Object appearance inspection apparatus and object appearance inspection method

Abstract

The object appearance inspection method includes: photographing an object appearance to obtain an image of the object appearance; using a rule check on the image of the appearance of the object in order to perform a preliminary defect determination; using deep learning on the image preliminarily determined to be defective in order to perform secondary defect determination; and performing a third defect determination on the image preliminarily determined to be defective or the image secondarily determined to be defective. During the preliminary defect determination, even if the object is changed to another object or the inspection condition of the object is changed, it is possible to preliminarily determine that the image of the appearance of the object is defective in real time.

Description

Technical Field

[0001] This embodiment relates to an object appearance inspection device and an object appearance inspection method. Background Art

[0002] Products manufactured in a manufacturing factory are transported after defect inspection in the product appearance.

[0003] As customers or consumers become more and more interested in the appearance of products, defect inspection in the product appearance becomes very important.

[0004] When performing defect inspection of the product appearance with the naked eye, since the defect detection in the product appearance varies depending on the experience and eyesight of the inspector, the reliability of the detection decreases.

[0005] Recently, technologies for detecting defects in the product appearance using images of the product appearance have been proposed.

[0006] However, there is still a problem that the detection reliability decreases because the defect detection varies depending on the quality of the camera that acquires the image of the product appearance and the quality of the analysis device that analyzes the acquired image.

[0007] In addition, the product target can be changed or the inspection conditions of the product can be changed. In this case, there is a problem as follows: the predefined manual makes it difficult to respond to the product appearance inspection in real time. When the product target is changed or the inspection conditions of the product are changed, parameters or algorithms must be optimized to match the changed product or the changed inspection conditions. However, when the optimization of the parameters or algorithms is performed manually, it takes a lot of time and the defect determination process stops, making it difficult to respond in real time.

[0008] Disclosure

[0009] [Technical Problem]

[0010] This embodiment aims to solve the above problems and other problems.

[0011] Another object of this embodiment is to provide an object appearance inspection device and an object appearance inspection method capable of improving the reliability of defect detection of the object appearance.

[0012] In addition, another object of this embodiment is to provide an object appearance inspection device and an object appearance inspection method capable of detecting defects in the object appearance in real time even when the object is replaced by another object or the inspection conditions of the object are changed.

[0013] The technical problems of this embodiment are not limited to the content described herein, but include the content that can be understood through the description of the present invention.

[0014] [Technical Solution]

[0015] According to one embodiment, for achieving the above or other purposes, an object appearance inspection method includes: obtaining an image related to an object appearance; performing a first defect determination on the image related to the object appearance using a rule-based inspection method; performing a second defect determination on the image determined to be defective in the first defect determination using deep learning; and performing a third defect determination on the image determined to be defective in the first defect determination or the image determined to be defective in the second defect determination, wherein the execution of the first defect determination includes performing the first defect determination on the image related to the object appearance in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

[0016] The real-time execution of the first defect determination may include selecting the best algorithm for the first defect determination in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

[0017] The selection of the best algorithm may include assigning weights to experience data related to at least one engineer working at each of multiple locations; and selecting the best algorithm by considering the weighted experience data.

[0018] The real-time execution of the first defect determination may include selecting the best parameters for the first defect determination in real time even if the object is replaced by another object or the inspection object conditions are changed.

[0019] The real-time execution of the first defect determination may include performing the first defect determination in real time using the selected best algorithm or the selected best parameters.

[0020] The real-time execution of the first defect determination may include monitoring the performance of the first defect determination.

[0021] The execution of the first defect determination may include: collecting verification data; grouping the collected verification data into multiple groups; and extracting verification data from each of the multiple groups.

[0022] The object appearance inspection method may include classifying the object according to the second defect determination result or the third defect determination result.

