JP2014020926A - Surface defect inspection device and surface defect inspection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a surface defect inspection device and a surface defect inspection method which can properly determine a type of a defect on a surface of a steel plate.SOLUTION: A surface defect inspection device captures a surface of a steel plate 10 by cameras 3 and 4 and extracts a defect image corresponding to a surface defect from the captured image. Then, the device performs a normalization processing on the defect image for converting the size of the defect image into the preset reference size, performs comparison by matching the normalized defect image with surface defect template images pre-registered with a defect library, and determines a type of the surface defect corresponding to the defect image. At this point, the same processing as the normalization processing is performed on the template images.

Description

  The present invention relates to a surface defect inspection apparatus and a surface defect inspection method for inspecting a surface defect of a steel sheet.

  In a conventional surface defect inspection apparatus, a feature amount of a defect is extracted from a two-dimensional defect image obtained by photographing a defect on the surface of a steel sheet with a camera or the like, and the type and degree of the defect are inspected using the feature amount. The feature amount is information obtained by digitizing the feature of the defect, and includes, for example, width, length, area, inclination, position information, contour (peripheral length, etc.), density (luminance) distribution, etc. (for example, patents). References 1 and 2).

JP 2002-195952 A Japanese Patent No. 3194996

By the way, for the defect determination, tree-like logic that is performed by repeatedly comparing the feature amount with a predetermined threshold value is widely adopted. In this case, if any one of the feature amounts is invalid, an unexpected route is used. Tracing will lead to unjustified detection results.
However, the defects generated on the surface of the steel sheet include complicated ones that cannot be expressed by simple features such as length and width. In addition, even the defect of the same generation source may change in size depending on the generation time and the generation location. That is, even if the defect is the same, the feature information is not fixed.

Therefore, if the determination method is fixed, the same defect may be erroneously determined as another defect or may not be detected as a defect.
Then, this invention makes it the subject to provide the surface defect inspection apparatus and surface defect inspection method which can discriminate | determine the surface defect of a steel plate appropriately.

  In order to solve the above problems, one aspect of the surface defect inspection apparatus according to the present invention is a surface defect inspection apparatus that detects a surface defect of a steel sheet, and includes a photographing unit that photographs a surface image of the steel sheet, and the photographing A defect image extracting means for extracting a defect image corresponding to the surface defect from the surface image photographed by the means, and the defect image extracted by the defect image extracting means is converted to a preset reference size. By comparing the defect image obtained by the image conversion means, the defect image obtained by the image conversion means, and the template image of the surface defect subjected to the same image conversion processing as the image conversion means by matching, the defect And a defect specifying means for specifying the type of surface defect corresponding to the image.

In this way, since the size of the defect image is normalized and compared with the template image, even if the defect of the same source has different sizes due to the influence of the occurrence time, occurrence location, etc., the defect type is appropriately Can be specified. In addition, the comparison between the defect image and the template image is performed by matching, and the matching of the defect is determined based on the size of the matching error when the two images are overlapped. Therefore, it is not possible to express the feature quantity such as simple length and width clearly. Even in the case of a simple defect, the defect can be determined appropriately.
In the above, the image conversion means performs a process of smoothing the luminance of the pixels on the defect image extracted by the defect image extraction means.

Thereby, even when the brightness of the defect gradually changes from dark to bright, even when the size of the defect cannot be clearly expressed as a numerical value, the defect determination can be performed appropriately.
Further, in the above, the image conversion means performs an averaging process on the defect image extracted by the defect image extraction means.
Thereby, since a defect image can be blurred and the number of pixels can be reduced, the dispersion | variation at the time of comparing a brightness | luminance per pixel can be suppressed.

Further, in the above, by comparing the feature point extracting means for extracting the feature point of the defect image extracted by the defect extracting means with the feature point threshold set in advance by the feature point extracted by the feature point extracting means, Characteristic point comparing means for determining the type and degree of the surface defect corresponding to the defect image.
As described above, in addition to the defect determination by matching the defect image with the template image, the defect determination using the feature point of the defect image enables more appropriate defect determination.

