Inspection program and inspection device

The inspection system addresses the limitation of existing AI-based abnormality detection by determining abnormality levels and sorting workpieces based on severity, enabling flexible and precise categorization and discharge of products through a machine learning model and conveyance system.

WO2024241551A1PCT designated stage expired Publication Date: 2024-11-28ROBIT INC
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2023/019408
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-24
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Existing inspection systems using AI for abnormality detection in workpieces only consider the presence or type of abnormalities, failing to account for the severity level, which can vary based on sorting purposes, requiring flexible sorting capabilities.

Method used

An inspection program and device that utilize a machine learning model to determine the abnormality level of workpieces by setting threshold values, allowing for flexible sorting regardless of the purpose, and include a conveyance system to differentiate normal, abnormal, and intermediate products based on predefined thresholds.

Benefits of technology

Enables accurate identification and sorting of workpieces into normal, abnormal, and intermediate categories, allowing for flexible selection and discharge of products based on set criteria, improving the precision and adaptability of the inspection process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023019408_28112024_PF_FP_ABST
    Figure JP2023019408_28112024_PF_FP_ABST
Patent Text Reader

Abstract

According to the present invention: threshold values (221) for identifying whether an object is a normal product or an abnormal product according to abnormality levels of the object are set in a computer for inspecting the object; the abnormality levels of the object are determined by using a machine learning model on an image of the object which has been acquired; and for each of a plurality of abnormality types that can occur in the same portion of the object, an object including a portion in which the determined abnormality level is at or above the threshold value (221) is identified as an abnormal product.
Need to check novelty before this filing date? Find Prior Art

Description

Inspection program and inspection device

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

[0002] When inspecting a workpiece, artificial intelligence (AI) is sometimes used to detect abnormalities (in other words, any peculiar areas) present in the photographed image of the workpiece.

[0003] When detecting anomalies using AI, the probability that an anomaly exists and the type of anomaly if one exists are output.

[0004] JP 2021-182225 A JP 2021-131364 A

[0005] However, there are cases where it is required to flexibly select whether a workpiece is abnormal or normal based not only on the presence or absence of an abnormality and the type of abnormality, but also on the level of the abnormality (in other words, the severity).

[0006] FIG. 1 is a diagram showing an image of the level of deflation of almonds.

[0007] FIG. 1 shows images of three almonds, with the levels of shrinkage of the almond surfaces increasing in the order of reference symbols A1, A2, and A3.

[0008] For example, as shown by reference symbol A2, there are almonds that are shriveled all over but only slightly, so they can be treated as normal products, while there are almonds that are significantly shriveled all over, so they can be treated as abnormal products, as shown by reference symbol A3.

[0009] The level of abnormality that is considered normal may differ depending on the purpose of the workpiece sorting, and there is a risk that multiple machine learning data sets will need to be prepared depending on the purpose of the workpiece sorting.

[0010] For example, almonds may be used as snacks to accompany alcoholic beverages, or as a single almond placed on top of pastries. When used as snacks, even slight shrinkage is expected to be tolerated and the almonds will be selected accordingly, whereas when used as a single almond placed on top of pastries, even slight shrinkage is expected to be unacceptable and the almonds will be selected accordingly.

[0011] In one aspect, the technology described in this specification aims to provide an inspection program and an inspection device that are capable of flexibly sorting workpieces regardless of the purpose of the workpiece sorting.

[0012] In one aspect, the inspection program causes a computer that inspects an object to execute a process that sets a threshold value for identifying whether the object is normal or abnormal depending on the abnormality level of the object, determines the abnormality level of the object using a machine learning model for a captured image of the object, and identifies, for each of multiple types of abnormalities that can occur in the same part of the object, the object that includes a part where the determined abnormality level is equal to or higher than the set threshold value as the abnormal product.

[0013] According to the disclosed technology, it is possible to provide an inspection program and an inspection device that are capable of flexibly sorting workpieces regardless of the purpose of the workpiece sorting.

