KR20230160754A - Defect inspection performance device using learning model, method, system and program - Google Patents

Abstract

An optional artificial intelligence engine-based object non-destructive inspection method, device, and system are provided. The electronic device that performs non-destructive testing on an object based on the selective artificial intelligence engine includes: a memory that stores a plurality of learning variables for defect inspection and a plurality of learning models corresponding to the individual learning variables; and a processor that inspects the object for defects, wherein the processor includes a processing unit that inspects the object for defects based on at least one learning model selected from among the plurality of stored learning models.

Description

Device, method, system and program for performing defect inspection using a learning model {DEFECT INSPECTION PERFORMANCE DEVICE USING LEARNING MODEL, METHOD, SYSTEM AND PROGRAM}

The present invention relates to an apparatus, method, system, and program for performing defect inspection using a learning model.

Product defects can cause deterioration of supply chain services and loss of automation equipment. Therefore, it is very important to properly inspect the product for defects.

Non-destructive testing, that is, non-destructive testing, which does not destroy the object by utilizing radiation, especially was detected. However, this conventional inspection method has limitations in defect detection performance, so there is a problem in that it cannot detect all defects from the X-ray image, that is, there are defects that are not detected. In addition, since the characteristics are different depending on the composition of the object, this can lead to differences in the results of the test, that is, accuracy, even in the case of allogenes, which is a problem that reduces the reliability of the test.

Republic of Korea Patent Publication No. 10-2249836 (2021.05.03)

The problem to be solved by the present invention is to provide a method, device, and system that selects the optimal artificial intelligence learning model for an inspection object and improves inspection reliability by improving non-destructive inspection accuracy while simultaneously increasing the efficiency of the inspection system.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood by those skilled in the art from the description below.

An electronic device that performs non-destructive testing on an object based on a selective artificial intelligence engine according to one aspect of the present invention to solve the above-described problem includes a plurality of learning variables for defect inspection and a plurality of learning variables corresponding to the individual learning variables. memory to store models; and a processor that inspects the object for defects, wherein the processor includes a processing unit that inspects the object for defects based on at least one learning model selected from among the plurality of stored learning models.

A non-destructive inspection method for an object based on a selective artificial intelligence engine in an electronic device according to one aspect of the present invention includes storing a plurality of learning variables for defect inspection and a plurality of learning models corresponding to the individual learning variables; Receiving image data of the object; determining a category for the input object through the stored learning model and selectively selecting at least one learning model from the plurality of stored learning models based on the determined category of the object; performing defect inspection of the object based on the selected at least one learning model; and providing the defect inspection results.

A selective artificial intelligence engine-based object non-destructive inspection system according to one aspect of the present invention includes an image acquisition device that acquires image data for an object by irradiating radiation; and an electronic device, wherein the electronic device includes: a memory storing a plurality of learning variables for defect inspection and a plurality of learning models corresponding to the individual learning variables; and a processor that performs defect inspection of the object based on at least one learning model selected from among the plurality of stored learning models.

Other specific details of the invention are included in the detailed description and drawings.

According to the present invention, the following effects can be achieved.

According to the present invention, the efficiency of the inspection system can be increased while improving inspection reliability by improving the accuracy of non-destructive inspection of the inspection object.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a block diagram illustrating an artificial intelligence-based object non-destructive inspection system according to an embodiment of the present invention.
Figure 2 is a flowchart illustrating a method for non-destructive inspection of an object according to an embodiment of the present invention.
Figure 3 is a block diagram of an electronic device according to an embodiment of the present invention.
Figure 4 is a block diagram of a learning unit according to an embodiment of the present invention.
Figure 5 is a block diagram of a processing unit according to an embodiment of the present invention.
Figure 6 is a block diagram of an electronic device according to another embodiment of the present invention.
Figure 7 is a block diagram of the processing unit of Figure 6.
Figures 8 and 9 are diagrams to explain the results of non-destructive testing of an object according to the present invention.
Figure 10 is a flowchart illustrating a non-destructive method of object according to another embodiment of the present invention.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are merely provided to ensure that the disclosure of the present invention is complete and to provide a general understanding of the technical field to which the present invention pertains. It is provided to fully inform the skilled person of the scope of the present invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for describing embodiments and is not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used in the specification, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other elements in addition to the mentioned elements. Like reference numerals refer to like elements throughout the specification, and “and/or” includes each and every combination of one or more of the referenced elements. Although “first”, “second”, etc. are used to describe various components, these components are of course not limited by these terms. These terms are merely used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may also be a second component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless clearly specifically defined.

