Quality inspection system, quality inspection device, quality inspection method, and quality inspection program

Abstract

To provide a technique capable of improving the precision of detection of a defect of a wall material.SOLUTION: A quality inspection system comprises: an acquisition part which acquires an image obtained by photographing a wall material; a division part which divides the image into a plurality of small images; a detection part which detects a defect included in the wall material by referring to a combination of at least two determination results obtained by inputting the small images to two machine learning models, having learned by different learning method, respectively; and an output part which outputs a detection result.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a quality inspection system, a quality inspection device, a quality inspection method, and a quality inspection program. [Background technology]

[0002] There are known techniques for detecting defects in building materials. For example, Patent Document 1 describes a decorative material repair device that detects defects in a decorative material that is a building material and has an uneven pattern on the surface of the base material. The decorative material repair device described in Patent Document 1 detects defects in the decorative material by comparing image data obtained by imaging the surface of the decorative material with pre-stored reference data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2008-201073 A Summary of the Invention [Problem to be solved by the invention]

[0004] The decorative material repair device described in Patent Document 1 is not designed to detect small dents on large wall materials, such as defects of about 1 mm (e.g., dirt or scratches) on a wall material measuring about 1 m x 3 m. In other words, the decorative material repair device described in Patent Document 1 has a problem with the accuracy of detecting defects in wall materials.

[0005] The present disclosure has been made in consideration of the above problems, and an exemplary objective thereof is to provide a technique for improving the accuracy of detecting defects in wall materials. [Means for solving the problem]

[0006] A quality inspection system according to an exemplary aspect of the present disclosure includes an acquisition means for acquiring an image of a wall material, a division means for dividing the image into a plurality of small images, and at least two machine learning models for inputting an image of the wall material and outputting a judgment result for judging whether the wall material is defective or not, the at least two machine learning models trained using different learning methods, each of which detects defects contained in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images into each of the at least two machine learning models, and an output means for outputting the detection result by the detection means.

[0007] A quality inspection method according to an exemplary aspect of the present disclosure includes: an acquisition process in which at least one processor acquires an image of a wall material; a division process in which the at least one processor divides the image into a plurality of small images; a detection process in which the at least one processor detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images into at least two machine learning models trained by different learning methods, the at least one processor inputting an image of the wall material and outputting a judgment result determining whether the wall material is defective or not; and an output process in which the at least one processor outputs the detection result in the detection process.

[0008] A quality inspection program according to an exemplary aspect of the present disclosure is a program that causes a computer to function as a quality inspection system, and causes the computer to function as: an acquisition means for acquiring an image of a wall material; a division means for dividing the image into a plurality of small images; at least two machine learning models that take an image of a wall material as input and output a judgment result determining whether or not the wall material is defective, each of which is trained using different learning methods, and which detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images into each of the at least two machine learning models; and an output means for outputting the detection result by the detection means.

[0009] A quality inspection device according to an exemplary aspect of the present disclosure includes an acquisition means for acquiring an image of a wall material, a division means for dividing the image into a plurality of small images, and at least two machine learning models for inputting an image of the wall material and outputting a judgment result for judging whether the wall material is defective or not, the at least two machine learning models trained using different learning methods, each of which detects defects contained in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images into each of the at least two machine learning models, and an output means for outputting the detection result by the detection means. Effect of the Invention

[0010] According to an exemplary aspect of the present disclosure, an exemplary effect is provided in that a technique for improving the accuracy of detecting defects in wall materials can be provided. [Brief description of the drawings]

[0011] [Figure 1] 1 is a block diagram showing a configuration of a quality inspection system according to the present disclosure. [Diagram 2] 1 is a flow diagram showing the flow of a quality inspection method according to the present disclosure. [Diagram 3] 1 is a block diagram showing a configuration of a quality inspection device according to the present disclosure. [Figure 4] 1 is a flow diagram showing the flow of a quality inspection method according to the present disclosure. [Diagram 5] FIG. 1 is a diagram illustrating an example of an inspection system using a quality inspection system according to the present disclosure. [Figure 6] FIG. 13 is a diagram showing an example of an image displayed on an input terminal according to the present disclosure. [Figure 7] FIG. 13 is a diagram showing an example of an image of a wall material captured in the present disclosure. [Figure 8] FIG. 13 is a diagram showing an example of an image displayed on an output terminal according to the present disclosure. [Figure 9] 13A and 13B are diagrams illustrating other examples of images displayed on the output terminal according to the present disclosure. [Figure 10]FIG. 13 is a diagram showing yet another example of an image displayed on the output terminal according to the present disclosure. [Figure 11] FIG. 13 is a diagram showing yet another example of an image displayed on the output terminal according to the present disclosure. [Figure 12] 1 is a block diagram showing a configuration of a quality inspection system according to the present disclosure. [Figure 13] 1 is a flow diagram showing the flow of a quality inspection method according to the present disclosure. [Figure 14] FIG. 4 is a diagram illustrating an example of processing executed by a preprocessing unit according to the present disclosure. [Figure 15] FIG. 13 is a diagram illustrating another example of processing executed by the preprocessing unit according to the present disclosure. [Figure 16] FIG. 13 is a diagram illustrating yet another example of processing executed by the preprocessing unit according to the present disclosure. [Figure 17] FIG. 11 is a diagram illustrating an example of a process executed by a division unit according to the present disclosure. [Figure 18] FIG. 13 is a diagram illustrating another example of the process executed by the division unit according to the present disclosure. [Figure 19] 1 is a block diagram showing a configuration of a quality inspection device according to the present disclosure. [Figure 20] 2 is a block diagram showing an example of a hardware configuration of a quality inspection system, a quality inspection device, an AP / DB server, and a determination server according to the present disclosure. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, an embodiment obtained by appropriately combining the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, an embodiment obtained by appropriately omitting a part of the technical means employed in each of the exemplary embodiments shown below may also be included in the scope of the present invention. In addition, the effects mentioned in each of the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, an embodiment that does not exhibit the effects mentioned in each of the exemplary embodiments shown below may also be included in the scope of the present invention.

[0013] First Exemplary Embodiment A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is the basic form of each exemplary embodiment described later. The scope of application of each technical means adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means adopted in this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs. In addition, each technical means shown in the drawings referred to for explaining this exemplary embodiment can also be adopted in other exemplary embodiments included in this disclosure to the extent that no particular technical obstacle occurs.

[0014] (Configuration of quality inspection system 1) The configuration of the quality inspection system 1 will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the quality inspection system 1. As shown in Fig. 1, the quality inspection system 1 includes an acquisition unit 11, a division unit 12, a detection unit 13, and an output unit 14. In this exemplary embodiment, the acquisition unit 11, the division unit 12, the detection unit 13, and the output unit 14 respectively realize an acquisition means, a division means, a detection means, and an output means.

[0015] The acquisition unit 11 acquires an image of a wall material. The acquisition unit 11 supplies the acquired image to the division unit 12.

[0016] The division unit 12 divides the image acquired by the acquisition unit 11 into a plurality of small images. The division unit 12 supplies the plurality of divided small images to the detection unit 13.

[0017] The detection unit 13 is at least two machine learning models that receive an image of a wall material as an input and output a judgment result of whether or not the wall material is defective, and detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 12 to each of the at least two machine learning models trained by different learning methods. The detection unit 13 supplies the detection result to the output unit 14.

[0018] The output unit 14 outputs the detection result from the detection unit 13 .

[0019] (Effects of quality inspection system 1) As described above, the quality inspection system 1 is configured to include an acquisition unit 11 that acquires an image of a wall material, a division unit 12 that divides the image acquired by the acquisition unit 11 into a plurality of small images, and at least two machine learning models that take an image of the wall material as input and output a judgment result determining whether the wall material is defective or not, each of the at least two machine learning models trained using different learning methods, a detection unit 13 that detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 12, and an output unit 14 that outputs the detection result by the detection unit 13.

[0020] Therefore, the quality inspection system 1 has the effect of improving the accuracy of detecting defects in wall materials.

[0021] (Flow of quality inspection method S1) The flow of the quality inspection method S1 will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the quality inspection method S1. As shown in Fig. 2, the quality inspection method S1 includes an acquisition process S11, a division process S12, a detection process S13, and an output process S14.

[0022] (Acquisition process S11) In the acquisition process S11, the acquisition unit 11 acquires an image of a wall material. The acquisition unit 11 supplies the acquired image to the division unit 12.

[0023] (Split process S12) In the division process S12, the division unit 12 divides the image acquired by the acquisition unit 11 into a plurality of small images. The division unit 12 supplies the plurality of divided small images to the detection unit 13.