[0014] 1 is a diagram showing an image of the level of puckering of almonds; FIG. 2 is a block diagram schematically showing an example of the hardware configuration and an example of the functional configuration of an inspection system according to an embodiment; FIG. 3 is a diagram showing an example of an abnormality level setting screen according to an embodiment; FIG. 4 is a diagram showing an example of a map of an input image according to an embodiment; (a) is a diagram showing an example of an abnormality degree map, (b) is a diagram showing an example of an abnormality level map, (c) is a diagram showing an example of an abnormality type #1 map, (d) is a diagram showing an example of an abnormality type #2 map, and (e) is a diagram showing an example of an abnormality type #3 map; (a) is a diagram explaining an evaluation process of the abnormality degree map, (b) is a diagram explaining an abnormality type selection process, and (c) is a diagram explaining a process of determining a final evaluation according to the abnormality level; FIG. 5 is a flowchart explaining an abnormality inspection process according to an embodiment; FIG. 6 is a flowchart explaining pre-processing of an input image according to an embodiment; and FIG. 7 is a flowchart explaining machine learning processing according to an embodiment.

[0015] An embodiment will be described below with reference to the drawings. However, the embodiment described below is merely an example, and is not intended to exclude various modifications and applications of techniques not explicitly stated in the embodiment. This embodiment can be implemented with various modifications within the scope of its purpose.

[0016] Furthermore, each figure does not necessarily include only the components shown in the figure, but may also include other functions, etc.

[0017] Hereinafter, in the drawings, parts with the same reference numerals indicate similar parts.

[0018] [A] Example of Embodiment [A-1] Example of System Configuration FIG. 2 is a block diagram that schematically shows an example of the hardware configuration and the functional configuration of an inspection system 100 according to an embodiment.

[0019] The inspection system 100 includes an inspection device 1, a camera 21, a display 22, and a conveying device 23.

[0020] The camera 21 photographs a workpiece (object; not shown) that is an object to be inspected by the inspection device 1, and inputs the photographed image into the inspection device 1. One or more cameras 21 may be provided in the inspection system 100. The workpiece may be, for example, a plant such as a fruit or vegetable, an animal, or an industrial product.

[0021] The display 22 displays various information for the operator of the inspection device 1. The display 22 displays, for example, a setting screen 220, which will be described later with reference to FIG.

[0022] The transport device 23 transports the workpiece to a position where it can be photographed by the camera 21. The transport device 23 may also include a discharge mechanism for discharging a defective workpiece.

[0023] The conveying device 23 may be equipped with an air blowing unit if the workpieces are lightweight and not easily damaged. The conveying device 23 uses the air blowing unit to blow and transport the workpieces, while the camera 21 photographs the workpieces. The air blowing unit may function as a discharge mechanism by adjusting the output air volume to blow normal workpieces and abnormal workpieces to different positions (in other words, distances). Furthermore, if the conveying device 23 is equipped with an air blowing unit for conveyance and an air blowing unit for discharge, the air blowing unit for discharge may blow air in a different direction from the air blowing unit for conveyance, thereby blowing normal workpieces and abnormal workpieces to different positions, thereby fulfilling the function as a discharge mechanism.

[0024] Furthermore, the transport device 23 may be equipped with a conveyor, and the work may be photographed by the camera 21 while being transported by the conveyor. In this case, a defective work may be ejected by an air blowing unit that blows air in a direction perpendicular to the transport direction of the conveyor, or by a robot arm for gripping the work or a flap for flicking the work.

[0025] The transport device 23 may be provided with an abnormality notification unit instead of the discharge mechanism. When a defective workpiece is detected, the abnormality notification unit may notify the operator of the inspection system 100 of the detection. The abnormality notification unit may notify the operator of the abnormality by, for example, displaying an image on the display 22, sounding an alert, emitting a lamp, or the like.

[0026] The inspection device 1 includes a central processing unit (CPU) 11 , a memory 12 , and a storage device 13 .

[0027] The storage device 13 is, for example, a device that stores data in a readable and writable manner, and may be, for example, a hard disk drive (HDD), a solid state drive (SSD), or a storage class memory (SCM). The storage device 13 stores a machine learning model for executing the anomaly inspection process in the embodiment.

[0028] The memory 12 is illustratively a storage device including a read-only memory (ROM) and a random access memory (RAM). The RAM may be, for example, a dynamic RAM (DRAM). Programs such as a basic input / output system (BIOS) may be written to the ROM of the memory 12. The software programs in the memory 12 may be read and executed by the CPU 11 as appropriate. The RAM of the memory 12 may also be used as a primary recording memory or a working memory.

