JP2010223597A - Surface flaw inspecting device and surface flaw inspecting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a surface flaw inspecting device, capable of further enhancing the inspection precision, based on the data from both sides of a metal strip, with respect to the flaw produced on the surface of the metal strip, and to provide a surface flaw inspecting method. <P>SOLUTION: The surface flaw inspecting device 10, for inspecting the flaw on the surface of the fed metal strip F. The device 10 has a first and second imaging parts 101 and 201 for imaging the first and second surfaces of the metal strip F, first and second extraction parts 103 for extracting the feature data containing the position data of the flaw candidate on the first or second surface side of the metal strip F, on the basis of the imaged first or second image; a closet approach specifying section131 for specifying the flaw candidate on either one of the first and second surface sides of the closest approach with respect to the flaw candidate on the other one of the first and second surface sides, on the basis of the position data of the flaw candidates on the first and second surface sides to allow both flaw candidates to correspond to each other and an one-side determining section132 for determining, at least one of the kind and degree of the flaw candidate on one side on the basis of the feature data of the flaw candidates on both sides which are allowed to correspond to each other. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a surface defect inspection apparatus and a surface defect inspection method for detecting defects on the surface of a metal strip.

  Various metal materials are used for products in various fields. Among these metal materials, metal strips that are relatively thin and formed into strips are widely available, but the importance of quality control of metal strips is increasing with the improvement of product technology and the development of science and technology. .

  Various types of surface defects such as inclusions and irregularities (hereinafter also simply referred to as “疵”) may occur on the surface of such a metal strip due to various causes. Therefore, it is very important to accurately inspect such wrinkles.

  In order to detect wrinkles formed on a metal band, for example, wrinkle inspection devices that use light, such as the wrinkle inspection device described in Patent Document 1, are often used.

Japanese Patent No. 3669256

  As an example of a metal band that can generate surface wrinkles, a thin plate such as a cold-rolled steel plate manufactured in the steel industry is raised. Among the thin plate wrinkles, what appears on both sides is a serious defect, so it is desirable to distinguish it from what appears only on one side. In particular, in a thin material such as a tin plate, a push rod generated by plastic deformation of a plate due to a foreign matter adhering to a conveyance roll appears on both sides. On the other hand, when an inspection device is provided in a portion that is rubbed against the roll, an image as if it is pushed by a small foreign matter attached to the lashing roll may be obtained. However, such a minute foreign substance remains in the range of elastic deformation and often does not remain as wrinkles, which is a cause of erroneous detection.

  Therefore, in the inspection apparatus disclosed in Patent Document 1, ridges having a deviation between the width direction position and the longitudinal direction position equal to or less than a set value are determined as the same ridge from the inspection results on both sides. Then, for the same bag, the type with the larger extracted feature value (such as the larger luminance change) is used to determine the type of the bag.

  On the other hand, for example, wrinkles caused by foreign matter in steel such as a bearded wrinkle are linear wrinkles on one side, but a part of them may appear in the form of dots on the opposite surface. In this case, there is a high possibility that the hole will be formed by processing, and it is harmful, so it is desired to detect it separately.

  However, as in the inspection apparatus disclosed in Patent Document 1, when the recognition of the same wrinkles on the front and back sides is made based on whether or not a fixed-size rectangle is entered regardless of the type of wrinkles, it becomes a relatively large wrinkle such as a beard. On the other hand, there is a possibility that the rectangular shape is too small and the dotted portions on the opposite surface are not regarded as the same. For relatively small wrinkles, neighboring wrinkles that are not identical because the rectangles are too large may be considered the same.

  Moreover, in the inspection apparatus of the said patent document 1, the feature-value is integrated about the same wrinkles, or it determines again based on a rule from the determination result of both surfaces. However, in such a method, a part of the feature amount is lost, or the determination is performed from discrete information such as the determination result of both sides instead of the feature value that is a continuous value, so that the determination accuracy is lowered.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide high inspection accuracy based on information from both sides of the metal band for wrinkles generated on the surface of the metal band. There is to inspect in.