[0024] (Detection process S13) In the detection process S13, the detection unit 13 detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting small images divided by the division unit 12 to at least two machine learning models that input an image of a wall material and output a judgment result of whether or not the wall material is defective, the at least two machine learning models trained by different learning methods. The detection unit 13 supplies the detection result to the output unit 14.

[0025] (Output process S14) In the output process S14, the output unit 14 outputs the detection result by the detection unit 13.

[0026] (Effect of quality inspection method S1) As described above, the quality inspection method S1 includes an acquisition process S11 in which the acquisition unit 11 acquires an image of a wall material, a division process S12 in which the division unit 12 divides the image acquired by the acquisition unit 11 into a plurality of small images, a detection process S13 in which at least two machine learning models, each of which is trained by a different learning method, detect defects contained in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 12 to at least two machine learning models that are trained by different learning methods, and an output process S14 in which the output unit 14 outputs the detection result by the detection unit 13. Therefore, the quality inspection method S1 can achieve the same effect as the quality inspection system 1 described above.

[0027] (Configuration of quality inspection device 2) The configuration of the quality inspection device 2 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the quality inspection device 2. As shown in Fig. 3, the quality inspection device 2 includes an acquisition unit 21, a division unit 22, a detection unit 23, and an output unit 24. In this exemplary embodiment, the acquisition unit 21, the division unit 22, the detection unit 23, and the output unit 24 respectively realize an acquisition means, a division means, a detection means, and an output means.

[0028] The acquisition unit 21 acquires an image of a wall material. The acquisition unit 21 supplies the acquired image to the division unit 22.

[0029] The division unit 22 divides the image acquired by the acquisition unit 21 into a plurality of small images. The division unit 22 supplies the plurality of divided small images to the detection unit 23.

[0030] The detection unit 23 is at least two machine learning models that receive an image of a wall material as an input and output a determination result of whether the wall material is defective or not, and detects defects in the wall material by referring to a combination of at least two determination results obtained by inputting the small images divided by the division unit 22 to each of the at least two machine learning models trained by different learning methods. The detection unit 23 supplies the detection result to the output unit 24.

[0031] The output unit 24 outputs the detection result from the detection unit 23 .

[0032] (Effect of quality inspection device 2) As described above, the quality inspection device 2 is configured to include an acquisition unit 21 that acquires an image of a wall material, a division unit 22 that divides the image acquired by the acquisition unit 21 into a plurality of small images, and at least two machine learning models that input an image of a wall material and output a determination result of whether the wall material is defective or not, the at least two machine learning models trained by different learning methods, a detection unit 23 that detects defects in the wall material by referring to a combination of at least two determination results obtained by inputting the small images divided by the division unit 22 to each of the at least two machine learning models, and an output unit 24 that outputs the detection result by the detection unit 23. Therefore, the quality inspection device 2 can achieve the same effect as the above-mentioned quality inspection system 1.

[0033] (Flow of quality inspection method S2) The flow of the quality inspection method S2 will be described with reference to Fig. 4. Fig. 4 is a flow diagram showing the flow of the quality inspection method S2. As shown in Fig. 4, the quality inspection method S2 includes an acquisition process S21, a division process S22, a detection process S23, and an output process S24.

[0034] (Acquisition process S21) In the acquisition process S21, the acquisition unit 21 acquires an image of a wall material. The acquisition unit 21 supplies the acquired image to the division unit 22.

[0035] (Split process S22) In the division process S22, the division unit 22 divides the image acquired by the acquisition unit 21 into a plurality of small images. The division unit 22 supplies the plurality of divided small images to the detection unit .

[0036] (Detection process S23) In the detection process S23, the detection unit 23 detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 22 to at least two machine learning models that input an image of the wall material and output a judgment result of whether or not the wall material is defective, the at least two machine learning models trained by different learning methods. The detection unit 23 supplies the detection result to the output unit 24.

[0037] (Output process S24) In the output process S24, the output unit 24 outputs the detection result by the detection unit 23.

[0038] (Effect of quality inspection method S2) As described above, the quality inspection method S2 includes an acquisition process S21 in which the acquisition unit 21 acquires an image of a wall material, a division process S22 in which the division unit 22 divides the image acquired by the acquisition unit 21 into a plurality of small images, a detection process S23 in which the detection unit 23 detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 22 to at least two machine learning models trained by different learning methods, the detection unit 23 dividing the image of the wall material into at least two small images and outputting a judgment result indicating whether the wall material is defective or not, and an output process S24 in which the output unit 24 outputs the detection result by the detection unit 23. Therefore, the quality inspection method S2 can achieve the same effect as the quality inspection system 1 described above.

[0039] Second Exemplary Embodiment A second exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be given the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technical means adopted in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means adopted in this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs. In addition, each technical means shown in each drawing referred to for explaining this exemplary embodiment can be adopted in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs.

[0040] (Overview of Quality Inspection System 1A) An overview of the quality inspection system 1A will be described with reference to Fig. 5. Fig. 5 is a diagram showing an example of an inspection system using the quality inspection system 1A. In the inspection system shown in Fig. 5, an input terminal IN_PC, an output terminal OUT_PC, scanner control terminals S_PC1 to S_PC5, and the quality inspection system 1A are communicatively connected.

[0041] The quality inspection system 1A is a system that detects defects in the wall material WM using images of the wall material WM. The quality inspection system 1A detects defects in the wall material WM using at least two machine learning models described below. The quality inspection system 1A also has a configuration for training at least two machine learning models.

[0042] As shown in FIG. 5, the quality inspection system 1A includes an AP (Application) / DB (Database) server 3 and a judgment server 4.

[0043] 5, the AP / DB server 3 acquires images P1 to P9 of the wall material WM from the scanner control terminals S_PC1 to S_PC5. The AP / DB server 3 processes and divides the acquired images P1 to P9 to generate a plurality of images S_P (small images) to be judged. The AP / DB server 3 outputs the plurality of images S_P to the judgment server 4, thereby requesting the judgment server 4 to judge whether or not the wall material WM is defective.

[0044] Moreover, the AP / DB server 3 acquires at least two judgment results output from the judgment server 4. The AP / DB server 3 detects defects contained in the wall material WM by referring to a combination of at least two judgment results output from the judgment server 4. The combination of judgment results referred to by the AP / DB server 3 will be described later.

[0045] Moreover, the AP / DB server 3 outputs the detection result to the output terminal OUT_PC. The AP / DB server 3 may output, in addition to the detection result, an image showing the defective portion to the output terminal OUT_PC.

[0046] The "image showing a defective part" output by the AP / DB server 3 is an image showing a part determined to be defective by the machine learning model. In other words, the "image showing a defective part" output by the AP / DB server 3 may be ultimately determined to be defective by a worker or a user of the wall material, or may be ultimately determined not to be defective by a worker or a user of the wall material.

[0047] The determination server 4 acquires a plurality of images S_P output from the AP / DB server 3. The determination server 4 acquires at least two judgment results by inputting each of the acquired plurality of images S_P into at least two machine learning models. The determination server 4 outputs the acquired at least two judgment results to the AP / DB server 3.

[0048] (Inspection system flow 1) An example of the flow of the inspection system shown in FIG. 5 will be described.

[0049] As shown in the upper left of Fig. 5, the worker moves the wall material WM to the conveyor in order to photograph the wall material WM that is the target of defect detection. Here, the worker inputs to the input terminal IN_PC whether to detect defects contained in the wall material WM or to train at least two machine learning models. An example of an image displayed on the input terminal IN_PC is shown in Fig. 6.

[0050] Fig. 6 is a diagram showing an example of an image displayed on the input terminal IN_PC. The image DP1 shown in Fig. 6 includes an interface IF1 for selecting whether to detect defects contained in the wall material WM ("AI inspection" in Fig. 6) or to train at least two machine learning models ("AI learning" in Fig. 6). The image DP1 also includes an interface IF2 for inputting the product number of the wall material WM. The flow when the worker selects "AI inspection" will be described below.

[0051] The wall material WM transferred to the conveyor has its top surface photographed by scanner SC_1 and scanner SC_2. Hereinafter, the 2D scanner will be simply referred to as the "scanner". Similarly, the 2D image will be simply referred to as the "image". Each of scanner SC_1 and scanner SC_2 is equipped with two photographing devices. Scanner SC_1 and scanner SC_2 supply the photographed images to scanner control terminal S_PC1 and scanner control terminal S_PC2, respectively. Scanner control terminal S_PC1 and scanner control terminal S_PC2 output the photographed images P1 to P4 to AP / DB server 3.

