Inspection method, computer program for inspection device, and manufacturing method of substrate
By generating input images with differentiated display modes for parts and boundaries, the method addresses the misjudgment of normal boundaries as abnormal, enhancing defect detection accuracy in inspection methods.
Patent Information
- Application Number
- JP2023079287
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-12
- Publication Date
- 2024-11-22
Smart Images

Figure 2024163556000001_ABST
Abstract
Description
[Technical field]
[0001] The technology disclosed herein relates to inspecting an object for abnormalities. [Background technology]
[0002] Patent Document 1 discloses an inspection device that inputs an image of a target product into a machine learning model and outputs an image indicating an estimation result of the likelihood of a defect in the target product. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent Publication No. 2022-003495 Summary of the Invention [Problem to be solved by the invention]
[0004] Object inspection using a machine learning model can also be used to inspect the shape of the boundary between the first and second parts. However, when a binarized image containing the boundary was input into the learning model, a problem occurred in which not only parts of the boundary that had an abnormality in shape were judged to be abnormal, but also parts of the boundary that had no abnormality.
[0005] This specification provides a technique for accurately determining whether or not there is an abnormality in the shape of the boundary portion between a first portion and a second portion. [Means for solving the problem]
[0006] The inspection method disclosed in this specification includes a generation step of processing an image of an object captured by an imaging device to generate an input image, and a determination step of inputting the input image into a learning model to determine whether an abnormality exists in the shape of a boundary portion between a first portion and a second portion of the object, wherein the generation step generates the input image in which the first portion and the boundary portion have different display modes, and in which the boundary portion and the second portion have different display modes.
[0007] By using the input image, it becomes difficult to erroneously determine that a portion of the boundary portion that is not abnormal is abnormal. By using the input image, it is possible to accurately determine an abnormality that exists in the shape of the boundary portion.
[0008] In a second aspect, in the above-mentioned first aspect, the generating step may generate the input image in which the first portion and the second portion have display modes different from each other.
[0009] According to the above aspect, by using an input image in which the first portion and the second portion are further distinguished, it is possible to accurately determine an abnormality present in the shape of the boundary portion.
[0010] In a third aspect, in the above second aspect, the generating step may generate the input image in which the first portion and the second portion are colored differently from each other.
[0011] According to the above embodiment, the first portion and the second portion can be distinguished from each other by utilizing the difference in color.
[0012] In a fourth aspect, in the second aspect, the generating step may generate the input image in which the first portion and the second portion have patterns different from each other.
[0013] According to the above embodiment, the first portion and the second portion can be distinguished from each other by utilizing the difference in the patterns.
[0014] In a fifth embodiment, in the fourth embodiment above, the pattern applied to the first portion may have a plurality of first diagonal lines parallel to each other, and the pattern applied to the second portion may have a plurality of second diagonal lines parallel to each other, the plurality of second diagonal lines extending in a different direction than the plurality of first diagonal lines.
[0015] According to the above embodiment, the first portion and the second portion can be distinguished by utilizing the difference in the extending direction of the first oblique line and the second oblique line.
[0016] In a sixth aspect, in the second aspect, the generating step may generate the input image in which a luminance of a color of the first portion and a luminance of a color of the second portion are different from each other.
[0017] According to the above embodiment, the first portion and the second portion can be distinguished from each other by utilizing the difference in color luminance.
[0018] In a seventh aspect, in the sixth aspect above, the generation step may generate the input image in which the brightness of the color of the boundary portion is higher than the brightness of the color of the first portion and the brightness of the color of the second portion.
[0019] According to the above embodiment, the boundary portion can be emphasized with a relatively high luminance more than the first portion and the second portion.
[0020] In an eighth aspect, in the first and second aspects above, the generation process may generate the input image in which the first portion and the boundary portion are colored differently from each other, and the boundary portion and the second portion are colored differently from each other.
[0021] According to the above embodiment, the first portion, the boundary portion, and the second portion are distinguished from each other by using the difference in color. By using the input image in which each portion is distinguished, it is possible to accurately determine an abnormality present in the shape of the boundary portion.
[0022] In a ninth aspect, in the first and second aspects above, the generation process may generate the input image in which the first portion and the boundary portion are patterned differently from each other, and the boundary portion and the second portion are patterned differently from each other.
[0023] According to the above embodiment, the first portion, the boundary, and the second portion are distinguished from each other by utilizing the difference in the patterns. By utilizing the input image in which each portion is distinguished, it is possible to accurately determine an abnormality present in the shape of the boundary.
