JP5151019B2 - Wrinkle detection device and wrinkle detection method - Google Patents

Description

  The present invention relates to a wrinkle detection device, a wrinkle detection method, a computer program, and a recording medium, and more particularly, to a technique suitable for use in wrinkle detection that detects harmful wrinkles in a finished product at an early stage with high accuracy.

  The steel plate production line includes, for example, a hot rolling process 1, a pickling process 2, a cold rolling process 3, an annealing process 4, a surface treatment process 5, an inspection process 6 and the like as shown in FIG. In many of these steps, a wrinkle detecting device is provided. The wrinkle detection device is a device that identifies and detects wrinkles by collecting an image of the surface of an object to be inspected with an image capturing device such as a CCD camera and processing the image. The role of the wrinkle detection device is to collect basic data for quality control and operation or process management of the steel sheet in the hot rolling process 1, the pickling process 2, and the cold rolling process 3. In other words, in these processes, this wrinkle detection device is used to ensure that wrinkles are inspected, find harmful flaws at the customer's site, and prevent defective products from flowing into the lower processes or change production plans. There is a demand to take. In the lower process, it is also important to determine the defective coil part for quality assurance to the customer, to guide the defective part to the inspector, and to collect data for quality control.

  In the wrinkle inspection in the steel plate production line, it is widely performed to use individual wrinkle detection devices installed in each process independently. In addition, for example, as described in Patent Document 1 and Patent Document 2, by displaying the wrinkle inspection images of a plurality of processes side by side for the same part, it is possible to grasp the history of wrinkles from the upper process. There is a technique that facilitates consistent quality control of steel sheets up to the lower process.

JP 2000-28547 A JP 2003-329600 A

  By the way, the soot detecting device needs to pick up soot from noises such as dust, water droplets, oil stains, etc. from the image of the surface of the object to be inspected, such as a steel plate, and to identify the detected soot as harmful or harmless. Each wrinkle detection device provided in each process performs image processing that facilitates wrinkle detection, and then identifies harmful wrinkles using discrimination logic. The target level of soot detection is to detect harmful soot without omission (detection sensitivity) and to eliminate harmless soot and the like as harmful soot (overdetection) (detection accuracy). However, in the prior art, the detection accuracy is not at a sufficient level, and an inspection by an inspector is often performed in parallel with the wrinkle detection device.

  In order to improve the detection sensitivity and detection accuracy of the wrinkle detection device installed in a certain process, images that are correctly named separately on the same inspected object are accumulated as learning data, and the image processing and discrimination logic of the detection device is stored. It is effective to improve.

  In the upper processes such as hot rolling process 1 and pickling process 2, in order to collect learning data, the inspector detects flaws that are harmful at the customer by slowing the plate speed during operation, or It is necessary for the inspector to visually inspect the surface defects on the steel sheet through the inspection line. Further, the wrinkle disappears in the cold rolling process 3, or the wrinkle disappears by plating, and the customer It has been difficult for an inspector to determine a flaw that is harmful to the former using the steel plate in the upper process, and it has been difficult to accumulate sufficient learning data.

  By the way, in the upper process such as the hot rolling process 1 and the pickling process 2, the defect detection device is used to reliably perform defect inspection, find harmful defects at the customer, and do not let defective products flow to the lower process. Or there is a demand to take early measures such as changing the production plan.

  In view of the above-described problems, the present invention provides a wrinkle detection apparatus and wrinkle detection method for accurately detecting wrinkles that are finally harmful in a finished product using appropriate learning data for wrinkle detection in the upper process. The purpose is to provide.