[0052] Examples of images P1 to P4 are shown in Fig. 7. Fig. 7 is a diagram showing examples of captured images of the wall material WM.

[0053] As described above, the scanner SC_1 and the scanner SC_2 each include two photographing devices. Therefore, the scanner SC_1 and the scanner SC_2 output four images P1 to P4 shown in Fig. 7 to the AP / DB server 3. The images P1 and P2 are images photographed by the scanner SC_1 from the center to the left along the longitudinal direction of the wall material WM. The images P3 and P4 are images photographed by the scanner SC_2 from the center to the right along the longitudinal direction of the wall material WM.

[0054] Next, the side of the long side of the upper surface of the wall material WM is photographed by scanner SC_3 and scanner SC_4. Scanner SC_3 and scanner SC_4 each supply the photographed image to scanner control terminal S_PC3. Scanner control terminal S_PC3 outputs photographed images P5 and P6 to AP / DB server 3. Examples of images P5 and P6 are shown in FIG. 7. Image P5 is an image photographed by scanner SC_3 from the left in the direction of travel of the conveyor. Image P6 is an image photographed by scanner SC_4 from the right in the direction of travel of the conveyor.

[0055] Next, the wall material WM is photographed by a 3D scanner SC_5 composed of multiple scanners. The 3D scanner SC_5 supplies the photographed image to a 3D scanner control terminal S_PC5. The 3D scanner control terminal S_PC5 generates an image P9 in which the wall material WM is rendered 3D based on the photographed image. The 3D scanner control terminal S_PC5 outputs the image P9 to the AP / DB server 3. An example of the image P9 is shown in FIG.

[0056] Furthermore, the side surface on the short side of the upper surface of the wall material WM is photographed by scanner SC_7 and scanner SC_8. Scanner SC_7 and scanner SC_8 each supply the photographed image to scanner control terminal S_PC4. Scanner control terminal S_PC4 outputs photographed images P7 and P8 to AP / DB server 3. Examples of images P7 and P8 are shown in FIG. 7. Image P7 is an image photographed by scanner SC_7 from the left in the conveyor travel direction. Image P8 is an image photographed by scanner SC_8 from the right in the conveyor travel direction.

[0057] When the AP / DB server 3 acquires the images P1 to P9, it processes and divides the images P1 to P9 to generate a plurality of images S_P to be judged. An example of the process in which the AP / DB server 3 processes and divides the images will be described later. The AP / DB server 3 outputs the generated plurality of images S_P to the judgment server 4.

[0058] The determination server 4 obtains at least two determination results by inputting each of the multiple images S_P output from the AP / DB server 3 into at least two machine learning models. The determination server 4 outputs the obtained at least two determination results to the AP / DB server 3.

[0059] The AP / DB server 3 acquires at least two judgment results output from the judgment server 4. The AP / DB server 3 detects defects contained in the wall material WM by referring to a combination of at least two judgment results output from the judgment server 4. The AP / DB server 3 outputs the detection result to the output terminal OUT_PC.

[0060] The output terminal OUT_PC outputs the detection result output from the AP / DB server 3. Examples of images displayed on the output terminal OUT_PC are shown in Figs. 8 to 11.

[0061] Fig. 8 is a diagram showing an example of an image displayed on the output terminal OUT_PC. The image DP2 shown in Fig. 8 includes the detection result IR1 output from the AP / DB server 3. Furthermore, when the output terminal OUT_PC acquires an image showing a defective part in addition to the detection result, the output terminal OUT_PC may display the image showing the defective part. An example of the image showing the defective part is shown in Fig. 9.

[0062] Fig. 9 is a diagram showing another example of an image displayed on the output terminal OUT_PC. The image DP3 shown in Fig. 9 is an image including the entire wall material WM as a subject. Furthermore, the image DP3 includes a detection result IR3 indicating that a defective portion is included within a rectangular frame.

[0063] Furthermore, the detection result output from the AP / DB server 3 may include at least one of information regarding the outer shape of the wall material WM (e.g., width, length, parallelism, perpendicularity, and thickness) and information regarding the color of the wall material WM. As one example, the detection result may include, as the information regarding the outer shape of the wall material WM, information indicating whether the outer shape of the wall material WM is within a threshold value. As another example, the detection result may include information indicating whether the pixel value of the wall material WM in the image is within a threshold value.

[0064] When the detection result includes information indicating whether the contour of the wall material WM is within a threshold value and information indicating whether the pixel value of the wall material WM in the image is within a threshold value, the image DP2 may include a result IR2 indicated by the information. Furthermore, the image DP2 may include an item MR indicating the contour of the wall material WM and the pixel value of the wall material WM in the image.

[0065] In this way, the quality inspection system 1A includes in the detection result at least one of information indicating whether the outer shape of the wall material WM is within a threshold value and information indicating whether the pixel value of the wall material WM is within a threshold value, so that it can detect and present defects other than scratches and dents.

[0066] The image DP2 may also include an interface IF3 for inputting the visual inspection results by the operator. An example of an image displayed on the output terminal OUT_PC when the operator selects the interface IF3 for inputting the visual inspection results is shown in FIG.

[0067] Fig. 10 is a diagram showing yet another example of an image displayed on the output terminal OUT_PC. The image DP4 shown in Fig. 10 includes an interface IF4 for inputting defective parts. The worker inputs the defective parts into the image of the wall material WM, which is the interface IF4.

[0068] (Inspection system flow 2) A description will be given of another example of the flow of the inspection system shown in Fig. 5. The following describes the flow when the operator selects "AI learning" on the input terminal IN_PC.

[0069] Even in this configuration, the AP / DB server 3 acquires images P1 to P9 by the method described above. An example of images displayed on the output terminal OUT_PC while images P1 to P9 are being captured is shown in FIG.

[0070] Fig. 11 is a diagram showing yet another example of an image displayed on the output terminal OUT_PC. An image DP5 shown in Fig. 11 includes text indicating that a learning image is being photographed.

[0071] When the wall material WM is photographed, the output terminal OUT_PC displays the image DP2 described above. When the inspection result is input as pass in the image DP2 (i.e., when the worker judges that it is not defective), the AP / DB server 3 generates a set of the photographed images P1 to P9 (small images obtained by processing and dividing each of the images P1 to P9) and a judgment result indicating that the wall material WM included as a subject in the images P1 to P9 is not defective (normal) as training data. The AP / DB server 3 outputs the generated training data to the judgment server 4. The judgment server 4 uses the training data output from the AP / DB server 3 to train the machine learning model.

[0072] On the other hand, when the inspection result is input as "fail" in the image DP2 (i.e., when the worker judges it to be defective), the output terminal OUT_PC displays an image DP4. The output terminal OUT_PC outputs to the AP / DB server 3 information indicating the input to the interface IF4 for inputting the defective part (input indicating which part of the wall material WM contains the defect).

[0073] The AP / DB server 3 generates, as training data, a set of small images obtained by dividing the captured images P1 to P9, the images including the areas containing the defects indicated by the information, and a judgment result indicating that the wall material WM included as a subject in the images is defective, based on the information output from the output terminal OUT_PC. The AP / DB server 3 outputs the generated training data to the judgment server 4. The judgment server 4 trains a machine learning model using the training data output from the AP / DB server 3. An example of a configuration for training a machine learning model will be described later.

[0074] The training data may be generated based on the result of a customer (the manufacturer of the wall material) judging whether or not the image is defective. In other words, a pair of an image and a judgment result indicating that the image is judged to be not defective by the customer, and a pair of an image and a judgment result indicating that the image is judged to be defective by the customer may each be used as training data.

[0075] (Configuration of quality inspection system 1A) The configuration of the quality inspection system 1A will be described with reference to Fig. 12. Fig. 12 is a block diagram showing the configuration of the quality inspection system 1A. As described above, the quality inspection system 1A includes the AP / DB server 3 and the judgment server 4.

[0076] The AP / DB server 3 and the determination server 4 are connected to each other so as to be able to communicate with each other via a network N. Although the specific configuration of the network N does not limit the present exemplary embodiment, examples thereof include a wireless LAN (Local Area Network) or a wired LAN. As other examples, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0077] (AP / DB Server 3 Configuration) As shown in FIG. 12, the AP / DB server 3 includes a control unit 31, a storage unit 32, and a communication unit 33.