In order to solve the above problems, according to one aspect of the present invention, there is provided a surface defect inspection apparatus for inspecting wrinkles formed on a surface of a metal band to be conveyed,
A first imaging unit that is disposed on the first surface side, which is one of both surfaces of the metal band, and images the first surface;
A first extraction unit that extracts feature information including at least position information representing a longitudinal position and a width direction position of the eyelid candidate on the first surface, based on the first image captured by the first imaging unit;
A second imaging unit that is disposed on the second surface side, which is the other surface of both surfaces of the metal strip, and images the second surface;
A second extraction unit that extracts feature information including at least position information indicating the position of the eyelid candidate on the second surface side based on a second image captured by the second imaging unit;
Based on the position information of the wrinkle candidates on the first surface side and the second surface side, the wrinkle on the other side closest to the flaw candidate on one side of the first surface side and the second surface side A closest identification unit that identifies candidates and associates the candidates with each other as closest candidates;
Based on the feature information of the one-side eyelid candidate and the feature information of the other-side eyelid candidate associated with the one-side eyelid candidate, at least one of the type and degree of the one-side eyelid candidate is determined. A one-side determination unit that
A surface defect inspection device is provided.

  In addition, the one-side determination unit, as the feature information of each of the one side and the other side cocoon candidates that are associated with each other as the nearest neighbor candidate, at least the displacement amount in the longitudinal direction between the heel candidates and the width direction Based on the positional deviation amount, at least one of the type and degree of the one-side eyelid candidate may be determined.

Further, the feature information of the wrinkle candidate includes feature amount information representing the feature amount of the wrinkle candidate that can be extracted from the first image or the second image.
The one-side determining unit, as the feature information of each of the one-side and other-side cocoon candidates associated with each other as the nearest neighbor candidate, based on at least the feature amount information of each of the heel candidates, You may determine at least one of the kind and grade of a cocoon candidate.

The metal strip is conveyed in the longitudinal direction,
The nearest neighbor specifying unit calculates a distance between the one side candidate and the plurality of other side candidates on a space where the unit distance in the longitudinal direction is longer than the unit distance in the width direction. Alternatively, the other side candidate with the shortest distance may be identified, and the candidates may be associated with each other as the closest candidate.

  Moreover, you may further have the other side determination part which determines at least one of the kind and grade of this other side wrinkle candidate based on the characteristic information of the said other side wrinkle candidate.

  In addition, the one-side determination unit further uses the determination result for the other-side wrinkle candidate by the other-side determination unit, and the one-side determination unit associated with the other-side wrinkle candidate as the closest-contact candidate You may determine at least one of the kind and grade of a cocoon candidate.

  In addition to the feature information of the other heel candidate, the other side determination unit further includes feature information of one heel candidate associated with the other heel candidate as the nearest neighbor candidate. Based on this, at least one of the type and the degree of the other wing candidate may be determined.

Further, the second imaging unit images the second surface on the downstream side in the transport direction from the imaging position of the first imaging unit,
The surface defect inspection apparatus may perform the first extraction at least during a delay period that is a time interval required when the metal strip is transported from the imaging range of the first imaging unit to the imaging range of the second imaging unit. A delay memory capable of recording the feature information of the first candidate on the first side extracted by the first unit.

  In addition, the surface defect inspection apparatus has at least about the length of the maximum wrinkle assumed as an inspection target, and the one side wrinkle during a reception buffer period which is a time interval required for transporting the metal strip. You may further have the receiving buffer memory which can record the feature information of a candidate as a search object of the nearest neighbor candidate with which the nearest neighbor specific part matches with the other side candidate.

