JP6218094B1 - Inspection method, inspection apparatus, inspection program, and recording medium - Google Patents

Abstract

An inspection method capable of suppressing the occurrence of at least one of false information and oversight and improving the detection accuracy of a defect. An inspection method is an inspection method for detecting a defect by performing image processing on a captured image of an inspection object, and registering a reference image of the inspection object and an inspection standard for detecting the defect (a) ), A step (b) of acquiring a captured image, and a step of detecting a defect by performing image processing for comparing the reference image and the captured image using the inspection standard registered in step (a) (c) And (d) recognizing an overdetected defect that should not be detected as a defect from among the defects detected in step (c), and the number of overdetected defects recognized in step (d) (E) for adjusting the inspection standard such that the inspection standard is reduced, and (f) for storing the inspection standard adjusted in step (e). [Selection] Figure 3

Description

  The present invention relates to an inspection method, an inspection apparatus, an inspection program, and a recording medium.

  In general, optical inspection using a camera is used for inspection of a printed circuit board having a copper foil wiring pattern formed on a resin film.

  In optical inspection, a method of detecting a defect by performing image processing for comparing a captured image of an inspection object with a reference image by a computer is widely used (for example, Patent Document 1).

Japanese Patent Laid-Open No. 05-264467

  However, in the above method, there is a possibility that so-called false information (overdetection) that erroneously recognizes a non-defective part as a defect may occur, and the image processing result by the computer must be rechecked by the operator. There is. Furthermore, in the above method, there is a possibility that a so-called oversight (undetected) in which a defect cannot be recognized may occur.

  The present invention has been made in view of the above circumstances, and can suppress the occurrence of at least one of false information and oversight and improve the detection accuracy of a defect, an inspection method, an inspection apparatus, an inspection program, and a recording The purpose is to provide a medium.

An inspection method according to the present invention that achieves the above-described object is an inspection method for detecting a defect by performing image processing on a captured image of an inspection object, the inspection image for detecting a reference image of the inspection object and the defect An image for comparing the reference image and the captured image using the inspection standard registered in the step (a) for registering a reference, the step (b) for acquiring the captured image, and the inspection standard registered in the step (a). Performing a process to detect the defect (c); recognizing an overdetected defect that should not be detected as a defect from the defects detected in the step (c); Adjusting the inspection criteria so that the number of over-detected defects recognized in the step (d) is reduced; and adjusting the inspection criteria in the step (e) (F) storing an inspection criterion, and between the step (c) and the step (e), an undetected undetected in the defect detected in the step (c) Recognizing defects, further comprising the step of (g), wherein in step (e) the inspection criteria are further adjusted so that the number of undetected defects recognized in step (g) is reduced; The step (b), the step (c), the step (d), and the step (g) are repeatedly performed on a plurality of captured images, and the step (e) is performed on the plurality of captured images. Whether or not it is an operator defect with respect to the defect of the same type and size after the step (f). Size And step (l) the result display different pair of the captured image, the step of accepting a change of the operator's judgment results or computer defect in a determination of whether the result of the (m), that have a.

An inspection apparatus according to the present invention that achieves the above object is an inspection apparatus that detects a defect by performing image processing on a captured image of an inspection object, and detects a reference image of the inspection object and the defect. A registration unit that registers the inspection standard, an acquisition unit that acquires the captured image, and an image process that compares the reference image with the captured image using the inspection standard registered by the registration unit. A detection unit that detects the defect; a first recognition unit that recognizes an overdetection defect that should not be detected as a defect from the defects detected by the detection unit; and the detection unit that detects the defect A first recognizing unit for recognizing an undetected defect that does not exist in the defect, and a first adjusting the inspection standard so that the number of the over-detected defects recognized by the first recognizing unit is reduced. Adjustment section The second adjustment unit that adjusts the inspection standard, the first adjustment unit, and the second adjustment unit adjusted so that the number of the undetected defects recognized by the second recognition unit decreases. A storage unit that stores inspection criteria, a display unit that displays a pair of the captured images with different determination results as to whether or not the defect is the same as the defect for the same type and size, and the worker A receiving unit that accepts a change in the determination result or whether the determination result is a computer defect, a process for acquiring the captured image by the acquisition unit, a process for detecting the defect by the detection unit, The process of recognizing the overdetected defect by the first recognizing unit and the process of recognizing the undetected defect by the second recognizing unit are repeatedly performed on a plurality of captured images, and the adjustment unit By using the defect of the defect and the undetected recognized the over detection for the plurality of captured images, to adjust the inspection standard.

