Apparatus for visual inspection and method for visual inspection

Abstract

This suppresses fluctuations in judgment during visual inspection of the external appearance of the object being inspected. [Solution] The visual inspection device comprises an anomaly detection unit, a feature quantity calculation unit, a quality determination unit, a visual inspection support unit, and a display unit. The anomaly detection unit detects anomalies in the inspection target image generated by photographing the inspection target. The feature quantity calculation unit calculates feature quantities related to the anomaly. The quality determination unit determines whether the inspection target is a good product or a defective product by applying the feature quantities calculated by the feature quantity calculation unit to a determination flow in which one or more conditions corresponding to the feature quantities related to the anomaly are set. The visual inspection support unit compares the first determination result obtained by the quality determination unit with a second determination result obtained from a visual inspection of the inspection target, which indicates whether the inspection target is a good product or a defective product. The display unit displays the determination flow when the first determination result and the second determination result are different from each other.

Description

Technical Field

[0001] The present invention relates to an apparatus and method for inspecting the appearance of an object.

Background Art

[0002] In the visual inspection of the appearance of products and the like, visual inspection by a person may be performed on defect candidates extracted by image processing. However, since visual inspection depends on the physical condition of the inspector (e.g., fatigue) and the skills of each individual, fluctuations (i.e., variations) in the pass / fail judgment may occur. Therefore, a method has been proposed to improve the efficiency of the visual inspection judgment in the appearance inspection of the inspected object and to prevent fluctuations in the inspection criteria caused by visual inspection (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] As described above, a method has been proposed to prevent fluctuations in the inspection criteria caused by visual inspection. However, in the proposed technology, it is difficult to set thresholds and the like for extracting defect candidates, and there is a risk that defect products cannot be extracted, or the number of extracted defect candidates increases, resulting in an increased burden on the inspector.

[0005] One aspect of the present invention aims to suppress fluctuations in the visual inspection judgment of the appearance of an inspection object.

Means for Solving the Problems

[0006] An appearance inspection apparatus according to one aspect of the present invention includes: an abnormality detection unit that detects abnormal locations in an inspection target image generated by photographing an inspection target; a feature quantity calculation unit that calculates feature quantities related to the abnormal locations; a quality determination unit that determines whether the inspection target is a good product or a defective product by applying the feature quantities to a determination flow in which one or more conditions corresponding to the feature quantities are set; a visual inspection support unit that compares a first determination result obtained by the quality determination unit with a second determination result obtained in a visual inspection of the inspection target, which indicates whether the inspection target is a good product or a defective product; and a display unit that displays the determination flow in which one or more conditions are set when the first determination result and the second determination result are different from each other. [Effects of the Invention]

[0007] According to the above-described embodiment, fluctuations in judgment during visual inspection of the appearance of the object being inspected are suppressed. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of a visual inspection device according to an embodiment of the present invention. [Figure 2] This is a flowchart (part 1) showing an example of the processing performed by a visual inspection device. [Figure 3] This is a flowchart (part 2) illustrating an example of the processing performed by a visual inspection device. [Figure 4] This figure shows an example of detecting an abnormal area. [Figure 5] This figure shows an example of a histogram for brightness values. [Figure 6] This figure shows an example of a decision tree used in quality determination. [Figure 7] This figure shows an example of a decision tree flow displayed on the display unit. [Figure 8] This figure shows an example of a confirmation screen that appears when the visual inspection result differs from the pass / fail judgment result from the pass / fail judgment unit. [Figure 9] This figure shows an example of the hardware configuration of a visual inspection device. [Modes for carrying out the invention]

[0009] Figure 1 shows an example of a visual inspection device according to an embodiment of the present invention. The visual inspection device 1 according to an embodiment of the present invention comprises an inspection unit 10, an inspection support unit 20, and an inspection result database 30. The visual inspection device 1 may further include other functions or devices not shown in Figure 1. For example, a camera 40 may be connected to the visual inspection device 1, and the visual inspection device 1 may be configured to include the camera 40.