In addition, the nearest neighbor specifying unit specifies the closest candidate for the first face to the second face side candidate, and associates the candidate candidates with each other as the closest candidate. With
The delay memory also serves as the reception buffer memory, and not only during the delay period, but also at least during the reception buffer period, the nearest neighbor specifying unit displays the feature information of the first-side cocoon candidate. It may be recordable as a search target of the nearest neighbor candidate associated with the second face side candidate.

Moreover, in order to solve the above problem, according to another aspect of the present invention, there is provided a surface defect inspection method for inspecting wrinkles formed on the surface of a metal band to be conveyed,
A first imaging step in which a first imaging unit arranged on the first surface side, which is one of both surfaces of the metal strip, images the first surface;
A first extraction step of extracting feature information including at least position information indicating a longitudinal position and a width direction position of the eyelid candidate on the first surface side based on the first image captured in the first imaging step;
A second imaging step in which a second imaging unit arranged on the second surface side, which is the other surface of both surfaces of the metal strip, images the second surface;
A second extraction step for extracting feature information including at least position information indicating the position of the eyelid candidate on the second surface side based on the second image captured in the second imaging step;
Based on the position information of the wrinkle candidates on the first surface side and the second surface side, the wrinkle on the other side closest to the flaw candidate on one side of the first surface side and the second surface side A closest identification step of identifying candidates and associating the candidate candidates with each other as closest candidate candidates;
Based on the feature information of the one-side eyelid candidate and the feature information of the other-side eyelid candidate associated with the one-side eyelid candidate, at least one of the type and degree of the one-side eyelid candidate is determined. A one-side determination step,
A surface defect inspection method is provided.

  As described above, according to the present invention, the first surface side wrinkle candidate that is one of both sides of the metal band and the second surface side wrinkle candidate that is the other surface are closest to each other. Can be associated with each other as closest candidates. Then, based on the feature information of the both, it is possible to determine the type and degree of the wrinkle candidate. Therefore, wrinkles generated on the surface of the metal band can be inspected with high inspection accuracy based on information from both sides of the metal band.

It is explanatory drawing for demonstrating the structure of the surface defect inspection apparatus which concerns on 1st Embodiment of this invention. It is explanatory drawing for demonstrating the nearest neighbor search process by the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the nearest neighbor search process by the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating operation | movement of the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the example of arrangement | positioning of the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the inspection example of the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the structure of the surface defect inspection apparatus which concerns on 2nd Embodiment of this invention. It is explanatory drawing for demonstrating the nearest neighbor search process by the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating operation | movement of the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the structure of the surface defect inspection apparatus which concerns on 3rd Embodiment of this invention. It is explanatory drawing for demonstrating operation | movement of the surface defect inspection apparatus which concerns on the same embodiment. It is explanatory drawing for demonstrating the surface defect inspection apparatus which concerns on a prior art. It is explanatory drawing for demonstrating the surface defect inspection apparatus which concerns on a prior art.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

  The surface defect inspection apparatus according to each embodiment of the present invention detects, for example, wrinkles formed on the surface of a metal strip such as a thin plate such as a cold-rolled steel plate manufactured in the steel industry or a thin plate of other non-ferrous metal material. can do. At this time, this surface defect inspection apparatus can detect wrinkles with high detection accuracy. Therefore, in order to facilitate understanding of the effects and the like of the surface defect inspection apparatus according to each embodiment of the present invention, first, the surface defect inspection apparatus according to the prior art as used so far will be described, and then The surface defect inspection apparatus according to the first embodiment will be described. In the first embodiment, an example of detection by the surface defect inspection apparatus will be described by taking as an example a bag that was difficult to detect so as to facilitate understanding of effects and the like. Then, 2nd Embodiment and 3rd Embodiment are described as a modification with respect to the surface defect inspection apparatus which concerns on 1st Embodiment. However, in this case, in the second embodiment and the third embodiment, detailed description of the contents described in the first embodiment will be omitted, and differences from the first embodiment will be mainly described. That is, below, it demonstrates in the following order.