[0023] According to another embodiment, an object appearance inspection device includes: a plurality of appearance inspectors configured to perform a first defect determination on an image related to an object appearance using a rule-based inspection method; a deep learning device configured to perform a second defect determination on an image determined to be defective in the first defect determination using deep learning; and an image determination device configured to perform a third defect determination on an image determined to be defective in the first defect determination or an image determined to be defective in the second defect determination, wherein each of the plurality of appearance inspectors includes a rule-based inspection unit configured to perform the first defect determination on the image related to the object appearance in real time even when the object is replaced by another object or the inspection conditions of the object are changed.

[0024] Even when the object is replaced by another object or the inspection conditions of the object are changed, the rule-based inspection unit can select the best algorithm for the first defect determination in real time.

[0025] The rule-based inspection unit can assign weights to experience data related to at least one engineer working at each of multiple locations, and select the best algorithm by considering the weighted experience data.

[0026] Even when the object is replaced by another object or the object inspection conditions are changed, the rule-based inspection unit can select the best parameters for the first defect determination in real time.

[0027] The rule-based inspection unit can perform the first defect determination in real time using the selected best algorithm or the selected best parameters.

[0028] Each of the plurality of appearance inspectors may include a performance monitoring unit configured to monitor the performance of the first defect determination.

[0029] The rule-based inspection unit can collect verification data, group the collected verification data into multiple groups, and extract verification data from each of the multiple groups.

[0030] Each of the plurality of appearance inspectors may classify the object according to the second defect determination result or the third defect determination result.

[0031] [Beneficial effects]

[0032] According to the embodiment, when the object is replaced by another object or the inspection conditions of the object are changed, at least one parameter value among a plurality of parameters can be optimized or the control value of the algorithm can be optimized or replaced by another algorithm, so that the defect inspection of the object appearance can continue in real time without stopping.

[0033] According to an embodiment, since the defect of the object appearance is determined three times, the accuracy of defect detection can be significantly improved, thereby improving the reliability.

[0034] According to an embodiment, by grouping the verification data of various distributions and extracting the verification data from each grouped set, the accuracy of verification can be improved by extensive sampling of the verification data.

[0035] According to the following detailed description, the additional scope of applicability of the embodiments will become apparent. However, since various changes and modifications within the spirit and scope of the embodiments can be clearly understood by those skilled in the art, it should be understood that specific embodiments such as the detailed description and the preferred embodiments are given only as examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Illustrates defect detection in an object appearance inspection device according to an embodiment.

[0037] Figure 2 Illustrates an object appearance inspection device according to an embodiment.

[0038] Figure 3 Is a flowchart illustrating an object appearance inspection method according to an embodiment.

[0039] Figure 4 Is an illustration of Figure 3 The flowchart of S220.

[0040] Figure 5 Describes a method for selecting the best parameters.

[0041] Figure 6 Describes a method for selecting the best algorithm.

[0042] Figure 7 Is a flowchart illustrating a method for improving the efficiency of verification data.

[0043] Figure 8 Is a schematic diagram illustrating a method for improving the efficiency of verification data.

[0044] Figure 9 Illustrates various parameters. DETAILED DESCRIPTION

[0045] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the accompanying drawings. Regardless of the reference numerals, the same or similar components will be assigned the same reference numerals, and redundant descriptions thereof will be omitted. For ease of writing the specification, the suffixes "module" or "component" for components used in the following description are assigned or used interchangeably, and do not have different meanings or roles by themselves. In addition, the drawings are intended to facilitate understanding of the embodiments disclosed in this specification, and the technical concepts disclosed in this specification are not limited by the drawings. Further, when an element such as a layer, a region, or a substrate is referred to as being "on" another element, this includes that it can be directly connected on another element or other intermediate elements can exist between them.

[0046] Figure 1 Describe defect detection in an object appearance inspection device according to an embodiment.

[0047] Refer to Figure 1 , an object appearance inspection device 100 according to an embodiment may include an appearance inspector 110, a deep learning device 120, and an image determination device 130. The deep learning device 120 may be included in the appearance inspector 110, but is not limited thereto.

[0048] The appearance inspector 110 may include a plurality of appearance inspectors 110a and 110b for mass inspection.