Spatially relative terms such as “below”, “beneath”, “lower”, “above”, “upper”, etc. are used as a single term as shown in the drawing. It can be used to easily describe the correlation between a component and other components. Spatially relative terms should be understood as terms that include different directions of components during use or operation in addition to the directions shown in the drawings. For example, if a component shown in a drawing is flipped over, a component described as “below” or “beneath” another component will be placed “above” the other component. You can. Accordingly, the illustrative term “down” may include both downward and upward directions. Components can also be oriented in other directions, so spatially relative terms can be interpreted according to orientation.

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

In this specification, 'image or image data' refers to still image or video data obtained through a tube or detector using radiation. In one embodiment, the image may be an X-ray image of an object through an X-ray tube or X-ray detector. At this time, the X-ray image is, for example, a 2D (Dimensional) image, a CT (Computed Tomography) image reconstructed from a continuous 2D image aggregation, and reconstructed CT volume data. Can include slice images.

In this specification, 'defect' refers to a part that is not a part that is defined or can be defined as normal for the object during non-destructive testing of an object that is the subject of defect inspection based on artificial intelligence according to the present invention. This can be expressed by various names such as defect or error. Depending on the embodiment, the present invention is not limited to such expressions, and may also include meanings that are the same as or similar to defects in the conventional sense.

1 is a block diagram illustrating an artificial intelligence-based object non-destructive inspection system according to an embodiment of the present invention.

Referring to FIG. 1, a system that performs artificial intelligence-based non-destructive inspection of an object according to an embodiment of the present invention may be configured to include an electronic device 100 and an image acquisition device 150. At this time, the configuration of the electronic device 100 and the image acquisition device 150 shown in FIG. 1 is an example and is not limited thereto, and one or more components are added in relation to performing the operation according to the present invention. It may be configured or vice versa.

The electronic device 100 may be configured to include a memory and a processor. The memory may correspond to or include the database 120 shown in FIG. 1, and the processor may include the control unit 110. ) and may include at least one of the AI engine 130. At this time, the AI engine 130 includes, but is not limited to, a deep learning network.

The electronic device 100 may be connected to the image acquisition device 150 through a network to receive image data about the object.

The image acquisition device 150 may be configured to include a detector 160, an X-ray tube 170, and a lighting source (not shown), and the detector 160 may be at least one of a 2D detector and a 3D detector. You can. In the above, the detector 160 and the In addition, the image acquisition device 150 may additionally include a device capable of photographing the motion of a moving object and a CT detector (not shown). The light source includes, but is not limited to, a terahertz transmissive light source.

As a component of the electronic device 100, the control unit 110 controls operations performed in the electronic device 100, and the database 120 controls the image of the object received from the image acquisition device 150 and the object. Data that is received and processed by the electronic device 100, such as a learning dataset used for defect inspection and a learning model corresponding to the learning dataset, is stored.

The control unit 110 inputs image data of the object and determines (classifies) the category of the object through a learning model stored in the database 120, and inspects the image data of the object that is the subject of defect inspection for defects. Various machine learning models that specify an operation mode, selectively select at least one learning model corresponding to the specified defect inspection operation mode, and perform defect inspection from image data of the object based on the selected learning model ( It may include a hardware unit with computing power to perform machine-learning model algorithms and related applications. For example, the control unit 110 may include at least one of a central processing unit, a microprocessor, and a graphics processor. Additionally, the control unit 110 may further include a separate memory (not shown) that stores a machine learning model algorithm or application.