[0010] Camera 40 is a digital camera (or electronic camera) that uses semiconductor elements such as an image sensor to convert input light into an electrical signal, and further converts that electrical signal into digital data. In this embodiment, camera 40 generates an inspection image by photographing the appearance of the object to be inspected 50. The object to be inspected 50 is not particularly limited, but for example, it is an industrial product produced in a factory. In this case, camera 40 photographs a number of objects to be inspected 50 one by one in sequence and generates an inspection image for each.

[0011] The inspection unit 10 includes an image acquisition unit 11, an anomaly detection unit 12, a feature quantity calculation unit 13, and a defective product candidate extraction unit 14. The inspection unit 10 may also include other functions not shown in Figure 1.

[0012] The image acquisition unit 11 includes a memory element such as a semiconductor memory and acquires and stores the image to be inspected generated by the camera 40. The image to be inspected may be a color image or a black and white image. The image to be inspected is composed of multiple pixels (i.e., picture elements), and each pixel is represented by luminance data. When the image to be inspected is a color image, each pixel may be represented by luminance data of the R component, G component, and B component.

[0013] The abnormality detection unit 12 detects abnormal locations in the inspection target image acquired by the image acquisition unit 11. That is, abnormalities appearing in the appearance of the inspection target object 50 are detected. Note that the method for detecting abnormal locations is not particularly limited and can be realized by known techniques.

[0014] The feature amount calculation unit 13 calculates feature amounts related to the abnormal locations detected by the abnormality detection unit 12. The feature amount may represent the number of abnormal locations detected by the abnormality detection unit 12. The feature amount may represent the area of each abnormal location detected by the abnormality detection unit 12. The area of an abnormal location is represented, for example, by the number of pixels in the region detected as an abnormal location in the inspection target image.

[0015] The defective product candidate extraction unit 14 determines whether the inspection target object 50 is a non-defective product or a defective product based on the feature amounts calculated by the feature amount calculation unit 13. Here, in order to reliably detect defective products or to prevent detection misses of defective products, the inspection unit 10 performs a pass / fail determination with strict criteria. That is, an inspection target object 50 that may be a defective product is determined to be a defective product by the defective product candidate extraction unit 14. Therefore, even an inspection target object 50 determined to be a defective product by the defective product candidate extraction unit 14 may actually be a non-defective product. Thus, in the following description, an inspection target object 50 determined to be a defective product by the defective product candidate extraction unit 14 may be referred to as a "defective product candidate".

[0016] When the inspection target object 50 is determined to be a non-defective product by the defective product candidate extraction unit 14, the inspection unit 10 may store the inspection target image of the inspection target object 50 in the inspection result database 30 in association with the determination result. On the other hand, when the inspection target object 50 is determined to be a defective product by the defective product candidate extraction unit 14 (that is, when a defective product candidate is extracted), the inspection unit 10 transmits the inspection target image of the inspection target object 50 and the feature amounts calculated by the feature amount calculation unit 13 to the inspection support unit 20.

[0017] Here, the inspection object 50 is inspected by the inspection unit 10 of the appearance inspection device 1 and is also visually inspected by a person. However, the visual inspection does not necessarily have to be performed on all inspection objects 50. For example, it may be performed only on the inspection objects 50 (i.e., defective product candidates) determined to be defective products by the defective product candidate extraction unit 14.

[0018] The inspection support unit 20 has a function for suppressing fluctuations (i.e., variations) in the pass / fail determination in visual inspection by a person. To realize this function, the inspection support unit 20 includes a feature quantity calculation unit 21, a pass / fail determination unit 22, a visual inspection support unit 23, and a display unit 24. Note that the inspection support unit 20 may further have other functions not shown in FIG. 1.

[0019] The inspection support unit 20 is activated when a defective product candidate is extracted in the inspection unit 10. Then, the inspection support unit 20 acquires the inspection target image of the inspection object 50 extracted as a defective product candidate and the feature quantity calculated by the feature quantity calculation unit 13.

