JP5842373B2 - Surface defect detection method and surface defect detection apparatus - Google Patents

Description

  The present invention relates to a surface defect detection method and a surface defect detection apparatus.

  In order to promote quality control or quality assurance in the manufacturing process of continuous metal strips such as steel plates and aluminum plates, it is important to continuously detect various surface defects occurring on these surfaces. Therefore, a surface defect inspection method using an optical method has been put into practical use (Patent Document 1).

  For example, the surface of the surface of the steel plate or the like is illuminated, and the surface wrinkles of the steel plate or the like is detected based on an image signal obtained by continuously photographing with an imaging device such as a CCD camera. . More specifically, for each pixel of the obtained surface image, it is determined whether or not the luminance value is equal to or higher than a predetermined luminance value, and pixels that are equal to or higher than the threshold value (or lower) are extracted as pixels that may constitute the eyelid. . After that, a connection process between adjacent pixels is performed on the extracted pixels.

If there are slight wrinkles on the mirror-finished surface, this method is also effective, but if there is a ground pattern or dirt on the inspection object, these will be detected as wrinkles. So-called “overdetection” may be caused.
When such over-detection occurs, it is difficult to accurately determine wrinkles with conventional surface defect detection devices, and it is necessary to rely on visual inspection that is inferior in detection rate, resulting in concerns such as the oversight of serious defects. .

JP 58-204353 A

  In view of the above circumstances, the present invention provides a surface defect detection method capable of discriminating and detecting harmful defects on a metal strip and overdetection, and a surface defect detection apparatus used in the method. With the goal.

  As a result of earnestly examining the problems of the prior art, the present inventors, as a result of over-detection due to the surface roughness of the metal band, the size of the wrinkle area to be detected, and the luminance value of the pixels in the wrinkle area We found that there is a certain relationship with the sum of The inventors of the present invention have further studied based on this finding, and found that the above problem can be solved by the following configuration.

(1) A surface defect detection method for detecting a surface defect of a traveling metal strip,
An image signal acquisition step of acquiring an image signal representing a two-dimensional image of the surface of the metal band by receiving light reflected from the surface of the metal band by receiving light from a light source on the surface of the traveling metal band,
The brightness represented by the image signal is normalized with reference to the brightness of the normal part of the surface of the metal band, and for each normalized brightness value, the brightness level indicating the normal part is used as a reference, and the polarity is positive and negative. A cocoon candidate point extracting step for extracting pixels that exceed a predetermined threshold as a cocoon candidate point;
A linking process step of linking a plurality of cocoon candidate points existing within a predetermined distance from each other as a continuous cocoon candidate region;
The sum of absolute values of the normalized luminance values of the wrinkle candidate points included in the wrinkle candidate region (detected integrated value), the number of pixels of the maximum length of the wrinkle candidate region in the traveling direction of the metal band (detection) A calculation step of calculating the number of pixels (detection width) of the maximum length of the wrinkle candidate region in a direction orthogonal to the traveling direction of the metal band,
A ratio (detection width / density width) between the detection width and the density width (detection integrated value / detection length) obtained by dividing the detection integrated value by the detection length is determined in advance according to the metal band. A surface defect comprising: determining that the detected wrinkle candidate region is overdetected when exceeding a set reference value; and determining the detected wrinkle candidate region as a defect if the detected value is equal to or less than the reference value Detection method.

(2) a light receiving unit that receives reflected light reflected by the surface of the traveling metal strip and acquires an image signal representing a two-dimensional image of the surface of the metal strip;
The brightness represented by the image signal is normalized with reference to the brightness of the normal part of the surface of the metal band, and for each normalized brightness value, the brightness level indicating the normal part is used as a reference, and the polarity is positive and negative.疵 candidate point extracting means for extracting pixels exceeding a predetermined threshold as 疵 candidate points;
A connection processing means for connecting and processing a plurality of eyelid candidate points existing within a predetermined distance from each other as a continuous eyelid candidate area;
The sum of absolute values of the normalized luminance values of the wrinkle candidate points included in the wrinkle candidate region (detected integrated value), the number of pixels of the maximum length of the wrinkle candidate region in the traveling direction of the metal band (detection) Length) and a calculation means for calculating the maximum number of pixels (detection width) of the wrinkle candidate region in a direction orthogonal to the traveling direction of the metal strip,
A ratio (detection width / density width) between the detection width and the density width (detection integrated value / detection length) obtained by dividing the detection integrated value by the detection length is determined in advance according to the metal band. A surface defect comprising: a determination unit that determines that the detected wrinkle candidate region is over-detected when larger than a set reference value, and determines the detected wrinkle candidate region as a defect when equal to or less than the reference value Detection device.