An inspection program for detecting a defect by performing image processing on a captured image of an inspection object, the step (a) of registering a reference image of the inspection object and an inspection standard for detecting the defect, Procedure (b) for acquiring a captured image, and procedure for detecting the defect by performing image processing for comparing the reference image and the captured image using the inspection standard registered in the procedure (a). (C), a procedure (d) for recognizing an overdetected defect that should not be detected as a defect from the defects detected in the procedure (c), and the procedure recognized in the procedure (d) Causing the computer to execute a step (e) of adjusting the inspection standard and a step (f) of storing the inspection standard adjusted in the step (e) so that the number of overdetected defects is reduced. The hand Between the step (c) and the step (e), the computer further executes a step (g) of recognizing an undetected defect that does not exist in the defect detected in the step (c), In the step (e), the inspection standard is further adjusted so that the number of the undetected defects recognized in the step (g) is reduced, and the step (b), the step (c), the step (D) and the procedure (g) are repeatedly performed on a plurality of captured images, and the procedure (e) includes the overdetection defect recognized on the plurality of captured images and the non-detection. A procedure of displaying a pair of the captured images with different determination results as to whether or not the defect is the operator's defect for the defect of the same type and size after the step (f). l) and the judgment result of the operator Or a procedure for accepting a change in the computer the defect is a determination of whether the result of the (m), Ru is executed further the computer to.

  A recording medium according to the present invention that achieves the above object is a computer-readable recording medium that records the above-described inspection program.

  According to the present invention, the inspection standard is adjusted so that the number of overdetected defects is reduced or the number of undetected defects is reduced. For this reason, generation | occurrence | production of at least one of a misinformation and an oversight can be suppressed, and the detection accuracy of a defect can be improved.

It is a perspective view which shows schematic structure of the inspection apparatus which concerns on this embodiment. It is a block diagram which shows schematic structure of the computer which concerns on this embodiment. It is a flowchart which shows the procedure of the inspection method which concerns on this embodiment. It is a flowchart which shows the procedure of a test inspection process. It is a flowchart which shows the procedure of a learning database construction process. It is a flowchart which shows the procedure of a machine learning process. It is a figure which shows a part of reference image. It is a figure which shows the state by which the defect candidate was displayed on the display part. It is a figure which shows the display part of the state in which normal detection, overdetection, and non-detection were selected. It is a flowchart which shows the procedure of an ambiguous inspection process. It is a figure which shows a pair of picked-up image from which the judgment result of whether it is a worker's defect differs. It is a figure which shows a mode that the several test target object is formed in one board | substrate. It is a figure which shows a mode that the same wiring pattern is formed in multiple numbers by one test target object.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant description is omitted. In addition, the dimensional ratios in the drawings are exaggerated for convenience of explanation, and may be different from the actual ratios.

  FIG. 1 is a perspective view showing a schematic configuration of an inspection apparatus 1 according to the present embodiment. FIG. 2 is a block diagram showing a schematic configuration of the computer 40 according to the present embodiment.

  The inspection apparatus 1 detects defects in the substrate W by performing image processing on a captured image of the substrate (corresponding to an inspection object) W. As shown in FIG. 1, the inspection apparatus 1 includes a placement unit 10, an illumination unit 20, an imaging unit 30, and a computer 40.

  The placement unit 10 places the substrate W thereon. The placement unit 10 is moved in the X direction by a drive stage (not shown) (see the arrow in FIG. 1). A known drive stage can be used. The substrate W is, for example, a printed board in which a copper foil wiring pattern is formed on a resin film. In the present embodiment, the substrate W is a single sheet type.

  The illumination unit 20 is provided above the placement unit 10 and irradiates the substrate W with light. Although it does not specifically limit as a light source, LED (Light Emitting Diode), a fluorescent lamp, an X ray, etc. can be used.

  The imaging unit 30 is provided above the placement unit 10 and images the substrate W. The imaging unit 30 preferably uses a line scan camera. By imaging with a line scan camera extended in the Y direction orthogonal to the X direction, a wide range of images can be acquired in a short time, and defect inspection can be performed quickly. The imaging unit 30 is moved in the Y direction by a driving unit (not shown) (see the arrow in FIG. 1).

  The computer 40 controls each part and performs various processes. As shown in FIG. 2, the computer 40 includes a CPU (Central Processing Unit) 41, a memory 42, a hard disk 43, a display unit 44, an input unit 45, and a communication unit 46.

  The CPU 41 executes control of each unit and various arithmetic processes according to programs stored in the memory 42 and the hard disk 43.

  The memory 42 includes a ROM (Read Only Memory) that stores various programs and various data, and a RAM (Random Access Memory) that temporarily stores programs and data as a work area.

  The hard disk 43 stores various programs including an operating system and various data.

  The display unit 44 is a liquid crystal display, for example, and displays various types of information to be described later.

  The input unit 45 includes a pointing device such as a mouse and a keyboard, and is used for performing various inputs.

  The communication unit 46 is an interface for communicating with other devices.

  The CPU 41 functions as a registration unit, an acquisition unit, a detection unit, a recognition unit, and an adjustment unit by executing various programs. Here, the registration unit registers the reference image of the substrate W and the inspection standard for detecting the defect. The acquisition unit acquires a captured image of the substrate W imaged by the imaging unit 30. The detection unit detects a defect on the substrate W by performing image processing for comparing the reference image and the captured image using the inspection standard registered by the registration unit. The recognition unit recognizes an overdetected defect that should not be detected as a defect from the defects detected by the detection unit, and recognizes an undetected defect that does not exist in the defect detected by the detection unit. . The adjustment unit adjusts the inspection standard so that the number of overdetected defects recognized by the recognition unit decreases and the number of undetected defects recognized by the recognition unit decreases. Further, the hard disk 43 stores the inspection standard adjusted by the adjustment unit as a storage unit.