[0020] The feature quantity calculation unit 21 calculates the feature quantity related to the abnormal location detected by the abnormality detection unit 12. Here, the feature quantity calculation unit 21 calculates a feature quantity different from the feature quantity calculated by the feature quantity calculation unit 13. For example, the feature quantity calculation unit 21 may calculate the luminance distribution within the region corresponding to the abnormal location detected by the abnormality detection unit 12. The region corresponding to the abnormal location detected by the abnormality detection unit 12 may be referred to as an "abnormal detection region".

[0021] The quality determination unit 22 determines whether the object to be inspected 50 is a good product or a defective product by providing the feature quantities calculated by the feature quantity calculation unit 13 and the feature quantity calculation unit 21 to a pre-created judgment flow. This judgment flow has one or more conditional expressions set that correspond to the feature quantities calculated by the feature quantity calculation unit 13 and the feature quantity calculation unit 21. Each conditional expression shall include a threshold corresponding to the feature quantity related to the defective part. The quality determination unit 22 performs quality determination based on more parameters than the defective product candidate extraction unit 14, and therefore has a higher judgment accuracy.

[0022] The visual inspection support unit 23 compares the judgment result obtained by the pass / fail judgment unit 22 with the judgment result obtained from the visual inspection. If the two judgment results match, the visual inspection support unit 23 determines that the judgment result from the visual inspection is correct. Conversely, if the two judgment results differ, the visual inspection support unit 23 determines that the judgment result from the visual inspection is incorrect, or that the judgment flow described above is inappropriate. In this case, the visual inspection support unit 23 may save the inspection target image of the inspection target object 50 as a visual judgment fluctuation image in the inspection result database 30.

[0023] When the judgment result obtained by the pass / fail judgment unit 22 differs from the judgment result obtained by the visual inspection, the display unit 24 displays the judgment flow used by the pass / fail judgment unit 22 on a display device (not shown). At this time, one or more conditional expressions constituting the judgment flow and the threshold values ​​set for each conditional expression are displayed. This allows the inspector performing the visual inspection to check whether their pass / fail judgment is wavering or not. Furthermore, if the judgment flow is inappropriate, the inspector can correct it.

[0024] Figures 2 and 3 are flowcharts illustrating an example of the processing performed by the visual inspection device 1. Note that the processing shown in the flowchart in Figure 2 is mainly performed by the inspection unit 10. The processing shown in the flowchart in Figure 3 is mainly performed by the inspection support unit 20.

[0025] In S1, the image acquisition unit 11 acquires an image of the object to be inspected. The image of the object to be inspected is generated when the camera 40 captures the external appearance of the object to be inspected 50.

[0026] In S2, the anomaly detection unit 12 detects abnormal areas in the inspection target image acquired by the image acquisition unit 11. That is, abnormalities appearing on the appearance of the inspection target object 50 are detected using the inspection target image. Abnormalities appearing on the appearance of the inspection target object 50 include, for example, dirt, stains, discoloration, and scratches. The method of detecting abnormal areas by image processing is not particularly limited and can be implemented using known technologies. For example, the anomaly detection unit 12 may perform image processing such as contrast conversion, morphological conversion, edge detection, and binarization on the inspection target image, and further perform pattern matching to detect abnormal areas. Alternatively, the anomaly detection unit 12 may use AI processing such as semantic segmentation to detect abnormal areas.

[0027] Figure 4 shows an example of abnormality detection. In this embodiment, the image acquisition unit 11 acquires the inspection target image shown in Figure 4A. This inspection target image is composed of multiple pixels, and each pixel has brightness information. When brightness is represented by 8 bits, the value range of the brightness information is "0" to "255".

[0028] The anomaly detection unit 12 detects abnormal areas by detecting "edges" in the image to be inspected. In this embodiment, as shown in Figure 4B, three abnormal areas X1 to X3 are detected.

[0029] In S3, the feature calculation unit 13 calculates features related to the anomaly detected by the anomaly detection unit 12. For example, the feature calculation unit 13 obtains the following values ​​as features related to the anomaly.