The wrinkle detection device of the present invention is a wrinkle detection device that performs wrinkle inspection using an image of a steel sheet surface in an upper step of a steel plate production process, and images the upper surface of the steel plate surface to obtain upper process wrinkle image data. a step flaw image detection means onto which detects the on the steps flaw image detection means step flaw image data after being detected by the lower step flaw is image data of a harmful flaw detected in the lower step of said steel sheet production processes Comparing means for comparing image data, 疵 image extracting means for extracting harmful 疵 image data from the upper process 疵 image data based on the result of comparison by the comparing means, and 疵 image extracting means means for storing the extracted harmful flaw image data as learning data, so as to match the defect coordinate position in the steel plate of the lower step, to correct the coordinate position of the upper step flaw image data And a correction means, said comparing means, and the coordinate position of the lower step flaw image data, compared by matching the coordinate position of the step flaw image data after being corrected by said correction means, the flaw image extraction The means detects the eyelid image data that matches the coordinate position of the lower process eyelid image data, and the comparing means further includes a feature amount indicating the feature of the lower process eyelid image data and the detected upper process eyelid. The wrinkle image extracting means compares the wrinkle image data having a feature amount similar to the wrinkle feature amount of the lower process 疵 image data with respect to the upper process 疵 image data. It is further characterized in that it is further detected from the image data and the learning data accumulated by the accumulating means is used for determining the wrinkle of a new steel sheet.

The wrinkle detection method of the present invention is a wrinkle detection method in which wrinkle inspection is performed using an image of a steel sheet surface in an upper step of a steel plate production process, and the upper step wrinkle image data is obtained by imaging the upper steel plate surface. Upper process 上 image detection process to detect, upper process 疵 image data detected by the upper process 疵 image detection process, and lower process あ る which is image data of harmful flaws detected in the lower process of the steel plate production process A comparison step of comparing image data, a haze image extraction step of extracting harmful haze image data from the upper step haze image data based on a result of comparison in the comparison step, and a haze image extraction step The storage process of storing the extracted harmful soot image data as learning data, and the coordinate position of the upper process soot image data is corrected so as to coincide with the soot coordinate position in the steel plate of the lower process And a repairing step, the comparing step includes a coordinate position of the lower step flaw image data, compared by matching the coordinate position of the step flaw image data after being corrected by the correction step, the defect image extraction The step detects haze image data that matches the coordinate position of the lower step hail image data, and the comparison step further includes a feature quantity indicating a haze feature of the lower step hail image data and the detected upper step haze. The wrinkle image extracting step compares the wrinkle image data having a feature amount similar to the wrinkle feature amount of the lower step wrinkle image data. Further detection is performed from image data, and the learning data accumulated in the accumulation step is used for determining the wrinkle of a new steel sheet.

  A computer program according to the present invention causes a computer to execute each step of the method described above.

  The recording medium of the present invention records the computer program described above.

  According to the present invention, the eyelid image data captured by the eyelid inspection in the lower process with high determination accuracy is compared with the eyelid image data imaged by the eyelid inspection in the upper process, and finally harmful eyelid image data. Is stored in the upper process as learning data, so that sufficient learning data can be stored in the wrinkle detection device in the upper process. Thereby, while the precision of the wrinkle inspection in an upper process can be improved markedly, the wrinkle determination precision of a steel plate can be improved. In addition, when a flaw that is harmful in the upper process is detected, it is possible to cope with the problem early, such as preventing a defective product from flowing into the lower process or changing the production plan.

(First embodiment)
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram showing an outline of a steel plate (thin plate) production process and an example of classification of wrinkle detection devices.
In FIG. 1, the steel sheet hot-rolled in the hot-rolling step 1 is pickled in the pickling step 2 in order to remove oxides generated by the adsorption of oxygen on the surface. To be extended. Next, the steel sheet is shipped as a product through the inspection process 6 after being subjected to continuous annealing in the annealing process 4 or surface treatment in the surface treatment process 5 depending on the product.

  In FIG. 1, the flaw inspection (hereinafter abbreviated as flaw inspection) is performed in a hot rolling process 1, a pickling process 2, a cold rolling process 3, an annealing process 4, a surface treatment process 5, and an inspection. In step 6, generally, the inspection of the upper process refers to the inspection upstream of the lower process. In the example of FIG. 1, the inspection in the upper process includes the hot rolling process 1 and the pickling process. 2, cold rolling process 3, and annealing process 4, inspection in the lower process refers to surface treatment process 5 and inspection process 6.