[0049] Each of the plurality of appearance inspectors 110a and 110b may perform a first defect determination on an image related to the object appearance using a rule inspection method. The rule inspection method may determine defects from an image using a predetermined algorithm based on a plurality of predetermined parameters ( Figure 9 ).

[0050] According to an embodiment, when the object is replaced with another object or the inspection conditions of the object are changed, at least one parameter value among the plurality of parameters may be optimized, or the control value of the algorithm may be optimized or replaced with a different algorithm, so that the defect inspection of the object appearance can continue in real time without stopping.

[0051] The deep learning device 120 may perform a second defect determination on the image determined to be defective using deep learning based on the first defect determination result. For example, a plurality of feature information may be extracted from the corresponding image, and the plurality of feature information may be input and learned to output determination information such as whether the product is defect-free or defective. The feature information includes the image brightness, color, saturation, size, and shape of the defective object extracted from the image. Deep learning can be applied to existing deep learning technologies and deep learning technologies developed or applied in the future.

[0052] The image determination device 130 may perform a third defect determination on an image that has been determined to be defective as a result of the first defect determination or the second defect determination. The image determination device 130 may determine whether an image is defective through visual observation by an inspector. For example, by an inspector visually observing the corresponding image and inputting information on whether it is defect-free or defective, the third defect determination result of the image can be obtained.

[0053] Meanwhile, the appearance inspector 110 may classify an object as a defect-free or defective product and discharge it based on the result of the defect determination for the appearance of the object. That is, although not shown, at least two discharge ports are provided, and a defect-free object may be classified and discharged through one discharge port, and a defective object may be classified and discharged through another discharge port.

[0054] When it is determined by the appearance inspector 110 that the appearance of the object is defect-free, the object may be classified and discharged as a defect-free product.

[0055] When it is determined by the appearance inspector 110 that the appearance of the object is defective, the corresponding image may be input to the deep learning device 120, and a second defect determination may be performed. Alternatively, when it is determined by the appearance inspector 110 that the appearance of the object is defective, the feature information extracted from the image may be input to the deep learning device 120 to determine secondary defects.

[0056] When it is determined by the deep learning device 120 that the appearance of the object is defect-free, the result determined to be defect-free may be sent to the appearance inspector 110, and the object may be classified as defect-free and discharged. The appearance inspector 110 may classify the object as defect-free and discharge it based on the determination result provided as defect-free by the deep learning device 120 or the image determination device 130.

[0057] When it is determined by the deep learning device 120 that the appearance of the object is defective, the image may be input to the image determination device 130, and a third defect determination may be performed.

[0058] According to this embodiment, since the defects in the appearance of the object are determined three times, the accuracy of defect detection can be significantly improved, and the reliability can be improved.

[0059] Meanwhile, the object appearance inspection device 100 according to the embodiment may include a data collection device 150 and an additional inspection unit 160.

[0060] The data collection device 150 may be a device that automatically collects, classifies, and stores data generated or obtained from the appearance inspector 110, the deep learning device 120, etc.

[0061] The additional inspection unit 160 can additionally inspect or re-inspect the determination result of the appearance of an object through the image determination device 130. The additional inspection unit 160 can visually inspect the object itself with the remaining manpower.

[0062] Meanwhile, conventionally, when an object inspected by an object appearance inspection device during operation is replaced with another object or the inspection conditions of the object are changed, it is necessary to stop defect determination, making real-time defect determination difficult.

[0063] However, in this embodiment, even when the object inspected by the object appearance inspection device 100 during operation is replaced with another object or the inspection conditions of the object are changed, defect determination of an image related to the object appearance can be performed in real time. The inspection conditions may include various parameter settings for inspecting the object ( Figure 9 ) or various algorithms.

[0064] Hereinafter, real-time defect determination that can be performed regardless of object replacement or change of the inspection conditions of the object will be described in detail.

[0065] Figure 2 An object appearance inspection device according to an embodiment is shown.