[0030] (1) Number of abnormal locations detected by the abnormality detection unit 12 (2) Area of ​​each abnormal location detected by the abnormality detection unit 12 In the example shown in Figure 4B, the number of detected anomalies is "3". Additionally, the area of ​​the anomaly is calculated by determining the number of pixels within each region identified as anomaly X1 to X3.

[0031] In steps S4 to S5, the defective product candidate extraction unit 14 determines whether the object to be inspected 50 is a good product or a defective product based on the feature quantities calculated by the feature quantity calculation unit 13. For example, if two or more abnormal locations are detected in the image to be inspected, the defective product candidate extraction unit 14 extracts the object to be inspected 50 as a non-good product candidate. Furthermore, even if only one abnormal location is detected in the image to be inspected, if the area of ​​that abnormal location (i.e., the number of pixels in the area corresponding to that abnormal location) exceeds a predetermined threshold, the defective product candidate extraction unit 14 may extract the object to be inspected 50 as a non-good product candidate.

[0032] When the object to be inspected 50 is identified as a candidate for a defective product, the visual inspection device 1 proceeds to S11. That is, the inspection support unit 20 is activated. At this time, the inspection unit 10 transmits the inspection target image of the object to be inspected 50 and the feature quantities calculated by the feature quantity calculation unit 13 to the inspection support unit 20. On the other hand, when the object to be inspected 50 is determined to be a good product, in S6, the inspection unit 10 saves the inspection target image of the object to be inspected 50 in the inspection result database 30 in association with the determination result.

[0033] In S7, the inspection unit 10 determines whether or not to terminate the visual inspection. If the visual inspection is to continue, the visual inspection device 1 returns to S1. In this case, the visual inspection device 1 acquires the inspection target image of the next object to be inspected 50 and executes the processes in S2 to S6. On the other hand, if the inspector of the object to be inspected 50 gives a termination instruction, the inspection unit 10 terminates the inspection.

[0034] Here, the inspector of the object to be inspected 50 visually inspects the appearance of the object to be inspected 50. However, the inspector is not required to visually inspect all of the objects to be inspected 50; they may only visually inspect the objects to be inspected 50 that have been determined to be defective by the defective product candidate extraction unit 14 (i.e., defective product candidates).

[0035] In S11, the inspection support unit 20 acquires the inspection target image of the inspection target object 50 determined to be defective by the inspection unit 10, and the feature quantities calculated by the feature quantity calculation unit 13 for the abnormal areas detected in the inspection target image. The inspection support unit 20 does not necessarily need to acquire pixel information for the entire inspection target image; it may receive only pixel information for each area identified as an abnormal area. The pixel information shall include brightness information.

[0036] In S12, the feature calculation unit 21 calculates feature quantities related to the abnormal location detected by the abnormality detection unit 12. In this visual inspection by the inspector, for example, the degree and size of dirt or stains attached to the surface of the object to be inspected 50 are confirmed. At this time, the inspector checks whether the dirt or stain is dark or light. Therefore, the feature calculation unit 21 creates a luminance histogram representing the distribution of luminance values ​​within the area corresponding to the abnormal location as a feature quantity related to the abnormal location.

[0037] Figure 5 shows an example of a histogram for luminance values. The horizontal axis represents luminance (luminance range). In this example, the luminance range is "0" to "255". The luminance range is divided into multiple luminance blocks. The size of each luminance block is "16". The vertical axis represents area. Area is represented by the number of pixels. In other words, this histogram represents the number of pixels (i.e., area) belonging to each luminance block. Note that a histogram is created for each anomaly.

[0038] In the example shown in Figure 5, within the region corresponding to the anomaly, for example, there are no pixels with a brightness value of "0 to 15," there are 45 pixels with a brightness value of "16 to 31," and there are 120 pixels with a brightness value of "32 to 47." Thus, the brightness histogram shows what brightness ranges of pixels make up the region corresponding to each anomaly.