  Hereinafter, an inspection method for detecting defects on the substrate W by the inspection apparatus 1 will be described with reference to FIGS.

  FIG. 3 is a flowchart showing the procedure of the inspection method by the inspection apparatus 1.

  In S01, the reference data is registered in the computer 40. Specifically, the operator inputs a reference image of the substrate W and an inspection standard for detecting defects into the computer 40, and the reference image and the inspection standard are registered in the computer 40.

  As a reference image of the substrate W, a design drawing image such as a CAD (Computer Aided Design) image or a captured image of the substrate W (golden sample) having no defect can be used. For example, it is registered as image data in RAW format. The As a reference image of the substrate W, for example, an image as shown in FIG. 7 can be acquired.

  Examples of the inspection standard for detecting a defect include a threshold value of luminance, an appearance shape, and an inspection standard. Here, the inspection standard is, for example, an allowable value of line width. In S01, the reference image is binarized using the threshold value of luminance registered as the inspection standard.

  In S02, the test board W for machine learning is inspected. Specifically, first, a test substrate W prepared in advance is imaged by the imaging unit 30, and the computer 40 acquires a captured image of the test substrate W. Then, the computer 40 performs image processing for comparing the captured image with the reference image using the inspection standard registered in S01, and detects a defect candidate on the test substrate W. A detailed description of the test inspection process S02 will be described later.

  In S03, a machine learning database is constructed. Specifically, the operator verifies the presence or absence of an overdetected defect and an undetected defect in the defect candidates detected in S02. Then, the verification result by the operator is stored in the computer 40. Detailed description of the learning database construction process S03 will be described later.

  In S04, it is determined whether or not a satisfactory level of database has been obtained. Specifically, the operator determines whether or not a sufficient amount of data has been accumulated for overdetected and undetected defects. If a satisfactory level of database is not obtained (S04: NO), the process returns to S02. Then, S02 to S04 are repeated for the newly prepared test substrate W until a satisfactory level database is obtained. On the other hand, if a satisfactory level of database is obtained (S04: YES), the process proceeds to S05.

  In S05, machine learning based on the database is performed. Specifically, the inspection standard is adjusted so that the computer 40 adjusts the threshold value of luminance and the inspection standard based on the database constructed in S03, so that the number of overdetected defects and the number of undetected defects are reduced. To optimize. A detailed description of the machine learning process S05 will be described later.

  In S06, a new test substrate W is inspected. Specifically, first, a newly prepared test substrate W is imaged by the imaging unit 30, and the computer 40 acquires a captured image of the new test substrate W. Then, the computer 40 performs image processing for comparing the captured image with the reference image using the inspection standard optimized in S05, and detects a defect candidate on the new test substrate W.

  In S07, it is determined whether or not satisfactory performance is obtained. Specifically, the operator verifies the presence or absence of an overdetected defect and an undetected defect for the defect candidates detected in S06, and determines whether or not a satisfactory inspection performance is obtained.

  If satisfactory performance is not obtained (S07: NO), the process returns to S02. Then, S02 to S07 are repeated for the newly prepared test substrate W until satisfactory performance is obtained. On the other hand, when satisfactory performance is obtained (S07: YES), the process proceeds to S08.

  In S08, the inspection apparatus 1 is applied to the inspection in the mass production process. Specifically, the inspection standard optimized in S05 is stored in the computer 40, and the inspection apparatus 1 performs the inspection in the mass production process of the substrate W using the optimized inspection standard. In the mass production process, the inspection described in S06 and S07 may be performed again on the first plurality of substrates W.

  As described above, in the inspection method according to the present embodiment, before the mass production process is started, first, a plurality of test substrates W are inspected, and the operator's verification results for the inspection results are stored as data. Then, machine learning based on the stored data is performed, and the inspection standard is optimized so that the number of overdetected and undetected defects is reduced. Therefore, according to the inspection method according to the present embodiment, it is possible to improve the defect detection accuracy by suppressing the occurrence of so-called false information and oversight.

  Hereinafter, the test inspection process S02, the learning database construction process S03, and the machine learning process S05 will be described in order with reference to FIGS.

  FIG. 4 is a flowchart showing the procedure of the test inspection process S02. In the test inspection process S02, the test board W for machine learning is inspected, and defect candidates on the test board W are detected. The test substrate W is a substrate whose position and type of defect have been grasped by an operator.

  In S021, the test substrate W is imaged by the imaging unit 30, and the computer 40 acquires a captured image of the test substrate W.

  In S022, the computer 40 binarizes the captured image acquired in S021, using the luminance threshold registered as the inspection standard in S01. As a result, the resin film portion on the surface of the test substrate W and the wiring pattern P portion of the copper foil are distinguished.

  In S023, the computer 40 performs image processing for comparing the captured image binarized in S022 with the reference image of the golden sample binarized in advance in S01, and detects a defect candidate. Specifically, the difference between the reference image and the captured image is detected in units of pixels. For example, when the size of the difference portion deviates from the inspection standard, the difference portion is detected as a defect candidate.