[0066] Refer to Figure 2 , the object appearance inspection device 100 according to an embodiment may include an optical system instrument 111, a rule-based inspection unit 112, a performance monitoring unit 113, and a data collection unit 114. The optical system instrument 111, the rule-based inspection unit 112, the performance monitoring unit 113, and the data collection unit 114 may be provided in each of the plurality of appearance inspectors 110a and 110b, but are not limited thereto.

[0067] Although not shown, the optical system instrument 111 may be composed of at least one camera, an optical system, and a control device for controlling the optical system. The optical system instrument 111 may be provided to each of the appearance inspectors 110a and 110b, but is not limited thereto. In an embodiment, the optical system instrument 111 may acquire an image of each of the six sides of the object, but is not limited thereto. The object may have, for example, a hexahedron shape. In this case, images of the lower surface, the upper surface, and the four side surfaces of the object can be acquired. The optical system can be moved along the x-axis, y-axis, and z-axis by the control device, and the distance between the optical system and the camera can be adjusted.

[0068] The rule-based inspection unit 112 can perform a first defect determination on an image related to the object appearance using a rule inspection method. The rule inspection device refers to inspecting for defects in the object appearance by using various predetermined parameters and an algorithm for performing defect determination based on these parameters.

[0069] Even if the object is replaced by another object or the object condition is changed, the rule-based inspection unit 112 can also perform a first defect determination on an image related to the object appearance in real time.

[0070] Even if the object is replaced by another object or the inspection object condition is changed, the rule-based inspection unit 112 can also select the best algorithm for the first defect determination in real time.

[0071] Even if the object is replaced by another object or the inspection object condition is changed, the rule-based inspection unit 112 can also select the best algorithm for the first defect determination in real time. Here, the selection of the best parameters can include selecting the set value of the best parameters or selecting the number of the best parameters.

[0072] The rule-based inspection unit 112 can perform a first defect determination on an image related to the object appearance using the best algorithm and / or the best parameters. The first defect determination can be performed using the best algorithm, the first defect determination can be performed using the best parameters, or the first defect determination can be performed using the best algorithm and the best parameters.

[0073] The rule-based inspection unit 112 can perform data verification efficiency. For example, the rule-based inspection unit 112 can collect verification data, group the collected verification data into multiple groups, and extract verification data from each of the multiple groups. Using the verification data extracted in this way, the accuracy of the first defect determination according to the best parameters and / or the best algorithm can be verified.

[0074] Meanwhile, the performance monitoring unit 113 can monitor the performance of the first defect determination. The performance monitoring unit 113 can receive data related to the first defect determination from each of the multiple appearance inspectors 110a and 110b through the data collection unit 114, and monitor the performance of the first defect determination based on these data. The data related to the first defect determination can include various parameters, algorithms, the first determination results, etc., but is not limited thereto.

[0075] The data collection unit 114 can collect, classify, and store data related to the first defect determination from each of the multiple appearance inspectors 110a and 110b. The data collection unit 114 can be a database (DB) or a storage server, but is not limited thereto.

[0076] Figure 3 is a flowchart illustrating an object appearance inspection method according to an embodiment.

[0077] As Figure 1 and Figure 3 shown, the multiple appearance inspectors 110a and 110b can each acquire an image of the object appearance (S210). The multiple appearance inspectors 110a and 110b are each equipped withFigure 2 The optical system device 111 shown enables an image of an object's appearance to be obtained through the optical system device 111.

[0078] Multiple appearance inspectors 110a and 110b can perform a first defect determination (S220) using a rule-based inspection method based on an image related to the object's appearance.

[0079] The deep learning device 120 can perform a second defect determination (S230) using deep learning based on an image determined to be defective based on the result of the first defect determination.