[0039] In S13, the quality determination unit 22 determines whether the object to be inspected 50 is good or defective by providing the feature quantities calculated by the feature quantity calculation unit 13 and the feature quantity calculation unit 21 to a pre-created judgment flow. In this embodiment, the quality determination unit 22 makes a quality determination using the number of detected abnormal locations, the maximum value of the area of ​​each abnormal location (maximum detected area), and a histogram of brightness information for each abnormal location.

[0040] The decision flow is created in advance before the visual inspection of the objects to be inspected 50 begins. For example, before the visual inspection begins, a predetermined number of objects to be inspected 50 are photographed using the camera 40 to obtain an image of each object to be inspected 50. Visual inspection is then performed on these objects to determine whether they are good or defective. The AI ​​model, which outputs a good / bad judgment result based on the inspected images, is given the inspected images of each object to be inspected 50, and the good / bad judgment result from the visual inspection is given as the "correct answer" for each inspected image. This optimizes each parameter in the AI ​​model. In this embodiment, a decision tree as a decision flow is created by the procedure described above. The decision tree includes one or more conditional expressions for determining whether the object to be inspected 50 is good or defective, and each conditional expression is set to a predetermined threshold. In this case, the content of one or more conditional expressions is determined by the procedure described above, and furthermore, the threshold values ​​set for each conditional expression are determined.

[0041] Figure 6 shows an example of a decision tree used in pass / fail judgment. In this example, the decision tree consists of four conditional expressions. Each conditional expression represents a condition for determining whether the object to be inspected 50 is a defective product or not. For example, condition 1 represents "If multiple abnormalities are detected in the image to be detected, the object to be inspected is a defective product." Condition 2 is used when condition 1 is not met. That is, condition 2 is used when only one abnormality is detected in the image to be inspected. Condition 2 represents "If the maximum detection area is greater than 800 pixels, the object to be inspected is a defective product." The maximum detection area represents the area of ​​the largest abnormality detected by the abnormality detection unit 12. However, in the decision tree of this embodiment, condition 2 is used when condition 1 is not met. That is, condition 2 is used when only one abnormality is detected by the abnormality detection unit 12. Therefore, condition 2 essentially represents "If the area of ​​the detected abnormality is greater than 800 pixels, the object to be inspected is a defective product."

[0042] Condition 3 is used when neither Conditions 1 nor 2 are met. Condition 3 means that "if the number of pixels with a brightness value of "32 to 47" in the area corresponding to the abnormal location is greater than 100, the item being inspected is defective." Condition 4 is used when neither Conditions 1 nor 3 are met. Condition 4 means that "if the number of pixels with a brightness value of "96 to 111" in the area corresponding to the abnormal location is greater than 200, the item being inspected is defective." Note that in Conditions 3 and 4, the number of pixels may be replaced with "area."

[0043] The quality determination unit 22 determines whether the object to be inspected 50 is a good product or a defective product by providing the features calculated by the feature calculation unit 13 and the feature calculation unit 21 to the judgment flow (i.e., the decision tree) described above. In this embodiment, the number of abnormal locations detected in the target image, the area of ​​the abnormal locations, and the brightness histogram for the abnormal locations are provided to the judgment flow.

[0044] In S14, the visual inspection device 1 acquires the judgment result of the visual inspection performed by the inspector. Specifically, the visual inspection device 1 acquires the judgment result of the visual inspection of the object to be inspected 50 that has been determined to be defective by the inspection unit 10. The judgment result of the visual inspection indicates whether the object to be inspected 50 is a good product or a defective product. The inspector then records the judgment result of the visual inspection in a predetermined memory area, and the visual inspection device 1 acquires the judgment result from that memory area.

[0045] In S15, the visual inspection support unit 23 compares the judgment result from the pass / fail judgment unit 22 with the judgment result of the visual inspection. If the two judgment results match, the visual inspection device 1 proceeds to S7 in Figure 2. That is, the inspection unit 10 checks whether or not to terminate the visual inspection. However, when the two judgment results match, although not specifically shown, the visual inspection device 1 may save the inspection target image of the inspection target object 50 in the inspection result database 30 in association with the judgment result. In this case, the visual inspection device 1 may save the inspection target image according to the inspector's instructions.