For defect inspection, the electronic device 100 can learn image data about an object and input it to the AI engine 130 to obtain (high-definition or improved) object image data. At this time, in a general sense, the image acquired through the AI engine 130 is not only image data of higher quality or improved overall than the object image data input from the image acquisition device 150, but also defects during non-destructive inspection of the object based on artificial intelligence. From an inspection perspective, it may represent all or some improved or new image data.

According to one embodiment, the electronic device 100 compares the inspection result of the image data for the object before learning with the inspection result for the image data for the object obtained through the AI engine 130 to model a new learning model. can also be additionally defined. Based on the added learning model, the electronic device 100 determines whether to use the AI engine 130 based on the image of the input object, that is, the object acquired through the AI engine 130 rather than the image data of the input object. It is also possible to decide whether to use image data to inspect for defects and perform actions accordingly.

Therefore, in relation to the present invention, the It can be.

Figure 2 is a flowchart illustrating a method for non-destructive inspection of an object according to an embodiment of the present invention. Figure 3 is a block diagram of the electronic device 100 according to an embodiment of the present invention. Figure 4 is a block diagram of the learning unit 220 according to an embodiment of the present invention. Figure 5 is a block diagram of the processing unit 230 according to an embodiment of the present invention. FIG. 6 is a block diagram of the electronic device 100 according to another embodiment of the present invention, and FIG. 7 is a block diagram of the processing unit 610 of FIG. 6. Figures 8 and 9 are diagrams to explain the results of non-destructive testing of an object according to the present invention. Figure 10 is a flowchart illustrating a non-destructive method of object according to another embodiment of the present invention.

The operations of FIGS. 2 and 10 may be performed through the electronic device 100 of FIG. 1 . Here, the operations of FIGS. 2 and 10 will be described with reference to the configuration of the electronic device 100 shown in FIGS. 3 to 7.

First, referring to FIG. 2, in operation 11, the electronic device 100 may store a plurality of learning variables and a corresponding learning model. Here, ‘learning variable’ refers to a training dataset (first and/or second training dataset) described later, and ‘corresponding learning model’ refers to a learning model created using the training dataset. .

In operation 12, the input unit 310 receives image data of an object obtained through radiation (eg, X-ray) irradiation from the image acquisition device 150.

Depending on the embodiment, the operation order of operations 11 and 12 may be defined differently from that shown in FIG. 2 . This can be applied not only to operations 11 and 12, but also to the sequence of operations shown in FIG. 2 (and FIG. 10, which will be described later).

The image data of the object input to the input unit 310 is raw data received from the image acquisition device 150, or is at least partially processed to be suitable for defect inspection through the above-described AI engine 130. It could be data.

In one embodiment, whether to perform defect inspection may be determined in advance based on received image data of the object, that is, raw data. Based on the result of determining whether the object is defective according to a separate operation or operation 16 described later, if there is an error in the raw data used above or the reliability of the inspection result is judged to be below a predetermined standard, it corresponds to ground truth. Not only is it difficult to obtain a similar result, but even if a similar result is obtained, the result cannot be trusted. In this case, as described above, improved image data is obtained through the AI engine 130 and used as basic data for defect inspection. Available.

The learning unit 320 takes the training dataset as an input and generates a learning model corresponding to the input training dataset through the preprocessing module 410, the feature extraction module 420, and the inspection processing module 430, and stores it in memory. It can be temporarily stored in .

In operations 13 and 14, the processing unit 330 determines the category of the object through a learning model based on image data of the input object, and specifies a defect inspection operation mode for the object based on the determined category. Thus, a learning model corresponding to the specified operation mode can be selected.