[0024] In a tenth form, in the first and second forms above, the generation process may generate the input image in which the luminance of the color of the first portion and the luminance of the color of the boundary portion are different from each other, and the luminance of the color of the boundary portion and the luminance of the color of the second portion are different from each other.
[0025] According to the above embodiment, the first portion, the boundary, and the second portion are distinguished from each other by utilizing the difference in color luminance. By utilizing the input image in which each portion is distinguished, it is possible to accurately determine an abnormality present in the shape of the boundary.
[0026] In an eleventh aspect, in the first and second aspects, the display manner may be a combination of at least two of a color, a pattern, and a brightness of a color.
[0027] According to the above embodiment, it is possible to distinguish the first part, the boundary, and the second part by combining two or more elements. By using an input image in which each part is distinguished, it is possible to accurately determine an abnormality present in the shape of the boundary.
[0028] In a twelfth form, in any of the first to eleventh forms above, the first portion may be an outer portion that is outside the object, the second portion may be an inner portion that is inside the object, and the boundary portion may be an edge of the object.
[0029] According to the above aspect, an abnormality (for example, a defect) present on the edge of an object can be determined with high accuracy.
[0030] In a thirteenth embodiment, in the first to twelfth embodiments described above, the method further includes a binarization step of binarizing the image of the object captured by the imaging device, and the generation step may process the image binarized by the binarization step to generate the input image.
[0031] According to the above aspect, the input image can be generated with higher accuracy, compared to a configuration in which the input image is generated directly from an image captured by an imaging device.
[0032] In a 14th form, in the above-mentioned 1st to 13th forms, the parameters of the learning model are adjusted using a plurality of good product images, and the good product images may be images in which no abnormalities exist in the boundary portion.
[0033] According to the above embodiment, it is sufficient to collect images of non-defective products in order to adjust the parameters of the learning model, and it is not necessary to collect images in which an abnormality exists at the boundary. The parameters of the learning model can be easily adjusted.
[0034] This specification further discloses a computer program for an inspection device, which causes a computer of the inspection device to function as a generation unit that processes an image of an object captured by an imaging device to generate an input image, and a determination unit that inputs the input image to a learning model and determines an abnormality present in the shape of a boundary portion between a first portion and a second portion of the object, and the generation unit generates the input image in which the first portion and the boundary portion have different display modes, and the boundary portion and the second portion have different display modes.
[0035] This specification further discloses a method for manufacturing a substrate, the manufacturing method including an inspection step for inspecting a substrate, the inspection step including a generation step for processing an image of the substrate captured by an imaging device to generate an input image, and a determination step for inputting the input image into a learning model to determine an abnormality present in the shape of a boundary portion between a first portion and a second portion of the substrate, the generation step generating the input image in which the first portion and the boundary portion have different display modes, and the boundary portion and the second portion have different display modes. [Brief description of the drawings]
[0036] [Figure 1] FIG. 2 is a block diagram of a substrate manufacturing apparatus. [Diagram 2] FIG. 1 is a diagram showing an inspection target for a non-defective product; [Diagram 3] FIG. 1 is a diagram showing an inspection target for defective products. [Figure 4] FIG. 11 is a flow chart showing an inspection process. [Diagram 5] 11 is a specific example of an output image obtained in the inspection process of the comparative example. [Figure 6] 11 is a specific example of an output image obtained in the inspection process of the embodiment. [Figure 7] This is a specific case A2 in which the display mode change process is performed on an image of another inspection target. [Figure 8] This is a specific case A3 in which the display mode change process is performed on an image of another inspection target. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0037] (Configuration of substrate manufacturing apparatus 2; Figures 1, 2, and 3) The board manufacturing apparatus 2 manufactures circuit boards. The circuit boards are, for example, printed circuit boards and LTCC boards. LTCC is an abbreviation for Low Temperature Co-fired Ceramics. The board manufacturing apparatus 2 includes a printer 4 and an inspection device 10. The printer 4 prints a wiring pattern on the board. The inspection device 10 inspects the wiring pattern printed by the printer 4 for abnormalities.
[0038] The inspection device 10 includes an imaging device 20, a display device 22, and a control unit 30. The imaging device 20 includes a camera. The imaging device 20 uses the camera to capture an image of a board. Specifically, when a board on which a wiring pattern is printed is carried in from the printing machine 4, the imaging device 20 captures an image of the board. Then, the imaging device 20 supplies the captured image of the board to the control unit 30.