[0078] The storage unit 32 stores data referenced by the control unit 31. An example of the data stored in the storage unit 32 is an image of the wall material WM. Examples of the storage unit 32 include, but are not limited to, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

[0079] The communication unit 33 is an interface that transmits and receives data via a network. For example, the communication unit 33 transmits data provided from the control unit 31 to the determination server 4, and provides data received from the determination server 4 to the control unit 31. Examples of the communication unit 33 include, but are not limited to, communication chips in various communication standards such as Ethernet (registered trademark), Wi-Fi (registered trademark), and wireless communication standards for mobile data communication networks, and USB-compliant connectors.

[0080] (Control unit 31) The control unit 31 controls each of the components included in the AP / DB server 3 .

[0081] 12, the control unit 31 includes an acquisition unit 311, a preprocessing unit 312, a division unit 313, a detection unit 314, and an output unit 315. In this exemplary embodiment, the acquisition unit 311, the preprocessing unit 312, the division unit 313, the detection unit 314, and the output unit 315 respectively realize an acquisition means, a preprocessing means, a division means, a detection means, and an output means.

[0082] The acquiring unit 311 acquires data. As one example, the acquiring unit 311 acquires images (for example, the above-mentioned images P1 to P9) of the wall material WM. As another example, the acquiring unit 311 acquires at least two determination results. The acquiring unit 311 stores the acquired data in the storage unit 32.

[0083] The pre-processing unit 312 processes the image before division (performs pre-processing on the image before division). As an example, the pre-processing unit 312 processes the image acquired by the acquisition unit 311. For example, the pre-processing unit 312 performs at least one of image combination, tilt correction of the wall material WM in the image, positioning of the wall material WM in the image, color processing, and edge processing. The pre-processing unit 312 stores the processed image in the memory unit 32. An example of the processing performed by the pre-processing unit 312 will be described later.

[0084] The division unit 313 divides an image into a plurality of small images. As an example, the division unit 313 divides an image, which is stored in the storage unit 32 and has been processed by the pre-processing unit 312, into a plurality of small images. The division unit 313 stores the divided plurality of small images in the storage unit 32. An example of the process executed by the division unit 313 will be described later.

[0085] The detection unit 314 detects defects contained in the wall material WM. As an example, the detection unit 314 detects defects contained in the wall material WM by referring to a combination of at least two judgment results output from the judgment server 4. The detection unit 314 stores the detection result in the storage unit 32.

[0086] Furthermore, the detection unit 314 changes the combination of at least two judgment results to be referred to depending on the type of the wall material WM. As an example, the detection unit 314 detects defects contained in the wall material WM by performing a logical operation on at least two judgment results. In this case, the detection unit 314 may change the logical operation to be performed on at least two judgment results depending on the type of the wall material WM. An example of the process executed by the detection unit 314 will be described later.

[0087] The detection unit 314 may also measure the outer shape and color of the wall material WM. In this case, the detection unit 314 may include information on the outer shape of the wall material WM and information on the color of the wall material WM in the detection result.

[0088] The output unit 315 outputs data. As one example, the output unit 315 outputs the detection result stored in the storage unit 32. As another example, the output unit 315 outputs an image showing a defective portion in addition to the detection result. As yet another example, the output unit 315 outputs a plurality of small images (for example, the above-mentioned image S_P) divided by the division unit 313 to the determination server 4 as images to be determined.

[0089] In addition, the output unit 315 generates and outputs training data used in machine learning of a first machine learning model M1 and a second machine learning model M2, which will be described later. An example of the method in which the output unit 315 generates training data is as described above.

[0090] (Configuration of Judgment Server 4) As shown in FIG. 12, the determination server 4 includes a control unit 41, a storage unit 42, and a communication unit 43.

[0091] The storage unit 42 stores data referenced by the control unit 41. Examples of the data stored in the storage unit 42 include a first machine learning model M1 and a second machine learning model M2, which will be described later. In this case, the storage unit 42 may store parameters that define the first machine learning model M1 and the second machine learning model M2, respectively. Another example of the data stored in the storage unit 42 is an image to be determined. Examples of the storage unit 42 include, but are not limited to, a flash memory, an HDD, an SSD, or a combination thereof.

[0092] The communication unit 43 is an interface that transmits and receives data via a network. For example, the communication unit 43 transmits data supplied from the control unit 41 to the AP / DB server 3, and supplies data received from the AP / DB server 3 to the control unit 41. Examples of the communication unit 43 include, but are not limited to, communication chips in various communication standards such as Ethernet, Wi-Fi, and wireless communication standards for mobile data communication networks, and USB-compliant connectors.

[0093] (Control unit 41) The control unit 41 controls each of the components included in the determination server 4 .

[0094] 12, the control unit 41 includes an acquisition unit 411, a determination unit 412, an output unit 413, and a learning unit 414. In this exemplary embodiment, the learning unit 414 realizes a learning means.

[0095] The acquisition unit 411 acquires data. As one example, the acquisition unit 411 acquires a plurality of divided small images. As another example, the acquisition unit 411 acquires teacher data. The acquisition unit 411 stores the acquired data in the storage unit 42.

[0096] The determination unit 412 determines whether the wall material WM is defective. As an example, the determination unit 412 makes a determination by inputting a plurality of images stored in the storage unit 42 to at least two machine learning models that input an image (for example, the above-mentioned image S_P) captured of the wall material WM and output a determination result of whether the wall material WM is defective or not, the at least two machine learning models being trained by different learning methods. The determination unit 412 stores in the storage unit 42 at least two determination results output from each of the at least two machine learning models.

[0097] The at least two machine learning models trained by different learning methods include at least one first machine learning model M1 and at least one second machine learning model M2.

[0098] The first machine learning model M1 is a machine learning model that is trained by machine learning a pair of an image of a normal wall material WM that does not include defects and a judgment result indicating that the wall material WM is normal, as training data. Hereinafter, the first machine learning model is also referred to as a one-class model.

[0099] The second machine learning model M2 is a machine learning model that is machine-learned using a pair of an image of a normal wall material WM and a judgment result indicating that the wall material WM is normal, and a pair of an image of a defective wall material WM and a judgment result indicating that the wall material WM is defective as teacher data. It is also called a multi-class model. Examples of the first machine learning model and the second machine learning model will be described later.

[0100] The output unit 413 outputs data. As an example, the output unit 413 outputs at least two determination results stored in the storage unit .

[0101] The learning unit 414 trains the first machine learning model M1 and the second machine learning model M2. As an example, the learning unit 414 trains the first machine learning model M1 and the second machine learning model M2 using teacher data output from the AP / DB server 3.

[0102] Specifically, as described above, the learning unit 414 trains the first machine learning model M1 using a pair of an image of a normal wall material WM that does not contain any defects and a judgment result indicating that the wall material WM is normal as training data.

[0103] In addition, the learning unit 414 trains a second machine learning model M2 using, as training data, a pair of an image of normal wall material WM and a judgment result indicating that the wall material WM is normal, and a pair of an image of defective wall material WM and a judgment result indicating that the wall material WM is defective.

[0104] With this configuration, the learning unit 414 can suitably train the first machine learning model M1 and the second machine learning model M2.

[0105] (Flow of quality inspection method S1A) The flow of the quality inspection method S1A executed by the quality inspection system 1A will be described with reference to Fig. 13. Fig. 13 is a flow diagram showing the flow of the quality inspection method S1A.

[0106] (Step S11A) In step S11A, the acquisition unit 311 of the AP / DB server 3 acquires an image of the wall material WM. The acquisition unit 311 of the AP / DB server 3 stores the acquired image in the storage unit 32.

[0107] (Step S12A) In step S12A, the preprocessing unit 312 performs preprocessing on the image stored in the storage unit 32. The preprocessing unit 312 stores in the storage unit 32 the image that has been preprocessed.

[0108] (Step S13A) In step S13A, the division unit 313 divides the image processed by the pre-processing unit 312 and stored in the storage unit 32 into a plurality of small images. The division unit 313 stores the divided plurality of small images in the storage unit 32.

[0109] (Step S14A) In step S14A, the output unit 315 outputs the divided small images stored in the storage unit 32 to the determination server 4.

[0110] (Step S15A) In step S15A, the acquisition unit 411 of the determination server 4 acquires the plurality of small images output from the AP / DB server 3. The acquisition unit 411 stores the acquired plurality of small images in the storage unit .

[0111] (Step S16A) In step S16A, the determination unit 412 obtains at least two determination results by inputting a plurality of small images stored in the storage unit 42 to each of at least one first machine learning model M1 and at least one second machine learning model M2, and determines whether the wall material WM is defective or not. The determination unit 412 stores the at least two determination results in the storage unit 42.

[0112] (Step S17A) In step S17A, the output unit 413 outputs at least two of the determination results stored in the storage unit 42 to the AP / DB server 3.