In operation 15, the processing unit 330 checks image data of the object for defects based on the selected learning model and generates defect test result data.

In operation 16, the output unit 340 provides defect inspection result data generated for the object by the processing unit 330. In this specification, 'provision' may be defined in various ways related to the output of results to the target object, such as direct or indirect output through a display, transmission to the target terminal, and/or output control.

The artificial intelligence-based object non-destructive inspection method according to the present invention will be described in more detail with reference to FIG. 10 as follows.

In operations 21 and 22, the electronic device 100 learns a model for defect inspection based on image data of the inspection object and receives image data of the object, which correspond to operations 11 and 12 of FIG. 2 described above, respectively. It is a movement that works.

In operation 23, the electronic device 100 determines a defect inspection operation mode for the object.

In operation 24-1, if the defect inspection operation mode determined in operation 23 is the first operation mode, the electronic device 100 selects a learning model corresponding to the operation mode through the operation mode selection module, and selects the learning model corresponding to the operation mode to select the image data of the object. Process, i.e. perform defect checking.

On the other hand, in operation 24-2, if the defect inspection operation mode determined in operation 23 is the second operation mode, the electronic device 100 selects a learning model corresponding to the operation mode through the operation mode selection module, and selects a learning model corresponding to the operation mode of the object. Image data is processed, i.e. defect inspection is performed.

In relation to operations 23 and 24, details on determining the defect inspection operation mode (first mode, second mode) and processing image data of the object according to the determined defect inspection operation mode are described in the description of FIGS. 4 and 5 below. This is described later. In order to facilitate understanding of the technical idea of the present invention, only two defect inspection operation modes, the first and second operation modes, are defined and explained in this specification, but the number of operation modes and definition of operation modes can be adjusted according to settings or requests. etc. may be implemented differently.

Referring to FIG. 3 , the electronic device 100 may be configured to include an input unit 310, a defect inspection unit, and an output unit 340. At this time, the defect inspection unit may include a learning part 320 and a processing part 330.

Referring to FIG. 4, the learning unit 320 constituting the defect inspection unit includes a preprocessing module 410 that receives training data (or dataset) for learning non-destructive inspection of an object, that is, defect inspection, from memory and preprocesses it, It may be configured to include a feature extraction module 420 that extracts features from the preprocessed data, and an inspection processing module 430 that generates a learning model to be used for defect inspection based on the extracted features. In the above, the learning unit 320 may use one AI engine for defect inspection learning, but is not limited to this. Depending on the embodiment, the learning unit 320 is implemented with a plurality of learning modules, and each learning module may use a different AI engine for defect inspection learning. At this time, the individual learning module may be configured to include at least one of the pre-processing module, feature extraction module, and inspection processing module of the learning unit 320 described above.

Referring to Figure 4, the training datasets, namely Prod #1, Prod #2,... , Prod #N (where N is a positive integer) can be defined as the first training dataset for a predefined individual test target. Additionally, the learning models Model #1, Model #2, … , Model #N (where N is a positive integer) can be defined as a learning model created corresponding to the first training dataset, that is, a first learning model.

In other words, the learning unit 320 may generate a plurality of corresponding individual first learning models using a plurality of learning variables, that is, individual first training datasets. The first learning model created in this way can also be defined as a learning model specialized for individual test subjects. That is, the learning unit 320 may generate N first learning models corresponding to N first training datasets (where N is a positive integer).

Meanwhile, a second training dataset that is a combination of at least two or more individual first training datasets or includes all of the first training datasets may be defined. This second training dataset may be named a full training dataset, a combined training dataset, or a united training dataset. The learning unit 320 may generate a second learning model using this second training dataset. Accordingly, the second learning model is a different learning model from the above-described first learning model, and the number of second learning models may be determined by the number of second training datasets defined above. Additionally, the individual training dataset used in the second training dataset above may or may not be the same as the above-described first training dataset. For example, the second training dataset may include at least one training dataset that is not classified as the first training dataset. From this perspective, the first training dataset may be viewed as a classified training dataset that is the basis for creating the first learning model.