<1. Surface defect inspection apparatus according to the prior art>
<2. First Embodiment>
(2-1. Configuration of surface defect inspection apparatus)
(2-2. Operation of surface defect inspection apparatus)
(2-3. Arrangement example of surface defect inspection device)
(2-4. Inspection example of surface defect inspection device)
(2-5. Examples of effects according to first embodiment)
<3. Second Embodiment>
(3-1. Configuration of surface defect inspection apparatus)
(3-2. Operation of surface defect inspection apparatus)
(3-3. Examples of effects according to second embodiment)
<4. Third Embodiment>
(4-1. Configuration of surface defect inspection apparatus)
(4-2. Operation of surface defect inspection apparatus)
(4-3. Examples of effects according to third embodiment)

<1. Surface defect inspection apparatus according to the prior art>
12 and 13 are explanatory views for explaining a surface defect inspection apparatus according to the prior art.

  In the surface defect inspection apparatus according to the related art, both sides of a metal strip are imaged by an imaging apparatus, and the captured image is analyzed to extract a wrinkle candidate from a change in luminance value or the like. Then, whether or not the two wrinkle candidates are the same is determined by whether or not the deviation between both of the width direction coordinates x and the deviation between both the longitudinal direction coordinates y are within a predetermined threshold. At this time, the state of the same wrinkle determination is conceptually shown in FIGS. FIG. 12 shows a determination process for a spot-like stain defect, and FIG. 13 shows a determination process for a bald defect.

  As shown in FIG. 12A, the deviation range 2Ws of the width direction coordinate x and the deviation range 2Ls of the longitudinal direction coordinate y are determined with the wrinkle candidate Du extracted from the upper surface as the center (position O). Then, the wrinkle candidate Dd extracted from the lower surface and included in the deviation range in the same range Ar is determined to be the same wrinkle as the wrinkle candidate Du. However, as shown in FIG. 12, when the same range Ar is too larger than the size of the wrinkle candidate, the upper surface wrinkle candidate Du and the lower surface wrinkle candidate Dd are the same even though they are different from each other. It may be recognized as a samurai. That is, the size of the wrinkles that can be detected is limited according to the same range Ar of the width direction coordinate x and the longitudinal direction coordinate y that are regarded as the same wrinkles. However, it is actually difficult to predict the size of the wrinkles. Setting the same range Ar in advance may determine that the same wrinkles are not the same, and it is difficult to detect wrinkles. May end up.

On the other hand, the same can be said when the same range Ar is too narrow.
That is, as shown in FIG. 13, for example, in the case of a cocoon having different sizes on one surface and the other surface, such as a bearded cocoon, the width direction coordinate x When the deviation range 2Ws and the deviation range 2Ls of the longitudinal coordinate y are determined, there is a possibility that the same wrinkle candidate Dd extracted from the lower surface does not fall within the same range Ar. In this case, both eyelid candidates Du and Dd are not determined to be the same, but are determined to be different eyelids. Therefore, contrary to the above, it may be determined that the same wrinkles are not the same wrinkles, which may make it difficult to detect wrinkles.

  In addition, in the surface defect inspection apparatus according to this prior art, the type and degree of wrinkle candidates are automatically specified for the upper and lower surface wrinkle candidates Du and Dd determined to be the same wrinkle. However, even if the same defect is identified without the occurrence of the above-described defects, the surface defect inspection apparatus according to the related art has, for example, the defect candidate Du, Dd, Thus, it is performed based only on the feature amount of the eyelid candidate with the larger amount of change of the feature amount such as the luminance value for the normal part (feature amount integration), or the judgment result such as the eyelid type for each surface Done using.

  When the feature values are integrated according to the rule that the larger length is taken as the feature amount of the bag having a length that differs greatly between the front and the back, such as the beaker shown in FIG. However, the information about whether it penetrates behind is lost, and the accuracy is lowered in that it correctly determines the harmfulness of the bag. In addition, when generating decision rules using statistical pattern recognition methods such as support vector machines, it is possible to use rules with better decision accuracy if continuous values are used rather than discrete values. When re-determination is performed based on the determination rule based on the result, a discreet value such as a moth type is used as input data, and the determination accuracy is reduced as compared with the case where a feature value that is a continuous value is used as it is. End up.