[0029] The CPU 11 is an example of a processor, and is illustratively a processing device that performs various controls and calculations, and realizes various functions by executing an operating system (OS) and programs read from the memory 12. That is, the CPU 11 may function as a learning model acquisition unit 111, an abnormality level setting unit 112, an abnormality level determination unit 113, and a discharge processing unit 114, as shown in FIG.

[0030] The programs for implementing the functions of the learning model acquisition unit 111, the abnormality level setting unit 112, the abnormality level determination unit 113, and the discharge processing unit 114 may be provided in a form recorded on a computer-readable recording medium, such as a flexible disk, CD (CD-ROM, CD-R, CD-RW, etc.), DVD (DVD-ROM, DVD-RAM, DVD-R, DVD+R, DVD-RW, DVD+RW, HD DVD, etc.), Blu-ray disk, magnetic disk, optical disk, or magneto-optical disk. The computer (CPU 11 in this embodiment) may read the program from the above-mentioned recording medium via a reading device (not shown), transfer it to an internal recording device or an external recording device, and store it for use. Alternatively, the program may be recorded on a storage device (recording medium), such as a magnetic disk, optical disk, or magneto-optical disk, and provided to the computer from the storage device via a communication path.

[0031] To realize the functions of the learning model acquisition unit 111, the abnormality level setting unit 112, the abnormality level determination unit 113, and the discharge processing unit 114, a program stored in an internal storage device (memory 12 in this embodiment) may be executed by a computer (CPU 11 in this embodiment). Alternatively, a program recorded on a recording medium may be read and executed by a computer.

[0032] The learning model acquisition unit 111 generates a machine learning model for executing the anomaly inspection process in the embodiment and stores it in the storage device 13. The learning model acquisition unit 111 also acquires the machine learning model stored in the storage device 13. In this embodiment, by appropriately adjusting the threshold, one machine learning model that can be used for multiple sorting applications is implemented.

[0033] The abnormality level setting unit 112 sets information on the judgment threshold, the test result threshold, and the ambiguous discharge rate based on input by the operator using a setting screen 220, which will be described later with reference to FIG.

[0034] The abnormality level determination unit 113 determines the abnormality level when an abnormality exists in each work based on the machine learning model acquired by the learning model acquisition unit 111 and the input from the camera 21.

[0035] The discharge processing unit 114 discharges the workpiece that is found to be defective based on the setting by the abnormality level setting unit 112 and the abnormality level determination result by the abnormality level determination unit 113 .

[0036] Furthermore, the rejection processing unit 114 may reject some of the workpieces that are unclear as to whether they are normal or abnormal. For example, if it is acceptable for up to 10% of the almonds to be mixed with large chips, abnormal levels 1 and 2 are considered normal, abnormal levels 2 and 3 are considered unclear and 90% are rejected, and abnormal levels 3 and above are all considered abnormal and rejected.

[0037] FIG. 3 is a diagram illustrating an example of an abnormality level setting screen 220 in the embodiment.

[0038] Values ​​are input into the setting screen 220 shown in FIG. 3 by the operator of the inspection system 100 using a keyboard and mouse (not shown).

[0039] The setting screen 220 displays, for each label indicating the type of abnormality, a judgment threshold, a judgment result threshold, and an ambiguous discharge rate.

[0040] In the example shown in FIG. 3, the labels indicate "shrunk", "worm-eaten", "chipped" and "foreign matter".

[0041] The judgment threshold is set as the probability that each workpiece is a defective product for each label.

[0042] The threshold value of the inspection result is set based on the abnormality level of each work, and determines whether each work is a normal product, an ambiguous product, or an abnormal product for each label.

[0043] In Figure 3, for example, for the label "Shrunk," the threshold values ​​of the inspection results indicate that an abnormality level of less than 1.5 indicates a normal product (see the white area of ​​indicator 221), an abnormality level of 1.5 or more but less than 3.5 indicates an ambiguous product (see the diagonal area of ​​indicator 221), and an abnormality level of 3.5 or more indicates an abnormal product (see the black area of ​​indicator 221).