[0039] The display device 22 displays various information, such as images captured by the imaging device 20, and results of inspection by the inspection device 10. The display device 22 is, for example, a liquid crystal display.
[0040] The control unit 30 includes a CPU 32 and a memory 34 configured by a non-volatile memory etc. The CPU 32 executes an inspection process (see FIG. 4) described later in accordance with an inspection program 40 stored in the memory 34.
[0041] In addition to the inspection program 40, the memory 34 stores a learning program 42, a learning model 44, and model parameters 46. The learning model 44 is a model (i.e., a mathematical formula) of a multilayer neural network. The multilayer neural network is a so-called deep learning model, such as a convolutional type or a fully connected type. The model parameters 46 are the values of each weight in the intermediate layer of the learning model 44. The multilayer neural network is a known technology, and a detailed description thereof will be omitted here.
[0042] An abnormality may occur in the shape of a wiring pattern. Figures 2 and 3 show an example of a wiring pattern 100 to be inspected. The wiring pattern 100 is printed on a substrate 110. Figure 2 shows an example of a good product, in which no defect exists at an edge 102 of the wiring pattern 100. On the other hand, Figure 3 shows an example of a defective product, in which a defect 104 exists at an edge 102 of the wiring pattern 100.
[0043] The learning program 42 adjusts the model parameters 46 so as to output a judgment image in which defects existing in the edge 102 are judged from an image of the wiring pattern 100. In the judgment image, a portion where a defect exists is displayed with a higher brightness than other portions. For example, the learning program 42 adjusts the model parameters 46 using a plurality of good product images. The good product images are images of good products in which the edge 102 does not have a defect 104. According to this embodiment, in order to adjust the model parameters 46, it is sufficient to collect good product images, and it is not necessary to collect defective product images in which the edge 102 has a defect 104. The model parameters 46 can be easily adjusted. In a modified example, the learning program 42 may adjust the model parameters 46 using both good product images and defective product images. In addition, the method of adjusting the model parameters 46 is a known technique, and detailed description thereof will be omitted here. For example, the adjustment method may be either supervised learning or unsupervised learning.
[0044] The display device 22 and the control unit 30 are, for example, a terminal device such as a PC, and the imaging device 20 is provided separately from the terminal device. In a modified example, the imaging device 20, the display device 22, and the control unit 30 may be integrally configured.
[0045] (Inspection process; Figure 4) The inspection process executed by the CPU 32 in accordance with the inspection program 40 will be described. In S10, the CPU 32 monitors whether an image of the board 110 on which the wiring pattern 100 is printed is acquired from the imaging device 20. If the CPU 32 acquires an image from the imaging device 20 (YES in S10), the CPU 32 proceeds to S12. Note that if the CPU 32 does not acquire an image from the imaging device 20 (NO in S10), the CPU 32 continues monitoring.
[0046] In S12, the CPU 32 binarizes the captured image acquired in S10. For example, by binarization, the inner portion that is the inside of the wiring pattern 100 is filled with a single color of white, and the outer portion that is the outside of the wiring pattern 100, i.e., the background, is filled with a single color of black.
[0047] In S14, the CPU 32 executes a display mode change process on the captured image binarized in S12. The display mode change process is image processing for changing the display mode of each part of the binarized captured image. Case A1 in FIG. 4 is a case in which the display mode change process is executed on the captured image of the wiring pattern 100 in FIG. 3. The captured image of the wiring pattern 100 includes an inside image 130 corresponding to the inside part of the wiring pattern 100, a background image 136 corresponding to the background of the wiring pattern 100, and a boundary image 132 corresponding to the boundary between the inside image 130 and the background image 136. The boundary image 132 corresponds to the edge 102 of the wiring pattern 100. The boundary image 132 includes a defect image 134 corresponding to a defect 104 present in the edge 102.
[0048] In the display mode change process, background image 136, boundary image 132, and inner image 130 are changed to have different luminances. In case A1, the luminances of background image 136, boundary image 132, and inner image 130 are "150," "255," and "75," respectively. That is, the luminance of boundary image 132 is higher than the luminance of inner image 130 and background image 136. Note that the above specific numerical values are merely examples.