  FIG. 2 is a diagram illustrating an example of a schematic configuration of the wrinkle detection device according to the present embodiment, and FIG. 3 is a block diagram illustrating an example of a functional configuration of the wrinkle detection device according to the present embodiment.

  2 and 3, the wrinkle detection device 11 according to the present embodiment includes an illumination processing unit 13 that performs a process for irradiating the steel plate 12, and reflected light obtained by irradiating the steel plate 12 with the illumination processing unit 13. And a captured image processing unit 14 for processing the captured image.

  The illumination processing unit 13 includes a light emitting unit that emits light having high luminance, such as a halogen lamp and a laser light generator, and a condensing unit such as a condensing lens, and generates the halogen lamp and the laser light. The light emitted from the apparatus is linearly collected by the condenser lens, so that the surface of the steel plate 12 is illuminated linearly in the width direction.

  On the other hand, the captured image processing unit 14 includes a movement status detection unit 15, an imaging control unit 16, an imaging unit 17, an image processing unit 18, a monitor 24, and a wrinkle detection unit 21.

  The movement status detection means 15 has a function of detecting the movement status of the steel sheet 12 moving in the rolling line direction (Y direction in FIG. 2) and generating a movement status detection signal.

  The imaging control means 16 has a function of determining the start and end of the wrinkle inspection of the steel sheet 12 based on the movement status detection signal and controlling the operations of the illumination processing unit 13 and the imaging means 17 according to the determination result.

  The imaging means 17 is, for example, a CCD camera, and as shown in FIG. 2, a steel plate disposed on the optical path of light irradiated from the illumination processing unit 13 and reflected by the steel plate 12 and illuminated by the illumination processing unit 13. It has a function of imaging 12 eyelid inspection ranges and generating an image signal.

  The image processing unit 18 takes in the image signal generated by the imaging unit 17 and performs predetermined image processing, for example, generates image data to be displayed on the monitor 24, and stores the image data in the storage unit 19 and the eyelid detection unit 21. It has a function of outputting image data.

  The storage unit 19 includes image data (upper image data) obtained by imaging by the imaging unit 17, and image data of a steel plate transmitted from another defect detection device (lower process defect detection device) via the LAN 23. (Image data of the lower process (including the detection result of the wrinkle and the wrinkle image data)).

  The coordinate correction means 20 is different in the length of the steel plate 12 when the inspection is performed in the upper process and the length of the steel plate 12 when the inspection is performed in the lower process. The rear end or the front and back may be reversed, and the coordinate position is corrected to match the length of the steel sheet after rolling (lower process) in order to correspond the positional relationship of the wrinkles. In the present embodiment, it is necessary to compare the image data of the upper process with the image data of the lower process, and to extract the image data of the upper process that is harmful at the customer.

  The wrinkle detection means 21 refers to the learning data stored in the database 22 and detects wrinkles that are harmful at the customer from the image data of the upper process. Then, the wrinkle detection result and the wrinkle image data are displayed on the monitor 24, and the wrinkle detection result and the wrinkle image data are stored in the storage unit 19.

  Further, in the present embodiment, the upper-process image data corresponding to the lower-process image data that is harmful at the customer from the upper-process image data and the lower-process image data stored in the storage unit 19. Are automatically extracted and stored in the database 22 as learning data. For the image data in the lower process, the image data stored in the removable external storage medium may be read out instead of reading out the image data stored in the storage unit 19 via the LAN 23.

  FIG. 4 is a diagram showing an example of wrinkle distribution and wrinkle images in the upper process and the lower process in the present embodiment. The wrinkle detection means 21 compares the image data of the upper process corrected by the coordinate correction means 20 with the image data of the lower process stored in the storage means 19, and the distribution map and the wrinkle image data are stored on the monitor 24. It can be displayed.