[0044] The questionable product may be referred to as an intermediate product having a quality intermediate between that of a normal product and a defective product.

[0045] For the threshold value of the test result, a value of "1.5" (second threshold) or "3.5" (first threshold) may be directly entered into the input box, or the value may be adjusted by operating indicator 221.

[0046] Also, in Figure 3, for example, for the label "worm-eaten," the threshold value of the inspection result is that an abnormality level of less than 2.5 is a normal product (see the white area of ​​indicator 221), and an abnormality level of 2.5 or more is an abnormal product (see the black area of ​​indicator 221), and there is no need to set an ambiguous product.

[0047] The ambiguous rejection rate is set to indicate what percentage of workpieces determined to be ambiguous items are to be rejected.

[0048] FIG. 4 is a diagram illustrating a map of an input image in the embodiment.

[0049] 4, the luminance of a monochrome image normalized to [0, 1] is expressed as a 9 × 9 map. Note that the input image may also be a color image.

[0050] In the example shown in Figure 4, 0 indicates a location where no workpiece is present, 0.25 indicates that a normal portion of the workpiece is visible, 0.8 indicates that there is abnormal portion A, 0.65 indicates that there is abnormal portion B, and 0.9 indicates that there is abnormal portion C.

[0051] (a) of Figure 5 is a diagram illustrating a map of abnormality degree, (b) of Figure 5 is a diagram illustrating a map of abnormality level, (c) of Figure 5 is a diagram illustrating a map of abnormality type #1, (d) of Figure 5 is a diagram illustrating a map of abnormality type #2, and (e) of Figure 5 is a diagram illustrating a map of abnormality type #3.

[0052] In the examples shown in Figures 5(a) to 5(e), the cell resolution is 3 x 3, which is one-third of the cell resolution of the map of the input image shown in Figure 4. Note that the cell resolution of each map in Figures 5(a) to 5(e) is not limited to 3 x 3, and may be lower or higher than 3 x 3, or may be the same as the map of the input image (9 x 9 in the example shown in Figure 4).

[0053] In Fig. 5A, the anomaly level represents the probability of an anomaly occurring as a result of evaluating an image using machine learning. In the example shown in Fig. 5A, the anomaly level is 0.71 at the location where anomaly A is located, 0.8 at the location where anomaly B is located, and 0.55 at the location where anomaly C is located.

[0054] In Fig. 5(b), the abnormality level indicates the degree of abnormality when an abnormality is detected as a result of evaluating the image using machine learning. In the example shown in Fig. 5(b), the abnormality level is 0.6 at the location of abnormal part A, 0.1 at the location of abnormal part B, and 0.9 at the location of abnormal part C.

[0055] In (c) to (e) of FIG. 5, the abnormality types correspond to the labels shown in FIG.

[0056] In the example shown in FIG. 5C, for abnormality type #1, the location where abnormal part A is located reacts as 0.51, the location where abnormal part B is located reacts as 0.2, and the location where abnormal part C is located reacts as 0.05.

[0057] In the example shown in FIG. 5(d), for abnormality type #2, the location where abnormal part A is located reacts as 0.6, the location where abnormal part B is located reacts as 0.55, and the location where abnormal part C is located reacts as 0.25.

[0058] In the example shown in FIG. 5C, for abnormality type #3, the location where abnormal part A is located reacts as 0.1, the location where abnormal part B is located reacts as 0.05, and the location where abnormal part C is located reacts as 0.45.

[0059] FIG. 6A is a diagram explaining the evaluation process of the abnormality degree map, FIG. 6B is a diagram explaining the selection process of the abnormality type, and FIG. 6C is a diagram explaining the process of determining the final evaluation according to the abnormality level.

[0060] Assume that the threshold of the degree of abnormality of abnormality type #1 is 0.7, the threshold of the degree of abnormality of abnormality type #2 is 0.6, and the threshold of the degree of abnormality of abnormality type #3 is 0.51.

[0061] In the anomaly degree map shown in Figure 6A, cells that exceed the minimum threshold value for each anomaly type (in this example, 0.51 for anomaly type #3) are selected. The location of 0.71 is designated as cell #1, the location of 0.8 as cell #5, and the location of 0.55 as cell #9.