[0049] In addition, in the display mode change process, different patterns are applied to the background image 136, the boundary image 132, and the inner image 130. In case A1, the pattern applied to the background image 136 has a plurality of first diagonal lines parallel to each other, and the pattern applied to the inner image 130 has a plurality of second diagonal lines parallel to each other. The second diagonal lines extend in a direction different from the first diagonal lines. Specifically, the first diagonal lines extend toward the upper right of the paper, and the second diagonal lines extend toward the upper left of the paper. Note that the boundary image 132 is not diagonally lined and is filled with a single color. Note that the above types of patterns are merely examples. For example, a dot pattern may be applied to one of the inner image 130 and the background image 136, and a lattice pattern may be applied to the other of the inner image 130 and the background image 136.
[0050] In addition, in the display mode change process, different colors are assigned to the background image 136, the boundary image 132, and the inner image 130. In case A1, the background image 136, the boundary image 132, and the inner image 130 are assigned green (G), blue (B), and red (R), respectively. Note that the above colors are merely examples. For example, the background image 136, the boundary image 132, and the inner image 130 may be assigned black, yellow, and white, respectively.
[0051] In the display mode change process of S14, the captured image binarized in S12 is processed to generate an input image 120 as shown in case A1.
[0052] In S16, the CPU 32 inputs the input image 120 generated in S14 to the learning model 44. This causes a judgment image to be output from the learning model 44. In S18, the CPU 32 acquires the judgment image from the learning model 44. Details of the judgment image will be described with reference to FIG.
[0053] In S20, the CPU 32 analyzes the judgment image to determine the presence or absence of a defect in the edge 102 of the wiring pattern 100, the position of the defect, the size of the defect, etc. In S22, the CPU 32 outputs the analysis result of S20. For example, the CPU 32 causes the display device 22 to display the analysis result of S20. A user of the inspection device 10 can see the analysis result displayed on the display device 22 and know the presence or absence of a defect, the position of the defect, the size of the defect, etc.
[0054] (Comparative example: Figure 5) For example, a comparative example is assumed in which the display mode change process of S14 in Fig. 4 is not executed. In this comparative example, the captured image binarized in S12 is input to the learning model 44. In this modification, the model parameters 46 of the learning model 44 are adjusted by using the binarized non-defective image.
[0055] FIG. 5 is an example of a judgment image 900 acquired in S18 in the comparative example. In the judgment image 900, a judgment point 902 where it is judged that a defect exists is displayed with a higher brightness than other points. In FIG. 5 and FIG. 6 described later, the edge 102 corresponding to the judgment image 900 is virtually displayed by a dashed line. In the comparative example, as shown in FIG. 5, the judgment point 902 includes not only a part of the edge 102 where the defect 104 exists, but also a part of the edge 102 where the defect 104 does not exist. In particular, if the ratio of the part where the defect 104 does not exist to the judgment point 902 increases, in the analysis of S20 in FIG. 4, it may be erroneously analyzed that the defect 104 exists in the part where the defect 104 does not exist. In addition, the position and size of the defect 104 may also be erroneously analyzed. Furthermore, it may be erroneously analyzed that a defect exists in the side of the edge 102 where the defect 104 does not exist.
[0056] (Effect of this embodiment; Figure 6) 4, in this embodiment, an input image 120 whose display mode has been changed by the display mode change process of S14 is input to the learning model 44. In this embodiment, the model parameters 46 of the learning model 44 are adjusted by using a non-defective image whose display mode has been changed by the display mode change process.
[0057] FIG. 6 is an example of a judgment image 140 acquired in S18 in this embodiment. In the judgment image 140, a judgment point 142 that is judged to have a defect is displayed with a higher brightness than other points. As shown in FIG. 6, the judgment point 142 in this embodiment includes some parts where the defect 104 does not exist, but the ratio of the parts where the defect 104 does not exist in the judgment point 142 is smaller than that in the comparative example in FIG. 5. That is, in this embodiment, compared to the comparative example, it is difficult for the learning model 44 to erroneously judge that a part of the edge 102 where the defect 104 does not exist is a defect. In this embodiment, compared to the comparative example, it is possible to reduce the probability of erroneously analyzing that a defect 104 exists in a part where the defect 104 does not exist in the analysis in S20 of FIG. 4. In addition, it is also possible to reduce the probability of erroneously analyzing the position and size of the defect 104.