  The inventors of the present invention have intensively studied on the wrinkle inspection and the like of metal bands, and as a result, completed the present invention capable of improving the detection accuracy. In the surface defect inspection apparatus according to each embodiment of the present invention, it is possible to appropriately associate both-side wrinkle candidates, and it is possible to accurately determine the type and degree of wrinkles in real time. Each embodiment of the surface defect inspection apparatus will be described in detail below.

<2. First Embodiment>
(2-1. Configuration of surface defect inspection apparatus)
First, the configuration and the like of the surface defect inspection apparatus according to the first embodiment of the present invention will be described with reference to FIG. FIG. 1 is an explanatory diagram for explaining the configuration of the surface defect inspection apparatus according to the first embodiment of the present invention.

  As shown in FIG. 1, the surface defect inspection apparatus 10 according to the present embodiment roughly includes an upstream inspection apparatus 100, a downstream inspection apparatus 200, a determination result storage apparatus 300, and a display apparatus 400.

  The upstream inspection device 100 is an example of a first detection device, and the downstream inspection device 200 is an example of a second detection device. Then, the upstream inspection apparatus 100 detects the soot from the metal band F conveyed with the longitudinal direction as the conveyance direction, upstream of the downstream inspection apparatus 200. The upstream inspection device 100 and the downstream inspection device 200 are arranged so as to detect different surfaces. That is, in the example shown in FIG. 1, the upstream inspection device 100 detects wrinkles from the upper surface of the metal band F on the upstream side, and the downstream inspection device 200 in the example shown in FIG.疵 is detected from the lower surface of

  It is also possible for the upstream inspection apparatus 100 to detect the lower surface of the metal band F and the downstream inspection apparatus 200 to detect the upper surface of the metal band F. Furthermore, the upstream inspection apparatus 100 and the downstream inspection apparatus 200 do not necessarily have to be shifted in the transport direction, and may be disposed to face each other across the same position of the metal band F. However, in this embodiment, for convenience of explanation, a case will be described in which the upstream inspection device 100 detects wrinkles upstream in the transport direction relative to the downstream inspection device 200.

  The upstream inspection apparatus 100 includes an imaging unit 101, an image processing unit 102, a feature extraction unit 103, a delay memory 110, a reception buffer memory 120, a closest identification unit 131, and a determination unit 132. On the other hand, the downstream inspection apparatus 200 includes an imaging unit 201, an image processing unit 202, a feature extraction unit 203, and a determination unit 232.

  The imaging unit 101 included in the upstream inspection device 100 is an example of a first imaging unit, and the imaging unit 201 included in the downstream inspection device 200 is an example of a second imaging unit. For example, a line sensor camera or an area sensor camera can be used as the imaging units 101 and 201, but is not particularly limited as long as the upper surface or the lower surface of the metal band F can be imaged.

  The imaging unit 101 is disposed on the upper surface of the metal band F (an example of a first surface that is one of both surfaces of the metal band) and images the upper surface, while the imaging unit 201 is the lower surface of the metal band F. It is arranged on the side (an example of the second surface which is the other surface of both surfaces of the metal band), and the lower surface is imaged. At this time, the imaging unit 101 and the imaging unit 201 are controlled to perform imaging every time the metal band F is transported by a certain distance by a distance measuring device such as a pulse generator attached to a transport roll (not shown). . Further, the imaging unit 101 and the imaging unit 201 are controlled so as to capture the front and back of the metal band F at the same position.

  The imaging unit 201 according to the present embodiment is arranged on the downstream side by an interval ΔL so that the lower surface is imaged downstream of the imaging position of the imaging unit 101 on the upper surface side in the transport direction. Therefore, in the present embodiment, after the imaging unit 101 images the upper surface of the longitudinal position where the metal band F is located, the imaging unit 201 determines that the longitudinal position of the metal band F after the metal band F is conveyed by the interval ΔL. The lower surface of the camera is imaged.