[0080] The image determination device 130 is capable of performing a third defect determination (S240) based on an image determined to be defective based on the result of the first or second defect determination. For example, an image determined to be defective can be displayed on the image determination device 130, and the third defect determination can be performed by an inspector based on the displayed image, but is not limited thereto. For example, the image determination device 130 can perform the third defect determination by comparing an image determined to be defective with a reference image for determining a non-defective product using a mapping method. The mapping method is a method in which the reference image and the image determined to be defective are mapped one-to-one, and when the degree of mapping is lower than a set value, the image can be determined to be defective.

[0081] Figure 4 is a detailed description Figure 3 of the flowchart of S220.

[0082] As Figure 2 and Figure 4 shown, each of the multiple appearance inspectors 110a and 110b (i.e., the rule-based inspection unit 112) can check whether the object has been replaced by another object or whether the inspection conditions of the object's appearance have been changed (S221).

[0083] In a manufacturing factory that mass-produces products, the products may change frequently, or even within the same product, the design, shape, etc. may change. In addition, the inspection conditions of the object's appearance can also be changed to enable more reliable and accurate defect determination.

[0084] In this way, when the object (product) is changed or the inspection conditions of the object's appearance are changed, the parameters or algorithms required for performing the first defect determination can be changed. For example, the number of parameters can be changed, and parameters can be newly added or excluded. When the object is changed or the inspection conditions of the object's appearance are changed, the accuracy of the output value of the algorithm (i.e., the first defect determination) can be reduced. To this end, the control value of the algorithm can be changed, or a new algorithm can be established instead of the existing algorithm.

[0085] The rule-based inspection unit 112 can select the best algorithm (S222) and select the best parameters (S223). In the drawings, the selection of the best algorithm is shown to be performed before the selection of the best parameters, but conversely it can be performed. That is, the best algorithm can be selected after the best parameters are selected.

[0086] As Figure 6 As shown, the process can proceed in the following order: image loading (S241), algorithm input using (S242), algorithm processing (S243), and selection of the best algorithm by defect detection values and artificial intelligence (AI) (S245). The algorithm processing (S243) can process the image loaded in S241 by the algorithm input in S242 to calculate the defect detection values. Here, the defect detection values can be area, ratio, position difference, etc. Based on the defect detection values, the best algorithm can be selected by artificial intelligence (S245). When the best algorithm is selected, the best algorithm selection process can be terminated (S246). When the best algorithm is not selected, the process can be executed by moving to S242 and changing the control value of the algorithm or inputting another algorithm to select the best algorithm.

[0087] Meanwhile, the rule-based inspection unit 112 can assign weights to the experience data related to at least one engineer working at each of multiple locations, and select the best algorithm by considering the weighted experience data. Here, the locations can be companies, workplaces within the same company, etc. The engineers can be engineers working in fields such as planning algorithms, implementing algorithms, validating algorithms, applying algorithms, etc. to determine the defects of an object. Engineers working at numerous companies or various workplaces can construct various algorithms developed according to their occupations, capabilities, or levels.

[0088] Weights can be assigned to each of the algorithms constructed in this way. The algorithms to which weights are assigned can be limited to the currently used algorithms, but not limited to this. The algorithm with the highest weight among the weighted algorithms can be selected as the best algorithm. The verification of defect detection is performed by the best algorithm, and when it is not satisfied, the algorithm with the second-highest weight can be selected as the best algorithm. This process can be repeated to determine the algorithm to be applied to the rule-based inspection unit 112 of the embodiment. The algorithms can be input into S242 in the order of their respective weights from highest to lowest.

[0089] As Figure 5As shown, the process can proceed in the order of image loading (S231), matching parameter input (S232), image correction (S233), image deviation calculation (S234), and optimal parameter selection by artificial intelligence (AI) (S235). Image correction (S233) can calculate the image deviation by processing the image loaded in S231 with the matching parameters input in S232. Here, the image deviation can be the difference between the main sample image and the current image. The optimal parameters can be selected by artificial intelligence (S235) to reduce or make the image deviation zero. When the optimal parameters are selected, the optimal parameter selection process can be terminated (S236). When the optimal parameters are not selected, the process can be executed by moving to S232 and changing the value of the parameters or using another parameter to select the optimal parameters.