[0113] (Step S18A) In step S18A, the acquisition unit 311 of the AP / DB server 3 acquires at least two determination results output from the determination server 4. The acquisition unit 311 stores in the storage unit 32 the acquired at least two determination results.

[0114] (Step S19A) In step S19A, the detection unit 314 detects defects contained in the wall material WM by referring to a combination of at least two judgment results stored in the storage unit 32. The detection unit 314 stores the detection result in the storage unit 32.

[0115] (Step S20A) In step S20A, the output unit 315 outputs the detection result stored in the storage unit 32.

[0116] (Example 1 of Processing Executed by Preprocessing Unit 312) An example of the process executed by the pre-processing unit 312 will be described with reference to Fig. 14. Fig. 14 is a diagram showing an example of the process executed by the pre-processing unit 312.

[0117] In the following, an example will be described in which the upper surface of the wall material WM is photographed by the scanner SC_1 and the scanner SC_2 each having two imaging devices, as in the inspection system in FIG. 5 described above.

[0118] In this case, in step S11A, the acquisition unit 311 acquires images P1 to P4 in Fig. 14. Note that the wall material WM included as a subject in the images P1 to P4 acquired by the acquisition unit 311 has the words "product image" printed on the surface so that the pre-processing can be easily understood.

[0119] The pre-processing unit 312 generates an image P_P1 including the entire wall material WM by combining the images P1 to P4 in Fig. 14. Here, the images P1 to P4 include overlapping areas (for example, the area R_P of the image P2 and the area L_P of the image P3). Therefore, the pre-processing unit 312 combines the images P1 to P4 so that the areas R_P and L_P do not overlap.

[0120] With this configuration, the pre-processing unit 312 can generate an image P_P1 that includes the entire wall material WM, even if the wall material WM is large, by combining images captured by multiple scanners.

[0121] (Processing example 2 executed by the pre-processing unit 312) Another example of the processing executed by the pre-processing unit 312 will be described with reference to Fig. 15. Fig. 15 is a diagram showing another example of the processing executed by the pre-processing unit 312. In Fig. 15, the left-right direction of the image P_P1 will be described as the x-axis direction, and the up-down direction as the y-axis direction.

[0122] The pre-processing unit 312 corrects the inclination of the wall material WM in the image P_P1 generated by the above-mentioned processing.

[0123] First, as shown in the upper left of FIG. 15, the preprocessing unit 312 scans from five points on the upper side of the image P_P1 toward the inside to detect the wall material WM, which is the subject. Next, the preprocessing unit 312 connects the five points where the wall material WM is detected with a straight line. That is, the preprocessing unit 312 detects the upper side of the wall material WM in the image P_P1. Similarly, the preprocessing unit 312 scans from five points toward the inside of the left side, right side, and bottom side of the image P_P1 to detect the wall material WM, and detects the left side, right side, and bottom side of the wall material WM. Then, the preprocessing unit 312 extracts the wall material WM in the image P_P1, as shown in the upper right of FIG. 15.

[0124] Next, the pre-processing unit 312 uses an inverse trigonometric function to calculate a rotation angle θ at which the upper and lower sides of the wall material WM in the image P_P1 are parallel to the x-axis and the left and right sides of the wall material WM in the image P_P1 are parallel to the y-axis, as shown in the lower right of Fig. 15. Then, the pre-processing unit 312 generates an image P_P2 by rotating the wall material WM in the image P_P1 by the rotation angle θ, as shown in the lower left of Fig. 15.

[0125] With this configuration, the pre-processing unit 312 can generate an image P_P2 in which the inclination of the wall material WM is the same, even if the inclination of the wall material WM differs from image to image in a plurality of images captured of the wall material WM.

[0126] (Processing Example 3 Executed by Preprocessing Unit 312) Another example of the processing executed by the pre-processing unit 312 will be described with reference to Fig. 16. Fig. 16 is a diagram showing yet another example of the processing executed by the pre-processing unit 312. In Fig. 16 as well, the left-right direction of image P10 will be described as the x-axis direction, and the up-down direction as the y-axis direction.

[0127] In the following, the process of aligning the wall material WM performed by the pre-processing unit 312 will be explained using an example of image P5 in which the wall material WM is photographed from the side, such as by scanners SC_3, SC_4, SC_7, and SC_8 in the inspection system in Figure 5 described above.

[0128] First, the preprocessing unit 312 detects the wall material WM, which is the subject, by scanning from the upper side of the image P5 toward the inside as shown in the upper left of Fig. 15. The preprocessing unit 312 sets the detected point as a point p1.

[0129] Next, as shown in the diagram on the right side of the diagram in the upper left of FIG. 15, the preprocessing unit 312 scans from point p1 toward the left along the x-axis to detect the corner of the wall material WM. When the preprocessing unit 312 detects the corner of the wall material WM, it scans from the corner of the wall material WM toward the upward direction along the y-axis to the upper side of the image P5. When the preprocessing unit 312 scans to the upper side of the image P5, it scans toward the right along the x-axis direction to detect the conveyor. The preprocessing unit 312 sets the detected point as point p2.

[0130] Next, as shown in the left diagram of the upper right diagram of FIG. 15, the preprocessing unit 312 sets the intersection point of a line passing through the point p1 and parallel to the x-axis and a straight line passing through the point p2 and parallel to the y-axis as p3. The preprocessing unit 312 also scans the lower side of the image P5 inwardly, as well as the upper side of the image P5, to detect the wall material WM, which is the subject. The preprocessing unit 312 sets the detected point as point p4. Then, the preprocessing unit 312 sets the intersection point of a line passing through the point P4 and parallel to the x-axis and a straight line passing through the point p2 and parallel to the y-axis as p5. The preprocessing unit 312 calculates the distance d between the straight line passing through the points p3 and p5 and the wall material WM. The preprocessing unit 312 may calculate the distance to the wall material WM at multiple points and calculate the average of the distances as the distance d.

[0131] Furthermore, the preprocessing unit 312 sets a straight line parallel to the y-axis and a distance d to the left along the x-axis from the line passing through the calculated points p3 and p5 to the right side of the wall material WM. Next, as shown in the upper right of FIG. 15, the preprocessing unit 312 sets the point on the right side of the wall material WM, which is the upper right corner of the wall material WM, to point p6, and sets a region r1 to be cut out with a predetermined length as a margin from point p6. Here, the preprocessing unit 312 may set the margin based on the division size (size of the small image) by the dividing unit 313. For example, the preprocessing unit 312 may set the region to be cut out by adding region r2 to region r1 so that the length of the region to be cut out is an integer multiple of the division size by the dividing unit 313.

[0132] Then, as shown in the lower right of FIG. 16, the pre-processing unit 312 cuts out the image P5 according to the set cut-out region, and generates an image P_P3 in which the wall material WM is aligned.

[0133] With this configuration, the pre-processing unit 312 can generate an image P_P3 in which the position of the wall material WM is aligned, even if the position of the wall material WM differs in each of the multiple images captured of the wall material WM.

[0134] (Processing example 1 executed by the division unit 313) An example of the process executed by the dividing unit 313 will be described with reference to Fig. 17. Fig. 17 is a diagram showing an example of the process executed by the dividing unit 313. Below, the process of dividing the image P_P2 generated by the pre-processing unit 312 in the above-mentioned process will be described. In Fig. 17 as well, the left-right direction of the image P_P2 will be described as the x-axis direction, and the up-down direction as the y-axis direction.

[0135] First, as shown in the upper right of FIG. 17, the division unit 313 divides the image P_P2 into two in a direction parallel to the y axis, generating an image P_P4 and an image P_5P5.

[0136] Next, the dividing unit 313 sets the upper left corner of the image P_P4 as the division position where division starts, and generates a small image S_P1 divided into a predetermined size. Next, the dividing unit 313 slides the image P_P4 to the right parallel to the x-axis by a predetermined length, sets the point as the division position, and generates a small image S_P2 divided into a predetermined size.

[0137] If the point (division position) slid to the right parallel to the x-axis a predetermined length exceeds image P_P4, the division unit 313 sets the point slid downward in the y-axis direction a predetermined length from the top left of image P_P4 as the division position, and generates an image divided to a predetermined size, as in the process described above.

[0138] The division unit 313 repeats these processes to generate a plurality of divided small images S_P from the image P_P4.

[0139] The dividing unit 313 sets the upper right corner of the image P_P5 as the dividing position, and generates a small image S_P3 divided into a predetermined size. Next, the dividing unit 313 slides the image to the left in parallel with the x-axis by a predetermined length, and generates a small image S_P4 divided into a predetermined size.