[0058] In this embodiment, the background image 136 and the boundary image 132 have different display modes (i.e., brightness, pattern, color) and are distinguished from each other. The boundary image 132 and the inside image 130 have different display modes and are distinguished from each other. By using the input image 120 in which adjacent portions are distinguished, defects present in the edge 102 can be determined with high accuracy.
[0059] In this embodiment, the background image 136 and the inner image 130 have different display modes, and are distinguished from each other. By using the input image 120 in which not only the adjacent portions but also the background image 136 and the inner image 130 are distinguished from each other, it is possible to accurately determine defects present in the edge 102.
[0060] Moreover, according to the configuration of this embodiment, it is possible to distinguish each part of the input image 120 by combining three elements, namely, brightness, pattern, and color. By using the input image 120 in which each part is distinguished, it is possible to accurately determine defects present in the edge 102.
[0061] In this embodiment, the first and second oblique lines extend in a direction different from the direction in which the boundary image 132 extends. This makes it possible to further distinguish the background image 136 from the background image 136 and the inner image 130.
[0062] Furthermore, in this embodiment, the brightness of the boundary image 132 is higher than the brightness of the inner image 130 and the brightness of the background image 136. The boundary image 132 can be emphasized more than the background image 136 and the inner image 130 with a relatively high brightness.
[0063] In this embodiment, the input image 120 is generated from a binarized captured image. Compared to a configuration in which the input image 120 is generated directly from a captured image, the input image 120 can be generated with higher accuracy.
[0064] (Substrate manufacturing method: Figures 1 and 4) The substrate manufacturing method realized by the substrate manufacturing apparatus 2 includes a printing step in which the printer 4 prints the wiring pattern 100 on the substrate, and an inspection step in which the inspection apparatus 10 inspects the wiring pattern 100 printed by the printing step. In the inspection step, the inspection apparatus 10 executes the inspection process of Fig. 4 and displays on the display device 22 the analysis results such as the presence or absence of defects in the edge 102, the position of the defects, the size of the defects, etc.
[0065] (Other case A2; Figure 7) FIG. 7 describes a case A2 in which a display mode change process is performed on a captured image of a wiring pattern different from the wiring pattern 100 in FIG. 3. The objects of imaging in this case include a first wiring pattern and a second wiring pattern adjacent to the first wiring pattern. The captured image in this case includes a background image 216, an inside image 210, a boundary image 212, an inside image 220, and a boundary image 222. The background image 216 corresponds to the background of the first and second wiring patterns. The inside image 210 and the boundary image 212 correspond to the inside part of the first wiring pattern and the edge of the first wiring pattern, respectively. The inside image 220 and the boundary image 222 correspond to the inside part of the second wiring pattern and the edge of the second wiring pattern, respectively.
[0066] The brightness, pattern, and color of background image 216 are similar to case A1 of Figure 4. The brightness and pattern of inner images 210 and 220 are similar to case A1 of Figure 4. The brightness and pattern of border images 212 and 222 are similar to case A1 of Figure 4.
[0067] In this case, background image 216 is colored green (G). Border image 212 is colored blue (B), and border image 222 is colored yellow (Y). In this embodiment, inner image 210 is colored red (R), and inner image 220 is colored pink (P). That is, five portions 210, 212, 220, 222, and 216 of the captured image are colored differently from one another.
[0068] In this case, the boundary image 212 includes a defect image 214 due to the presence of a defect at the edge of the first wiring pattern. In this case, the learning model 44 is also less likely to erroneously determine that a portion of the edges of the first and second wiring patterns that does not have a defect is defective. This makes it possible to reduce the probability of erroneously analyzing that a portion that does not have a defect is defective.
[0069] (Other case A3; Figure 8) In this case, two types of wiring patterns 300 and 310 are printed on a substrate 320. When the substrate 320 is viewed from above, the wiring pattern 310 overlaps with the wiring pattern 300. The material, color, etc. of the wiring pattern 310 are different from those of the wiring pattern 300. In this case, a defect 314 exists in a boundary portion 312 between the wiring patterns 300 and 310.
[0070] The captured image in this case includes a first inside image 350 corresponding to a first portion inside the wiring pattern 300, a boundary image 362 corresponding to a boundary portion 312 between the wiring pattern 300 and the wiring pattern 310, and a second inside image 360 corresponding to a second portion inside the wiring pattern 310. The boundary image 362 in the captured image includes a defect image 364 corresponding to the defect 314.