[0062] In the selection of anomaly types shown in Figure 6(b), maps for anomaly types #1 to #3 are arranged, and a value exceeding the threshold (0.5 in this example) is selected in each map at the same coordinates as the coordinates of the cell selected in Figure 6(a). In the example shown in Figure 6(b), 0.51 is selected as a value exceeding the threshold of 0.5 in the map for anomaly type #1, and 0.6 and 0.55 are selected as values ​​exceeding the threshold of 0.5 in the map for anomaly type #2. On the other hand, in the map for anomaly type #3, there is no value exceeding the threshold of 0.5 at the same coordinates as the coordinates of the cell selected in Figure 6(a), so the maximum value, 0.45, is selected.

[0063] In determining the final evaluation according to the abnormality level shown in FIG. 6(c), if for abnormality type #1 an abnormality level of 0.5 or higher is abnormal, for abnormality type #2 an abnormality level of 0.05 to 0.56 is considered ambiguous and anything above that is abnormal, and for abnormality type #3 anything below 0.75 is considered normal, then cell #1 will be judged as abnormal for both abnormality types #1 and #2, cell #5 will be judged as ambiguous for abnormality type #2, and cell #9 will be judged as normal for abnormality type #3.

[0064] [A-2] Operation Example The abnormality inspection process in this embodiment will be described with reference to the flowchart shown in FIG. 7 (steps S1 to S12).

[0065] The abnormality level determination unit 113 performs inference on the image using AI (step S1).

[0066] The abnormality level determination unit 113 determines whether there is a cell having an abnormality degree exceeding the smallest abnormality degree threshold value for each abnormality type (step S2).

[0067] If there is no cell having an abnormality degree exceeding the smallest abnormality degree among the thresholds for each abnormality type (see No route in step S2), the abnormality inspection process ends.

[0068] On the other hand, if there is a cell with an abnormality level exceeding the smallest threshold for each anomaly type (see the Yes route in step S2), the processes of steps S4 to S11 are repeated to start evaluation of all cells with an abnormality level exceeding the smallest threshold (step S3). Note that evaluation may also be performed for all cells other than the cell with an abnormality level exceeding the smallest threshold.

[0069] The abnormality level determination unit 113 determines the abnormality type of the cell (step S4).

[0070] The abnormality level determination unit 113 determines whether the degree of abnormality is equal to or greater than the threshold value of any of the determined abnormality types (step S5).

[0071] If the degree of abnormality is not equal to or greater than the threshold value of any of the determined abnormality types (see the No route in step S5), the process proceeds to step S12.

[0072] On the other hand, if the degree of abnormality is equal to or greater than the threshold value of any of the determined abnormality types (see the Yes route in step S5), the processes of steps S7 to S10 are repeated, and evaluation of all determined abnormality types begins (step S6).

[0073] The abnormality level determination unit 113 determines whether the abnormality level is ambiguous or higher (step S7).

[0074] If the abnormality level is not above the ambiguous level (see the No route in step S7), the process proceeds to step S11.

[0075] On the other hand, if the abnormality level is ambiguous or higher (see the Yes route in step S7), it is determined whether the abnormality level is abnormal or higher (step S8).

[0076] If the abnormality level is abnormal or higher (see the Yes route in step S8), the process proceeds to step S10.

[0077] On the other hand, if the abnormality level is not abnormal or higher (see the No route in step S8), the discharge processing unit 114 determines whether discharge has been selected with a predetermined probability (step S9).

[0078] If discharge is not selected with a predetermined probability (see the No route in step S9), the process proceeds to step S11.

[0079] On the other hand, if discharge is selected with a predetermined probability (see the Yes route in step S9), the discharge processing unit 114 executes the discharge process (step S10).

[0080] When the evaluation of all the determined abnormality types has been repeatedly performed in steps S7 to S10, the evaluation of all the determined abnormality types is completed (step S11).

[0081] When all cell evaluations have been repeatedly performed in steps S4 to S11, evaluation of all cells having an abnormality degree exceeding the minimum threshold is completed (step S12), and the abnormality inspection process is then completed.