  For example, as an inspection standard, when an inspection standard is registered in which a defect is determined if there is an error of 30% or more of the wiring pattern P with respect to the reference image, the width of the wiring pattern P in the reference image is 10 μm. Those having a width of the wiring pattern P of less than 7 μm and those having the width of the wiring pattern P exceeding 13 μm are detected as defect candidates.

  In S024, the computer 40 stores the position and type of the defect candidate detected in S023. As the position of the defect, a coordinate value with the alignment mark on the substrate W as a reference position is stored.

  As described above, according to the test inspection process S02, first, a captured image of the test substrate W is acquired. Then, image processing for comparing the captured image with the reference image is performed, and a defect candidate on the test substrate W is detected.

  FIG. 5 is a flowchart showing the procedure of the learning database construction process S03. In the learning database construction process S03, the inspection result of the test substrate W is verified, and a machine learning database is constructed.

  In S031, the computer 40 displays the defect candidates detected in S02 on the display unit 44. FIG. 8 is a diagram illustrating a state in which defect candidates are displayed on the display unit 44. As shown in FIG. 8, an image having a predetermined size (for example, 100 pixels × 100 pixels or 200 pixels × 200 pixels) is displayed on the display unit 44, with the coordinates of the defect candidate being substantially at the center. In addition, a check box for selecting “normal detection”, “overdetection”, or “not detected” is provided below the defect candidate image.

  In S032, the computer 40 accepts registration of an undetected defect that does not exist in the defects detected in S02. Specifically, the operator refers to the defect candidate image displayed on the display unit 44 and confirms whether or not all known defects are displayed on the display unit 44. If all the known defects are not displayed on the display unit 44, it is determined that there is an undetected defect, and the coordinates and type of the undetected defect are registered in the computer 40. When the coordinates of the undetected defect are registered, the computer 40 additionally displays an image of a predetermined size centered on the registered coordinates on the display unit 44.

  In S033, the computer 40 accepts selection of “normal detection”, “overdetection”, or “not detected” by the operator. FIG. 9 is a diagram illustrating the display unit 44 in a state where “normal detection”, “overdetection”, or “not detected” is selected.

  If the operator determines that the defect is correctly detected in the defect candidate image displayed on the display unit 44, the worker selects “normal detection”. That is, when “normal detection” is selected, the defect candidates displayed on the display unit 44 are determined to be “defects”. If the worker determines that a part that is not originally a defect is detected in the defect candidate image displayed on the display unit 44, the worker selects “overdetection”. That is, if “overdetection” is selected, it is determined that the defect candidate displayed on the display unit 44 is “not a defect”. For the image added in S032, the operator selects “not detected”. That is, if “undetected” is selected, it is determined that the defect candidate added in S032 is “defect”.

  In FIG. 9, among the six defect candidate images displayed in S <b> 031, “normal detection” is selected for four images, and “overdetection” is selected for two images. . In addition, “not detected” is selected for the two images added in S032.

  In S034, the computer 40 stores the selection result received in S033. Specifically, the selection result of “normal detection” or “overdetection” is stored for the defect candidates detected by the image processing. For undetected defects, the selection result of “undetected” and the coordinates and types of undetected defects are stored.

  As described above, according to the learning database construction process S03, the verification result of the operator with respect to the inspection result of the test substrate W is stored, and a machine learning database is constructed.

  FIG. 6 is a flowchart showing the procedure of the machine learning process S05. In the machine learning process S05, an optimal inspection standard is calculated from the database constructed in S03.

  In S051, the computer 40 selects a luminance threshold value from inspection standards (luminance threshold value, inspection standard), for example, according to a predetermined priority order.

  In S052, the computer 40 searches for a threshold range in which the number of overdetected defects is reduced by image processing. Specifically, the computer 40 first reads the brightness threshold value registered in S01. Next, the computer 40 repeats the image processing by increasing / decreasing the luminance threshold by a predetermined amount from the initial value, thereby determining the relationship between the luminance threshold and the number of detected defect candidates for a predetermined type of defect. Ask. Then, the computer 40 calculates, for example, a range in which the number of defect candidates matches the number of “normally detected” defects stored in the database as a threshold range in which overdetected defects are reduced.

  For example, when the initial value of the luminance threshold is 120, 1000 defect candidates are detected by image processing, and as a result of verification by the operator, the number of overdetections among the 1000 defect candidates is 990 (that is, Assume that the number of “normally detected” defects is known to be 10). In this case, for example, the computer 40 repeatedly performs image processing while increasing or decreasing the luminance threshold value by 10 to calculate defect candidates. For example, if the number of defect candidates is 10 when the threshold is 70, the computer 40 calculates the range of the thresholds 70 to 120 as the threshold range in which overdetected defects are reduced.

