Inspection device and inspection method

Abstract

To provide an inspection device that can more accurately inspect the appearance of an object. [Solution] An inspection apparatus according to one aspect of the present disclosure includes: an acquisition unit that acquires an image of an object; a difference calculation unit that generates a difference map based on a comparison image which is an image of the object stored in advance and the image; a learning unit that generates a learning model that outputs the degree of abnormality of the object shown in the image in response to the input of the image by performing machine learning using a plurality of the comparison images; an abnormality calculation unit that generates an abnormality map for the object using the learning model; a coupling unit that generates a combined abnormality map based on the difference map and the abnormality map; and a determination unit that determines the state of the object based on the feature portion included in the combined abnormality map.

Description

Technical Field

[0001] The present disclosure relates to an inspection apparatus and an inspection method.

Background Art

[0002] Techniques for detecting abnormalities such as defects included in an evaluation object by image processing are known. In Patent Document 1, a reference cell image in which an image of a non-defective product of an object is photographed in advance and divided into an arbitrary number of cells is generated, and a plurality of learning data in which elements on the data of the reference cell image are changed within a certain range are used. A technique for detecting whether an object is non-defective based on a determination criterion generated thereby is disclosed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When detecting an abnormality by image processing, the resolution of the image processed for determination may not be good with respect to the resolution of the captured image, and the size and shape of an abnormal region such as a defect may become unclear. Therefore, when inspecting an object based on an inspection criterion regarding the size and shape of an abnormal region, there is a risk that an accurate result cannot be obtained.

[0005] An object of the present disclosure is to provide an inspection apparatus capable of more accurately inspecting the appearance of an object.

Means for Solving the Problems

[0006] An inspection apparatus according to one aspect of the present disclosure includes: an acquisition unit that acquires an image of an object; a difference calculation unit that generates a difference map based on a comparison image which is an image of the object stored in advance and the image; a learning unit that generates a learning model that outputs the degree of abnormality of the object shown in the image in response to the input of the image by performing machine learning using a plurality of the comparison images; an abnormality calculation unit that generates an abnormality map for the object using the learning model; a coupling unit that generates a combined abnormality map based on the difference map and the abnormality map; and a determination unit that determines the state of the object based on the feature portion included in the combined abnormality map. [Effects of the Invention]

[0007] The inspection device described herein allows for more accurate inspection of the appearance of an object. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing the functional configuration of an inspection device according to one embodiment of the present disclosure. [Figure 2] This figure shows an example of a convolutional neural network. [Figure 3] These figures illustrate examples of the figures generated by an inspection apparatus according to one embodiment of this disclosure. [Figure 4] This figure shows the hardware configuration of an inspection device according to one embodiment of the present disclosure. [Figure 5] This is a flowchart showing the processing flow in the learning phase performed by an inspection device according to one embodiment of the present disclosure. [Figure 6] A flowchart illustrating the flow of an inspection method performed by an inspection apparatus according to one embodiment of this disclosure. [Figure 7] This figure illustrates the process of generating a coupling abnormality map performed by the coupling section of an inspection device relating to one embodiment of this disclosure. [Figure 8]This figure illustrates the effect of using a mask image in the coupling abnormality map generation process according to one embodiment of the present disclosure. [Figure 9] A flowchart showing the flow of the determination process executed by the determination unit of an inspection device according to one embodiment of this disclosure. [Figure 10] This figure shows an example of a binarized image according to this embodiment. [Figure 11] This figure shows an example of a feature list in this embodiment. [Figure 12] This figure shows an example of a detection target region in an image captured according to one embodiment of this disclosure. [Figure 13] This figure shows an example of a display screen showing the result of determining the state of an object displayed by the display unit of an inspection device according to one embodiment of this disclosure. [Modes for carrying out the invention]

[0009] The embodiments for carrying out the invention will be described below with reference to the drawings. In each drawing, the same reference numerals are used for identical components, and redundant explanations may be omitted.

[0010] [Embodiment] <Configuration of inspection device 10> Figure 1 is a block diagram showing the functional configuration of an inspection device 10 according to one embodiment of the present disclosure. The inspection device 10 comprises an acquisition unit 11, a difference calculation unit 12, a learning unit 13, an abnormality calculation unit 14, a coupling unit 15, a determination unit 16, a display unit 17, and a storage unit 18.