Meanwhile, in FIGS. 4, 5, and 7, for convenience of explanation, only one second training dataset including all first training datasets is defined, but the present invention is not limited thereto. In other words, the second training dataset may be defined as a plurality of second training datasets depending on the configuration method, and there may also be a plurality of corresponding learning models.

In the above, the generated first and/or second learning model may have different parameters, weights, etc. depending on the test target.

As will be described later, in the present invention, it is possible to determine which learning model to use depending on whether the object is specified, and to check whether the object is defective according to the determined learning model. At this time, whether or not the object is specified may mean, for example, a category or classification determined for the object. Depending on the embodiment, whether or not the object is specified may be determined not by the electronic device 100 but by an external input, for example, an input or request from a defect inspection requester or terminal for the object. Accordingly, whether the object is specified may be determined by whether a learning model available for defect inspection can be specified for the object that is the target of defect inspection. That is, when the electronic device 100 receives the external input, it can specify and selectively select a learning model based on it. When the object of the object is specified, the electronic device 100 selects and uses a specific first learning model corresponding to the specified object, but otherwise, uses a second learning model to check whether the object is defective. You can. At this time, if there are a plurality of second learning models, the electronic device 100 determines which of the second learning model(s) to use or all of the plurality of second learning models to use, and provides the object accordingly. Defect inspection can be performed.

Meanwhile, depending on the embodiment, if the previously performed defect inspection result of the object is less than or equal to a predetermined standard value and the pre-selected learning model among the plurality of stored learning models is one of the plurality of first learning models, the electronic device 100 Can be controlled to select a second learning model rather than another first learning model belonging to the first learning model among the plurality of stored learning models. Depending on the embodiment, regardless of the target specification of the object or in cases where target specification of the object is difficult, the electronic device 100 may detect the target based on at least one of the first learning model or the second learning model set as default. Defect inspection can also be performed. In this case, the electronic device 100 may determine whether to define an additional learning model or retest based on the object defect inspection result according to a specific learning model set as default and perform the corresponding operation. For example, the electronic device 100 performs defect inspection on the object based on a specific first learning model set as default, and if the performance result is less than a predefined standard, defect inspection on the object based on the second learning model can be re-performed. Depending on the embodiment, the electronic device 100 selects at least two specific first learning models among the first learning models by default to perform defect inspection on the object, and performs defect inspection based on the second learning model according to the performance results. You can decide whether to re-perform. At this time, the selected specific first learning model will be determined based on information about the object subject to defect inspection, previous defect inspection performance history, settings of the electronic device 100, information input by the inspection requestor or terminal, etc. You can. This can also be used in the same or similar form to specify an object.

The operation of the learning unit 320 shown in FIG. 4 is i) when image data of an object subject to defect inspection is input through the input unit 310 shown in FIG. 3, ii) in the previous period of i). or irregularly, iii) may be performed regularly or irregularly, regardless of i) above.

Referring to FIG. 5 , the processing unit 330 may include a model selection module 510, a pre-processing module 520, a feature extraction module 530, and an inspection processing module 540.

Here, the preprocessing module 520, the feature extraction module 530, and the inspection processing module 540 may be separate components from the corresponding components shown in FIG. 4. Accordingly, the preprocessing module, feature extraction module, and inspection processing module can be viewed as individually implemented in the learning unit 320 of FIG. 4 and the processing unit 330 of FIG. 5 described above.

Meanwhile, depending on the embodiment, there is one processing unit as shown in FIG. 6, and in this case, the pre-processing module, feature extraction module, and inspection processing module are, as shown in FIG. 7, the learning unit of FIG. 4 ( 320) and the processing unit 330 of FIG. 5 may be implemented in a shared form.