  As shown in FIG. 4, the wrinkle distribution 41 in the lower process and the wrinkle distribution 42 in the upper process are displayed, and the vertical axis indicates the position from the front end to the end of the steel sheet after the coordinate correction is performed, The axis | shaft has shown the position of the horizontal direction (width direction) of a steel plate. When the area 45 in which the position of the eyelid matches in the upper process and the lower process is extracted, the eyelid image data 43 in the area 45 in the upper process and the eyelid image data 44 in the area 45 in the lower process are displayed. As described above, in the present embodiment, the upper process eyelid image data corresponding to the lower process eyelid image data is automatically extracted.

  As a result, there are cases where the wrinkles detected in the upper process are harmless flaws in the lower process and cases in which the wrinkles are harmful flaws. It is possible to know if it will be a harmful trap in the process. That is, it becomes possible to determine whether it is harmful or harmless to the customer from the image data generated in the upper process stage, and finally accumulate the image data that is harmful to the customer as learning data To be able to.

  On the other hand, when the area 45 is designated, the soot image data in the area 45 is displayed. However, when a large amount of soot image data is displayed, the harmful soot image data is individually stored as learning data. Or delete the harmless 疵 image data individually from the learning data. Therefore, in the present embodiment, a feature amount range is specified in order to determine whether detected wrinkles are harmful or harmless, and the corresponding wrinkle image data is automatically stored or deleted as learning data. Can be done. Specific examples are shown in Table 1.

  In Table 1, data of features 1 to 3 are recorded. For example, if feature 1 is the length of the eyelid, feature 2 is the area of the eyelid, feature 3 is the color of the eyelid (luminance), and the feature amount (value of x) in the length of the eyelid, the area of the eyelid and the brightness of the eyelid Is set. Whether it is harmful or harmless is determined according to the conditions of features 1 to 3, respectively. A specific determination method will be described later with reference to the flowchart of FIG.

Next, operation | movement of the wrinkle detection apparatus 11 of this Embodiment is demonstrated using the flowchart of FIGS.
FIG. 5 is a flowchart showing an example of the procedure of the screening in the present embodiment.

  In the first step S1, the imaging control means 16 waits until the steel plate 12 is detected based on the movement status detection signal output from the movement status detection means 15. When the steel plate 12 is detected, the process proceeds to step S2, and illumination is performed. The halogen lamp in the processing unit 13 is caused to emit light, and the CCD camera that is the imaging means 17 is operated.

Next, in step S3, the illumination processing unit 13 emits light to illuminate the steel plate 12.
Next, in step S4, the steel plate 12 moving in the rolling line direction (Y direction in FIG. 2) is imaged using the imaging means 17.

  Next, in step S5, the image processing unit 18 stands by until an image signal is input from the imaging unit 17, and when the image signal is input, the process proceeds to step S6, where the image data is generated, and the storage unit 19 and the eyelid detection are detected. Output to the means 21 and output to the monitor 24. Based on this image data, an actual image of the steel plate 12 is displayed on the monitor 24 in real time.

  Next, in step S <b> 7, the wrinkle detection means 21 refers to the learning data stored in the database 22, detects wrinkles from the image data output from the image processing means 18, and obtains the detection result and the wrinkle image data. Displayed on the monitor 24.

  Next, in step S <b> 8, the imaging control unit 16 determines whether the wrinkle inspection of the steel sheet 12 has been completed based on the movement status detection signal detected by the movement status detection unit 15.

  As a result of the determination, if the eyelid inspection is completed, the process proceeds to step S9, and the imaging control means 16 stops the light emission of the halogen lamp and the operation of the CCD camera. On the other hand, as a result of the determination in step S8, if the eyelid inspection is not completed, the process returns to step S5, and the processes from step S5 to step S7 are repeated until the eyelid inspection is completed.

FIG. 6 is a flowchart showing an example of a procedure for accumulating learning data in the present embodiment.
First, in step S <b> 21, the storage unit 19 acquires image data of the steel plate 12 (upper image data) from the image processing unit 18.