  In S053, the computer 40 searches for a threshold range in which the number of undetected defects decreases by image processing. Specifically, for a predetermined type of defect, the computer 40 repeats image processing by increasing / decreasing the brightness threshold by a predetermined amount from the initial value for a captured image of a portion having an undetected defect, The relationship between the brightness threshold and the number of undetected defects is obtained. Then, for example, the computer 40 calculates a range in which the number of undetected defects is zero as a threshold range in which overdetected defects are reduced.

  For example, it is assumed that the initial value of the brightness threshold is 120 and there are 20 undetected defects. In this case, for example, the computer 40 repeatedly performs image processing on a captured image of a portion with an undetected defect while increasing or decreasing the luminance threshold value by 10 to check whether an undetected defect is detected. . For example, if the number of undetected defects is zero when the threshold value is 60, the computer 40 calculates the range of the threshold values 60 to 120 as the threshold value range in which the number of undetected defects decreases.

  In S054, the computer 40 determines an optimum threshold value. Specifically, the computer 40 optimizes the threshold for reducing both the overdetected defect and the undetected defect based on the threshold range calculated in S052 and the threshold range calculated in S053. Determine as value.

  As described above, according to the machine learning process S05, the inspection standard is adjusted based on the database of the worker verification results, and the inspection standard is optimized so as not to detect an overdetected defect. The inspection standard optimized in the machine learning process S05 is stored in the computer 40 in association with the inspection object (inspected product name, lot name, etc.), and when inspecting the same inspection object again, the computer 40 Stored, optimized inspection criteria can be used.

  As described above, the inspection method according to the present embodiment is an inspection method for detecting a defect by performing image processing on a captured image of the substrate W, and has an inspection standard for detecting a reference image of the substrate W and a defect. Step (a) for registering, step (b) for acquiring a captured image, and using the inspection standard registered in step (a), image processing for comparing the reference image and the captured image is performed, and the defect is detected. A step (c) for detecting, a step (d) for recognizing an overdetected defect that should not be detected as a defect from the defects detected in the step (c), and an error detected in the step (d). Adjusting the inspection criteria so that the number of detected defects is reduced (e) and storing the inspection criteria adjusted in step (e) (f). According to this inspection method, in step (e), the inspection standard is adjusted so that the number of overdetected defects is reduced. For this reason, generation | occurrence | production of an overdetection (false report) can be suppressed and the detection accuracy of a defect can be improved.

  Further, the method further includes a step (g) of recognizing an undetected defect that does not exist in the defects detected in the step (c) between the step (c) and the step (e). Then, the inspection standard is further adjusted so that the number of undetected defects recognized in step (g) is reduced. Therefore, the occurrence of undetected (missing) can be suppressed, and the defect detection accuracy can be improved.

  Step (b), step (c), step (d), and step (g) are repeatedly executed for a plurality of captured images, and step (e) is recognized for the plurality of captured images. It is performed for overdetected and undetected defects. For this reason, the amount of data in the database used in the machine learning process S05 increases. Therefore, the inspection standard calculated in the machine learning process S05 can be made more optimal. Therefore, the inspection accuracy is improved.

  The inspection standard is a luminance threshold value. In step (e), the threshold value registered in step (a) is changed by a predetermined amount and image processing is performed using the changed threshold value to detect a defect. The threshold is adjusted by repeating the process. According to this inspection method, the defect detection accuracy is reliably improved.

  Further, after step (f), the image obtained by imaging the substrate W in step (h) using the step (h) for imaging the substrate W by the imaging unit 30 and the inspection standard stored in step (f). (I) detecting an image by subjecting the captured image to image processing. For this reason, it is possible to suitably inspect defects for mass-produced products.

<Modification 1>
Hereinafter, Modification 1 of the above-described embodiment will be described. The inspection method according to the modified example 1 is different from the inspection method according to the above-described embodiment in that an ambiguity inspection process S09 is provided between the machine learning process S05 and the confirmation inspection process S06 (see FIG. 3). Hereinafter, the ambiguous inspection process S09 will be described in detail.

  FIG. 10 is a flowchart showing the procedure of the ambiguity check process S09. FIG. 11 is a diagram illustrating a pair of captured images P1 and P2 that are different in the determination result of whether or not the defect is an operator's defect in S033. In the ambiguous inspection process S09, when the determination result of whether or not the defect is the same as the defect of the operator is different from the defect of the same type and size formed in the pair of captured images P1 and P2, the determination result is correct. Let the operator determine whether or not.

  In S091, as shown in FIG. 11A, a pair of captured images P1 and P2 having different judgment results as to whether or not the defect is the operator's defect in S033 with respect to a defect candidate of the same type and size. Are displayed on the display unit 44. In the first modification, a case where the defect is a pinhole H will be described as an example. Note that the types of defects are not limited to pinholes, but include defects and shorts. As shown in FIG. 11A, a pinhole H of the same size is formed on the wiring pattern P in one captured image P1 and another captured image P2. Note that the same size includes not only completely the same size but also substantially the same size.

  In S <b> 033, the operator determines that the pinhole H formed in one captured image P <b> 1 is defective (“normal detection” or “not detected” is selected). On the other hand, it is determined that the pinhole H formed in the other captured image P2 is not a defect (“overdetection” is selected).

