Inspection device, inspection method, and program

Abstract

An inspection device (500) comprises: a defect candidate extraction unit (563a) that extracts a defect candidate from an image obtained by imaging a lens that is a transparent body; a classification unit (563b) that classifies the defect candidate into a type corresponding to the shape by using an image analysis method using artificial intelligence; and a defect determination unit (563c) that determines, on the basis of a determination criterion corresponding to the type classified by the classification unit (563b), whether the defect candidate is a lens defect.

Description

Inspection device, inspection method, and program

[0001] The present invention relates to an inspection device, an inspection method, and a program.

[0002] 2. Description of the Related Art Inspection devices are known that inspect a transparent body such as a lens for defects such as scratches and stains by analyzing the image obtained by capturing the image of the transparent body with a camera.

[0003] The inspection device described in Patent Document 1 captures an image of interference fringes that appear at a predetermined position in the thickness direction of a transparent body to obtain hologram data, and then creates reconstructed image data of the transparent body at multiple different reconstruction positions from the obtained hologram data.The hologram data and reconstructed image data are then analyzed to determine whether or not there is a defect (a singular point in Patent Document 1) in the transparent body.

[0004] Japanese Patent Application Laid-Open No. 2018-54529

[0005] In an inspection device such as that described in Patent Document 1, an inspector or the inspection device determines the presence or absence of defects from an image based on predetermined rules (criteria), but it is difficult to accurately identify defects because the length, size, color, etc. of interference fringes that appear around defects in an image are not constant. For example, in the case of inspecting eyeglass lenses, it is difficult to prevent overdetection, in which minute scratches or stains that would not be detected in a final inspection before shipment are detected as defects because the lenses are later hard coated.

[0006] The present invention has been made in consideration of the above-mentioned situation, and aims to provide an inspection device, an inspection method, and a program that can improve inspection accuracy compared to conventional methods when inspecting defects in transparent bodies using images.

[0007] In order to achieve the above object, an inspection device according to a first aspect of the present invention comprises: a defect candidate extraction unit that extracts defect candidates from an image obtained by capturing an image of a transparent body; a classification unit that classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect determination unit that determines whether the defect candidates are defects in the transparent body using determination criteria according to the types classified by the classification unit.

[0008] In order to achieve the above object, an inspection method according to a second aspect of the present invention includes a defect candidate extraction step in which a defect candidate extraction unit extracts defect candidates from an image obtained by capturing an image of a transparent body; a classification step in which a classification unit classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect determination step in which a defect determination unit determines whether the defect candidate is a defect in the transparent body using determination criteria according to the type classified by the classification unit.

[0009] In order to achieve the above object, the program according to the third aspect of the present invention causes a computer to function as: a defect candidate extraction unit that extracts defect candidates from an image obtained by capturing an image of a transparent body; a classification unit that classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect judgment unit that judges whether the defect candidates are defects in the transparent body using judgment criteria according to the types classified by the classification unit.

[0010] According to the present invention, in the inspection of defects in a transparent body using an image, the inspection accuracy can be improved compared to the conventional method.

[0011] FIG. 1 is a diagram showing a general manufacturing process for eyeglass lenses. FIG. 2 is a diagram schematically showing the overall configuration of an inspection system according to an embodiment of the present invention. FIG. 3 is a block diagram showing an example of the configuration of an inspection device. FIG. 4 is a diagram showing an example of a defect candidate. FIG. 5 is a diagram showing an example of a defect candidate (part 1). FIG. 6 is a diagram showing an example of a defect candidate (part 2). FIG. 7 is a diagram showing an example of a defect candidate (part 3). FIG. 8 is a diagram showing an example of a defect candidate (part 4). FIG. 9 is a cross-sectional view of a lens. FIG. 10 is a diagram showing an example of creating an inspection image from a reproduced image. FIG. 11 is a flowchart showing an example of a defect determination process. FIG. 12 is a flowchart showing an example of an image inspection process.

[0012] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings, in which the same or corresponding parts are designated by the same reference numerals.