  Next, in step S <b> 22, the storage unit 19 obtains image data of the steel plate 12 of the inspection from another flaw detection device (lower step flaw detection device) via the LAN 23.

  Next, in step S23, the coordinate correction means 20 reads out the image data of the desired upper process from the storage means 19 in which the flaw detection results of the coils that have been passed through in the past are stored, and makes the positional relationship of the eyelids correspond. The coordinate position of the image data of the upper process is corrected so as to match the size of the steel plate 12 after rolling (lower process).

  Next, in step S24, the eyelid detection means 21 compares the eyelid image data of the upper process whose coordinate position has been corrected by the coordinate correction means 20 with the eyelid image data of the lower process stored in the storage means 19. . Then, it is determined whether the soot detected from the image data in the upper process is harmful at the customer.

  If the result of this determination is harmless, the process proceeds to step S25, and the eyelid detection means 21 deletes the corresponding upper process eyelid image data. If the result of determination in step S24 is harmful, the process proceeds to step S26, and the eyelid detection means 21 stores the corresponding upper process eyelid image data in the database 22 as learning data. Note that it is also possible to separately store the cocoon image determined to be harmless.

  Next, in step S27, it is determined whether or not the determination has been completed for all the wrinkles detected from the image data in the upper process. If the result of this determination is that it is finished, the processing is finished. If not completed, the process returns to step S24, and the subsequent processing is repeated until the determination for all the bags is completed.

  FIG. 7 is a flowchart showing an example of a procedure for determining whether it is harmful or harmless in the process of step S24 of FIG. In this flowchart, an example of determining whether it is harmful or harmless based on the feature amount range shown in Table 1 is shown.

  First, in step S31, the wrinkle detection means 21 includes wrinkles detected from the lower process image data corresponding to the wrinkles detected from the upper process image data (including those in which the wrinkles disappeared in the lower process). When feature amount x is set, x> 0.3 for feature 1 (the length of the heel), x> 0.3 for feature 2 (the area of the heel), and about feature 3 (the brightness of the heel) It is determined whether or not the condition x> 0.5 is satisfied.

  If the condition is satisfied as a result of the determination, the process proceeds to step S36, and the wrinkle detection means 21 determines that the wrinkle detected from the image data in the upper process is a harmful wrinkle. If the result of determination in step S31 does not satisfy the condition, the process proceeds to step S32.

  Next, in step S32, the wrinkle detection means 21 detects wrinkles detected from the image data of the lower process corresponding to the wrinkles detected from the image data of the upper process (those in which wrinkles disappeared in the lower process). Whether or not the condition 2 satisfies the condition x> 0.5 for the feature 2 (the area of the eyelids) and x> 0.3 for the feature 3 (the brightness of the eyelids) when the feature amount x is set. Determine.

  If the condition is satisfied as a result of the determination, the process proceeds to step S35, and the wrinkle detection means 21 determines that the wrinkle detected from the image data in the upper process is a harmless wrinkle. If the result of determination in step S32 does not satisfy the condition, the process proceeds to step S33.

  Next, in step S33, the wrinkle detecting means 21 detects wrinkles detected from the image data of the lower process corresponding to the wrinkles detected from the image data of the upper process (those in which wrinkles disappeared in the lower process). However, when the feature amount x is set, whether or not the condition of feature 1 (the length of the eyelid) x> 0.5 and the feature 3 (the brightness of the eyelid) satisfies the condition x> 0.3. Determine whether.

  If the condition is satisfied as a result of the determination, the process proceeds to step S35, and the wrinkle detection means 21 determines that the wrinkle detected from the image data in the upper process is a harmless wrinkle. If the result of determination in step S33 does not satisfy the condition, the process proceeds to step S34.