  The image processing unit 102 performs predetermined image processing for facilitating the extraction of a wrinkle candidate on a captured image on the upper surface side (hereinafter referred to as “first image”) captured by the imaging unit 101. On the other hand, the image processing unit 202 performs predetermined image processing on the lower-side captured image (hereinafter, referred to as “second image”) captured by the imaging unit 201 so that it is easy to extract a wrinkle candidate. . Examples of image processing performed by the image processing units 102 and 202 include, for example, shading correction for correcting luminance unevenness, spatial filtering processing for detecting a luminance change in a specific direction, and the like.

  Then, after the image processing, the image processing units 102 and 202 perform detection processing for detecting a wrinkle candidate as another image processing. Examples of the detection process include binarization of image luminance. In the binarization processing, the image processing units 102 and 202 set both high and low threshold values for the luminance value (for example, an average value) of a region where no wrinkles are formed in advance. Based on this, the corresponding first image or second image is binarized. More specifically, the luminance value between the high and low thresholds is set to “0” as “normal”, and the luminance value exceeding the high threshold or less than the low threshold is set to “1” as “abnormal”. As a result, the image processing units 102 and 202 can extract the wrinkle candidate. Here, “cancellation candidate” means a region having a luminance value expected to be a flaw formed on the metal band F, and includes not only wrinkles but also those that can be imaged such as dirt. It will be.

  The feature extraction unit 103 is an example of a first extraction unit, and acquires the first image processed by the image processing unit 102. On the other hand, the feature extraction unit 203 is an example of a second extraction unit, and acquires the second image processed by the image processing unit 202. At this time, it is desirable that the feature extraction units 103 and 203 acquire the first image and the second image before and after the detection process (for example, binarization) for detecting the wrinkle candidate.

  Then, the feature extraction units 103 and 203 extract information (hereinafter also referred to as “feature information”) in each image in the region that is determined as a wrinkle candidate by the image processing units 102 and 202. As the feature information, for example, position information (width-direction coordinate x and longitudinal-direction coordinate y) representing the position of the wrinkle candidate (the center of the circumscribed rectangle of the wrinkle candidate or the center of gravity of the binary image) on each surface of the metal band F And feature amount information representing the feature amount of the wrinkle candidate that can be extracted from each image. The feature amount information includes, for example, the width of the wrinkle candidate in the width direction coordinate x, the length of the wrinkle candidate in the longitudinal direction coordinate y, the shape in the surface of the wrinkle candidate, the highest luminance, the lowest luminance of the wrinkle candidate position, A histogram of the distribution of candidate luminances can be cited.

  The surface defect inspection apparatus 10 according to the present embodiment uses the feature information of the upper and lower surface wrinkle candidates extracted as described above to accurately determine the same wrinkle, and based on the feature information of both surfaces of the same wrinkle Thus, it is possible to determine the type and degree of cocoon. For this reason, the processing after the feature information extraction differs between the feature information of the eyelid candidate for the upper surface and the feature information of the eyelid candidate for the lower surface. Therefore, first, the feature information of the lower surface will be described below.

(Lower surface feature information)
The feature information extracted by the feature extraction unit 203 of the downstream inspection device 200 is sent to the upstream inspection device 100 and recorded in the reception buffer memory 120 of the upstream inspection device 100. The reception buffer memory 120 functions as a buffer that holds the lower surface side feature information necessary for specifying the same wrinkle candidate from the double-sided wrinkle candidates until the closest identifying unit 131 uses it. On the other hand, in the downstream inspection apparatus 200, the feature information extracted by the feature extraction unit 203 is sent to the determination unit 232.