[0090] The above optimal algorithm and / or the above optimal parameters can be applied to the appearance inspector 110. That is, the first defect determination in the appearance inspector 110 can be performed by the above optimal algorithm and / or the above optimal parameters. To this end, the existing parameters of the appearance inspector 110 can be set to the above optimal parameters (or their values), and the existing algorithm can be replaced with the above optimal algorithm, or the control value of the above existing algorithm can be replaced with the control value of the above optimal algorithm.

[0091] The rule-based inspection unit 112 can use the above optimal algorithm and / or the above optimal parameters to perform the first defect determination (S224) in real time. For example, the above optimal algorithm can be used to perform the first defect determination in real time for an image regarding the appearance of an object. For example, the first defect determination can be performed in real time by applying the above optimal parameters to a previous algorithm for an image regarding the appearance of an object. For example, the first defect determination can be performed in real time by applying the above selected optimal parameters to the above selected optimal algorithm for an image regarding the appearance of an object.

[0092] Meanwhile, the performance monitoring unit 113 can monitor the performance of the first defect determination (S225). That is, the performance monitoring unit 113 can monitor the accuracy of the first defect determination in real time based on the first defect determination result and the corresponding image using the above selected optimal algorithm and / or the above selected optimal parameters.

[0093] The first defect determination result and the corresponding image made by the appearance inspector 110 can be collected and stored by the data collection unit 114, and then the first defect determination result and the corresponding image can be sent to the performance monitoring unit 113 in response to a request from the performance monitoring unit 113. Alternatively, the first defect determination result and the corresponding image made by the appearance inspector 110 can be directly sent to the performance monitoring unit 113.

[0094] Meanwhile, the rule-based inspection unit 112 can improve the efficiency of verifying data. When selecting the best parameters or the best algorithm, the accuracy of the first defect in the best parameters or the lowest algorithm must be verified. Therefore, when selecting the best parameters or the best algorithm, the best parameters or the best algorithm can be applied to the appearance inspector 110, and then the first defect can be determined for the appearance of each object within a given time. As a result of this defect determination, the accuracy of the first defect determination can be verified based on the determination data (verification data) such as defect-free, defective, false positive, or false negative. The term "false positive" may refer to a case where an object is determined to be defective by the object appearance inspection device 100 according to an embodiment but is subsequently verified as defect-free. The term "false negative" may refer to a case where an object is determined to be defect-free by the object appearance inspection device 100 according to an embodiment but is subsequently verified as defective.

[0095] In this way, a large amount of determination data can be obtained, and the determination data can be widely distributed according to the x-axis and y-axis variables. In this case, since it takes too much time to verify all the determination data, some of the determination data can be sampled, and the sampled data can be used to perform the verification. At this time, when extracting and verifying sample data in a limited area among the determination data widely distributed along the x-axis variable and the y-axis variable, there is a problem of reduced reliability of the accuracy of the first defect determination verified using the extracted sample data.

[0096] Hereinafter, a method for improving the reliability by increasing the accuracy of the first defect determination is described.

[0097] Figure 7 is a flowchart illustrating a method for improving the efficiency of verification data. Figure 8 is a schematic diagram illustrating a method for improving the efficiency of verification data.

[0098] As Figure 7 illustrated, the rule-based inspection unit 112 can collect inspection data (S310), group the collected inspection data into multiple groups (S320), and extract inspection data from each of the multiple groups (S330).

[0099] As Figure 8 shown, the defect-free data, defective data, and false positive data can be distributed according to the size of the foreign matter extracted from the object appearance and the luminance deviation of the image regarding the object appearance.

[0100] The variously distributed defect-free data, defective data, and false positive data can be grouped into multiple groups. As Figure 8As shown, they can be grouped into multiple groups according to the size of foreign objects. In this case, defect-free data, defective data, and false positive data can be distributed to each of the multiple groups.