[0011] The inspection device 10 inspects the appearance of an object, such as a product that has been manufactured, processed, etc., based on the presence or absence of scratches, painting defects, etc. on the surface of the object. More specifically, the inspection device 10 generates a difference map based on the difference between the captured image of the object and a comparison image stored in advance, and generates an abnormality degree map using a learning model. The inspection device 10 multiplies the difference map and the mask data obtained by performing threshold processing on the abnormality degree map to generate a combined abnormality degree map. Then, the inspection device 10 determines the object based on the feature amount included in the combined abnormality degree map.

[0012] Hereinafter, each part included in the inspection device 10 will be described. The acquisition unit 11 acquires a captured image of the object. The object is a product manufactured and processed as described above, but is not limited thereto and can be various inspection objects.

[0013] The difference calculation unit 12 generates a difference map based on a comparison image that is an image of the object stored in advance and the captured image of the object acquired by the acquisition unit 11. More specifically, the difference calculation unit 12 generates a difference map based on the difference between the averaged image 20 obtained by averaging a plurality of comparison images and the captured image. The comparison image is assumed to be an image in which the appearance of the object is in a normal state, but is not limited thereto. Note that a larger number of comparison images for generating the averaged image 20 contributes more to the generation of an accurate difference map, but it is desirable to select the number of images in consideration of the processing time. The difference map is generated by calculating the difference in pixel units included in each of the captured image and the comparison image. Also, the difference map may be generated using a background difference method. The difference map may be a two-dimensional map indicating the position of abnormal portions in the captured image.

[0014] The learning unit 13 generates a learning model 19 that outputs the degree of abnormality of the object shown in the captured image for the input of the captured image by performing machine learning using a plurality of comparison images. The learning model 19 is a model that has been pre-learned to compress the dimensions of the input data. For example, it is generated using a convolutional neural network (CNN: Convolutional Neural Network) that is useful for extracting the feature amounts of the input image.

[0015] FIG. 2 is a diagram showing an example of a convolutional neural network. The learning model 19 is generated by learning a convolutional neural network as illustrated by a supervised learning method using a plurality of comparison images as learning data. The convolutional neural network includes an input layer, one or more convolutional layers and pooling layers, a fully-connected layer, and an output layer using a softmax function as an activation function.

[0016] In FIG. 1, the abnormality degree calculation unit 14 generates an abnormality degree map for the object using the learning model 19 generated by the learning unit. More specifically, the abnormality degree calculation unit 14 inputs a plurality of comparison images included in the learning data to the convolutional neural network that is the learning model 19 and obtains the output result. The abnormality degree map is calculated as the distance from the feature amount distribution extracted by the convolutional neural network from a plurality of comparison images. The abnormality degree map may be a two-dimensional map showing the positions of abnormal portions in the captured image. Also, although a larger number of comparison images for generating the abnormality degree map contributes more to the generation of a more accurate abnormality map, it is desirable to select the number of images in consideration of the processing time.

[0017] The merging unit 15 generates a combined anomaly score map based on the difference map and anomaly score map of the target object. The merging unit 15 generates a mask image by processing the anomaly score of regions in the anomaly score map where the anomaly score is below a threshold to 0, and generates a combined anomaly score map by multiplying the mask image and the difference map. Through this process, noise contained in the difference map is suppressed and defects contained in the target object are extracted.

[0018] The determination unit 16 determines the state of the object based on the feature portions included in the coupling anomaly score map generated by the coupling unit 15. More specifically, the determination unit 16 extracts feature portions from the coupling anomaly score map whose anomaly score is above a threshold, and determines the state of the object by considering the portion of the extracted feature portion whose feature quantity is above a predetermined value as a defective portion. In other words, feature portion A can be said to be a portion that is a candidate for a defective portion.

[0019] The determination unit 16 creates a feature list, which is a list of feature parts extracted from the defective parts. Among the feature parts, the parts whose feature quantity is greater than or equal to a predetermined value are considered defective parts. The determination unit 16 determines that there is an abnormality in the object if the number of defective parts in the combined abnormality map is greater than or equal to a predetermined number. The feature quantity may be at least one of the pixel value and the area of ​​the feature part, but other data such as color data such as RGB values ​​may also be used. The pixel value is calculated based on the spectral sensitivity characteristics of the imaging device that captured the object and the brightness of each image.

[0020] The display unit 17 displays the result of the determination made by the determination unit 16. The display unit 17 may include a display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display that displays the result of the determination using characters and images.

