1. A training data creation assistance device that assists creation of training data used for reexamination of an object to be examined, comprising:

an abnormal data acquisition unit that acquires abnormal data representing an image of an inspection object previously determined to be defective;

a normal data acquisition unit that acquires normal data indicating an image of a normal inspection object so as to correspond to the abnormal data acquired by the abnormal data acquisition unit;

a difference data generation unit configured to generate difference data indicating a difference between the abnormal data acquired by the abnormal data acquisition unit and the normal data acquired by the normal data acquisition unit so as to correspond to the abnormal data; and

and a presentation unit that presents an image of a portion of the inspection target object based on the difference data generated by the difference data generation unit.

2. The training data creation assistance apparatus according to claim 1, further having:

an accepting unit that accepts selection of the difference data presented by the presenting unit; and

and a registration unit that registers the difference data selected by the receiving unit as training data.

3. The training data creation assistance apparatus according to claim 2,

the receiving unit further receives the correction of the difference data presented by the presenting unit,

the registration unit registers the corrected difference data as training data.

4. The training data creation assistance apparatus according to any one of claims 1 to 3,

the normal data acquisition unit acquires a plurality of normal data in a manner corresponding to the abnormal data,

the differential data generation unit generates a plurality of differential data based on abnormal data and a plurality of normal data corresponding to the abnormal data.

5. The training data creation assistance apparatus according to any one of claims 1 to 3,

the difference data generation unit generates difference data based on an average of abnormal data and a plurality of normal data corresponding to the abnormal data.

6. The training data creation assistance apparatus according to any one of claims 1 to 3,

the normal data acquisition unit acquires normal data so as to correspond to a plurality of abnormal data,

the differential data generation unit generates differential data based on each abnormal data and normal data corresponding to the abnormal data.

7. The training data creation assistance apparatus according to any one of claims 1 to 6,

the normal data acquired by the normal data acquiring unit includes image data representing an image of the inspection object that has not been determined to be defective in advance.

8. The training data creation assistance apparatus according to any one of claims 1 to 7,

the normal data acquired by the normal data acquisition unit includes main data representing a design drawing of the object to be inspected.

9. The training data creation assistance apparatus according to claim 8,

the normal data acquisition unit acquires main data corrected based on the machining accuracy of the inspection target as normal data.

10. The training data creation assistance apparatus according to any one of claims 1 to 9,

a non-inspection target region is set in the abnormal data acquired by the abnormal data acquiring unit and the normal data acquired by the normal data acquiring unit,

the difference data generating unit generates difference data excluding the set non-inspection target region.

11. The training data creation assistance apparatus according to any one of claims 1 to 10,

the abnormal data acquiring unit further acquires defect information indicating a type of a defect of the inspection target corresponding to the acquired abnormal data,

the difference data generation unit adds the defect information acquired by the abnormal data acquisition unit to the generated difference data.

12. The training data creation assistance apparatus according to any one of claims 1 to 11,

the abnormal data acquiring section acquires abnormal data after binarization processing,

the normal data acquisition unit acquires normal data after binarization processing.

13. A training data creation assistance method for assisting creation of training data used for reexamination of an object to be examined, comprising:

acquiring abnormal data indicating an image of an inspection object previously determined to be defective;

acquiring normal data indicating a normal image of the inspection object so as to correspond to the acquired abnormal data;

generating difference data indicating a difference between the acquired abnormal data and normal data acquired in a manner corresponding to the abnormal data; and

and presenting an image of a portion of the inspection target based on the generated difference data.