[0013] An inspection system 1 according to an embodiment of the present invention will be described. The inspection system 1 is a system used in the appearance inspection process S4 in the general manufacturing process of eyeglass lenses 10 shown in FIG. 1 . In the appearance inspection process S4, the presence or absence of defects in the lens 10 after the lens formation process S1, polishing process S2, and cleaning process S3 is inspected. Note that defects here refer to defects such as scratches, dirt, and air bubbles in the lens 10 that do not meet the standards at the time of shipping. Lenses 10 determined to be free of defects in the appearance inspection process S4 proceed to the next hard coating process (HC process) S5. Lenses 10 determined to be defective in the appearance inspection process S4 are discarded and excluded from subsequent processes. The HC process S5 is a process in which a hard coating is applied to the lens 10 to prevent scratches. Note that the manufacturing process of eyeglass lenses 10 shown in FIG. 1 is merely a general example, and the details of the manufacturing process will naturally vary depending on the manufacturer, etc.

[0014] 2 is a schematic diagram showing the overall configuration of an inspection system 1 according to an embodiment of the present invention. The inspection system 1 is a system for inspecting a transparent object, that is, an eyeglass lens 10, for defects. The inspection system 1 includes a conveying unit 100, an irradiation unit 200, a PLC (Programmable Logic Controller) 300, an imaging unit 400, and an inspection device 500.

[0015] The transport unit 100 is a mechanism for transporting the lens 10 to an appropriate position where it can be inspected. The transport unit 100 includes a transport shaft 101, a stage 102 slidably mounted on the transport shaft 101, and an actuator 103 which is a power source for sliding the stage 102. The lens 10 to be inspected is placed on the stage 102, and the stage 102 slides along the transport shaft 101, thereby transporting the transparent object in the left-right direction in FIG. 1. Note that the configuration of the transport unit 100 is not limited to this, and the transport unit 100 may be configured using other transport mechanisms such as a belt conveyor, a ball screw, etc.

[0016] The irradiation unit 200 is disposed below the lens 10 and irradiates the lens 10 with light having high coherence (for example, laser light) in accordance with commands from the inspection device 500 .

[0017] The PLC 300 is connected to the inspection device 500 via a LAN (Local Area Network) 600. Based on commands from the inspection device 500, the PLC 300 controls the driving of the actuator 103 of the transport unit 100, thereby moving the stage 102, which is slidably fixed on the transport shaft 101, to an appropriate position. In addition, based on commands from the inspection device 500, the PLC 300 controls the on / off of irradiation by the irradiation unit 200.

[0018] The imaging unit 400 is installed in a position where it can image the surface (top surface) of the lens 10 opposite to the surface irradiated with light from the irradiation unit 200. The imaging unit 400 includes a CCD image sensor or the like, and, in accordance with commands from the inspection device 500, images interference fringes that are generated by interference between the light (reference light) irradiated onto the lens 10 from the irradiation unit 200 and the object light that is the light from the lens 10 irradiated with the reference light, and acquires hologram data that is the image data thereof.

[0019] The inspection device 500 is a computer such as a PC (Personal Computer) or a server. The inspection device 500 analyzes the hologram data acquired by the imaging unit 400 and inspects the lens 10 for defects. As shown in FIG. 3 , the inspection device 500 includes an IF (interface) unit 510, a communication unit 520, an input unit 530, an output unit 540, a storage unit 550, a control unit 560, and a system bus 570 that interconnects these units.

[0020] The IF unit 510 includes, for example, a capture board, and is an interface for connecting the inspection device 500 to the imaging unit 400. The communication unit 520 is a communication device that complies with the standards of the LAN 600, and communicates with the PLC 300 via the LAN 600 under the control of the control unit 560.

[0021] The input unit 530 is an input device such as a keyboard or a mouse, and outputs a signal corresponding to an input operation received from a user to the control unit 560. For example, the input unit 530 receives an input operation from the user instructing the start of a defect determination process, which will be described later.

[0022] The output unit 540 is an output device such as a liquid crystal display or a speaker, and outputs various information based on instructions from the control unit 560. For example, when a defect is detected in the lens 10 in the defect determination process described below, the output unit 540 may notify the fact by displaying a picture or by sound.