[0082] Next, the pre-processing of the input image in this embodiment will be described with reference to the flowchart shown in FIG. 8 (steps S21 to S23).

[0083] The abnormality level determination unit 113 functions as an image input unit that processes an input image as necessary (step S21).

[0084] The abnormality level determination unit 113 functions as a latent space projection unit that projects the input into a low-dimensional latent space to enable high-speed processing (step S22).

[0085] The abnormality level determination unit 113 functions as an output unit that interprets the information in the latent space, converts it into a predetermined format, and outputs it (step S23).Then, the pre-processing of the input image is completed.

[0086] Next, the machine learning process in this embodiment will be described with reference to the flowchart (steps S31 to S38) shown in FIG.

[0087] When a learning sample is imaged by placing the sample in the inspection system 100 and capturing an image with the camera 21, the learning model acquisition unit 111 performs instruction on the captured data (step S31). Specifically, the learning model acquisition unit 111 instructs the abnormality type and abnormality level to the abnormal portion in the image, and performs processing called semantic segmentation and bounding box only on the abnormal portion.

[0088] The learning model acquisition unit 111 converts the image and teaching data into a format suitable for learning, such as a [0, 1] map (step S32).

[0089] The processes of steps S34 to S37 are repeatedly executed, and learning is started until sufficient accuracy is achieved (step S33).

[0090] The learning model acquisition unit 111 inputs the converted image to the AI ​​(step S34).

[0091] The learning model acquisition unit 111 acquires the output of the AI ​​(step S35).

[0092] The learning model acquisition unit 111 evaluates the output of the AI ​​against the converted teaching data (step S36).

[0093] The learning model acquisition unit 111 backpropagates the evaluation results and updates the AI ​​parameters (step S37).

[0094] When the learning is repeated in steps S34 to S37 until sufficient accuracy is achieved, the learning is completed (step S38), and the machine learning process is terminated.

[0095] [A-3] Effects According to the inspection program and the inspection device 1 in the example of the embodiment described above, for example, the following effects can be achieved.

[0096] The abnormality level setting unit 112 sets a threshold value for determining whether an object is normal or abnormal according to the abnormality level of the object. The abnormality level determination unit 113 determines the abnormality level of the object using a machine learning model for a captured image of the object. The abnormality level determination unit 113 identifies, as an abnormal object, an object including a portion where the determined abnormality level is equal to or greater than the set threshold value for each of multiple types of abnormalities that may occur in the same portion of the object.

[0097] This enables flexible work sorting using a single machine learning model regardless of the purpose of work sorting, and also enables appropriate work sorting according to the abnormality level for each of multiple anomaly types.

[0098] The states of the object include not only normal and abnormal products, but also intermediate products having quality intermediate between normal and abnormal products. The thresholds include a first threshold for identifying whether the object is a normal product or an intermediate product, and a second threshold for identifying whether the object is an intermediate product or normal product. The abnormality level determination unit 113 identifies, as a normal product, an object including a portion whose determined abnormality level is equal to or greater than the first threshold, identifies, as an intermediate product, an object including a portion whose determined abnormality level is equal to or greater than the second threshold but less than the first threshold and that has not been identified as a normal product, and identifies, as a normal product, an object including a portion whose determined abnormality level is equal to or greater than the second threshold and that has not been identified as a normal product or an intermediate product.

[0099] This makes it possible to appropriately identify intermediate products (in other words, ambiguous products) that have intermediate quality between normal products and defective products.

[0100] The abnormality level setting unit 112 sets a rejection rate for intermediate products for each of a plurality of abnormality types. The rejection processing unit 114 rejects all of the abnormal products from the plurality of targets, as well as a portion of the intermediate products according to the rejection rate.

[0101] This allows intermediate products to be discharged at an appropriate rate.

[0102] The abnormality level setting unit 112 sets a plurality of thresholds for each abnormality type. The abnormality level determination unit 113 determines a plurality of abnormality levels for each abnormality type. The abnormality level determination unit 113 identifies, as an abnormal product, an object that includes a portion where each of the determined abnormality levels is equal to or greater than a corresponding threshold among the set thresholds.

[0103] This allows for more accurate identification of defective products by determining multiple abnormality levels for each type of abnormality.