  Further, as a result of performing the image processing using the threshold value determined in S054, the computer 40 determines that the pinhole H formed in the pair of captured images P1 and P2 is “not defective”. In this case, as shown in FIG. 11A, the outer frames of the pair of captured images P1 and P2 are displayed by solid lines. On the other hand, when the computer 40 determines that the pinhole H formed in the pair of captured images P1 and P2 is “defective” as a result of performing image processing using the threshold value determined in S054, FIG. As shown in (B), the outer frame of the pair of captured images P1 and P2 can be displayed with dotted lines. Note that the determination result of the computer 40 may be identified by the difference in color.

  In S092, a change in the determination result of the computer 40 by image processing using the determination result of the worker in S033 or the threshold value determined in S054 is accepted. For example, in FIG. 11A, when the operator determines that the determination result of the operator “defective” is incorrect, the operator presses the command button of “change visual determination” at the lower left to determine the operator's determination. Change to be “not a defect”. In addition, when the operator determines that the determination result “not defective” of the computer 40 is incorrect, the command button of “CP determination change” on the upper left is pressed, and the determination of the computer 40 is “defective”. Change to In addition, as shown in the right diagram of FIG. 11A, when the judgment of the worker and the computer 40 is the same and the worker judges that there is no mistake in the judgment, the command buttons on the upper right and lower right are Do not push.

  In S093, when the change of the determination result of the computer 40 is accepted in S092, the feature amount of the peripheral portion of the defect is extracted, and the change of the determination result of the computer 40 in S092 is reflected in the inspection processes of S06 and S08. Thus, the change of the judgment result of the computer 40 is stored. In the first modification, for example, the line width of the wiring pattern P can be given as the feature amount of the peripheral portion of the defect. In one captured image P1, the line width W1 of the wiring pattern P is short, whereas in the other captured image P2, the line width W2 is relatively larger than the line width W1 of the one captured image P1. At this time, even if the size of the pinhole H is the same, the computer 40 learns that it is “defect” if the line width is relatively short, and “not defective” if the line width is relatively long. Remember.

  As described above, the inspection method according to the first modification includes a pair of imaging in which, after step (f), the determination result as to whether or not the defect is the operator's defect with respect to the defect of the same type and size. A step (l) for displaying an image, and a step (m) for accepting a change in the judgment result of the operator or the judgment result as to whether the computer 40 is defective or not. According to this inspection method, even if an operator selects in error in S033, a pair of captured images with different determination results as to whether or not the defect is the same as the defect with respect to the defect of the same type and size. Is displayed and a change in the operator's determination result is accepted, so that an operator's determination error can be suitably prevented.

  In addition, when a change in the determination result of the computer 40 is accepted in step (m), the feature amount of the peripheral portion of the defect is extracted, and the change in the determination result of the computer 40 in step (m) is reflected in the future inspection. (N) is further stored. Therefore, the inspection can be performed with higher accuracy.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims.

  In the above-described embodiment, in S05, the luminance threshold is adjusted to optimize the luminance threshold so that the number of overdetected defects and the number of undetected defects are reduced, thereby detecting the defect detection accuracy. Improved. However, the accuracy of defect detection may be improved based on the ratio of normally detected defects calculated based on one luminance threshold without adjusting the luminance threshold.

  Specifically, first, by using the brightness threshold registered as the inspection standard, the processes of S02 and S03 are performed, so that the selection results of “normal detection”, “overdetection”, and “not detected” Coordinates and defect types are stored. Then, these steps are repeated for a plurality of (for example, 1000) captured images, and a database in which feature amounts, selection results, coordinates, and types are associated and stored is constructed. Here, the feature amount is, for example, the depth of a groove cut into a V shape, and can be appropriately set according to the type of defect. Then, the above process is repeated for a plurality of captured images. Then, the ratio of overdetection is calculated from the selection result associated with the same coordinate in the database, and if it is equal to or greater than the predetermined ratio, it is not displayed as overdetection at that coordinate in the confirmation inspection process S06 or the mass production inspection process S08. Can be. In S033, when the computer 40 accepts the selection of “normal detection”, “overdetection”, or “not detected” by the operator, the above three selection results are stored in association with the position and type of the defect. Then, the ratio of “normal detection” among the accumulated selection results may be displayed on the display unit 44.

  In the above-described embodiment, in S023, the computer 40 performs image processing for comparing the captured image binarized in S022 with the reference image of the golden sample binarized in advance in S021, and the defect is detected. Candidates have been detected. However, when the reference image is a design drawing image such as a CAD image, a defect candidate may be detected by performing image processing for comparing the binarized captured image with the design drawing image in S022. . At this time, the design drawing image as a reference image is not binarized.

  In the above-described embodiment, the threshold is adjusted so that the number of overdetected defects and the number of undetected defects are reduced. However, the threshold may be adjusted so that the number of over-detected defects or the number of undetected defects is reduced.

  In the above-described embodiment, a line scan camera is used as the imaging unit 30. However, an area scan camera may be used as the imaging unit. Further, a 3D camera may be used as the imaging unit. In this way, by using a 3D camera as the imaging unit, it is possible to detect a defect in three dimensions.