Furthermore, one aspect of the surface defect inspection method according to the present invention is a surface defect inspection method for detecting a surface defect of a steel plate, wherein a surface image of the steel plate is photographed, and the surface defect is taken from the photographed surface image. The corresponding defect image is extracted, and the extracted defect image is subjected to processing for converting the image size to a preset reference size, and then the defect image and the surface defect subjected to the same image conversion processing are subjected to the processing. By comparing the template image with matching, the type of surface defect corresponding to the defect image is specified.
Thereby, the defect determination of the complicated defect which cannot be expressed clearly as the feature amount can be appropriately performed.

  According to the present invention, it is possible to easily and appropriately determine a defect even for a complicated defect that cannot be clearly expressed as a feature amount or a defect whose size changes depending on the generation time. Therefore, it is possible to improve surface defect determination accuracy as compared with the case of performing only defect determination using feature quantities such as a simple length and width.

It is a figure which shows the structure of the surface defect inspection apparatus in this embodiment. It is a flowchart which shows the defect determination processing procedure performed with a signal processing part. It is a figure which shows the example of normalization size. It is a figure which shows the example normalized by the length. It is a figure which shows the example normalized by the width | variety. It is a figure which shows the example normalized by length and width. It is a figure which shows the example normalized by the brightness | luminance and the size. It is a figure which shows the flow of a defect determination. It is a figure which shows the example of a defect erroneous determination.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram showing a configuration of a surface defect inspection apparatus in the present embodiment.
In the figure, reference numeral 1 denotes a surface defect inspection apparatus for inspecting surface defects of a steel sheet 10 traveling on a process line (thin steel sheet production line). Here, the steel plate 10 is a tin steel plate, an automobile steel plate, a silicon steel plate, a stainless steel plate, or the like.
The surface defect inspection apparatus 1 includes a projector 2, two sets of line cameras (hereinafter simply referred to as cameras) 3, 4, a signal processing unit 5, and a display unit 6. Here, the projector 2 and the cameras 3 and 4 correspond to photographing means.

The projector 2 is composed of a line-type light source, and is arranged above the steel plate 10 to irradiate line-shaped light over the entire width direction of the steel plate surface. The cameras 3 and 4 are disposed above the steel plate 10 and take an image of a light irradiation position by the projector 2. The cameras 3 and 4 are arranged at different attachment positions at different attachment angles, and output captured image data to the signal processing unit 5, respectively.
The signal processing unit 5 combines the image data input from the cameras 3 and 4. In this way, the steel plate surface is photographed by two cameras 3 and 4 having different attachment positions and attachment angles, and by combining these two photographed images, the entire surface of the steel plate where the surface defects 11 are emphasized is synthesized. Two-dimensional image data can be obtained.

Here, a case has been described in which line cameras are used as the cameras 3 and 4, and two-dimensional image data is generated from a one-dimensional image signal. However, image recognition of a steel sheet surface is performed using a two-dimensional camera. Also good.
And the signal processing part 5 performs the defect determination process mentioned later, extracts the defect image corresponding to the surface defect of the steel plate 10 from the surface image of the steel plate 10, and determines the kind and grade of a defect. The determination result of the defect determination process is displayed on the display unit 6 such as a monitor.

Next, the defect determination process executed by the signal processing unit 5 will be specifically described.
FIG. 2 is a flowchart showing a defect determination processing procedure.
First, in step S1, the signal processing unit 5 acquires a surface image of the steel plate 10 taken by the cameras 3 and 4, and proceeds to step S2.
In step S2, the signal processing unit 5 performs threshold processing on the surface image acquired in step S1. In this threshold value process, the brightness change of each pixel of the surface image is compared with a preset threshold value, and if the brightness change is less than or equal to the threshold value, it is determined that the noise is noise, and the part is removed. .

Next, in step S3, the signal processing unit 5 erases a minimal defect portion having a size smaller than a preset size from the image data obtained in step S2, and proceeds to step S4.
In step S4, the signal processing unit 5 performs a proximity defect combining process on the image data obtained in step S3. Here, the defect candidate images are individually enlarged, and after these images are combined, the image is reduced. As a result, it is possible to combine the adjacent discontinuous images.