[0021] The memory unit 18 stores the learning model 19 generated by the learning unit 13. The memory unit 18 also stores an averaged image 20 obtained by averaging multiple comparison images of the target object that were acquired in advance. The memory unit 18 is composed of, for example, ROM (Read Only Memory) and RAM (Random Access Memory).

[0022] Figure 3 is a diagram illustrating examples of the figures generated by an inspection apparatus 10 according to one embodiment of the present disclosure. In Figure 3, (a) is the image of the object, (b) is the difference map, (c) is the anomaly score map, (d) is the mask image, and (e) is the combined anomaly score map.

[0023] In Figure 3(a), A represents the characteristic part of the object. As shown in Figure 3(b), the difference map contains noise indicated by B in the figure. Such noise B is caused by ambient light at joints between parts of the object, for example. The difference map generated by the difference calculation unit 12 is generated by calculating the difference between the comparison image and the captured image, so it reflects the shape information of the characteristic part A, but it may contain noise B such as variations in the shape of the object and ambient light.

[0024] Furthermore, the anomaly score map generated by the anomaly score calculation unit 14 has an unclear range for feature portion A due to the effects of convolution processing, etc., making it difficult to accurately extract relatively small feature portion A. As a result, as shown in Figure 3(c), the feature portion A displayed in the anomaly score map is larger than its actual range, and the shape of feature portion A is unclear. On the other hand, since the anomaly score map calculates the anomaly score based on a feature distribution derived from multiple comparison images that include variability, it is less susceptible to noise and tends to be a map that includes only feature portion A.

[0025] Therefore, a mask image like the one shown in Figure 3(d) is generated. As described above, the mask image is generated by processing the anomaly score of the region in the anomaly score map that is below a threshold to 0. As a result, only feature A included in the anomaly score map is displayed in the mask image.

[0026] The difference map and mask image are combined to generate the combined anomaly score map shown in Figure 3(e). In the combined anomaly score map, the noise B displayed in the difference map is not included, and feature A is displayed more clearly in a state closer to its actual shape than in the anomaly score map.

[0027] Therefore, the joining unit 15 generates a joined anomaly score map by multiplying the difference map and the mask image based on the anomaly score map. The joined anomaly score map displays feature portion A with a more accurate shape.

[0028] Figure 4 shows the hardware configuration of an inspection device 10 according to one embodiment of the present disclosure. The inspection device 10 includes an input device 101, a display device 102, an external I / F (Interface) 103, RAM 104, ROM 105, a CPU (Central Processing Unit) 106, a communication I / F 107, and an HDD 108, each connected by bus B.

[0029] The input device 101 and the display device 102 may be connected as needed. In the inspection device 10, the RAM 104, ROM 105, and CPU 106 realize the functions of each part of the inspection device 10.

[0030] The input device 101 includes a touch panel, operation keys and buttons, a keyboard and mouse, etc., used by the user to input various signals. The display device 102 consists of a display such as a liquid crystal or organic EL that displays the screen, and a speaker that outputs sound data such as voice and music. The communication I / F 107 is an interface for the inspection device 10 to communicate data via an external network.

[0031] Furthermore, the HDD108 is an example of a non-volatile storage device that stores programs and data. The programs and data stored include the OS, which is the basic software that controls the entire inspection device 10, and applications that provide various functions on the OS. The inspection device 10 may also use a drive device that uses flash memory as the storage medium instead of the HDD108. For example, an SSD could be used as the drive device.

[0032] External I / F 103 is an interface to an external device. An external device could be a recording medium 103a. This allows the inspection device 10 to read from and write to the recording medium 103a via the external I / F 103. The recording medium 103a could be a flexible disk, CD, DVD, SD memory card, USB memory, etc.

[0033] RAM104 is an example of volatile semiconductor memory that temporarily holds programs and data. ROM105 is an example of non-volatile semiconductor memory that can retain programs and data even when the power is turned off. ROM105 stores programs and data such as the BIOS, OS settings, and network settings that are executed when the test device 10 is started up.

[0034] The CPU 106 is a computing unit that realizes the overall control and functions of the inspection device 10 by reading programs and data from storage devices such as the ROM 105 and HDD 108 onto the RAM 104 and executing processing.

[0035] <Processing flow executed by inspection device 10> The inspection method performed by the inspection device 10 will be described below with reference to Figures 5 and 6. Figure 5 is a flowchart showing the processing flow in the learning phase performed by the inspection device 10 according to one embodiment of the present disclosure. Figure 6 is a flowchart showing the flow of the inspection method performed by the inspection device 10 according to one embodiment of the present disclosure.