  The determination unit 232 is an example of the other side determination unit, and determines at least one of the type and degree of the wrinkle candidate on the lower surface side based on the feature information on the lower surface side (an example of the other side). In other words, the determination unit 232 determines the wrinkle candidate on the lower surface side based only on the lower surface side feature information. The “type of heel candidate” means a type or cocoon type classified according to the cause or shape of the heel. For example, a push heel, a hole, an uneven heel, a three heel, a hege heel・ Spot type such as dot-like ridges, line-shaped folds, etc., and dross and foreign substances that are attached are also included. The “degree of wrinkle” means a degree or degree of harmfulness such as the influence of the wrinkle on the quality expressed in numerical value or grade.

  The determination by the determination unit 232 can be realized by methods using various classification algorithms. For example, examples of the determination process include if the rule, neural network, decision tree, support vector machine, and the like. Of course, other classification algorithms may be used. In these classification algorithms, the type and degree of wrinkle candidates can be output with a predetermined algorithm using feature information as an input value. In order to establish this classification algorithm, the determination unit 232 inputs a plurality of pieces of teacher information including feature information for a wrinkle candidate whose type and degree are known and a correct determination result for the feature information, and the classification Establish an algorithm.

  Note that the classification algorithm establishment method, the characteristics of each algorithm, and the like are different for each algorithm, and thus detailed description thereof is omitted here. However, regardless of which classification algorithm is used, the determination unit 232 can output at least one of the types and degrees of the wrinkle candidates (hereinafter also simply referred to as “determination results”) with the feature information on the lower surface side as input. A simple classification algorithm.

  The determination result by the determination unit 232 is recorded in the reception buffer memory 120 in association with the feature information of the corresponding wrinkle candidate. The reception buffer memory 120 is a memory for storing data that is a target of nearest neighbor search by the nearest neighbor specifying unit 131 (= detection result of the downstream inspection device 200). The reception buffer period in which the characteristic information of the downstream side candidate is stored in the reception buffer memory 120 will be described in detail in the description of the nearest neighbor search by the nearest neighbor specifying unit 131. Similarly, the determination result storage device 300 is also associated with the feature information of the wrinkle candidate.

(Top surface feature information)
Next, the upper surface characteristic information on the upstream inspection apparatus 100 side will be described.
The feature information extracted by the feature extraction unit 103 of the upstream inspection device 100 is recorded in the delay memory 110 of the upstream inspection device 100.

  The delay memory 110 is a memory for waiting until the feature information of the upstream heel candidate is recorded and the inspection result of the downstream inspection device 200 is stored in the reception buffer memory 120. The delay memory 110 requires at least a delay period of ΔL shown in FIG. 2 in order to associate the upper surface with the lower surface, but in the present embodiment, it is desirable to have a delay period longer than that. A desirable delay period in which the delay memory 110 delays the feature information will be described in detail in the description of the nearest neighbor specifying process by the nearest neighbor specifying unit 131.

  The nearest neighbor specifying unit 131 acquires the upstream cocoon candidate feature information stored in the delay memory 110 and the downstream cocoon candidate feature information stored in the reception buffer memory 120. Then, the nearest neighbor specifying unit 131 performs a nearest neighbor search based on both feature information, and specifies the closest other eyelid candidate to the one side eyelid candidate. As a result, the nearest neighbor specifying unit 131 associates the one side candidate with the closest candidate candidate as the closest candidate. In this way, it is highly likely that the closest wing candidates are the same heel. Therefore, the nearest neighbor specifying unit 131 associates the closest neighbor candidates with each other as the closest candidate.

  In the present embodiment, the closest identifying unit 131 uses the upstream side cocoon candidate, that is, the upper side cocoon candidate, as a reference, and selects the downstream side cocoon candidate, that is, the lower side cocoon candidate closest to the cocoon candidate. Identify.

  Further, as described above, the metal band F is conveyed in the longitudinal direction, and the wrinkles formed on the front and back sides may be formed longer in the longitudinal direction than the width direction of the metal band F. In such a case, the longitudinal coordinate y may vary greatly even if the position of the front and back wrinkle candidates is the same wrinkle.