[0140] If the point (division position) reached by sliding the image to the left parallel to the x-axis a predetermined length exceeds image P_P5, the division unit 313 sets the point reached by sliding the image downward in the y-axis direction a predetermined length from the top right of image P_P5 as the division position, and generates an image divided to a predetermined size, as in the process described above.

[0141] The division section 313 repeats these processes to generate divided small images S_P from the image P_P5.

[0142] (Processing example 2 executed by the division unit 313) Another example of the process executed by the dividing unit 313 will be described with reference to Fig. 18. Fig. 18 is a diagram showing another example of the process executed by the dividing unit 313. Below, the process of dividing the image P_P3 generated by the pre-processing unit 312 in the above-mentioned process will be described. Also in Fig. 18, the left-right direction of the image P_P3 will be described as the x-axis direction, and the up-down direction as the y-axis direction.

[0143] The dividing unit 313 sets the upper left corner of the image P_P3 as the dividing position, and generates a small image S_P5 divided into a predetermined size. Next, the dividing unit 313 slides the image downward parallel to the y-axis by a predetermined length, sets the point as the dividing position, and generates a small image P_P6 divided into a predetermined size.

[0144] The division unit 313 repeats these processes to generate divided small images S_P from the image P_P5.

[0145] (Example of a first machine learning model M1 and a second machine learning model M2) An example of the first machine learning model M1 and the second machine learning model M2 will be described.

[0146] As described above, the first machine learning model M1 is a machine learning model that is trained by machine learning a pair of an image of a normal wall material WM that does not include defects and a judgment result indicating that the wall material WM is normal, as training data. The following four examples are given as examples of the first machine learning model M1. A first machine learning model M1_1 that has been machine-trained using as training data a pair of an image including the wall material WM as a subject captured by a scanner, the small images divided by the division unit 313, and a determination result indicating that the wall material WM is normal. A first machine learning model M1_2 that has been machine-trained using as training data a set of a small image that is an image including the wall material WM as a subject captured by a scanner, that is divided by the division unit 313, and that is further edge-processed by the pre-processing unit 312, and a determination result that indicates that the wall material WM is normal. A first machine learning model M1_3 that has been machine-trained using as training data a pair of a small image, which is an image captured by a scanner and includes the wall material WM as a subject, divided by the division unit 313, and further has a specific color emphasized by the pre-processing unit 312, and a determination result indicating that the wall material WM is normal. A first machine learning model M1_4 that is machine-learned using as training data a pair of a 3D image including the wall material WM as a subject captured by a 3D scanner, the 3D small image segmented by the segmentation unit 313, and a determination result indicating that the wall material WM is normal. The second machine learning model M2 is a machine learning model that is trained by using as training data a pair of an image of a normal wall material WM and a judgment result indicating that the wall material WM is normal, and a pair of an image of a defective wall material WM and a judgment result indicating that the wall material WM is defective. The following two are examples of the first machine learning model M1. A second machine learning model M2_1 that has been machine-trained using as training data a pair of an image including the wall material WM as a subject captured by a scanner, the pair of small images divided by the division unit 313, and a determination result indicating that the wall material WM is normal, and a pair of an image including the wall material WM as a subject captured by a scanner, the pair of small images divided by the division unit 313, and a determination result indicating that the wall material WM is defective. A second machine learning model M2_2 that has been machine-trained using as training data a 3D image including the wall material WM as a subject captured by a 3D scanner, a pair of 3D small images segmented by the segmentation unit 313, and a determination result indicating that the wall material WM is normal, and a 3D image including the wall material WM as a subject captured by a 3D scanner, a pair of 3D small images segmented by the segmentation unit 313, and a determination result indicating that the wall material WM is defective. In this way, the quality inspection system 1A uses a machine learning model that is trained by machine learning various images as training data, such as images captured by a 2D scanner, images captured by a 3D scanner, images without preprocessing, images with preprocessing, images of normal wall material WM, and images of defective wall material WM. The machine learning model also includes at least one one-class model (first machine learning model M1) and at least one multi-class model (second machine learning model M2). Therefore, the quality inspection system 1A can detect various defects (scratches, dents, etc.) on the wall material WM with high accuracy.

[0147] (Examples of combinations of judgment results) A description will be given of an example of a combination of the determination results referred to by the detection unit 314. An example of a combination of the determination results of the above-mentioned machine learning models will be described below. In addition, each machine learning model and the determination result ("OK" or "NG") will be referred to as follows. - Judgment result R1 by the first machine learning model M1_1 -R2: Judgment result by the first machine learning model M1_2 - Judgment result R3 by the first machine learning model M1_3 - Judgment result R4 by the second machine learning model M2_1 - Judgment result R5 by the first machine learning model M1_4 - Judgment result R6 by the second machine learning model M2_2 As described above, the detection unit 314 detects defects in the wall material WM by performing a logical operation on at least two judgment results. As an example, the following formula (1) is used to detect defects in the wall material WM. Detection result = R1and(R2orR3)orR4orR5orR6...(1) When the detection result according to the above formula (1) is "OK", the detection unit 314 stores the detection result indicating that the wall material WM does not contain any defects in the storage unit 32. On the other hand, when the detection result according to the above formula (1) is "NG", the detection unit 314 stores the detection result indicating that the wall material WM contains any defects in the storage unit 32.

[0148] As described above, the detection unit 314 may change the combination of at least two judgment results to be referred to depending on the type of wall material WM. For example, the detection unit 314 may change the combination by referring to a table that associates the type of wall material WM with an equation indicating the combination. As an example, the detection unit 314 identifies the type of wall material WM based on the product number input to the interface IF2 of the above-mentioned image DP1.

[0149] For example, when the wall material WM is product A, the detection unit 314 detects defects contained in the wall material WM using the above-mentioned formula (1). Detection result = R1and(R2orR3)orR4orR5orR6...(1) Furthermore, when the wall material WM is the product B, the detection unit 314 detects defects contained in the wall material WM using the following formula (2). Detection result = (R1andR2andR3)orR4orR5orR6...(2) Furthermore, when the wall material WM is the product C, the detection unit 314 detects defects contained in the wall material WM using the following formula (3). Detection result = R2 or (R3 and R1) or R4 or R5 or R6 (3) Similarly, when the detection result according to the above formula is "OK", the detection unit 314 stores the detection result indicating that the wall material WM does not contain any defects in the storage unit 32. On the other hand, when the detection result according to the above formula is "NG", the detection unit 314 stores the detection result indicating that the wall material WM contains any defects in the storage unit 32.

[0150] In this way, the detection unit 314 changes the combination of at least two judgment results to be referred to depending on the type of the wall material WM. Therefore, the detection unit 314 can detect defects depending on the type of the wall material WM. Furthermore, the detection unit 314 detects defects contained in the wall material WM by performing a logical operation on at least two judgment results. Therefore, the detection unit 314 can detect defects depending on the contents judged by each judgment result.

[0151] (Effects of Quality Inspection System 1A) In this way, the quality inspection system 1A detects defects contained in the wall material WM by referring to a combination of at least two judgment results obtained by dividing the wall material WM into a plurality of small images and inputting the divided small images S_P into a first machine learning model M1 and a second machine learning model M2 trained by different learning methods. Therefore, the quality inspection system 1A detects defects using small images S_P divided from an image including a large-sized wall material WM as a subject, and is therefore able to detect small defects in the wall material WM.

[0152] Furthermore, the quality inspection system 1A divides the wall material WM into a plurality of small images. Then, the quality inspection system 1A outputs an image showing the defective portion. Therefore, the quality inspection system 1A can not only show the user whether or not the wall material WM has a defect, but also show the user where the defect is.

[0153] Third Exemplary Embodiment A third exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be given the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technical means employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technical means employed in this exemplary embodiment can be employed in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs. In addition, each technical means shown in each drawing referred to for explaining this exemplary embodiment can be employed in other exemplary embodiments included in this disclosure, as long as no particular technical hindrance occurs.

[0154] (Configuration of quality inspection device 2A) The configuration of the quality inspection device 2A will be described with reference to Fig. 19. Fig. 19 is a block diagram showing the configuration of the quality inspection device 2A.

[0155] As shown in FIG. 19, the quality inspection device 2A includes a control unit 20, a storage unit 26A, and a communication unit 27A.

[0156] The storage unit 26A stores data referenced by the control unit 20. Examples of data stored in the storage unit 26A include a first machine learning model M1 and a second machine learning model M2. In this case, parameters that define the first machine learning model M1 and the second machine learning model M2 may be stored in the storage unit 26A. Examples of the storage unit 26A include, but are not limited to, a flash memory, a HDD, an SSD, or a combination thereof.