  ADVANTAGE OF THE INVENTION According to this invention, the surface defect detection method which can discriminate and detect the harmful | toxic harm on a metal strip, and an overdetection, and the surface defect detection apparatus used for this method can be provided.

It is a schematic block diagram of 1st Embodiment of the surface defect detection apparatus of this invention. It is a figure which shows one embodiment of the two-dimensional image of the surface of the metal strip obtained by the eyelid candidate extraction means, the connection processing means, and the calculation means. It is a figure which shows the flow which performs the surface defect detection method of this invention. It is a figure which shows an example of a cocoon candidate area | region. It is a schematic block diagram of 2nd Embodiment of the surface defect detection apparatus of this invention. It is a figure which shows the detection result of the galvannealed steel plate surface used in the Example. It is a figure which shows the detection result of the hot dip galvanized steel plate surface used in the Example.

  Hereinafter, a surface defect detection method and a surface defect detection apparatus according to the present invention will be described with reference to the drawings.

<Surface Defect Detection Device (First Embodiment)>
First, a first embodiment of a surface defect detection device of the present invention will be described below with reference to the drawings.
FIG. 1 is a schematic configuration diagram of a first embodiment of a surface defect detection apparatus of the present invention.
In FIG. 1, the metal strip 100, the light source 12, the surface defect detection apparatus 10, and the display means 24 are shown. The surface defect detection apparatus 10 includes a light receiving unit 14, a wrinkle candidate point extraction unit 16, a connection processing unit 18, a calculation unit 20, and a determination unit 22.
Below, each structure is explained in full detail.

(Metal strip 100)
The type of the metal strip 100 is not limited as long as it is a long steel plate. Steel plate, etc.), stainless steel plate, aluminum plate and the like.
The method for transporting the metal strip 100 is not particularly limited. For example, the metal strip 100 on the transport line is transported by the rotation of a plurality of transport rolls arranged at regular intervals. In order to take an image of the surface of the metal strip 100, a light source 12 and a light receiving unit 14, which will be described later, may be disposed above the transport line.

(light source)
The light source 12 is a device that irradiates the metal strip 100 with light. The light source 12 is arranged so that light enters the entire width direction of the metal strip 100 at a constant incident angle with respect to the metal strip 100. For example, as shown in FIG. 1, one linear light source is disposed over the entire width of the metal strip 100 in the upper position of the transport path of the metal strip 100.
As the light source 12, it is preferable to use a linear light source having diffusion characteristics, that is, a linear diffused light source. In addition, when the specular reflection component and the specular diffuse reflection component are extracted separately from the regular reflection direction of the metal strip 100, it is preferable to use polarized light. The effect of the linear diffused light source is as described in paragraphs [0050] to [0092] of Japanese Patent No. 3275811.
A fluorescent lamp can also be used as the linear diffused light source. A fiber light source in which the exit ends of the bundle fiber are aligned on a straight line can also be used.
The location and number of the light sources 12 can be appropriately selected according to the type of the metal strip 100.

(Light receiving section)
The light receiving unit 14 is a device that receives the reflected light reflected by the metal strip 100 and obtains an image signal. For example, a CCD camera is used, and a field of view is set at a position irradiated with light on the surface of the metal strip 100. If one light receiving unit 14 cannot measure all the width direction of the metal strip 100 (the direction perpendicular to the transport direction), a plurality of light receiving units 14 may be installed in the width direction.
The analog image signal acquired by the light receiving unit 14 is converted into a digital image signal by an A / D conversion unit arranged inside the light receiving unit 14 (not shown). The generated image signal is output to the eyelid candidate point extracting means 16 described later.
Note that the image signal acquired by the light receiving unit 14 is subjected to luminance unevenness correction processing (shading correction) in a preprocessing unit (not shown) as necessary, and is then output to the eyelid candidate point extracting unit 16.