[0036] In the learning phase, a learning model 19 and an averaged image 20 are generated using comparison images, which are images of the target object taken in advance to generate difference maps and anomaly maps. First, comparison images are obtained in which the appearance of the target object is in a normal state (S101).

[0037] The learning unit 13 generates a learning model 19 that outputs the degree of abnormality of an object shown in an captured image, based on the input image of the captured image, by performing machine learning using multiple comparison images (S102). The difference calculation unit 12 generates an averaged image 20 by averaging the multiple comparison images (S103). The comparison images used to generate the averaged image 20 may be, for example, around 100 images, but are not limited to this.

[0038] Once the learning phase is complete and the learning model 19 and averaged image 20 are generated, the process moves on to the main process. As shown in Figure 6, first the acquisition unit 11 acquires the captured image of the target object (S201).

[0039] The anomaly calculation unit 14 generates an anomaly map for the target object using the learning model generated by the learning unit 13 (S202). The difference calculation unit 12 generates a difference map based on the difference between the averaged image 20, which was generated based on a previously stored comparison image, and the captured image (S203). Note that the processing in step 202 and the processing in step S203 may be performed in reverse order.

[0040] The combining unit 15 generates a combined anomaly score map by multiplying the difference map and the mask image generated from the anomaly score map (S204). The determination unit 16 extracts feature parts A in the combined anomaly score map whose anomaly score is above a threshold and creates a feature list (S205). The determination unit 16 determines the state of the object based on the feature quantities in the feature parts included in the combined anomaly score map (S206). The display unit 17 displays the result of determining the state of the object (S207).

[0041] These steps carry out the inspection method according to one embodiment of the present invention. However, the inspection method according to one embodiment of the present invention may include other steps as appropriate, depending on the measurement conditions, measurement environment, etc.

[0042] <Generation process for linkage anomaly map: S204> Here, the process for generating a coupling abnormality map will be explained. Figure 7 is a diagram illustrating the coupling abnormality map generation process performed by the coupling unit 15 of the inspection device 10 according to one embodiment of this disclosure. Below, the details of the process in step 204 in the flowchart shown in Figure 6 will be explained.

[0043] In Figure 7, (a) is a graph showing the distribution of anomalies in the anomaly score map, (b) is a graph showing the distribution of anomalies in the difference map, (c) is a graph showing the distribution of anomalies in the mask image, and (d) is a graph showing the distribution of anomalies in the combined anomaly score map. In each graph, the vertical axis represents the anomaly score and the horizontal axis represents the position. In the figure, th represents the anomaly score threshold.

[0044] As shown in Figure 7(a), the anomaly map shows that the anomaly score exceeds the threshold th in a region C that is wider than the actual feature region A. Furthermore, as shown in Figure 7(b), the difference map shows that noise B is included in regions other than feature region A.

[0045] The coupling unit 15 generates a mask image by processing the abnormality of the region where the abnormality is below the threshold th in the abnormality map to set the abnormality to 0. As a result, the abnormality of the region where the abnormality is below the threshold th, as shown in Figure 7(a), becomes 0, and in the mask image, the abnormality is 0 in all regions except region C, as shown in Figure 7(c).

[0046] The merging unit 15 then performs a process of multiplying the difference map and the mask image. As a result, the distribution of anomalies in the merged anomaly map, which is generated by multiplying the anomaly stats in Figures 7(b) and (c), is as shown in Figure 7(d).

[0047] As a result of the coupling unit 15 performing the above processing, the coupling anomaly score map shows that the anomaly score exceeds the threshold th only in feature portion A. Therefore, in the coupling anomaly score map, noise B is removed, and feature portion A, which is a narrower region than region C, is accurately represented.

[0048] Figure 8 is a diagram illustrating the effect of using a mask image in the coupling anomaly map generation process according to one embodiment of the present disclosure. The captured image on which the image shown in Figure 8 is based is different from the captured image shown in Figure 3. In Figure 8, (a) is the coupling anomaly map according to the example, and (b) is the coupling anomaly map according to the comparative example. In the coupling anomaly map according to the example, there is a difference in the degree of anomaly between feature portion A and noise B. In contrast, in the coupling anomaly map according to the comparative example, the degree of anomaly is approximately the same for feature portion A and noise B.