[0023] The storage unit 550 includes a nonvolatile semiconductor memory such as a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (Electrically Erasable Programmable ROM), and serves as a so-called auxiliary storage device. The storage unit 550 stores a program 551 for the control unit 560 to execute a defect determination process (described later), a defect classification model 552, a determination criteria DB 553, and a defect record DB 554. In addition to the above, the storage unit 550 also stores various data necessary in advance for executing the program 551.

[0024] The defect classification model 552 is an artificial intelligence (AI) model for classifying defect candidates detected from an inspection image of the lens 10 (described later) into types according to their shapes. Specifically, the defect classification model 552 is an AI model that outputs the type of defect candidate when an image of the defect candidate and its surrounding interference fringes is input. The types of defect candidates that the defect classification model 552 can output in this embodiment are shown in FIG. 4 . The types shown in this table, "dot_deep," "dot_thin," "line_deep," "line_thin," "line_dot_deep," "line_dot_thin," "dot_deep_W," "line_deep_W," and "line_dot_deep_W," are classified according to the shape of the defect portion detected as a defect candidate from the inspection image and the intensity and color system of the interference fringes occurring around the defect portion. 4 is an example of classification, and defect candidates may be classified by other characteristics. Also, the "rough" category in this table indicates a false defect that was detected as a defect candidate in the inspection image but is not actually a defect. For example, a defect candidate that is hard coated in the next HC process S5 and therefore is not recognized as a defect in the final appearance inspection process S8 is determined to be "rough."

[0025] 5A, 5B, 5C, and 5D show examples of defect candidates that appear in inspection images. The defect candidate shown in FIG. 5A is classified as "dot_deep" because the defect is dot-like and has strong black interference fringes around it. The defect candidate shown in FIG. 5B is classified as "dot_thin" because the defect is dot-like and has weak black interference fringes around it. The defect candidate shown in FIG. 5C is classified as "line_dot_deep" because the defect is linear and has strong black interference fringes around it. The defect candidate shown in FIG. 5D is classified as "line_dot_thin" because the defect is linear and has weak black interference fringes around it. Although omitted in FIGS. 5A, 5B, 5C, and 5D, actual interference fringes are formed in stages across multiple layers centered on the defect.

[0026] 3 , the defect classification model 552 is generated by learning a large amount of training data using a machine learning technique such as deep learning or a neural network, and is stored in advance in the storage unit 550. The training data here is, for example, data indicating the results of an actual defect inspection of the lens 10 using images of captured interference fringes. Note that instead of data from an actual inspection, data artificially created by simulation or the like may also be used as training data for learning.

[0027] The criterion DB 553 stores criterion data for determining whether a defect candidate is a true defect for each of the above-described types of defect candidate. For example, in the case of the criterion data for the defect candidate type "dot_deep," the criterion data includes a threshold value for the size of the dot-like portion detected as the defect candidate and a threshold value for the size of the interference fringes around the defect candidate. If the size of the defect candidate determined to be "dot_deep" and the interference fringes around it are equal to or smaller than this threshold value, the defect candidate is determined not to be a defect. Here, it is desirable to set the criterion data stored in the criterion DB 553 taking into consideration the influence of the subsequent HC process S5. For example, let the thickness of the film coated on the lens 10 in the HC process S5 be X. In this case, since defects corresponding to scratches or stains with a thickness of X or less are expected to disappear in the HC process S5, criterion data may be set so that the defect is not determined to be a defect.

[0028] The defect record DB 554 is a database that records information about defects detected in the lens 10. Specifically, the defect record DB 554 records the type of the detected defect, the position and size of the defect within the inspection image, and so on.

[0029] The control unit 560 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The control unit 560 operates by the CPU reading out a program 551 stored in the storage unit 550 and executing the program using the RAM as a work area. The control unit 560 includes a reproduction image creation unit 561, an inspection image creation unit 562, and an image inspection unit 563 as main functional components according to the present invention.