(Mass candidate point extraction means)
The cocoon candidate point extracting unit 16 is a unit that extracts the cocoon candidate point based on the image signal acquired by the light receiving unit 14.
More specifically, the eyelid candidate point extraction unit 16 first normalizes the luminance represented by the image signal acquired by the light receiving unit 14 with reference to the luminance of the normal part of the surface of the metal band 100, and relative to the normal part. It converts into the luminance signal (luminance value) which shows a deviation. In other words, the image signal is normalized so that the luminance of the normal part of the surface of the metal band 100 becomes a predetermined luminance of all gradations, and converted into a luminance signal indicating deviation. Pixels with high brightness that are brighter than the normal part are detected as positive-polarity luminance signals, and pixels with low brightness that are darker than normal parts are detected as negative-polarity luminance signals.
Next, the eyelid candidate point extraction means 16 uses, for each of the normalized luminance values of each pixel, the type of the metal strip 100 for the positive polarity and the negative polarity on the basis of the luminance level indicating the normal part. Accordingly, a pixel area exceeding a predetermined threshold is extracted as a wrinkle candidate area. That is, with reference to the luminance value indicating the normal portion, pixels that exceed a predetermined threshold for the positive polarity luminance signal and the negative polarity luminance signal are extracted as wrinkle candidate points. In other words, pixels that are larger than the predetermined deviation from the normal part set according to the metal band 100 are extracted as wrinkle candidate points. The data obtained by the heel candidate point extraction means 16 is output to the connection processing means 18 described later.

(Consolidation processing means)
The connection processing unit 18 is a unit that generates a cocoon candidate region based on the cocoon candidate points extracted by the cocoon candidate point extracting unit 16.
More specifically, the connection processing means 18 uses a plurality of eyelid candidate points existing within a predetermined distance range among the eyelid candidate points extracted by the eyelid candidate point extracting means 16 as a continuous eyelid candidate region. It is a means to perform connection processing.
An example of the connection process is shown in FIG. FIG. 2A is a diagram in which wrinkle candidate points are extracted from the image signal obtained by the light receiving unit 14 by the wrinkle candidate point extracting means 16. One square in FIG. 2A corresponds to one pixel, and the colored portion is a pixel extracted as a wrinkle candidate point. Next, when the linking candidate point located at a position two pixels away from a certain cocoon candidate point is connected, the cocoon candidate region X surrounded by the thick black line in FIG. 2B becomes the cocoon candidate region. Note that the predetermined distance between the wrinkle candidate points to be connected is appropriately changed depending on the type of the metal band 100. For example, one wrinkle candidate point is adjacent to either one of the wrinkle candidate points vertically or horizontally diagonally. May be.
Data obtained by the connection processing means 18 is output to the calculation means 20 described later.

(Calculation means)
The calculating unit 20 is a unit that calculates a feature value related to the wrinkle candidate region obtained by the connection processing unit 18.
More specifically, the calculation means 20 calculates the sum of absolute values (detected integrated values) of normalized luminance values of the wrinkle candidate points included in the wrinkle candidate area obtained by the connection processing means 18, the metal band 100. Calculate the maximum number of pixels (detection length) of the heel candidate region in the traveling direction and the maximum number of pixels (detection width) of the heel candidate region in the direction orthogonal to the traveling direction of the metal strip 100. It is means to do. Data on the detected integrated value, detection length, and detection width obtained is output to the determination means 22 described later.

(Judgment means)
The determining means 22 is a density value calculated from the feature values (detected integrated value, detected length, detected width) of the wrinkle candidate area calculated by the calculating means 20, and a reference value set according to the metal band 100 Is a means for determining that the wrinkle candidate region is overdetected or defective.

(Display means 24)
The display unit 24 is a device that displays the determination data obtained by the determination unit 22 together with the two-dimensional image of the surface of the metal strip 100. If there is a defect in the display image range, the outer frame display, blinking display, color change display, etc. of the estimation unit will be performed, alerting the inspector, and warning sounds, announcements to prevent missed inspections Etc. may be performed.

<Surface defect detection method>
An example of the surface defect detection method of the present invention will be described with reference to the flowchart of FIG.
First, the metal band 100 traveling from the light source 12 described above is irradiated with light, and reflected light from the surface of the metal band 100 is received by the light receiving unit 14 (for example, CCD), and an image signal on the surface of the metal band 100 is acquired. (Step S101). In order to convert the acquired image signal from an analog signal to a digital signal, A / D conversion is performed, and a digital image signal representing a two-dimensional image of the surface of the metal strip 100 is acquired (step S102).