[0049] As described above, in this embodiment, the mask image does not change the anomaly score in regions where the anomaly score in the anomaly score map is equal to or greater than the threshold th. When generating the combined anomaly score map, if a difference map and an anomaly score map that has undergone binarization processing are combined without using a mask image, it is difficult for a difference in the anomaly score between defective area A and noise B to occur.

[0050] Therefore, when generating a combined anomaly score map, it is effective to use a mask image in which anomalies above a threshold th in the anomaly score map remain unchanged, from the viewpoint of removing noise B and accurately displaying feature parts A.

[0051] <Decision process: S206> Figure 9 is a flowchart showing the flow of the determination process performed by the determination unit 16 of the inspection device 10 according to one embodiment of this disclosure. The details of the process in step 206 in the flowchart shown in Figure 6 will be described below.

[0052] The determination unit 16 creates a feature list, which is a list of feature parts included in the object (S301). In creating the feature list, the determination unit 16 generates a binarized image by performing a binarization process on the joint anomaly score map and extracts feature part A.

[0053] Figure 10 shows an example of a binarized image according to this embodiment. In Figure 10, (a) shows an example of a concatenation error map that forms the basis of the binarized image, and (b) shows an example of a binarized image. Note that the captured image on which the image shown in Figure 10 is based is different from the captured image shown in Figures 3 and 8. In the illustrated example, there are three feature areas A. The binarized image is an image displayed in black and white, and the parts displayed in white are feature areas A. Note that binarized images are not limited to black and white, and various display methods can be used.

[0054] Figure 11 shows an example of a feature list in this embodiment. As shown in the figure, the feature list displays the area, position, and pixel value of each of the three feature parts A as feature quantities, but various parameters such as color data may also be used for display. The feature quantities of feature parts A may be measured using a binarized image. In addition, the position of each feature part A may be displayed using an XY coordinate system.

[0055] Next, the determination unit 16 performs a determination process to determine whether each feature portion shown in the feature list corresponds to a defective portion. For the determination process, a detection target region R, which is the area to be determined, is set in the captured image.

[0056] Figure 12 shows an example of a detection target area R in an image captured according to one embodiment of the present disclosure. As shown in the figure, the area shown in white in the captured image of the object is the detection target area R. The inspection device 10 according to this embodiment does not include feature portions A in areas other than the detection target area R in the determination process. Note that the method of displaying the detection target area R is not limited to the example shown.

[0057] As shown in the flowchart of Figure 9, if feature portion A is in an area other than the detection target area R (Yes in S302), feature portion A is excluded from the feature list (S303). Also, if feature portion A is in the detection target area R (No in S302), the determination unit 16 calculates the feature quantity of each feature portion A and compares it with a threshold (S304). In the following example, the feature quantity threshold is assumed to be a pixel value of 100 and the area of ​​the missing feature portion A is 40px, but it is not limited to these examples.

[0058] If the pixel value of feature part A is 100 or less, or if the area of ​​feature part A is 40px or more (Yes in S304), feature part A is deemed to be a defective part by assigning an "NG label" to it, etc. (S305). If the pixel value of feature part A is greater than 100 and the area of ​​feature part A is less than 40px (No in S304), feature part A is assigned an "OK label" and is not deemed a defective part, remaining as a feature part (S306).

[0059] The determination unit 16 determines that there is an abnormality in the object if the number of defective parts in the bonding abnormality map is greater than or equal to a predetermined number. In the inspection apparatus 10 according to this embodiment, the predetermined number is set to one, but it is not limited to this and any number can be set. If there is one or more defective parts with the "NG label" assigned to them among the three feature parts A (Yes in S307), the determination unit 16 determines that there is an abnormality in the object (S308). If there are no defective parts (No in S307), the determination unit 16 determines that there is no abnormality in the object (S309) and completes the process.

[0060] <Displaying the result: S207> Figure 13 is a diagram showing an example of a display screen showing the result of determining the state of an object displayed by the display unit 17 of an inspection device 10 according to one embodiment of this disclosure. The details of the process in step 207 in the flowchart shown in Figure 6 will be explained below.

[0061] As shown in the figure, the display screen includes the judgment result, a feature list, and a judgment image, but the items displayed are not limited to these. The judgment result is the state of the object as determined by the judgment unit 16. In the illustrated example, the judgment unit 16 determined that there was an abnormality in the object.