[0157] The communication unit 27A is an interface that transmits and receives data via a network. For example, the communication unit 27A transmits data provided from the control unit 20 to other devices, and provides data received from other devices to the control unit 20. Examples of the communication unit 27A include, but are not limited to, communication chips in various communication standards such as Ethernet, Wi-Fi, and wireless communication standards for mobile data communication networks, and USB-compliant connectors.

[0158] (Control unit 20) The control unit 20 controls each of the components included in the quality inspection device 2A.

[0159] 19, the control unit 20 includes an acquisition unit 21A, a preprocessing unit 312A, a division unit 22A, a detection unit 23A, an output unit 24A, and a learning unit 414A. In this exemplary embodiment, the acquisition unit 21A, the preprocessing unit 312A, the division unit 22A, the detection unit 23A, the output unit 24A, and the learning unit 414A respectively realize an acquisition means, a preprocessing means, a division means, a detection means, an output means, and a learning means.

[0160] The acquisition unit 21A has the configuration of the above-mentioned acquisition unit 21 and acquisition unit 311, and acquires an image of the wall material WM. The acquisition unit 21A stores the acquired image in the storage unit 26A.

[0161] The preprocessing unit 312A has the configuration of the preprocessing unit 312 described above, and processes (preprocesses) the image. An example of the process in which the preprocessing unit 312A preprocesses the image is as described above. The preprocessing unit 312A stores the processed image in the storage unit 26A.

[0162] The dividing unit 22A has the configuration of the dividing unit 22 and the dividing unit 313 described above, and divides the image processed by the preprocessing unit 312A into a plurality of small images. An example of the process in which the dividing unit 22A divides an image is as described above. The dividing unit 22A stores the divided plurality of small images in the storage unit 26A.

[0163] The detection unit 23A has the configuration of the detection unit 23 and the detection unit 314 described above, and detects defects contained in the wall material WM by referring to a combination of at least two judgment results obtained by inputting the small images divided by the division unit 22A to at least one first machine learning model M1 and at least one second machine learning model M2. An example of the process in which the detection unit 23A detects defects is as described above. The detection unit 23A stores the detection result in the storage unit 26A.

[0164] Output section 24A has the configuration of output section 24 and output section 315 described above, and outputs the detection result by detection section 23A.

[0165] The learning unit 414A has the configuration of the learning unit 414 described above, and trains the first machine learning model M1 and the second machine learning model M2. An example in which the learning unit 414A trains the first machine learning model M1 and the second machine learning model M2 is as described above.

[0166] (Effect of quality inspection device 2A) In this way, the quality inspection device 2A detects defects contained in the wall material WM by referring to a combination of at least two judgment results obtained by dividing the wall material WM into a plurality of small images and inputting the divided small images into a first machine learning model M1 and a second machine learning model M2 trained by different learning methods. Therefore, the quality inspection device 2A also detects defects using images divided from an image including a large-sized wall material WM as a subject, so that it is possible to detect small defects in the wall material WM.

[0167] [Software implementation example] Some or all of the functions of the quality inspection system 1, quality inspection equipment 2, 2A, AP / DB server 3, and judgment server 4 (hereinafter also referred to as "the above-mentioned equipment") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0168] In the latter case, each of the above devices is realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in Fig. 20. Fig. 20 is a block diagram showing the hardware configuration of computer C that functions as each of the above devices.

[0169] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for causing the computer C to operate as each of the above-mentioned devices. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing the functions of each of the above-mentioned devices.

[0170] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0171] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0172] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0173] [Appendix A] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0174] (Appendix A1) An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; A quality inspection system equipped with

[0175] (Appendix A2) The at least two machine learning models include: At least one first machine learning model that is machine-trained using a pair of an image of a normal wall material that does not include defects and a determination result that indicates that the wall material is normal as training data; At least one second machine learning model that has been machine-learned using as training data a pair of an image of a normal wall material and a determination result indicating that the wall material is normal, and a pair of an image of a defective wall material and a determination result indicating that the wall material is defective; Contains, Quality inspection system as described in Appendix A1.

[0176] (Appendix A3) The detection means changes a combination of the at least two determination results to be referred to depending on the type of the wall material. A quality inspection system as described in Appendix A1 or A2.

[0177] (Appendix A4) The detection means includes: and detecting defects in the wall material by performing a logical operation on the at least two judgment results. A quality inspection system as described in any of Appendices A1 to A3.

[0178] (Appendix A5) The method further includes a pre-processing unit that performs at least one of combining the images acquired by the acquisition unit, correcting the inclination of the wall material in the images acquired by the acquisition unit, aligning the position of the wall material in the images acquired by the acquisition unit, processing the color of the images acquired by the acquisition unit, and processing the edges of the images acquired by the acquisition unit; The division means divides the image processed by the pre-processing means into a plurality of small images. A quality inspection system as described in any of Appendices A1 to A4.

[0179] (Appendix A6) The output means outputs an image showing the defective portion. A quality inspection system as described in any of Appendices A1 to A5.

[0180] (Appendix A7) The detection result includes at least one of information indicating whether or not an outer shape of the wall material is within a threshold value, and information indicating whether or not a pixel value of the wall material in the image acquired by the acquisition means is within a threshold value. Quality inspection system as described in Appendix A6.

[0181] (Appendix A8) a learning means for learning the first machine learning model and the second machine learning model; Further equipped A quality inspection system as described in any of Appendices A1 to A7.

[0182] (Appendix A9) An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; Equipped with quality inspection equipment.

[0183] (Appendix A10) An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; A learning means for learning at least two machine learning models by different learning methods, the machine learning models inputting an image of a wall material and outputting a judgment result of whether the wall material is defective or not; A learning system comprising:

[0184] (Appendix A11) An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; A learning means for learning at least two machine learning models by different learning methods, the machine learning models inputting an image of a wall material and outputting a judgment result of whether the wall material is defective or not; A learning device comprising:

[0185] [Appendix B] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0186] (Appendix B1) An acquisition process in which at least one processor acquires an image of the wall material; a segmentation process in which the at least one processor segments the image into a plurality of sub-images; a detection process in which the at least one processor detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output process in which the at least one processor outputs a detection result in the detection process; Including quality inspection methods.

[0187] (Appendix B2) The at least two machine learning models include: At least one first machine learning model that is machine-trained using a pair of an image of a normal wall material that does not include defects and a determination result that indicates that the wall material is normal as training data; At least one second machine learning model that has been machine-learned using as training data a pair of an image of a normal wall material and a determination result indicating that the wall material is normal, and a pair of an image of a defective wall material and a determination result indicating that the wall material is defective; Contains, Quality inspection method described in Appendix B1.

[0188] (Appendix B3) In the detection process, the at least one processor changes a combination of the at least two determination results to be referred to depending on the type of the wall material. Quality inspection methods described in Appendix B1 or B2.

[0189] (Appendix B4) In the detection process, the at least one processor and detecting defects in the wall material by performing a logical operation on the at least two judgment results. A quality inspection method as described in any of Appendices B1 to B3.

[0190] (Appendix B5) The at least one processor further includes a pre-processing step of performing at least one of the following: combining the images acquired in the acquisition process; correcting the inclination of the wall material in the images acquired in the acquisition process; aligning the position of the wall material in the images acquired in the acquisition process; color processing the images acquired in the acquisition process; and edge processing the images acquired in the acquisition process; In the division process, the at least one processor divides the image processed in the pre-processing into a plurality of small images. A quality inspection method as described in any of Appendices B1 to B4.

[0191] (Appendix B6) In the output process, the at least one processor outputs an image showing a defect location. A quality inspection method as described in any of Appendices B1 to B5.

[0192] (Appendix B7) The detection result includes at least one of information indicating whether an outer shape of the wall material is within a threshold value and information indicating whether a pixel value of the wall material in the image acquired by the acquisition process is within a threshold value. Quality inspection method described in Appendix B6.

[0193] (Appendix B8) a learning process in which the at least one processor trains the first machine learning model and the second machine learning model; Also includes A quality inspection method as described in any of Appendices B1 to B7.

[0194] (Appendix B9) An acquisition process in which at least one processor acquires an image of the wall material; a segmentation process in which the at least one processor segments the image into a plurality of sub-images; A learning process in which the at least one processor learns at least two machine learning models using different learning methods, the machine learning models inputting an image of a wall material and outputting a determination result of whether the wall material is defective or not; Learning methods including:

[0195] [Appendix C] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0196] (Appendix C1) A program for causing a computer to function as a quality inspection system, The computer, An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; This acts as a quality inspection program.