Next, the digital image signal is subjected to shading correction or the like for correcting the luminance unevenness caused by the illumination light from the light source 12, and the image signal is converted to the normal level signal with the normal level signal as the reference level. Normalization is performed so that the central tone value of the key becomes “128” (step S103). More specifically, the reference value “128” is set to “0” for the luminance of each pixel, and a pixel (bright pixel ) having a luminance higher than the reference value is detected as a positive luminance signal (for example, “+10”). A pixel having a luminance lower than the reference value (dark pixel ) is detected as a negative luminance signal (for example, “−10”).
Note that the gradation at the time of normalization is 256 gradations above, but the gradation is not particularly limited, and may be 128 gradations, 512 gradations, 1024 gradations. For example, normalization may be performed so that the signal of the normal part becomes the central gradation value “64” of 128 gradations.
In the above aspect, the normal part of the surface of the metal band is normalized so as to have the central luminance of all gradations, and the positive luminance signal and the negative luminance signal are obtained based on the central level of all gradations indicating the normal part. Detected. When the luminance change from the normal part is biased in one direction from the normal part, the brightness of the normal part is set in addition to the center of all gradations in consideration of the bias.

Next, the normalized luminance value is compared with a threshold value set in advance according to the metal band 100 (step S104). The threshold is appropriately set depending on the type of the metal strip 100, but is usually provided for each of the positive signal and the negative signal. For example, the signal of the normal part of the alloyed hot-dip galvanized steel sheet is normalized so as to have a central gradation value “128” of 256 gradations, the luminance value of the normal part is set to “0”, and “40” and “−40” Is set as a threshold value, the determination is made based on whether or not the deviation based on the normal part exceeds “40”.
A pixel whose normalized luminance value does not exceed the threshold value is determined to be a normal part (step S105). Pixels whose normalized luminance values exceed the threshold are extracted as wrinkle candidate points.

If it is determined that there is a wrinkle candidate point, a plurality of wrinkle candidate points existing within a predetermined distance range are connected as a single continuous wrinkle candidate region (step S106).
In FIG. 4, the points extracted as wrinkle candidate points in step S104 are represented by colored pixel portions, and regions obtained by connecting them are indicated as wrinkle candidate regions. The numerical value shown in the colored pixel portion is a normalized luminance value of each colored pixel portion calculated by setting the luminance value of the normal portion as “0”.

Next, the sum of the absolute values of the normalized luminance values of the heel candidate points included in the heel candidate area (detected integrated value), the number of pixels of the maximum length of the heel candidate area in the traveling direction of the metal strip 100 (detection) Length) and the maximum number of pixels (detection width) of the heel candidate region in the direction orthogonal to the traveling direction of the metal strip 100 (step S107).
For example, the vertical direction in FIG. 4 corresponds to the traveling direction. In this case, the detection length is calculated as “10” and the detection width is calculated as “4”. Further, in FIG. 4, the detected integrated value is calculated as 1000 by adding the normalized luminance value of each cocoon candidate point in the cocoon candidate region.

Next, a ratio (detection width / density width) between the detection width and the density width (detection integrated value / detection length) obtained by dividing the detection integrated value by the detection length is calculated, and according to the metal band 100 Is compared with the reference value set in step S108. For example, based on FIG. 4, the density width is calculated as “100”.
The reference value to be set differs depending on the type of the metal band 100 and the gradation level at the time of normalization. For example, normalization is performed so that the signal of the normal part becomes the central gradation value “128” of 256 gradations. When a hot-dip galvanized steel sheet is used as the metal strip 100, the reference value is calculated as “10”.

When the ratio (detection width / density width) is larger than a reference value set according to the metal strip 100, the detected wrinkle region is determined as overdetection (step S109).
If the ratio (detection width / density width) is equal to or less than a reference value set according to the metal strip 100, the detected wrinkle region is determined as a defect (harmful flaw) (step S110).

The wrinkle candidate area determined to be defective has a luminance value greatly different from the normal part, and is configured by connecting a large number of pixels that are closely adjacent to each other.
On the other hand, a wrinkle candidate region determined to be over-detected tends to be configured as a wrinkle having a large detection width and detection length by connecting disperse wrinkle candidate points.
The present inventors have found that overdetection and defect can be determined by comparing the density width and the detection width based on such knowledge.