  Further, the inspection apparatus 1 may further include a color camera separately from the imaging unit 30. The color camera selectively captures a micro area centered on the coordinates of a defect candidate detected by image processing, and acquires a color image of the micro area. The operator can accurately determine whether or not the defect candidate is an overdetected defect by referring to the color captured image displayed on the display unit. Acquiring a color captured image with a color camera in this way improves the accuracy in determining “normal detection” or “overdetection” for a defect candidate. Then, by using the color captured image in this way, the determination accuracy of “normal detection” and “overdetection” is improved, and therefore, a more suitable inspection standard is calculated in the machine learning process S05.

  In the above-described embodiment, the luminance threshold is selected as the inspection standard in S051. However, an inspection standard may be selected as the inspection standard in S051. At this time, the inspection standard is calculated so that the number of overdetected and undetected defects is reduced. In S051, a luminance threshold and an inspection standard may be selected as the inspection standard.

  In the above-described embodiment, the illumination unit 20 is provided above the substrate W, and the imaging unit 30 images reflected light from the illumination unit 20. However, the illumination unit may be provided below the substrate W, and the imaging unit may capture the transmitted light from the illumination unit.

  In the above-described embodiment, overdetection and non-detection are determined in S033. However, in S033, at least one of over-detection and non-detection may be determined.

  In the above-described embodiment, the position and type of an undetected defect are registered in S052, and the inspection standard is adjusted based on the registered position and type. However, in S052, only the position of an undetected defect may be registered. In this case, the computer 40 predicts the type of defect and inquires the operator of the predicted type of defect.

  In the above-described embodiment, the case where the printed circuit board is inspected is described as an example. However, the inspection apparatus 1 of the present invention is not limited to the inspection of a printed circuit board, but is applied to various inspections for comparing a captured image with a reference image.

  In the above-described embodiment, the substrate W has been described by taking a single sheet type as an example. However, a form in which the roll-shaped substrate W is inspected by the inspection apparatus 1 while being drawn out may be employed.

  In the above-described embodiment, the inspection standard is adjusted by the computer 40 using the captured image of the substrate W imaged by the imaging unit 30 connected to the computer 40. However, the inspection standard may be adjusted by the computer 40 using the captured image of the substrate W captured by the imaging unit of another inspection apparatus. At this time, the substrate W may be re-captured by the color camera, and the operator may check the color-captured image and select an overdetection defect. By using a color captured image in this way, the accuracy in determining “normal detection” or “overdetection” for a defect candidate is improved.

  In the above-described embodiment, the case where one inspection object is formed on one substrate W has been described. However, as shown in FIG. 12, a plurality of inspection objects T may be configured on one substrate W. At this time, it is preferable to construct a large number of learning databases for one part by performing coordinate transformation so that the same part has the same coordinates in different inspection objects T. According to this inspection method, more learning databases are constructed in the learning database construction process S03, so that a satisfactory level of database can be obtained in a shorter time. Note that the same parts of different inspection objects T may be associated with each other without performing coordinate conversion.

  Further, as shown in FIG. 13, when a plurality of the same wiring patterns P are formed on one inspection object T, the same part has the same coordinates in the same wiring pattern P existing at different coordinates. In this way, it is preferable to construct a large number of learning databases for one part by performing coordinate transformation. According to this inspection method, more learning databases are constructed in the learning database construction process S03, so that a satisfactory level of database can be obtained in a shorter time. Note that the same parts of the same wiring pattern P existing at different coordinates may be associated with each other without performing coordinate conversion.

  The means and method for performing various processes in the inspection apparatus 1 according to the above-described embodiment can be realized by either a dedicated hardware circuit or a programmed computer. The program may be provided by a computer-readable recording medium such as a CD-ROM (Compact Disc Read Only Memory), or may be provided online via a network such as the Internet. In this case, the program recorded on the computer-readable recording medium is usually transferred to and stored in a storage unit such as a hard disk. The program may be provided as a single application software, or may be incorporated in the software of the apparatus as one function of the inspection apparatus 1.

1 Inspection device,
10 Placement part,
20 Illumination part,
30 imaging unit,
40 computers,
41 CPU,
42 memory,
43 Hard disk,
44 display,
45 Input section,
46 Communication Department,
P1, P2 A pair of captured images.

Claims (10)