Next, in step S5, the signal processing unit 5 extracts a defect image corresponding to the surface defect of the steel plate 10 from the image data obtained in step S4, and proceeds to step S6. Here, for example, a defect image is extracted by binarization processing or the like.
In step S6, the signal processing unit 5 performs normalization processing on the defect image extracted in step S5. In this embodiment, the size and brightness of the defect image are normalized. The size normalization is a process of enlarging or reducing the defect image so as to become an image having a preset reference size (width W, length L) as shown in FIG.

  That is, as shown in FIGS. 4A and 4B, when the length of the defect image is shorter than the reference size, it is enlarged in the length direction, and the defect of the reference size shown in FIG. Try to get an image. Further, as shown in FIGS. 5A and 5B, when the width of the defect image is shorter than the reference size, the defect image having the reference size shown in FIG. To get. Further, as shown in FIGS. 6A to 6E, when both the length and width of the defect image are different from the reference size, both the length direction and the width direction are changed, and FIG. A defect image having a reference size shown in f) is obtained.

Also, the luminance normalization is a process of smoothing the brightness of each pixel of the defect image so as to have a preset reference luminance. That is, as shown in FIGS. 7A to 7D, when the length and width of the defect image are different from the reference size and the luminance is different from the reference luminance, the length and width are normalized. After that, the luminance is normalized. In this way, a defect image having a reference size and a reference luminance shown in FIG. 7E is obtained.
Here, for example, by using a 5 × 5 Gaussian filter, the luminance difference of the defect image is normalized by the average of luminance, the maximum or minimum luminance, or inclination. Note that the luminance is gray scaled from 1 to 256.

  In step S7, the signal processing unit 5 performs an averaging process on the defect image normalized in step S6. Here, in the averaging process, for example, a luminance average value is calculated with 4 × 4 pixels, and the 4 × 4 pixels are regarded as one pixel whose luminance is the luminance average value, thereby reducing the number of pixels (image Blurring) processing. The average size can be appropriately selected according to the complexity of the defect. For example, a relatively simple one is averaged by a large 8 × 8, and a complicated one is averaged by a Gaussian filter.

  In step S8, the signal processing unit 5 compares the defect image averaged in step S7 with a template image registered in advance in the defect library. In the defect library, captured images of a plurality of defects (holes, scrapes, rust, dust, etc.) are registered in advance as template images in association with defect names. The template image is subjected to image conversion processing similar to the normalization and averaging described above.

As a result of comparison between the defect image and the template image, when matching is performed (when the matching rate is equal to or higher than a predetermined value), the process proceeds to step S9. When matching is not performed (when the matching rate is lower than the predetermined value), The process proceeds to step S10 described later.
In step S9, the signal processing unit 5 acquires the defect name of the template image matched with the defect image from the defect library. Then, after the acquired defect name is displayed on the display unit 6, the process proceeds to step S11 described later. As described above, the defect name is specified by matching the defect image whose size and brightness are normalized with the template image registered in the defect library.

In step S10, the signal processing unit 5 determines whether or not all of the plurality of template images registered in the defect library have been compared with the defect images. If the comparison with all the template images has not been completed, the process proceeds to step S8 to compare the defect image with the template image that has not been compared yet. On the other hand, when the comparison with all the template images has been completed, the process proceeds to step S11.
In step S11, the signal processing unit 5 extracts the feature amount of the defect image before normalization extracted in step S5. Here, the feature amount is a numerical value of the feature of the defect. For example, the length, width, area, circumference, inclination, position, density distribution, etc. of the defect are used.

In step S12, the signal processing unit 5 performs defect determination using a so-called tree-like logic that repeatedly compares the feature amount extracted in step S11 with a predetermined feature amount threshold value. If the feature amount meets the condition determined by the feature amount threshold value, the process proceeds to step S13 to determine the type and degree of the defect, display the determination result on the display unit 6, and then terminate the defect determination process. .
If the defect name is specified in step S9, only the degree of the defect may be determined in step S12 based on the feature amount indicating the size of the defect.