  Next, in step S34, the wrinkle detection means 21 detects wrinkles detected from the lower-step image data corresponding to the wrinkles detected from the upper-step image data (the wrinkles disappeared in the lower step). However, when the feature amount x is set, x> 0.3 for feature 1 (the length of the eyelid), x> 0.3 for feature 2 (the area of the eyelid), and feature 3 (the brightness of the eyelid) ) Is determined whether or not the condition x> 0.05 is satisfied.

  If the condition is satisfied as a result of the determination, the process proceeds to step S36, and the wrinkle detection means 21 determines that the wrinkle detected from the image data in the upper process is a harmful wrinkle. If the result of determination in step S34 does not satisfy the condition, the process proceeds to step S35, and the wrinkle detection means 21 determines that the wrinkle detected from the image data in the upper process is a harmless wrinkle.

  As described above, in the present embodiment, the eyelid image data captured in the lower process with high determination accuracy is compared with the eyelid image data captured in the upper process, and finally, Since the image data that becomes harmful is stored as image data (learning data) that is harmful at the customer in the upper process, the upper process defect detection device does not have to learn with ambiguous image data. Learning data can be accumulated sufficiently, the accuracy of inspection in the upper process can be remarkably improved, and the accuracy of determination of defects on the steel sheet can be greatly improved. As a result, in the case where wrinkles that are harmful in the upper process are detected, it is possible to respond quickly such as preventing defective products from flowing into the lower process or changing the production plan.

(Second Embodiment)
In the first embodiment, it is determined whether it is harmful or harmless by designating a feature amount range as shown in Table 1, but in this embodiment, a lower process is performed from within the extracted range. An example will be described in which a wrinkle that has a similar feature amount to a wrinkle that is harmful in the case (a wrinkle detected in the above process) is determined as a harmful wrinkle, and the corresponding wrinkle image data is accumulated as learning data. 1 to 6 are the same in the present embodiment, and therefore, the description of the overlapping parts is omitted.

FIG. 8 is a diagram illustrating an example of the relationship between the feature amounts of the wrinkles detected in the upper process and the wrinkles detected in the lower process in the present embodiment.
When the area 45 of the distribution map shown in FIG. 4 is designated, the feature amount of the wrinkle corresponding to the area 45 is calculated, and a wrinkle distribution map is generated as shown in FIG. The mark ● indicates the distribution of wrinkles detected in the upper process, and the mark ■ indicates the feature quantity distribution of the wrinkle database detected in the lower process. Note that the feature values (such as the length of the wrinkles) of the wrinkle data in the upper process were corrected to match the length of the steel sheet after rolling (lower process) in order to match the positional relationship of the wrinkles. Equivalent to the feature value of the heel data.

  In the present embodiment, a distribution diagram is shown in which the horizontal axis is the feature amount of feature 1 (the length of the heel) and the vertical axis is the feature amount of feature 2 (the heel area). In this case, since the wrinkles detected in the upper process, which are close to the distribution of wrinkles detected in the lower process, can be considered similar, it can be determined as a harmful wrinkle. On the other hand, since the wrinkles detected in the upper process that are distributed at positions away from the distribution of wrinkles detected in the lower process can be considered to be dissimilar, it can be determined as harmless wrinkles.

  As described above, it is possible to determine a wrinkle similar to the wrinkle detected in the lower process as a harmful wrinkle, and to store the corresponding wrinkle image data as learning data. Further, in the present embodiment, the comparison with the wrinkles detected in the lower process is performed, but when the similar determination is difficult, it is based on the harmful wrinkle image data of the lower process accumulated as learning data in the database. By supplementing the comparison with the feature amount, it can be accumulated as learning data more precisely.

(Third embodiment)
In the first and second embodiments, comparison is made with the lower-stage soot image data, but in any case, it is assumed that the learning data of the lower-process soot detection device is sufficiently accumulated. However, in some cases, the determination of wrinkles may not be sufficient. Therefore, in the present embodiment, an example will be described in which similarity is separately determined based on a feature amount in addition to collation with the eyelid image data in the lower process. Note that FIGS. 1 to 6 are the same in the present embodiment, and therefore, the description of the overlapping portions is omitted.