[0030] Based on the hologram data captured by the imaging unit 400, the reconstructed image creation unit 561 generates reconstructed images of the lens 10 for each reconstruction position in the thickness direction of the lens 10 using known arithmetic processing. FIG. 6 shows a cross-sectional view of the lens 10 to be inspected. In this figure, the top-to-bottom direction represents the imaging direction of the imaging unit 400, and the bottom-to-top direction represents the irradiation direction of light from the irradiation unit 200. As shown in this figure, the lens 10 has a curvature, and the reconstructed image creation unit 561 generates reconstructed images D1 to D7 at different reconstruction positions X1 to X7 in the thickness direction of the lens 10. Note that the positions X1 to X7 are arbitrary, and the reconstruction positions may be set appropriately depending on the thickness of the lens 10, etc. The reconstructed image creation unit 561 may also generate reconstructed images at more reconstruction positions.

[0031] Test image creation unit 562 extracts, from among the reproduced images D1 to D7 created by reproduced image creation unit 561 for each reproduction position, the image with the highest degree of focus, which indicates the degree to which the defective portion is in focus, on a pixel-by-pixel basis. Test image creation unit 562 then synthesizes the extracted pixels to create a single test image. As a result, the test image becomes an image in which the defective portion is in focus over the entire range (all pixels) of the image.

[0032] Assume that there is a defect P1 in the center of the lens 10 and a defect P2 at its edge, as shown in FIG. 6 . The lens 10 has a curvature, and the defects P1 and P2 occur at different positions in the thickness direction of the lens 10. Therefore, as shown in FIG. 7 , in the reconstructed image D1 generated at the reconstruction position X1 (see FIG. 6 ), the defect P1 is in focus and appears clearly, but the defect P2 is out of focus and appears unclear. On the other hand, in the reconstructed image D6 generated at the reconstruction position X6, the defect P2 is in focus and appears clearly, but the defect P1 is out of focus and appears unclear. In contrast, in the inspection image generated by combining the pixels with the highest degree of focus (i.e., the pixels most in focus on the defect portion) from the reconstructed images D1 to D7, both defects P1 and P2 appear clearly.

[0033] 3 , the image inspection unit 563 analyzes the inspection image created by the inspection image creation unit 562 to inspect the lens 10 for defects. As described above, the inspection image clearly shows both defects P1 and P2, which occur at different positions in the thickness direction of the lens 10, so the image inspection unit 563 can accurately inspect the defects from the inspection image alone. As a more detailed functional configuration, the image inspection unit 563 includes a defect candidate extraction unit 563a, a classification unit 563b, and a defect determination unit 563c.

[0034] The defect candidate extraction unit 563a extracts, from the inspection image, as defect candidates, locations that may be defective in the lens 10. As described above, interference fringes are formed around the defect candidates.

[0035] The classification unit 563b classifies the defect candidates extracted by the defect candidate extraction unit 563a into types according to their shapes using an image analysis method using artificial intelligence. Specifically, the defect type discrimination unit inputs images of the defect candidates and their surrounding interference fringes into the defect classification model 552 to classify the defect candidates. As a result, the defect candidates are classified into one of the types defined in the table shown in FIG. 4.

[0036] The defect determination unit 563c determines whether or not a defect candidate is a real defect based on determination criteria set according to the type of defect candidate classified by the classification unit 563b. Specifically, the defect determination unit 563c reads determination criteria data according to the type of defect candidate classified by the classification unit 563b from the determination criteria DB 553, and determines whether or not the defect candidate is a defect by comparing a determination threshold indicated by the determination criteria data with feature quantities (such as size) of the defect candidate and the interference fringes around it.

[0037] Next, the operation of the defect determination process executed by the inspection device 500 of the inspection system 1 will be described with reference to the flowchart of Fig. 8. For example, the defect determination process is started when the user inputs an instruction to start the process from the input unit 530.

[0038] First, the control unit 560 of the inspection device 500 moves the lens 10 to be inspected into the imaging range of the imaging unit 400 (step S101). Specifically, the control unit 560 instructs the PLC 300 via the LAN 600 to drive the actuator 103. Upon receiving this instruction, the PLC 300 drives the actuator 103. As a result, the stage 102 slides along the transport axis 101, and the lens 10 on the stage 102 is positioned within the imaging range of the imaging unit 400.

[0039] Next, the control unit 560 instructs the irradiation unit 200 to irradiate light via the PLC 300 (step S102). As a result, the irradiation unit 200 irradiates the lens 10 with light.