[0046] When the two judgment results differ from each other, the visual inspection support unit 23 determines that the visual inspection judgment result is incorrect, or that the judgment flow described above is inappropriate. For example, the visual inspection judgment result may fluctuate if the inspector lacks experience. Similarly, the visual inspection judgment result may fluctuate if the inspector is not in good health. On the other hand, if the conditional expression constituting the judgment flow or the threshold set within that conditional expression is inappropriate, the judgment result by the pass / fail judgment unit 22 may be inappropriate.

[0047] Here, if the judgment result from the pass / fail judgment unit 22 and the judgment result from the visual inspection differ from each other, it is important to make the inspector aware of the reason or cause for the difference between the two judgment results in order to suppress fluctuations in the visual inspection judgment. Therefore, when the two judgment results differ from each other, in S16, the display unit 24 displays a decision tree representing the judgment flow on the display device. At this time, it is preferable that the path taken by the pass / fail judgment unit 22 to determine whether the inspected object 50 is a "good product" or a "defective product" is highlighted on the judgment flow.

[0048] Figure 7 shows an example of a decision tree displayed by the display unit 24. In this embodiment, the following feature quantities related to the abnormal location are provided to the decision tree. (1) Number of abnormal areas detected in the target image: 1 (2) Maximum detection area (area of ​​the largest abnormal area detected in the target image): 605 pixels (3) Histogram of brightness information for abnormal locations (3-1) Number of pixels with a brightness value of "0 to 15": 0 (3-2) Number of pixels with a brightness value of "16-31": 45 (3-3) Number of pixels with a brightness value of "32-47": 120 (3-4) Number of pixels with a brightness value of "48-63": 90 (3-5) Number of pixels with a brightness value of "64-79": 95 (3-6) Number of pixels with a brightness value of "80-95": 85 (3-7) Number of pixels with a brightness value of "96-111": 75 (3-8) Number of pixels with a brightness value of "112-127": 45 (3-9) Number of pixels with a brightness value of "128-143": 30 (3-10) Number of pixels with a brightness value of "144-159": 20 (3-11) Number of pixels with a brightness value of "160-175": 0 (3-12) Number of pixels with a brightness value of "176-191": 0 (3-13) Number of pixels with a brightness value of "192-207": 0 (3-14) Number of pixels with a brightness value of "208-223": 0 (3-15) Number of pixels with a brightness value of "224~239": 0 (3-16) Number of pixels with a brightness value of "240~255": 0

[0049] In the decision tree's condition 1, the number of detected anomalies is compared with the threshold "1". According to feature (1), the number of anomalies is 1, which is below the threshold for condition 1. Therefore, condition 1 is not met. Thus, condition 2 is invoked.

[0050] In condition 2, the maximum detected area is compared with the threshold "800". According to feature (2), the area of ​​the anomaly is 605, which is below the threshold for condition 2. In other words, condition 2 is not met. Therefore, condition 3 is invoked.

[0051] In condition 3, the number of pixels with a brightness value of 32 to 47 within the area corresponding to the abnormality is compared with the threshold value of "100". According to feature (3-3), the number of pixels with a brightness value of 32 to 47 is 120, which exceeds the threshold value of condition 3. In other words, condition 3 is satisfied. Therefore, the quality determination unit 22 determines that the object being inspected is a defective product.

[0052] In this case, the display unit 24 displays the decision tree shown in Figure 7. At this time, the path that leads the inspected object to being classified as a "defective product" in the judgment flow is highlighted. Specifically, it is displayed that the judgments for condition 1 and condition 2 are "No" and the judgment for condition 3 is "Yes". In this way, the judgment flow displayed by the display unit 24 indicates whether or not one or more conditions used in the judgment by the pass / fail judgment unit 22 have been met.

[0053] The inspector performing the visual inspection can view this decision tree. Therefore, the inspector can understand why their judgment differs from the judgment made by the pass / fail judgment unit 22. In the cases shown in Figures 6 and 7, the inspector can recognize that their visual inspection judgment differs from the judgment made by the pass / fail judgment unit 22 regarding condition 3. In this case, the inspector can recognize that attention is needed when identifying areas with low brightness values.