FIG. 9 is a diagram illustrating an example of dispersion of feature amounts in the cocoon detected in the upper process in the present embodiment.
In FIG. 9, the 疵 deviating from the central value of the feature value (dispersion value is 2σ or more, etc.) is regarded as dissimilar 疵. A determination may be made. In the present embodiment, the feature amount of feature 1 (the length of the heel) has been described. Of course, similar or dissimilar determination may be performed using other features (for example, the area of the heel). Good. Moreover, the value based on statistical processing, such as Mahalanobis distance, may be used for the determination of similarity between features.

  As described above, in the present embodiment, similarity can be determined from the cocoon detected in the upper process, and accumulated in the learning data as significant data when determining the type, grade, etc. of the cocoon Will be able to.

(Another embodiment according to the present invention)
Software program for realizing the functions of the above-described embodiments for an apparatus or a computer in the system connected to the various devices so that the various devices are operated to realize the functions of the above-described embodiments. What was implemented by supplying the code and operating the various devices in accordance with a program stored in a computer (CPU or MPU) of the system or apparatus is also included in the scope of the present invention.

  In this case, the program code of the software itself realizes the functions of the above-described embodiments, and the program code itself and means for supplying the program code to the computer, for example, the program code The stored recording medium constitutes the present invention. As a recording medium for storing the program code, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

  Further, by executing the program code supplied by the computer, not only the functions of the above-described embodiments are realized, but also the OS (operating system) or other application software in which the program code is running on the computer. It goes without saying that the program code is also included in the embodiment of the present invention even when the functions of the above-described embodiment are realized in cooperation with the above.

  Further, after the supplied program code is stored in the memory provided in the function expansion board of the computer or the function expansion unit connected to the computer, the CPU provided in the function expansion board or function expansion unit based on the instruction of the program code Needless to say, the present invention also includes a case where the functions of the above-described embodiment are realized by performing part or all of the actual processing.

It is a figure which shows an example of the outline of a steel plate production process, and the classification | category of a wrinkle detection apparatus. It is a figure which shows an example of schematic structure of the wrinkle detection apparatus in the 1st Embodiment of this invention. It is a block diagram which shows an example of a function structure of the wrinkle detection apparatus in the 1st Embodiment of this invention. In 1st Embodiment of this invention, it is a figure which shows an example of the distribution of wrinkles in the upper process and a lower process, and a wrinkle image. It is a flowchart which shows an example of the procedure of the censorship in the 1st Embodiment of this invention. 5 is a flowchart illustrating an example of a procedure for accumulating learning data in the first embodiment of the present invention. It is a flowchart which shows an example of the procedure which determines whether it is harmful or harmless in the process of step S24 of FIG. In the 2nd Embodiment of this invention, it is a figure which shows an example of the relationship of the feature-value between the wrinkles detected by the upper process, and the wrinkles detected by the lower process. In the 3rd Embodiment of this invention, it is a figure which shows an example of dispersion | distribution of the feature-value in the soot detected at the upper process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Hot-rolling process 2 Pickling process 3 Cold-rolling process 4 Annealing process 5 Surface treatment process 6 Inspection process 11 Wrinkle detection apparatus 12 Steel plate 13 Illumination processing part 14 Captured image processing part 15 Moving condition detection means 16 Imaging control means 17 Imaging means 18 Image processing means 19 Storage means 20 Coordinate correction means 21 Haze detection means 22 Database 23 LAN
24 Monitor

Claims (10)