The model selection module 510 determines a defect inspection operation mode for the object and selectively selects at least one of the learning models generated based on the training dataset in FIG. 4 corresponding to the determined defect inspection operation mode. Depending on the embodiment, the model selection module 510 is based on the training dataset in FIG. 4 according to the defect inspection operation mode already determined by the components in the control unit 110 of FIG. 1 or the learning unit 320 of FIG. 4. You can also simply select at least one of the created learning models.

Referring to FIG. 5, the processing unit 330 performs pre-processing in the pre-processing module 520 on the image data of the object input through the input unit 310, and the pre-processed image data of the object is selected in the model selection module 510. The defect is inspected from the image data of the object through the feature extraction module 530 and the inspection processing module 540 based on the learning model selected as the learning model corresponding to the defect inspection operation mode, and defect inspection result data is based on the inspection result. is generated and delivered to the output unit 340. Thereafter, the output unit 340 provides defect inspection result data of the received object.

Meanwhile, in the present specification, the first training dataset may be generated based on data that has already been classified for examination of the object, and the second training dataset may be generated based on or including unclassified unknown data. It can be.

Referring to Table 1 below, in relation to the present invention, it can be seen as showing the defect inspection results according to the learning model.

Prod#1 Prod#2 Unknown 2nd AI learning model 80.9% 84.2% 79.3% Prod#1First AI Learning Model 95.3% 61.9% 38.1% Prod#2
1st AI learning model 52.1% 97.6% 50.7%

Therefore, as described above, it is desirable to selectively select a learning model for a specified object and perform defect inspection on the object. That is, in the case of an unknown object in which the object is not specified, defect inspection is performed using the second learning model, and for Prod#1 among the previously stored data, the first learning model corresponding to Prod#1 is used among the first learning models. model, and for Prod#2 among the previously stored data, performing defect inspection by selectively selecting the first learning model corresponding to Prod#2 among the first learning models has the best performance and efficiency, and is the desired Defect inspection results can be obtained.

According to another embodiment, referring to FIGS. 6 and 7 , the defect inspection unit 610 of the electronic device 100 may have a different configuration from the defect inspection unit shown in FIG. 3 . That is, in FIG. 3, the defect inspection unit shows the learning unit 320 and the processing unit 330 as individual components, and each component is implemented in such a way that it individually includes a pre-processing module, a feature extraction module, and an inspection processing module. However, in FIG. 6 and In 7, the pre-processing module 720, the feature extraction module 740, and the inspection processing module 750 are shared by the defect inspection unit 610 to perform the functions of the learning unit and processing unit of FIG. 3. At this time, the defect inspection processing unit 610 of FIGS. 6 and 7 includes a model selection module 710 as shown in FIG. 5.

Meanwhile, as in operation 16 of FIG. 2, the defect inspection result data is a defect portion detected on the image data of the object in the same manner as, for example, Figure 8 or Figure 9 (c) to Figure 9 (d). It can be provided in a way that displays . However, the present invention is not limited to this provision method. For example, text, audio, images (graphs), etc. related to the type of defect detected in the above manner, degree of defect, number of defects, defect ratio, etc. may be added or provided separately.

FIGS. 8(a) to 8(b) show defect inspection results from image data of an object using the above-described second defect inspection operation mode, that is, the second learning model described in FIGS. 4 to 5. On the other hand, FIGS. 8(c) to 8(d) show defect inspection results from image data of the object using the above-described first defect inspection operation mode, that is, the first learning model specialized for the object described in FIGS. 4 to 5. represents.

9(a) to 9(b) show the results of inspecting for defects in image data of an object using only one pre-generated learned model. On the other hand, Figures 9(c) to 9(d) show the results of inspecting for defects in the image data of the object by selectively selecting a learning model from a plurality of learning models according to the present invention described above. .