On the other hand, if the feature amount does not meet the conditions determined by the feature amount threshold value in step S12, the process proceeds to step S14, where the defect is determined not to be a harmful defect, and the determination result is displayed on the display unit. 6 is displayed, the defect determination process is terminated.
In FIG. 2, steps S2 to S5 correspond to defect image extraction means, steps S6 and S7 correspond to image conversion means, and steps S8 to S10 correspond to defect identification means. Step S11 corresponds to feature point extraction means, and steps S12 to S14 correspond to feature point comparison means.

Next, the operation of the present embodiment will be described with reference to FIG.
While the steel plate 10 is traveling on the process line, the projector 2 of the surface defect inspection apparatus 1 irradiates the surface of the steel plate 10 with line-shaped light, and the two line-type cameras 3 and 4 photograph the surface of the steel plate 10. To do. Thereby, as shown to Fig.8 (a), the two-dimensional surface image by which the surface defect of the steel plate 10 was image | photographed is obtained. The surface defect inspection apparatus 1 inspects the type and degree of surface defects based on the obtained surface image of the steel plate 10.

  First, pre-processing (steps S2 to S4 in FIG. 2) is performed on the surface image, and a defect image corresponding to the surface defect portion is extracted (step S5). Next, the size and brightness of the defect image are normalized and converted to a preset reference size and reference brightness (step S6). Furthermore, since there is a large variation in the luminance comparison of one pixel unit, an averaging process is performed to reduce the number of pixels (step S7). In this way, a defect image for surface defect inspection as shown in FIG. 8B is generated.

The defect image for surface defect inspection is compared with template images of various surface defects registered in advance in the defect library. Here, a plurality of template images registered in the defect library are read one by one and matched with a defect image for surface defect inspection.
As a result of the conformity confirmation, as shown in FIG. 8, when the defect image for surface defect inspection matches the template image 1 registered in the defect library (Yes in step S8), the defect image is the same as the template image 1. It is determined that the images are images of the same defect type, and the defect name is specified (step S9). The identified defect name is displayed on the display unit 6.

By the way, as a method for performing defect determination based on a two-dimensional image obtained by photographing the surface of a steel plate, feature values such as length, width, inclination, and luminance are simply extracted from the defect image, and the extracted feature values and a predetermined threshold value are extracted. A method of determining the type and degree of defects by comparing with the above is widely used. However, when this method is adopted, results such as erroneous detection and non-detection are caused.
This is because even if the defects are generated from the same source, the size of the defects may change with the generation time, and the brightness of the reflected light of the light emitted from the projector 2 may change. In such a case, even if the defect determination is performed based on the feature amount of the defect image that is simply taken as described above, the defect of the same type is erroneously determined as a different type of defect. .

Furthermore, even if the feature is clear, the brightness may gradually change from dark to bright. In this case as well, the size of the defect cannot be clearly expressed as a numerical value. In some cases, the determination of defects using the error may result in an erroneous determination.
In addition, the shape of the surface defect is complicated and often cannot be represented by simple features such as length and width. For example, as shown in FIG. 9A, when a length and a width are extracted as feature quantities for a defect such as “circle” or “square” that cannot express the outline of the defect clearly, length = L3, width = W3. However, the feature quantity of the defect shown in FIG. 9B, which is completely different in shape and generation source (defect type) from the defect shown in FIG. 9A, similarly has length = L3 and width = W3. Therefore, if a defect is determined based on a simple feature amount such as length or width, the defect illustrated in FIG. 9A is erroneously determined to be the same as the defect illustrated in FIG. 9B.

  As the feature amount, in addition to information such as the length and width of the defect, a length item of the defect outer periphery can be used. The defect shown in FIG. 9 (a) and the defect shown in FIG. 9 (b) have completely different outer lengths, so it seems that they can be distinguished from each other by comparing the outer lengths. When the averaging process is performed, the adjacent defect and the outer periphery are averaged, so that it is generally impossible to distinguish only by the outer peripheral feature amount.

  On the other hand, in the present embodiment, a method is used in which a defect image that has been normalized so that the size and brightness become a preset reference value and a template image that has been registered in advance are sequentially overlapped and compared. To determine the defect. Therefore, it is possible to appropriately determine the defect type for a surface defect whose size and brightness change with the generation time and a surface defect that cannot be quantified.