[0071] In the display mode change process in this case, the display mode of the image corresponding to the background of the wiring patterns 300 and 310 is not changed, and the display mode of each image 350, 362, and 360 other than the background is changed. Specifically, the luminance of the first inner image 350, the border image 362, and the second inner image 360 is changed to "150", "255", and "75", respectively. In addition, the pattern applied to the first inner image 350 has a first oblique line, and the pattern applied to the second inner image 360 has a second oblique line. In addition, green (G), blue (B), and red (R) are applied to the first inner image 350, the border image 362, and the second inner image 360, respectively.
[0072] In this case, in the display mode change process of S14 in Fig. 4, the captured images of the two wiring patterns 300 and 310 are processed to generate an input image 340 as shown in Fig. 8. Then, in S16, the input image 340 is input to the learning model 44. The learning model 44 outputs a judgment image in which a judgment location where it has been judged that a defect 314 exists is displayed with a higher brightness than other locations. Although the judgment image in this case also contains some parts where the defect 314 does not exist, the ratio of the parts where the defect 314 does not exist among the judgment locations can be kept small. This makes it possible to reduce the probability of erroneously analyzing a part where the defect 314 does not exist as being defective.
[0073] (Correspondence) The inspection device 10 and the imaging device 20 are examples of an "inspection device" and an "imaging device", respectively. The CPU 32, the inspection program 40, and the learning model 44 are examples of a "computer", a "computer program", and a "learning model", respectively. The input image 120 is an example of an "input image".
[0074] S12, S14, and S16 in FIG. 4 are examples of processes for realizing the "binarization step", "generation step", and "determination step", respectively.
[0075] In case A1 of FIG. 4, the wiring pattern 100 is an example of an "object." The outer portion of the wiring pattern 100 is an example of a "first portion (and outer portion)." The inner portion of the wiring pattern 100 is an example of a "second portion (and inner portion)." The edge 102 is an example of a "boundary portion (and edge)."
[0076] In case A2 of Fig. 7, the first and second wiring patterns are an example of an "object". The background corresponding to background image 216 is an example of a "first portion (and outer portion)". The inner portion corresponding to inner images 210 and 220 is an example of a "second portion (and inner portion)". The edges corresponding to boundary images 212 and 222 are an example of a "boundary portion (and edge)".
[0077] In case A3 of Fig. 8, wiring patterns 300 and 310 are an example of an "object." The inner part of wiring pattern 300 excluding the inner part of wiring pattern 310 is an example of a "first part." The inner part of wiring pattern 310 is an example of a "second part." Boundary part 312 is an example of a "boundary part."
[0078] Although specific examples of the technology disclosed in this specification have been described above, these are merely examples and do not limit the scope of the claims. The technology described in the claims includes various modifications and variations of the specific examples exemplified above. For example, the following modifications may be adopted.
[0079] (Variation 1) In the display mode change process of S14 in Fig. 4, the brightness may be changed, but the color and the pattern may not be changed. Also, in the display mode change process, the brightness and the color may be changed, but the pattern may not be changed. In this variation, it is sufficient that at least one of the brightness, the color, and the pattern is changed in the display mode change process.
[0080] (Variation 2) In input image 120 in Fig. 4, the display mode of inner image 130 and the display mode of background image 136 may be the same. In this variation, for example, inner image 130 and background image 136 may be white, and boundary image 132 may be black. Generally speaking, it is sufficient for the "input image" to have a first portion and a boundary portion having different display modes, and a boundary portion and a second portion having different display modes.
[0081] (Modification 3) In the input image 120, the luminance of the portion other than the boundary image 132, for example, the inner image 130, may be higher than the luminance of the other portions.
[0082] (Modification 4) The process of S12 in Fig. 4 does not have to be executed. In this modification, the "binarization step" can be omitted.
[0083] (Modification 5) The "target object" is not limited to a wiring pattern on a substrate, but may be a structure having a predetermined shape, such as a glass plate, a glass tube, or the like.
[0084] (Variation 6) The inspection apparatus 10 does not need to include the imaging apparatus 20. For example, there may be another imaging apparatus independent of the inspection apparatus 10, and the inspection apparatus 10 may receive captured images from the other imaging apparatus via wireless communication such as a wireless LAN.