<Surface Defect Detection Device (Second Embodiment)>
First, a second embodiment of the surface defect detection device of the present invention will be described below with reference to the drawings. FIG. 5 (A) is a side view of the vicinity of the light receiving portion of the second embodiment of the surface defect inspection apparatus of the present invention, and FIG. 5 (B) is a top view of the apparatus.
5A and 5B, the metal strip 100, the linear diffused light source 30, the cylindrical lens 32, the polarizing plate 34, the surface defect detection device 50, and the display means 24 are shown. The surface defect detection device 50 includes a light receiving unit 36, a wrinkle candidate point extraction unit 16, a connection processing unit 18, a calculation unit 20, and a determination unit 22. The light receiving unit 36 includes analyzers 38a, 38b, and 38c and light receiving cameras 40a, 40b, and 40c.
FIGS. 5A and 5B have the same configuration as that shown in FIG. 1 except for the linear diffused light source 30, the cylindrical lens 32, the polarizing plate 34, and the light receiving unit 36. Therefore, the same reference numerals are assigned to the same components, and the description thereof is omitted, and the linear diffused light source 30, the cylindrical lens 32, the polarizing plate 34, and the light receiving unit 36 will be mainly described.

In the second embodiment, by detecting the specular reflection component and the specular diffuse reflection component included in the reflected light from the surface of the metal strip 100, the surface of the surface to be inspected is cracked, curled or turned up. Such a pattern-like bald defect having no remarkable unevenness can be reliably detected, and higher defect detection accuracy can be exhibited.
The detection mechanism of the specular reflection component and the specular diffuse reflection component is as described in paragraphs [0022] to [0092] of JP-A No. 11-183398.

The linear diffused light source 30 projects the light of the metal halide light source from both ends of the transparent light guide rod, which is partially coated with diffuse reflection paint, to obtain uniform emitted light in the width direction.
The incident light with respect to the metal strip 100 emitted from each position of the linear diffused light source 30 has an incident angle θ of 60 °, for example, with respect to the entire width of the metal strip 100 in the traveling state via the cylindrical lens 32 and the polarizing plate 34. Irradiate. The azimuth angle (polarization angle) α of the polarizing plate 34 is set to 45 °.

The light receiving unit 36 includes light receiving cameras 40a and 40b including three linear array cameras having analyzers 38a, 38b, and 38c in which the light detection angles β are set to −45 °, 45 °, and 90 °, respectively, in front of the lens. , 40c.
The optical axes of the light receiving cameras 40a, 40b, and 40c are maintained parallel to each other. When the optical axes of the light receiving cameras 40a, 40b, and 40c are maintained in parallel as described above, the pixels of the three light receiving cameras 40a, 40b, and 40c correspond one-to-one with the same visual field size.

  The luminance value of each pixel for one line in the width direction of the metal band 100 in the reflected light received by each light receiving camera 40a, 40b, 40c is normalized to the luminance signal a, b, c, respectively, and the eyelid candidate point extracting means 16 is transmitted.

The eyelid candidate point extraction means 16 normalizes the luminance represented by the luminance signals a to c on the basis of the luminance of the normal part of the surface of the metal band 100 and converts it into an image signal indicating a relative deviation from the normal part.
Thereafter, the wrinkle is determined according to a predetermined wrinkle candidate determination criterion.

  Thereafter, “defect” or “overdetection” is detected through the above-described connection processing step, calculation step, and determination step.

The surface defect detection apparatus shown in FIG. 1 was used to detect surface defects in the alloyed hot-dip galvanized steel sheet and hot-dip galvanized steel sheet. In the wrinkle candidate point extracting step, the obtained image signal is normalized so that the signal of the normal part becomes the central gradation value “128” of 256 gradations, and the reference value “128” is set to “0”. As a luminance signal indicating a relative deviation from the normal part. A pixel having a luminance higher than the reference value (bright pixel ) was detected as a positive luminance signal, and a pixel having a luminance lower than the reference value (dark pixel ) was detected as a negative luminance signal. Thereafter, the predetermined deviation from the normal part is set to “40” (in other words, the luminance values “+40” and “−40” are set as threshold values), and the normalized luminance value exceeds the threshold value (deviation). Pixels were extracted as wrinkle candidate points.