  When the defect type is specified by matching the defect image with the template image, next, the feature amount (length, width, inclination, luminance, etc.) of the defect image is extracted (step S11), and the extracted feature amount and a predetermined amount are extracted. The type and degree of the defect are determined by comparing with the threshold value (step S12). At this time, if the defect determination is performed under the determination condition (feature amount threshold value or determination order) according to the defect type roughly specified by comparison with the template image, more detailed defect type determination is possible. In addition, it is possible to appropriately determine the degree of defects that cannot be confirmed from the normalized defect image. The determination result is displayed on the display unit 6.

On the other hand, as a result of the conformity check between the defect image and the template image, when all the template images are not conforming, such as matching between the defect image shown in FIG. 8 and the template image 2 registered in the defect library. (No in step S8, Yes in step S10), only defect determination using feature quantities such as length and width is performed (steps S11 and S12).
As described above, in the present embodiment, the defect image size and luminance are normalized, and the defect image is determined by comparing the normalized defect image with the template image, so that the defect size and luminance change with the occurrence time. It is possible to suppress the erroneous determination of the defect type and the non-detection of the defect due to the defect, and the accuracy of the defect determination can be improved. At this time, since the comparison with the template image is performed after performing the averaging process to blur the image, it is possible to reduce the variation in the result of comparing the luminance in units of one pixel.

  Furthermore, since the defect determination is performed by superimposing the defect image and the template image, it is possible to appropriately specify the defect type even for a complex defect that cannot be clearly expressed as a feature amount. Further, since each normalized image is used for comparing the defect image and the template image, the matching rate can be obtained by simple subtraction, and no complicated matching process is required.

In addition, after performing defect determination by overlaying the defect image and the template image, since defect determination using tree-like logic based on feature quantities such as length and width, after roughly specifying the defect type, Detailed determination of the defect type and determination of the degree of defect (harmful / harmless) can be performed in detail.
In the above embodiment, the case where the defect determination is performed by superimposing the defect image and the template image before the defect determination using the feature amount such as the length and the width has been described. However, these two defect determinations are performed. May be performed simultaneously.

  DESCRIPTION OF SYMBOLS 1 ... Surface defect inspection apparatus, 2 ... Light projector, 3, 4 ... Line type camera, 5 ... Signal processing part, 6 ... Display part, 10 ... Steel plate

Claims (5)

A surface defect inspection device for detecting a surface defect of a steel sheet,
Photographing means for photographing a surface image of the steel sheet;
A defect image extracting means for extracting a defect image corresponding to the surface defect from the surface image photographed by the photographing means;
Image conversion means for performing processing for converting the image size to a preset reference size for the defect image extracted by the defect image extraction means;
By comparing the defect image obtained by the image conversion unit with a template image of a surface defect subjected to the same image conversion process as the image conversion unit by matching, the type of surface defect corresponding to the defect image A surface defect inspection apparatus comprising: a defect identification unit that identifies the defect.   The surface defect inspection apparatus according to claim 1, wherein the image conversion unit performs a process of smoothing luminance of a pixel on the defect image extracted by the defect image extraction unit.   The surface defect inspection apparatus according to claim 1, wherein the image conversion unit performs an averaging process on the defect image extracted by the defect image extraction unit. Feature point extraction means for extracting feature points of the defect image extracted by the defect extraction means;
A feature point comparing means for determining the type and degree of the surface defect corresponding to the defect image by comparing the feature point extracted by the feature point extracting means with a preset feature point threshold. The surface defect inspection apparatus according to any one of claims 1 to 3. A surface defect inspection method for detecting a surface defect of a steel sheet,
The surface image of the steel sheet was photographed, a defect image corresponding to the surface defect was extracted from the photographed surface image, and the extracted defect image was subjected to processing for converting the image size to a preset reference size. Then, by comparing the defect image with the template image of the surface defect subjected to the same image conversion processing by matching, the type of the surface defect corresponding to the defect image is specified. Inspection method.

Images