[0085] The technical elements described in this specification or drawings have technical utility either alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. In addition, the technologies illustrated in this specification or drawings achieve multiple objectives simultaneously, and achieving one of these objectives is itself technically useful. [Explanation of symbols]
[0086] 2: Board manufacturing equipment 4:Printing machine 10: Inspection equipment 20: Imaging device 22:Display device 30: Control section 32 :CPU 34: Memory 40: Inspection Program 42: Study Program 44: Learning model 46: Model parameters 100: Wiring pattern 102: Edge 104: Defect 110: Substrate 120: Input image 130: Inside image 132 :Border image 134: Defect image 136: Background image 140: Judgment image 142: Judgment point 210:Inside image 212 :Border image 214: Defect image 216: Background image 220: Inside image 222 :Border image 300: Wiring pattern 310: Wiring pattern 312: Boundary part 314: Defect 320: Substrate 340: Input image 350: First inside image 360: Second inside image 362: Boundary image 364: Defect image 900: Judgment image 902: Judgment point
Claims
1. a generating step of processing an image of an object captured by an imaging device to generate an input image; a determination step of inputting the input image into a learning model and determining an abnormality present in the shape of a boundary portion between a first portion and a second portion of the object; Equipped with An inspection method, wherein the generation step generates the input image in which the first portion and the boundary portion have different display modes, and the boundary portion and the second portion have different display modes.
2. The inspection method according to claim 1 , wherein the generating step generates the input image in which the first portion and the second portion have different display modes.
3. The inspection method according to claim 2 , wherein the generating step generates the input image in which the first portion and the second portion are colored differently from each other.
4. The inspection method according to claim 2 , wherein the generating step generates the input image in which the first portion and the second portion have patterns different from each other.
5. The pattern on the first portion has a plurality of first oblique lines parallel to each other, The inspection method according to claim 4 , wherein the pattern applied to the second portion includes a plurality of second oblique lines parallel to each other, the plurality of second oblique lines extending in a direction different from that of the plurality of first oblique lines.
6. The inspection method according to claim 2 , wherein the generating step generates the input image in which a luminance of a color of the first portion and a luminance of a color of the second portion are different from each other.
7. The inspection method according to claim 6 , wherein the generating step generates the input image in which a luminance of a color of the boundary portion is higher than a luminance of a color of the first portion and a luminance of a color of the second portion.
8. The inspection method according to claim 1 or 2, wherein the generating step generates the input image in which the first portion and the boundary portion are colored differently from each other, and the boundary portion and the second portion are colored differently from each other.
9. The inspection method according to claim 1 or 2, wherein the generating step generates the input image in which the first portion and the boundary portion are patterned differently from each other, and the boundary portion and the second portion are patterned differently from each other.
10. The inspection method according to claim 1 or 2, wherein the generation step generates the input image in which the brightness of the color of the first portion and the brightness of the color of the boundary portion are different from each other, and the brightness of the color of the boundary portion and the brightness of the color of the second portion are different from each other.
11. The inspection method according to claim 1 , wherein the display mode is a combination of at least two of a color, a pattern, and a color brightness.
12. The first portion is an outer portion that is an outer side of the object, The second portion is an inner portion that is inside the object, The inspection method according to claim 1 , wherein the boundary portion is an edge of the object.
13. a binarization step of binarizing the image of the object captured by the imaging device, The inspection method according to claim 1 , wherein the generating step generates the input image by processing the image binarized in the binarization step.
14. The parameters of the learning model are adjusted using a plurality of images of non-defective products; The inspection method according to claim 1 , wherein the non-defective image is an image in which no abnormality exists in the boundary portion.
15. A computer program for an inspection apparatus, comprising: The computer program causes a computer of the inspection device to a generation unit that processes an image of an object captured by an imaging device to generate an input image; a determination unit that inputs the input image into a learning model and determines an abnormality present in a shape of a boundary portion between a first portion and a second portion of the object; Function as a The generation unit generates the input image in which the first portion and the boundary portion have different display modes, and the boundary portion and the second portion have different display modes.
16. A method for manufacturing a substrate, comprising an inspection step of inspecting the substrate, The inspection step includes: a generating step of processing an image of the substrate captured by an imaging device to generate an input image; a determination step of inputting the input image into a learning model and determining whether an abnormality exists in the shape of a boundary portion between a first portion and a second portion of the substrate; Equipped with A method for manufacturing a substrate, wherein the generation process generates an input image in which the first portion and the boundary portion have different display modes, and the boundary portion and the second portion have different display modes.
Citation Information
Patent Citations
Inspection device, unit selection device, inspection method, and inspection program
JP2022003495A