An inspection method for detecting defects by performing image processing on a captured image of an inspection object,
(A) registering a reference image of the inspection object and an inspection standard for detecting the defect;
Obtaining the captured image (b);
(C) detecting the defect by performing image processing for comparing the reference image and the captured image using the inspection standard registered in the step (a);
Recognizing an overdetected defect that should not be detected as a defect from the defects detected in step (c);
Adjusting the inspection criteria so that the number of overdetected defects recognized in step (d) is reduced (e);
Storing the inspection criteria adjusted in the step (e) (f);
Have
Between step (c) and step (e),
Recognizing undetected defects that are not present in the defects detected in step (c),
In the step (e), the inspection standard is further adjusted so that the number of the undetected defects recognized in the step (g) is reduced,
The step (b), the step (c), the step (d), and the step (g) are repeatedly performed on a plurality of captured images,
The step (e) is performed for the overdetected defect and the undetected defect recognized for the plurality of captured images ,
After step (f)
A step (l) of displaying a pair of the captured images having different determination results as to whether or not the defect is an operator defect with respect to the defect of the same type and size;
Inspecting how having a, and step (m) that receives a change of the operator's judgment results or computer defect in which whether the result of determination. The inspection standard is a luminance threshold,
In the step (e), the threshold value registered in the step (a) is changed by a predetermined amount, image processing is performed using the changed threshold value, and the defect detection process is repeated. The inspection method according to claim 1, wherein the threshold value is adjusted. Between step (c) and step (d),
Imaging the inspection object with a color camera (h);
A step (i) of displaying a color captured image obtained by imaging the inspection object with the color camera in the step (h) and receiving selection of the overdetected defect by an operator; Have
The inspection method according to claim 1, wherein in the step (d), the overdetected defect accepted in the step (i) is recognized. After step (f)
Step (j) of imaging the inspection object by an imaging unit;
A step of detecting the defect by performing the image processing on the captured image obtained by imaging the inspection object in the step (j) using the inspection standard stored in the step (f) (k) The inspection method according to any one of claims 1 to 3, further comprising: When a change in the determination result of the computer is accepted in the step (m), the feature amount of the peripheral portion of the defect is extracted, and the change in the determination result of the computer in the step (m) is reflected in the future inspection The inspection method according to any one of claims 1 to 4, further comprising a step (n) of storing as described. An inspection apparatus that detects a defect by performing image processing on a captured image of an inspection object,
A registration unit for registering a reference image of the inspection object and an inspection standard for detecting the defect;
An acquisition unit for acquiring the captured image;
Using the inspection standard registered by the registration unit, a detection unit that detects the defect by performing image processing for comparing the reference image and the captured image;
A first recognition unit for recognizing an overdetection defect that should not be detected as a defect from the defects detected by the detection unit;
A second recognition unit for recognizing an undetected defect that does not exist in the defect detected by the detection unit;
A first adjustment unit that adjusts the inspection standard such that the number of overdetected defects recognized by the first recognition unit decreases;
A second adjustment unit that adjusts the inspection standard so that the number of undetected defects recognized by the second recognition unit decreases;
A storage unit for storing the inspection standard adjusted by the first adjustment unit and the second adjustment unit;
A display unit that displays a pair of the captured images with different determination results as to whether or not the defect is the same as the defect of the same type and size;
A reception unit that receives a change in the determination result of the worker or the determination result of whether or not the defect is a computer;
Have
The process of acquiring the captured image by the acquisition unit, the process of detecting the defect by the detection unit, the process of recognizing the overdetected defect by the first recognition unit, and the undetected by the second recognition unit The process of recognizing a defect is repeatedly performed on a plurality of captured images,
The first adjustment unit and the second adjustment unit adjust the inspection standard using the overdetected defect and the undetected defect recognized for the plurality of captured images. A color camera for imaging the inspection object;
A display unit for displaying a color-captured image obtained by imaging the inspection object by the color camera;
An input unit that receives an input of selection of the overdetected defect by an operator based on the color captured image displayed on the display unit;
The inspection apparatus according to claim 6 , wherein the first recognition unit recognizes the overdetection defect whose input is received by the input unit. An inspection program for detecting defects by performing image processing on a captured image of an inspection object,
A procedure (a) for registering a reference image of the inspection object and an inspection standard for detecting the defect;
A procedure (b) of acquiring the captured image;
Using the inspection standard registered in the step (a), performing image processing for comparing the reference image and the captured image, and detecting the defect;
A step (d) of recognizing an overdetected defect that should not be detected as a defect from the defects detected in the step (c);
Adjusting the inspection criteria so that the number of over-detected defects recognized in step (d) is reduced;
A procedure (f) for storing the inspection standard adjusted in the procedure (e);
To the computer,
Between the procedure (c) and the procedure (e),
Causing the computer to further execute a step (g) of recognizing an undetected defect that does not exist in the defect detected in the step (c);
In the step (e), the inspection standard is further adjusted so that the number of the undetected defects recognized in the step (g) is reduced.
The procedure (b), the procedure (c), the procedure (d), and the procedure (g) are repeatedly performed on a plurality of captured images,
Step (e) is performed for the overdetected defect and the undetected defect recognized for the plurality of captured images ,
After step (f)
A procedure (l) for displaying a pair of the captured images with different determination results as to whether or not the defect is the same as the defect of the same type and size;
The operator of the determination result or the computer defect in a determination of whether the result of the procedure for accepting a change and (m), a further test program Ru is executed on the computer. Between the procedure (c) and the procedure (d),
A procedure (h) of imaging the inspection object with a color camera;
A step (i) of displaying a color captured image obtained by imaging the inspection object with the color camera in the step (h) and receiving selection of the overdetected defect by an operator; Have
The inspection program according to claim 8 , wherein the procedure (d) recognizes the overdetection defect received in the procedure (i). A computer-readable recording medium on which the inspection program according to claim 8 is recorded.