[0062] The feature list in the figure includes a column that, in addition to the table shown in Figure 11, displays the result of the judgment process for each feature part A, indicating whether it should remain a feature part or be classified as a defect. In the illustrated example, of feature parts A, No. 1 and No. 3 are determined to remain feature parts based on their feature quantities, while No. 2 is determined to be a defect.

[0063] Furthermore, in the judgment image shown in the figure, each feature portion A is displayed superimposed on the captured image, which is the image of the object. In the judgment image, "Defect 1," "Defect 2," and "Defect 3" correspond to "NO.1," "NO.2," and "NO.3" in the feature list, respectively. The position where defect portion A is displayed may be indicated based on coordinates that indicate the position in the feature list. As shown in the figure, feature portion A and the defect portion may be depicted by circles and outlines, etc., at the position corresponding to the coordinates that indicate the position, or feature portion A and the defect portion may be displayed in different colors. Note that the method of displaying defects in the judgment image is not limited to the example shown.

[0064] <Effects> The inspection apparatus 10 according to this disclosure generates a combined anomaly score map using a difference map generated using a comparison image, which is an image of the object in a normal state, and an anomaly score map generated using a learning model. The combined anomaly score map removes noise displayed in the difference map and can display defect areas of a more accurate size than the anomaly map, thus displaying defect areas of a more accurate shape. Therefore, the inspection apparatus 10 according to this embodiment can inspect the appearance of the object more accurately.

[0065] Furthermore, since the feature quantities used in the determination process for determining whether or not to identify a characteristic part A of the object as a defective part can be arbitrarily set, it is possible to set determination rules according to the shape of various objects. Therefore, it becomes possible to apply the determination process performed by the determination unit 16 to processes with complex inspection criteria, and the basis for the determination in the determination process becomes clear.

[0066] As described above, embodiments have been explained, but these embodiments are presented as examples only, and the present invention is not limited by these embodiments. The above embodiments can be implemented in various other forms, and various combinations, omissions, substitutions, and modifications are possible without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of Symbols]

[0067] 10 Inspection equipment 11 Acquisition Department 12 Difference calculation part 13. Learning Department 14 Abnormality calculation unit 15 Joint 16 Judgment section 17 Display 18 Memory section 19. Learning Models 20 Averaged images A. Feature section

Claims

1. An acquisition unit that acquires an image of the target object, A difference calculation unit generates a difference map based on a comparison image, which is a pre-stored image of the object, and the captured image. A learning unit generates a learning model that outputs the degree of abnormality of the object shown in the captured image in response to the input of the captured image by performing machine learning using multiple comparison images. An anomaly calculation unit that generates an anomaly map for the object using the learning model, A coupling unit that generates a combined anomaly map based on the difference map and the anomaly map, A determination unit that determines the state of the object based on the characteristic portion included in the coupling abnormality map, An inspection device equipped with the following features.

2. The coupling unit generates a mask image by processing the abnormality of the region in the abnormality map where the abnormality is below a threshold to set the abnormality to 0, and generates the combined abnormality map by processing the mask image and the difference map together. The inspection apparatus according to claim 1.

3. The difference calculation unit generates the difference map based on the difference between the averaged image, which is obtained by averaging the multiple comparison images, and the captured image. The inspection apparatus according to claim 1.

4. The determination unit extracts feature portions from the bond abnormality map whose abnormality level is equal to or greater than a threshold, and determines the state of the object by considering the portion of the extracted feature portions whose feature quantity is equal to or greater than a predetermined value as a defective portion. The inspection apparatus according to claim 1.

5. The determination unit, If the number of defective portions in the bonding abnormality map is greater than or equal to a predetermined number, it is determined that there is an abnormality in the object. The inspection apparatus according to claim 4.

6. The aforementioned feature quantity is at least one of the pixel value and the area of ​​the defect portion. The inspection apparatus according to claim 4.

7. The system further includes a display unit that displays the result of the determination made by the determination unit, The inspection apparatus according to claim 1.

8. An inspection method performed by an inspection device, The steps include: acquiring an image of the object, The steps include generating a difference map based on a comparison image, which is an image of the object stored in advance, and the captured image, The steps include: generating a learning model that outputs the degree of abnormality of the object shown in the captured image in response to the input of the captured image by performing machine learning using multiple comparison images; The steps include generating an anomaly map for the object using the aforementioned learning model, A step of generating a combined anomaly map based on the difference map and the anomaly map, The steps include determining the state of the object based on the feature portion included in the coupling abnormality map, Testing methods including those mentioned.

Images