[0040] Next, the control unit 560 causes the imaging unit 400 to perform an imaging operation, thereby acquiring hologram data of the lens 10 (step S103).

[0041] Next, the control unit 560 creates reconstructed images D1 to D7 from the acquired hologram data at each of a plurality of different reconstruction positions X1 to X7 in the thickness direction (Z direction) of the lens 10 that have been preset (see Figure 6) (step S104).

[0042] Next, the control unit 560 selects one pixel from the hologram image represented by the hologram data (step S105). In the following description, the pixel selected in step S105 will also be referred to as a selected pixel.

[0043] Next, the control unit 560 calculates the degree of focus (focus level) of the defective portion of the pixel at the same position as the selected pixel in each of the reproduced images D1 to D7 created in step S104 (step S106). It is generally known that a pixel region where the defective portion is in focus has a high contrast (brightness gradient) of brightness changes within the region. Therefore, for example, the control unit 560 may add up the absolute values ​​of the differences in brightness values ​​between adjacent pixels in a neighborhood of the target pixel (including the target pixel) and calculate this sum as the focus level of the target pixel. Note that the method for calculating the focus level is not limited to the above, and other methods may also be used to calculate the focus level.

[0044] Next, the control unit 560 extracts the pixel with the highest calculated focus degree from each of the reproduced images D1 to D7 (step S107). The pixel data of the extracted pixel is temporarily stored in the storage unit 550.

[0045] Next, the control unit 560 determines whether all pixels of the hologram image have been selected in step S105 (step S108). If all pixels have not been selected (step S108; No), the process returns to step S105.

[0046] On the other hand, if all pixels of the hologram image have been selected (step S108; Yes), the control unit 560 combines the pixels with the highest focus levels extracted in step S109 and generates a single test image that has been subjected to black and white binarization processing to make it easier to identify defects (step S109). Note that the test image may be generated by performing other image processing, such as color subtraction processing, instead of binarization processing.

[0047] Next, the control unit 560 analyzes the inspection image and executes an image inspection process to inspect the lens 10 for defects (step S110). If a defect is detected in the image inspection process, information about the defect (such as the type and location of the defect) is recorded in the defect record DB 554, and the control unit 560 performs a process to discard the lens 10. For example, the control unit 560 performs a process to remove defective products by switching and controlling the conveying unit 100. If no defect is detected in the image inspection process, the lens 10 is sent to the next HC process S5. This completes the defect determination process.

[0048] Next, the image inspection process of the defect determination process (step S110 in FIG. 8) will be described in detail with reference to the flowchart in FIG.

[0049] When the image inspection process starts, the control unit 560 first extracts possible defect locations in the lens 10 as defect candidates from the inspection image based on predetermined criteria (step S201). Specifically, if the inspection image contains a portion of a predetermined color (black or white) that has a length or area equal to or greater than a predetermined threshold, the control unit 560 extracts that portion as a defect candidate. In step S201, multiple defect candidates may be extracted. Note that the purpose of the process in step S201 is to widely extract possible defect locations as defect candidates, regardless of whether they actually are defects, so it is desirable to set the extraction threshold to a relatively small value.

[0050] If no defect candidate can be extracted in step S201 (step S202; No), it is determined that there is no defect in the lens 10, and the image inspection process ends.

[0051] On the other hand, if a defect candidate can be extracted (step S202; Yes), the control unit 560 selects one defect candidate (step S203). Then, the control unit 560 inputs an image of the selected defect candidate and the interference fringes formed around it into the defect classification model 552, and classifies the defect candidate into a type according to its shape (step S204). As a result, the defect candidate is classified into one of the types defined in the table shown in FIG. 4.

[0052] If the type of the defect candidate classified in step S204 is a false defect (rough) (step S205; Yes), the defect candidate does not correspond to a defect, and the process proceeds to step S209. On the other hand, if the type of the defect candidate is other than a false defect (rough) (step S205; No), the control unit 560 determines whether the defect candidate is a defect using a determination criterion corresponding to the classified type (step S206). Specifically, the control unit 560 acquires determination criterion data for the classified type from the determination criterion DB 553, and compares the determination threshold indicated by the determination criterion data with the feature amount of the defect candidate and the surrounding interference fringes, thereby determining whether the defect candidate is a defect.