[0101] Thereafter, verification data can be extracted from each of the multiple grouped groups. For example, false positive data can be extracted from each of the multiple groups illustrated in Figure 8 . Using the extracted false positive data, the accuracy of the first defect determination according to the optimal parameters and / or optimal algorithm can be verified.

[0102] According to the embodiment, by grouping the verification data of various distributions and extracting the verification data from each grouped set, the accuracy of verification can be improved by extensive sampling of the verification data.

[0103] The above detailed description should not be construed as restrictive in all respects and should be considered exemplary. The scope of this embodiment should be determined by a reasonable interpretation of the appended claims, and all changes within the equivalent scope of the embodiment are included within the scope of the embodiment.

Claims

1. An object appearance inspection method, comprising: Obtaining an image related to the appearance of the object; Performing a first defect determination on the image related to the object appearance using a rule-based inspection method; Performing a second defect determination on the image determined to be defective in the first defect determination using deep learning; And Performing a third defect determination on the image determined to be defective in the first defect determination or the image determined to be defective in the second defect determination; Wherein, the execution determination of the first defect includes: performing the first defect determination on the image related to the object appearance in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

2. The object appearance inspection method according to claim 1, wherein, The real-time execution of the first defect determination includes: selecting the best algorithm for the first defect determination in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

3. The object appearance inspection method according to claim 2, wherein, The selection of the best algorithm includes: Assigning weights to the experience data related to at least one engineer working at each of multiple locations; and Selecting the best algorithm by considering the weighted experience data.

4. The object appearance inspection method according to claim 2, wherein The real-time execution of the first defect determination includes: selecting the best parameters for the first defect determination in real time even if the object is replaced by another object or the inspection object conditions are changed.

5. The object appearance inspection method according to claim 4, comprising: Performing the first defect determination in real time using the selected best algorithm or the selected best parameters.

6. The object appearance inspection method according to claim 5, comprising: Monitoring the performance of the first defect determination.

7. The object appearance inspection method according to claim 1, wherein, The execution of the first defect determination includes: Collecting verification data, Grouping the collected verification data into multiple groups; and Extracting verification data from each of the multiple groups.

8. The object appearance inspection method according to claim 1, comprising: Classifying the object according to the second defect determination result or the third defect determination result.

9. An object appearance inspection device, comprising: Multiple appearance inspectors configured to perform a first defect determination on an image related to the object appearance using a rule inspection method; A deep learning device configured to perform a second defect determination on the image determined to be defective in the first defect determination using deep learning; And An image determination device configured to perform a third defect determination on the image determined to be defective in the first defect determination or the image determined to be defective in the second defect determination; Wherein, each of the multiple appearance inspectors includes a rule-based inspection unit configured to perform the first defect determination on the image related to the object appearance in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

10. The object appearance inspection device according to claim 9, wherein, The rule-based inspection unit is configured to select the best algorithm for the first defect determination in real time even if the object is replaced by another object or the inspection conditions of the object are changed.

11. The object appearance inspection device according to claim 10, wherein, The rule-based inspection unit is configured to assign weights to the experience data related to at least one engineer working at each of multiple locations and select the best algorithm by considering the weighted experience data.

12. The object appearance inspection device according to claim 10, wherein, The rule-based inspection unit is configured to select the best parameters for the first defect determination in real time even if the object is replaced by another object or the inspection object conditions are changed.

13. The object appearance inspection device according to claim 12, wherein, The rule-based inspection unit is configured to perform the first defect determination in real time using the selected best algorithm or the selected best parameters.

14. The object appearance inspection device according to claim 13, comprising: A performance monitoring unit, the performance monitoring unit being configured to monitor the performance of the first defect determination.

15. The object appearance inspection device according to claim 9, wherein, The rule-based inspection unit is configured to collect verification data, group the collected verification data into a plurality of groups, and extract verification data from each of the plurality of groups.

16. The object appearance detection device according to claim 9, wherein, Each of the plurality of appearance inspectors is configured to classify the object according to the second defect determination result or the third defect determination result.

Images