  In the determination step, the reference value for determining “overdetection” or “defect” is “10” in this case.

  While plotting each wrinkle candidate area with the ratio (detection width / density width) obtained in the calculation step as the vertical axis and the detection width as the horizontal axis, each checkered area is identified as “overdetected” or “defect” by visual confirmation. Determination was made (FIG. 6: galvannealed steel plate, FIG. 7: hot dip galvanized steel plate).

As can be confirmed from FIG. 6, it was confirmed that the wrinkle candidate region when the ratio (detection width / density width) is equal to or less than the reference value “10” is “defect” from the visual confirmation. On the other hand, when the ratio (detection width / density width) is greater than the reference value “10”, it was confirmed from visual confirmation that “overdetection” occurred.
That is, it was confirmed that “overdetection” or “defect” can be determined by comparing the ratio (detection width / density width) with the reference value “10”.

  Similarly in FIG. 7, it was confirmed that “overdetection” or “defect” can be determined by comparing the ratio (detection width / density width) with the reference value “10”.

DESCRIPTION OF SYMBOLS 100 Metal strips 10, 50 Surface defect detection apparatus 12 Light source 14 Light-receiving part 16 Haze candidate point extraction means 18 Connection processing means 20 Calculation means 22 Determination means 24 Display means 30 Linear diffused light source 32 Cylindrical lens 34 Polarizing plate 36 Light-receiving part 38a, 38b, 38c Analyzer 40a, 40b, 40c Light receiving camera

Claims (2)

A surface defect detection method for detecting a surface defect of a traveling metal strip,
An image signal acquisition step of acquiring an image signal representing a two-dimensional image of the surface of the metal band by receiving light reflected from the surface of the metal band by receiving light from the light source on the traveling metal band surface;
The luminance represented by the image signal is normalized on the basis of the luminance of the normal portion of the surface of the metal band, and for each normalized luminance value, the luminance level indicating the normal portion is used as a reference, and the positive polarity and the negative polarity are obtained. A wrinkle candidate point extracting step for extracting pixels that exceed a predetermined threshold as wrinkle candidate points;
A connection processing step of connecting a plurality of wrinkle candidate points existing within a predetermined distance from each other as a continuous wrinkle candidate region;
The sum of absolute values of the normalized luminance values of the wrinkle candidate points included in the wrinkle candidate area (detected integrated value), the number of pixels of the maximum length of the wrinkle candidate area in the traveling direction of the metal band (detection) Length), and a calculation step of calculating the number of pixels (detection width) of the maximum length of the wrinkle candidate region in a direction orthogonal to the traveling direction of the metal band,
A ratio (detection width / density width) between the detection width and the density width (detection integrated value / detection length) obtained by dividing the detection integrated value by the detection length is determined in advance according to the metal band. A surface defect comprising: determining that the detected wrinkle candidate area is overdetected if it exceeds a set reference value; and determining the detected wrinkle candidate area as a defect if it is equal to or less than the reference value Detection method. A light receiving unit that receives reflected light reflected from the surface of the traveling metal band and obtains an image signal representing a two-dimensional image of the surface of the metal band;
The luminance represented by the image signal is normalized on the basis of the luminance of the normal portion of the surface of the metal band, and for each normalized luminance value, the luminance level indicating the normal portion is used as a reference, and the positive polarity and the negative polarity are obtained.疵 candidate point extracting means for extracting pixels exceeding a predetermined threshold as 疵 candidate points;
A connection processing means for connecting and processing a plurality of eyelid candidate points existing within a predetermined distance from each other as a continuous eyelid candidate area;
The sum of absolute values of the normalized luminance values of the wrinkle candidate points included in the wrinkle candidate area (detected integrated value), the number of pixels of the maximum length of the wrinkle candidate area in the traveling direction of the metal band (detection) Calculating means for calculating the number of pixels (detection width) of the maximum length of the wrinkle candidate region in a direction orthogonal to the traveling direction of the metal band,
A ratio (detection width / density width) between the detection width and the density width (detection integrated value / detection length) obtained by dividing the detection integrated value by the detection length is determined in advance according to the metal band. A surface defect comprising: a determination unit that determines that the detected wrinkle candidate region is overdetected when exceeding a set reference value, and determines the detected wrinkle candidate region as a defect when equal to or less than the reference value Detection device.