[0053] If it is not determined to be a defect in step S206 (step S207; No), the process proceeds to step S209. On the other hand, if it is determined to be a defect (step S207; Yes), the control unit 560 records information about the defect, such as its type, its position in the inspection image, its size, etc., in association with the identification information of the lens 10 in the defect record DB 554 (step S208). Then, the process proceeds to step S209.

[0054] In step S209, the control unit 560 determines whether or not all of the extracted defect candidates have been selected in step S203.

[0055] If all defect candidates have not been selected (step S209; No), the process returns to step S203, and the series of processes (steps S203 to S208) for classifying the unselected defect candidates into types according to their shapes and determining whether or not they correspond to defects based on the determination criteria according to the types are repeated. On the other hand, if all defect candidates have been selected (step S209; Yes), the image discrimination process ends.

[0056] As described above, according to this embodiment, from among a plurality of reconstructed images generated based on hologram data capturing an image of the lens 10, each of which has a different reconstruction position, the image with the highest focus level, which indicates the degree to which the defective portion is in focus, is extracted on a pixel-by-pixel basis, and an inspection image is created by combining these images. In this inspection image, all defects are in focus and clearly displayed without being affected by the defect location within the lens 10. Therefore, defects can be inspected accurately simply by analyzing this inspection image. In other words, according to this embodiment, when inspecting defects in transparent bodies using hologram data, inspection speed can be improved while maintaining the same inspection accuracy as conventional methods.

[0057] Furthermore, according to this embodiment, whether a defect candidate is a defect is determined based on a determination criterion corresponding to the type classified by an image analysis method using artificial intelligence. Therefore, compared to conventional defect inspections that determine the presence or absence of defects without such classification, defect determination can be performed based on more detailed determination criterion corresponding to the type, thereby improving inspection accuracy. For example, in conventional inspections, overdetection occurs, whereby minute scratches or stains that would not be detected in a final inspection before shipment due to subsequent coating are detected as defects. However, by appropriately setting determination criterion corresponding to the type of defect candidate, this can be prevented.

[0058] (Modifications) The present invention is not limited to the above-described embodiment, and various modifications and applications are possible. For example, the inspection system 1 according to the above-described embodiment does not need to have all of the technical features described above, and may have some of the configurations described in the above-described embodiment so as to solve at least one problem in the prior art. Furthermore, at least a portion of each of the following modifications may be combined.

[0059] For example, in the above embodiment, the lens 10 for glasses is the object to be inspected, but the present invention is also applicable to lenses 10 other than those for glasses, and transparent bodies other than the lens 10, such as transparent plates.

[0060] In the above embodiment, the defect determination process involves extracting the image with the highest degree of focus from among multiple reconstructed images in pixel units and combining them to create an inspection image. However, the number of pixels may not be limited to one, and any number (first number) of pixels may be extracted from among multiple reconstructed images in pixel units and combined to create an inspection image. Increasing the first number increases the size of the composite unit, which may reduce the accuracy of defect determination, but it can shorten the processing time required to create an inspection image. Alternatively, the control unit 560 may accept a designation of the first number via the input unit 530 or the communication unit 520, and set the first number in accordance with the designation. In this case, the control unit 560 corresponds to the composite unit setting unit of the present invention.

[0061] Furthermore, the user may be allowed to freely set the criterion data stored for each type of defect candidate in the criterion DB 553. For example, the user may specify criterion data via the input unit 530 or the communication unit 520, and the control unit 560 may update the criterion DB 553 in accordance with the specification. In this case, the control unit 560 corresponds to the criterion setting unit of the present invention.

[0062] In the above embodiment, the image inspection process (FIG. 9) is performed on the test image, but the image that is the target of the image inspection process is not limited to the test image. For example, the image inspection process of this embodiment may also be used to inspect defects in conventional hologram images or reproduced images.

[0063] In the above embodiment, the program 551 executed by the control unit 560 of the inspection device 500 has been described as being stored in advance in the storage unit 550. However, the present invention is not limited to this, and the above program 551 may be implemented in an existing general-purpose computer or the like, so that the computer functions as a device equivalent to the inspection device 500 according to the above embodiment.