In the upper step of the steel plate production process, a wrinkle detection device that performs a wrinkle inspection using an image of the steel plate surface,
Upper process haze image detection means for detecting the upper process haze image data by imaging the steel sheet surface of the upper process,
Comparison means for comparing the upper process soot image data detected by the upper process soot image detecting means with the lower process soot image data that is image data of harmful flaws detected in the lower process of the steel sheet production process,
疵 image extraction means for extracting harmful moth image data from the upper process 疵 image data based on the result of comparison by the comparison means;
Accumulating means for accumulating harmful habit image data extracted by the habit image extracting means as learning data;
Correction means for correcting the coordinate position of the upper process heel image data so as to match the heel coordinate position in the steel sheet of the lower process,
The comparison means matches the coordinate position of the lower process 疵 image data with the coordinate position of the upper process 疵 image data corrected by the correction means,
The haze image extraction means detects haze image data that matches the coordinate position of the lower step haze image data,
The comparing means further compares the feature amount indicating the feature of the wrinkle of the lower process soot image data with the feature amount indicating the feature of the detected upper process soot image data,
The wrinkle image extracting means further detects the wrinkle haze image data having a feature amount similar to the wrinkle feature amount of the lower process hail image data from the upper process hail image data,
A wrinkle detection device using the learning data accumulated by the accumulating means for flaw determination of a new steel plate.   The hail image extracting means detects hail haze image data having a feature quantity similar to the haze feature quantity of the lower process hail image data stored as learning data in the database from the upper process haze image data. The wrinkle detection device according to claim 1. Furthermore, an extraction means for extracting a wrinkle having a feature value belonging to a predetermined range from the feature value of the wrinkle detected by the upper process wrinkle image detection means,
The accumulation means accumulates, as learning data, the haze image data extracted by the extraction means in addition to the harmful haze image data extracted by the haze image extraction means. 2. The wrinkle detection device according to 2.   The wrinkle detection apparatus according to claim 3, wherein the extraction unit extracts a wrinkle having a variance value of the feature amount smaller than a predetermined value. In the upper process of the steel sheet production process, a wrinkle detection method for performing a wrinkle inspection using an image of the steel sheet surface,
An upper process 疵 image detection step of detecting the upper process 疵 image data by imaging the steel plate surface of the upper process,
A comparison process for comparing the upper process soot image data detected by the upper process soot image detection process with the lower process soot image data that is image data of harmful flaws detected in the lower process of the steel sheet production process,
Based on the result of the comparison by the comparison step, the soot image extraction step for extracting harmful soot image data from the above step soot image data,
An accumulation step of accumulating harmful haze image data extracted by the haze image extraction step as learning data;
A correction step of correcting the coordinate position of the upper process heel image data so as to match the heel coordinate position in the steel plate of the lower process,
In the comparison step, the coordinate position of the lower process 疵 image data and the coordinate position of the upper process 疵 image data corrected by the correction process are matched and compared,
The haze image extraction step detects haze image data that matches the coordinate position of the lower step haze image data,
The comparison step further compares the feature amount indicating the feature of the wrinkle of the lower step image data with the feature amount indicating the feature of the detected upper step image data.
The wrinkle image extraction step further detects wrinkle wrinkle image data having a feature amount similar to the wrinkle feature amount of the lower step wrinkle image data from the upper step wrinkle image data,
A wrinkle detection method characterized by using learning data accumulated in the accumulation step for flaw determination of a new steel plate.   The wrinkle image extraction step detects wrinkle wrinkle image data having a feature amount similar to the wrinkle feature amount of the lower step wrinkle image data stored as learning data in the database from the upper step wrinkle image data. The wrinkle detection method according to claim 5. Furthermore, an extraction step of extracting a wrinkle having a feature amount belonging to a predetermined range from the feature amount of the wrinkle detected by the upper step wrinkle image detection step,
The said accumulation | storage process accumulate | stores the cocoon cocoon image data extracted by the said extraction process as learning data in addition to the harmful cocoon image data extracted by the said cocoon image extraction process. 6. The wrinkle detection method according to 6.   The wrinkle detection method according to claim 7, wherein the extraction step extracts a wrinkle having a variance value of the feature amount smaller than a predetermined value.   A computer program for causing a computer to execute each step of the method according to claim 5.   A computer-readable recording medium having recorded thereon the computer program according to claim 9.

Images