[0197] (Appendix C2) The at least two machine learning models include: At least one first machine learning model that is machine-trained using a pair of an image of a normal wall material that does not include defects and a determination result that indicates that the wall material is normal as training data; At least one second machine learning model that has been machine-learned using as training data a pair of an image of a normal wall material and a determination result indicating that the wall material is normal, and a pair of an image of a defective wall material and a determination result indicating that the wall material is defective; Contains, Quality inspection program described in Appendix C1.

[0198] (Appendix C3) The detection means changes a combination of the at least two determination results to be referred to depending on the type of the wall material. A quality inspection program as described in Appendix C1 or C2.

[0199] (Appendix C4) The detection means includes: and detecting defects in the wall material by performing a logical operation on the at least two judgment results. A quality inspection program as described in any of Appendices C1 to C3.

[0200] (Appendix C5) The computer, The image processing device further functions as a pre-processing device that performs at least one of combining the images acquired by the acquisition device, correcting the inclination of the wall material in the image acquired by the acquisition device, aligning the position of the wall material in the image acquired by the acquisition device, processing the color of the image acquired by the acquisition device, and processing the edges of the image acquired by the acquisition device; The division means divides the image processed by the pre-processing means into a plurality of small images. A quality inspection program as described in any of Appendices C1 to C4.

[0201] (Appendix C6) The output means outputs an image showing the defective portion. A quality inspection program as described in any of Appendices C1 to C5.

[0202] (Appendix C7) The detection result includes at least one of information indicating whether or not an outer shape of the wall material is within a threshold value, and information indicating whether or not a pixel value of the wall material in the image acquired by the acquisition means is within a threshold value. Quality Inspection Program as described in Appendix C6.

[0203] (Appendix C8) The computer, a learning means for learning the first machine learning model and the second machine learning model; Further functioning as A quality inspection program as described in any of Appendices C1 to C7.

[0204] (Appendix C9) A program for causing a computer to function as a learning system, The computer, A capture means for capturing images of the wall material, the capture means being operated by at least one processor; said at least one processor having a segmentation means for segmenting said image into a plurality of sub-images; A learning means for learning at least two machine learning models by different learning methods, the machine learning models inputting an image of a wall material and outputting a judgment result of whether the wall material is defective or not, A study program that includes:

[0205] [Appendix D] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0206] (Appendix D1) at least one processor, the at least one processor comprising: An acquisition process for acquiring an image of the wall material; a segmentation process for segmenting the image into a plurality of small images; a detection process for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output process for outputting a detection result obtained by the detection process; Carry out quality inspection system.

[0207] (Appendix D2) The at least two machine learning models include: At least one first machine learning model that is machine-trained using a pair of an image of a normal wall material that does not include defects and a determination result that indicates that the wall material is normal as training data; At least one second machine learning model that has been machine-learned using as training data a pair of an image of a normal wall material and a determination result indicating that the wall material is normal, and a pair of an image of a defective wall material and a determination result indicating that the wall material is defective; Contains, Quality inspection system as described in Appendix D1.

[0208] (Appendix D3) In the detection process, the at least one processor changes a combination of the at least two determination results to be referred to depending on the type of the wall material. A quality inspection system as described in Appendix D1 or D2.

[0209] (Appendix D4) In the detection process, the at least one processor and detecting defects in the wall material by performing a logical operation on the at least two judgment results. 1. A quality inspection system according to any of Appendices D1 to D3.

[0210] (Appendix D5) the at least one processor: Further performing a pre-processing process of performing at least one of combining the images acquired by the acquisition process, correcting the inclination of the wall material in the image acquired by the acquisition process, aligning the position of the wall material in the image acquired by the acquisition process, color processing the image, and edge processing of the image; In the division process, the at least one processor divides the image processed by the pre-processing process into a plurality of small images. 1. A quality inspection system according to any of appendices D1 to D4.

[0211] (Appendix D6) In the output process, the at least one processor outputs an image showing a defect location. 1. A quality inspection system according to any of Appendices D1 to D5.

[0212] (Appendix D7) The detection result includes at least one of information indicating whether an outer shape of the wall material is within a threshold value and information indicating whether a pixel value of the wall material in the image acquired by the acquisition process is within a threshold value. Quality inspection system as described in Appendix D6.

[0213] (Appendix D8) a learning process in which the at least one processor trains the first machine learning model and the second machine learning model; Run more 1. A quality inspection system according to any of Appendices D1 to D7.

[0214] (Appendix D9) at least one processor, the at least one processor comprising: An acquisition process for acquiring an image of the wall material; a segmentation process for segmenting the image into a plurality of small images; a detection process for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output process for outputting a detection result obtained by the detection process; Perform quality inspection equipment.

[0215] (Appendix D10) at least one processor, the at least one processor comprising: An acquisition process for acquiring an image of the wall material; a segmentation process for segmenting the image into a plurality of small images; A learning process for learning at least two machine learning models using different learning methods, the machine learning models inputting an image of a wall material and outputting a determination result of whether the wall material is defective or not; A learning system that executes.

[0216] (Appendix D11) at least one processor, the at least one processor comprising: An acquisition process for acquiring an image of the wall material; a segmentation process for segmenting the image into a plurality of small images; A learning process for learning at least two machine learning models using different learning methods, the machine learning models inputting an image of a wall material and outputting a determination result of whether the wall material is defective or not; A learning device that performs the following:

[0217] [Appendix E] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0218] (Appendix E1) A program for causing a computer to function as a quality inspection system, The computer includes: An acquisition process for acquiring an image of the wall material; a segmentation process for segmenting the image into a plurality of small images; a detection process for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output process for outputting a detection result obtained by the detection process; A quality inspection program is executed on a non-transient recording medium. [Explanation of symbols]

[0219] 1. 1A quality inspection system 2. 2A quality inspection device 3 AP / DB Server 4 Judgment Server 11, 21, 21A, 311, 411 Acquisition Department 12, 22, 22A, 312, 313 division part 13, 23, 23A, 314 Detector 14, 24, 24A, 315, 413 Output section 312, 312A Pretreatment section 412 Judgment section 414, 414A Learning Department M1 The first machine learning model M2 Second machine learning model

Claims

1. An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; A quality inspection system equipped with

2. The at least two machine learning models include: At least one first machine learning model that is machine-learned using a pair of an image of a normal wall material that does not include defects and a determination result that indicates that the wall material is normal as training data; At least one second machine learning model that has been machine-learned using as training data a pair of an image of a normal wall material and a determination result indicating that the wall material is normal, and a pair of an image of a defective wall material and a determination result indicating that the wall material is defective; Contains, The quality inspection system according to claim 1 .

3. the detection means changes a combination of the at least two determination results to be referred to depending on the type of the wall material. The quality inspection system according to claim 1 or 2.

4. The detection means includes: performing a logical operation on the at least two judgment results to detect defects contained in the wall material; The quality inspection system according to claim 1 or 2.

5. The method further includes a pre-processing unit that performs at least one of combining the images acquired by the acquisition unit, correcting the inclination of the wall material in the images acquired by the acquisition unit, aligning the position of the wall material in the images acquired by the acquisition unit, processing the color of the images acquired by the acquisition unit, and processing the edges of the images acquired by the acquisition unit; The division means divides the image processed by the pre-processing means into a plurality of small images. The quality inspection system according to claim 1 or 2.

6. The output means outputs an image showing the defective portion. The quality inspection system according to claim 1 or 2.

7. The detection result includes at least one of information indicating whether an outer shape of the wall material is within a threshold value and information indicating whether a pixel value of the wall material in the image acquired by the acquisition means is within a threshold value. The quality inspection system according to claim 6.

8. an acquisition process in which at least one processor acquires an image of the wall material; a segmentation process in which the at least one processor segments the image into a plurality of sub-images; a detection process in which the at least one processor detects defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; and an output process in which the at least one processor outputs a detection result in the detection process; Including quality inspection methods.

9. A program for causing a computer to function as a quality inspection system, The computer, An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; This acts as a quality inspection program.

10. An acquisition means for acquiring an image of a wall material; a dividing means for dividing the image into a plurality of small images; a detection means for detecting defects in the wall material by referring to a combination of at least two judgment results obtained by inputting the small image into at least two machine learning models that input an image of the wall material and output a judgment result that judges whether the wall material is defective or not, the at least two machine learning models being trained by different learning methods; an output means for outputting a detection result by the detection means; Equipped with quality inspection equipment.

Images