[0064] Such program 551 may be provided in any manner. For example, the program may be stored in a computer-readable recording medium (a flexible disk, a CD (Compact Disc)-ROM, a DVD (Digital Versatile Disc)-ROM, an MO (Magneto-Optical Disc), a memory card, a USB memory, or the like) and distributed, or the program 551 may be stored in storage on a network such as the Internet and provided by being downloaded.

[0065] Furthermore, when the above-mentioned processing is performed by sharing the work between an OS (Operating System) and an application program, or by cooperation between the OS and the application program, only the application program may be stored in a recording medium or storage. Furthermore, the program 551 may be superimposed on a carrier wave and distributed via a network. For example, the program 551 may be posted on a bulletin board system (BBS) on the network and distributed via the network. The above-mentioned processing may be performed by starting up the program 551 and executing it under the control of the OS in the same way as other application programs.

[0066] The present invention allows various embodiments and modifications without departing from the broad spirit and scope of the present invention. Furthermore, the above-described embodiments are intended to illustrate the present invention and do not limit the scope of the present invention. That is, the scope of the present invention is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of the invention equivalent thereto are considered to be within the scope of the present invention.

[0067] This application is based on Japanese Patent Application No. 2023-212740 filed on December 18, 2023. The entire specification, claims, and drawings of Japanese Patent Application No. 2023-212740 are incorporated herein by reference.

[0068] 1 Inspection system, 10 Lens, 100 Conveying unit, 101 Conveying axis, 102 Stage, 103 Actuator, 200 Irradiation unit, 300 PLC, 400 Imaging unit, 500 Inspection device, 510 IF unit, 520 Communication unit, 530 Input unit, 540 Output unit, 550 Memory unit, 551 Program, 552 Defect classification model, 553 Judgment criteria DB, 554 Defect recording DB, 560 Control unit, 561 Reproduced image creation unit, 562 Inspection image creation unit, 563 Image inspection unit, 563a Defect candidate extraction unit, 563b Classification unit, 563c Defect judgment unit, 570 System bus, 600 LAN

Claims

1. An inspection device comprising: a defect candidate extraction unit that extracts defect candidates from an image obtained by capturing an image of a transparent body; a classification unit that classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect judgment unit that judges whether the defect candidate is a defect in the transparent body using judgment criteria according to the type classified by the classification unit.

2. The inspection device according to claim 1, wherein the transparent body is a lens, and the inspection device is used in a visual inspection process in a manufacturing process of the lens.

3. The inspection device according to claim 2, wherein the defect determination unit does not determine that a defect is to be eliminated by processing in a process subsequent to the appearance inspection process.

4. The inspection device according to claim 3, wherein the subsequent step is a step of applying a coating to the lens.

5. The inspection device according to any one of claims 1 to 4, wherein the classification unit classifies the defect candidates by inputting images of the defect candidates and interference fringes occurring around them into an artificial intelligence model.

6. The inspection device according to any one of claims 1 to 4, wherein the classification unit classifies the defect candidates into types according to the shape of the defect candidates and characteristics of interference fringes occurring around the defect candidates.

7. The inspection device according to claim 1, wherein the defect determination unit records the type and position within the image of the defect candidate determined to be the defect.

8. The inspection device according to claim 1, further comprising a judgment criterion setting section that sets the judgment criterion used by the defect judgment section for judgment.

9. An inspection method comprising: a defect candidate extraction step in which a defect candidate extraction unit extracts defect candidates from an image obtained by imaging a transparent body; a classification step in which a classification unit classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect judgment step in which a defect judgment unit judges whether or not the defect candidate is a defect in the transparent body using judgment criteria according to the type classified by the classification unit.

10. A program that causes a computer to function as: a defect candidate extraction unit that extracts defect candidates from an image obtained by capturing an image of a transparent body; a classification unit that classifies the defect candidates into types according to their shapes using an image analysis method using artificial intelligence; and a defect judgment unit that judges whether the defect candidate is a defect in the transparent body using judgment criteria according to the type classified by the classification unit.

Images