[0054] Furthermore, the inspector can verify whether the content of each conditional expression that makes up the decision tree, or the threshold set within each conditional expression, is appropriate. If the conditional expression or threshold is not appropriate, the inspector or the administrator of the visual inspection device 1 may input an instruction to update the decision tree.

[0055] Returning to the flowchart shown in Figure 3, the inspection support unit 20 determines in S17 whether or not an instruction to update the decision tree is input when the decision tree is displayed. If an instruction to update the decision tree is input, the inspection support unit 20 updates the decision tree in S18 according to the instruction. For example, the threshold value in the specified condition expression is updated. On the other hand, if no instruction to update the decision tree is input, S18 is skipped. After this, the processing of the visual inspection device 1 proceeds to S7 in Figure 2. That is, the inspection unit 10 determines whether or not to terminate the visual inspection.

[0056] If the judgment result from the quality determination unit 22 and the judgment result from the visual inspection differ, the visual inspection device 1 may, although not specifically shown in the figures, save the inspection target image of the inspection target object 50 in the inspection result database 30. In this case, information indicating that the judgment result from the quality determination unit 22 and the judgment result from the visual inspection differ may be recorded in association with this inspection target image.

[0057] Figure 8 shows an example of a confirmation screen displayed when the visual inspection result differs from the pass / fail judgment result of the pass / fail judgment unit 22. In this example, the confirmation screen includes the inspection target image, the anomaly detection image, the feature quantities of the anomaly location, the decision tree flow, and various function buttons.

[0058] The inspection target image is an image of the exterior of the inspection target object 50 captured by the camera 40. The anomaly detection image represents the contour of the anomaly detected in the inspection target image by the anomaly detection unit 12. The contour of the anomaly may be displayed superimposed on the inspection target image. The feature quantities of the anomaly include the number of detected anomalies, the maximum detected area (number of pixels) representing the area of ​​the largest detected anomaly, and the brightness histogram within the area corresponding to the anomaly. In cases where multiple anomalies are detected, the brightness histogram corresponding to the anomaly specified by the inspector or the anomaly with the largest area may be displayed. The decision tree flow represents the judgment process by the pass / fail judgment unit 22. That is, the decision tree flow shows which conditional expressions were met and which were not met in the judgment by the pass / fail judgment unit 22. The normal image registration button is used to save the inspection target image along with information indicating that the inspection target object 50 is a good product when the inspection target object 50 is a good product. The abnormal image registration button is used to save the image of the inspected object 50 along with information indicating that the inspected object is defective, when the inspected object 50 is found to be defective. The decision tree update button is used to update the decision tree. When the decision tree update button is selected, the visual inspection device 1 may allow the user to change the content of the condition expression or the threshold value in the condition expression on the confirmation screen shown in Figure 8.

[0059] As described above, the decision tree flow shows the path taken by the inspection target 50 to arrive at the judgment of "good product" or "defective product" by the quality judgment unit 22. In addition, in the brightness histogram, the part related to the conditional expression in the decision tree may be highlighted.

[0060] If the inspector performing a visual inspection finds that the judgment result from the pass / fail judgment unit 22 differs from their own visual inspection result, they can refer to the confirmation screen described above to recognize or infer the reason or cause for the difference between the two judgment results. Therefore, fluctuations in the judgment of visual inspections can be suppressed.

[0061] <Hardware Configuration> Figure 9 shows an example of the hardware configuration of the visual inspection device 1. The visual inspection device 1 is implemented by a computer 100 which includes a processor 101, memory 102, storage device 103, input / output device 104, recording medium reader 105, and communication interface 106.

[0062] The processor 101 executes the visual inspection support program stored in the storage device 103. The execution of the visual inspection support program by the processor 101 provides the functions of the anomaly detection unit 12, feature quantity calculation unit 13, defective product candidate extraction unit 14, feature quantity calculation unit 21, pass / fail judgment unit 22, visual inspection support unit 23, and display unit 24 shown in Figure 1. Memory 102 is used as the working area for the processor 101. The storage device 103 stores the visual inspection support program and other programs. The inspection result database 30 may also be implemented by the storage device 103.