[0104] [B] Others The disclosed technology is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present embodiment. The configurations and processes of the present embodiment can be selected or combined as needed.

[0105] 100: Inspection system 1: Inspection device 11: CPU 111: Learning model acquisition unit 112: Abnormal level setting unit 113: Abnormal level determination unit 114: Discharge processing unit 12: Memory 13: Storage device 21: Camera 22: Display 220: Setting screen 221: Indicator 23: Conveyance device

Claims

1. An inspection program that causes a computer that inspects an object to execute the following processes: set a threshold value for identifying whether the object is normal or abnormal depending on the level of abnormality in the object; determine the level of abnormality in the object using a machine learning model for a captured image of the object; and identify, as the abnormal object, the object that includes a portion where the determined level of abnormality is equal to or higher than the set threshold value for each of multiple types of abnormalities that may occur in the same part of the object.

2. The inspection program according to claim 1, which causes the computer to execute the following processes: the state of the object includes, in addition to the normal product and the abnormal product, an intermediate product having a quality intermediate between the normal product and the abnormal product; the thresholds include a first threshold for identifying whether the object is the abnormal product or the intermediate product, and a second threshold for identifying whether the object is the intermediate product or the normal product; identifying the object including a portion where the determined abnormality level is equal to or greater than the first threshold as the abnormal product, identifying the object including a portion where the determined abnormality level is equal to or greater than the second threshold and less than the first threshold and which has not been identified as the abnormal product as the intermediate product, and identifying the object including a portion where the determined abnormality level is less than the second threshold and which has not been identified as the abnormal product or the intermediate product as the normal product.

3. The inspection program according to claim 2, which causes the computer to execute the following process: setting a rejection rate for the intermediate products for each of the plurality of abnormality types; and rejecting, from among the plurality of objects, a portion of the intermediate products according to the rejection rate in addition to all of the abnormal products.

4. The inspection program of any one of claims 1 to 3, which causes the computer to execute the following processes: setting a plurality of thresholds for each type of abnormality; judging a plurality of abnormality levels for each type of abnormality; and identifying the object including a portion where each of the judged abnormality levels is equal to or higher than a corresponding one of the set thresholds as the abnormal product.

5. An inspection device for inspecting an object, comprising a processor: setting a threshold for identifying whether the object is normal or abnormal depending on the level of abnormality in the object; determining the level of abnormality in the object using a machine learning model for a captured image of the object; and identifying the object as the abnormal item including a portion where the determined level of abnormality is equal to or higher than the set threshold for each of a plurality of types of abnormality that may occur in the same part of the object.

6. The inspection device of claim 5, wherein the states of the object include, in addition to the normal product and the abnormal product, an intermediate product having a quality intermediate between the normal product and the abnormal product, and the thresholds include a first threshold for identifying whether the object is the abnormal product or the intermediate product, and a second threshold for identifying whether the object is the intermediate product or the normal product, and the processor identifies the object including a portion where the determined abnormality level is equal to or greater than the first threshold as the abnormal product, identifies the object including a portion where the determined abnormality level is equal to or greater than the second threshold and less than the first threshold and which has not been identified as the abnormal product as the intermediate product, and identifies the object including a portion where the determined abnormality level is less than the second threshold and which has not been identified as the abnormal product or the intermediate product as the normal product.

7. The inspection device according to claim 6, wherein the processor sets a rejection rate for the intermediate products for each of the plurality of abnormality types, and rejects, from among the plurality of objects, a portion of the intermediate products according to the rejection rate in addition to all of the abnormal products.

8. The inspection device described in any one of claims 5 to 7, wherein the processor sets a plurality of thresholds for each type of abnormality, determines a plurality of abnormality levels for each type of abnormality, and identifies the object including a portion where each of the determined abnormality levels is equal to or higher than a corresponding one of the set thresholds as the abnormal product.

Citation Information

Patent Citations

  • Optical sorter and method for sorting by optical sorter

    JP2012143670A

  • Inspection method

    JP2015203586A

  • Inspection device, abnormality detection method, computer program, learning model generation method, and learning model

    JP2020153765A

  • Food product inspection system, food product inspection learning device and food product inspection method

    JP2022160796A