Rectangular portions 811 to 844 in the image data shown in FIGS. 8(a) to 8(d) and rectangular portions 911 in the image data shown in FIGS. 9(a) to 9(d) to 941) indicate a defect detection part.

In relation to the scope of the present invention, the processor may represent all of the input unit, learning unit, processing unit, and output unit shown in FIGS. 3 to 7, or may represent at least one or more of them, depending on the context.

The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or a combination thereof. The software module may be RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, hard disk, removable disk, CD-ROM, or It may reside on any type of computer-readable recording medium well known in the art to which the present invention pertains.

Above, embodiments of the present invention have been described with reference to the attached drawings, but those skilled in the art will understand that the present invention can be implemented in other specific forms without changing its technical idea or essential features. You will be able to understand it. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive.

100: electronic device
150: image acquisition device
310: input unit
320: Learning Department
330: processing unit
340: output unit
510: model selection module

Claims (10)

In a device for performing defect inspection using a learning model,
Memory; and
Including a processor that inspects the object for defects,
The processor,
a processing unit that performs defect inspection of the object based on at least one learning model selected from among a plurality of learning models for defect inspection of the object stored in the memory;
a preprocessing module that preprocesses a training dataset corresponding to a plurality of learning variables for defect inspection of the object stored in the memory;
a feature extraction module that extracts features from the preprocessed data; and
A processing module that generates an individual learning model corresponding to the training dataset based on the extracted features,
The plurality of learning models are,
For Prod#1 among the previously stored data, a defect check is performed by selecting the first learning model corresponding to Prod#1,
For Prod#2 among the unfiltered data, select a first learning model corresponding to Prod#2 and perform defect inspection,
A device characterized in that the defect type, defect degree, number of defects, and defect ratio corresponding to the defect portion are displayed in a rectangular area in the image data of the object. According to paragraph 1,
The plurality of learning models are,
A first training dataset that includes individual training data for pre-classified learning variables among the plurality of learning variables, and a first training dataset or a learning variable that is not pre-classified among the plurality of learning variables. 2. An apparatus comprising a training dataset. According to paragraph 2,
The plurality of learning models are,
An apparatus comprising a first learning model corresponding to the first training dataset, and a second learning model corresponding to the second training dataset. According to paragraph 3,
The plurality of learning models are,
In the case of an unknown object in which the object is not specified, the device is characterized in that defect inspection is performed using a second learning model. According to paragraph 4,
The processor,
an input unit that receives image data of the object; and
an output unit providing defect inspection results of the object; A device further comprising: In a method of performing defect inspection using a learning model performed by a device,
performing defect inspection of an object based on at least one learning model selected from among a plurality of learning models for defect inspection of the object;
Preprocessing a training dataset corresponding to a plurality of learning variables for defect inspection of the object;
extracting features from the preprocessed data; and
generating an individual learning model corresponding to the training dataset based on the extracted features; Including,
The plurality of learning models are,
For Prod#1 among the previously stored data, a defect check is performed by selecting the first learning model corresponding to Prod#1,
For Prod#2 among the unfiltered data, select a first learning model corresponding to Prod#2 and perform defect inspection,
A method characterized in that the defect type, defect degree, defect number, and defect ratio corresponding to the defect portion are displayed in a rectangular area in the image data of the object. According to clause 6,
The plurality of learning models are,
A first training dataset that includes individual training data for pre-classified learning variables among the plurality of learning variables, and a first training dataset or a learning variable that is not pre-classified among the plurality of learning variables. 2 Method, characterized in that it includes a training dataset. In clause 7,
The plurality of learning models are,
A method comprising a first learning model corresponding to the first training dataset, and a second learning model corresponding to the second training dataset. According to clause 8,
The plurality of learning models are,
In the case of an unknown object in which the object is not specified, the method is characterized in that defect inspection is performed using a second learning model. According to clause 9,
Receiving image data of the object; and
providing defect inspection results of the object; A method further comprising:

Images