[0063] The input / output device 104 may include input devices such as a keyboard, mouse, touch panel, and microphone. The input / output device 104 may also include output devices such as a display device and speaker. The recording medium reader 105 can acquire data and information recorded on the recording medium 110. The recording medium 110 is a removable recording medium that can be attached to and detached from the computer 100. The recording medium 110 can be implemented, for example, by semiconductor memory, a medium that records signals by optical action, or a medium that records signals by magnetic action. The visual inspection support program may be provided from the recording medium 110 to the computer 100. The communication interface 106 provides the function of connecting to a network. When the visual inspection support program is stored on the program server 120, the computer 100 may acquire the visual inspection support program from the program server 120. [Explanation of Symbols]

[0064] 1. Visual inspection device 10. Inspection Department 11 Image acquisition unit 12 Anomaly detection unit 13 Feature Calculation Unit 14 Defective product candidate extraction section 20. Inspection Support Department 21 Feature Calculation Unit 22. Quality Determination Unit 23 Visual Inspection Support Department 24 Display 30. Inspection Result Database 40 Cameras 50 Items to be inspected

Claims

1. An anomaly detection unit that detects abnormal areas in an image of the object to be inspected, which is generated by photographing the object to be inspected, A feature calculation unit that calculates feature quantities related to the aforementioned abnormal location, A quality determination unit that determines whether the object to be inspected is good or defective by applying the feature quantity to a determination flow in which one or more conditions corresponding to the feature quantity are set, A visual inspection support unit compares a first judgment result obtained by the quality determination unit with a second judgment result obtained in a visual inspection of the object to be inspected, which indicates whether the object to be inspected is a good product or a defective product. When the first determination result and the second determination result are different from each other, a display unit displays the determination flow in which the one or more conditions are set, A visual inspection device equipped with the following features.

2. The aforementioned feature quantity includes a luminance histogram representing the distribution of luminance within the abnormality detection region where the abnormality is detected in the image to be inspected. At least one of the one or more conditions described above is expressed using the number of pixels having a predetermined brightness value within the anomaly detection area. The quality determination unit determines whether the object being inspected is good or defective by inputting the luminance histogram into the determination flow. The visual inspection apparatus according to feature 1.

3. The display unit displays the judgment flow and the brightness histogram. The visual inspection apparatus according to feature 2.

4. The judgment flow displayed by the display unit indicates whether each of the one or more conditions used in the judgment by the pass / fail judgment unit has been met. The visual inspection apparatus according to feature 1.

5. The system further includes a defective product candidate extraction unit that extracts defective product candidates from among multiple inspected objects using stricter criteria than the aforementioned good / bad judgment unit. For the object to be inspected that has been extracted as a candidate for a defective product by the defective product candidate extraction unit, the quality determination unit determines whether the object to be inspected is a good product or a defective product. The visual inspection support unit compares the first determination result obtained by the quality determination unit with the second determination result obtained in the visual inspection for the object to be inspected. The display unit displays the determination flow when the first determination result and the second determination result for the object to be inspected are different from each other. The visual inspection apparatus according to feature 1.

6. Using a computer, A detection procedure for detecting abnormal areas in an image of an object to be inspected, which is generated by photographing the object to be inspected, A calculation procedure for calculating feature quantities related to the aforementioned abnormal location, A determination procedure for determining whether the object to be inspected is a good product or a defective product by applying the aforementioned feature quantity to a determination flow in which one or more conditions corresponding to the aforementioned feature quantity are set, A comparison procedure for comparing a first determination result obtained by the above determination procedure with a second determination result obtained by visual inspection of the object to be inspected, which indicates whether the object to be inspected is a good product or a defective product. When the first determination result and the second determination result are different from each other, a display procedure for displaying the determination flow in which the one or more conditions are set, A visual inspection method that performs this task.

Images