  The closest identifying unit 131 according to the present embodiment includes an upstream cocoon candidate (an example of one cocoon candidate) and a downstream side so that matching is accurately performed for such cocoon candy candidates. The distance (deviation in the plane of the metal band F) d between the heel candidate (an example of the other heel candidate) d is calculated in a state where the width direction and the longitudinal direction are not equivalent. Then, the nearest neighbor specifying unit 131 specifies the downstream side candidate having the shortest distance d and associates the candidate candidates with each other. In other words, the distance d between two points (front and back eyelid candidates) in a space where the unit distance in the longitudinal direction is longer in the real space than the unit distance in the width direction (also referred to as the short direction). It can also be said that the distance between.

More specifically, this nearest neighbor soot search process will be described.
Using the width direction coordinate x and the longitudinal direction coordinate y, the position information of the upstream eyelid candidate is (xu, yu), and the position information of the downstream eyelid candidate is (xd, yd). In this case, the distance d between the two points can be calculated, for example, as shown in Equation 1 below.

式１

Formula 1

  It is possible to simplify the calculation of the distance d by using the following formula 2 instead of the formula 1.

式２

Formula 2

  Here, α is a variable that can be set as appropriate and determines the degree of reduction in the longitudinal direction, and a value of 1 or less is set. For example, when there is a wrinkle where the position of the wrinkle on one side and the position of the wrinkle on the other side are greatly different in the longitudinal coordinate y and the density of the other wrinkle candidates is also high, Since no bias is applied in the longitudinal direction, even if the closest wrinkle candidate is specified, the wrinkle candidates adjacent in the width direction may be associated with each other. In this case, if α is set to be smaller than 1 and the deviation of the longitudinal coordinate y is allowed, the correct identical eyelid can be associated. As a specific value of α, for example, when the metal strip F is a thin steel plate and the product plate thickness is 1 mm with respect to a thickness of 250 mm at the time of casting, α is set to around 1/250. However, the value of α is not limited to this example, and can be appropriately set according to the characteristics of the metal band F, the degree of occurrence of wrinkles, and the like. Further, it is also possible to use a distance d other than Equations 1 and 2 for calculating the distance. For example, it is possible to apply a reverse bias (α> 1) to the width direction coordinate x instead of applying a bias (α <1) to the longitudinal direction coordinate y.

  Then, the nearest neighbor specifying unit 131 calculates the distance d between each upstream side candidate stored in the delay memory and the downstream side candidate stored in the reception buffer memory 120 as described above. Based on the result, the downstream side candidate closest to the upstream side candidate is specified.

  Here, with reference to FIGS. 2 and 3, the nearest neighbor search process by the nearest neighbor specifying unit 131 including the distance d biased in the longitudinal direction will be conceptually described. 2 and 3 are explanatory diagrams for explaining a nearest neighbor search process by the surface defect inspection apparatus according to the present embodiment.

  The feature information (position information) of the wrinkle candidate extracted from the feature extraction unit 103 of the upstream inspection device 100 is stored in the delay memory 110. The concept of the stored upstream cocoon candidate Du is shown in FIG. On the other hand, the imaging unit 201 of the downstream inspection apparatus 200 is arranged offset with respect to the imaging unit 101 by an interval ΔL in the transport direction. Therefore, if the time point at which the candy candidate Du is stored is “upstream detection time Tu”, the time point “downstream detection time Td” at which the corresponding downstream candy candidate Dd is stored is delayed by the interval ΔL. Therefore, the delay memory 110 holds the upstream wing candidate Du for a delay period corresponding to at least the interval ΔL. In other words, the delay period is a time interval required when the metal band F is transported from the imaging range of the imaging unit 101 to the imaging range of the imaging unit 201. The feature information is stored in the delay memory 110 for at least the delay period (which may include a time during which other data is delayed). On the other hand, in order to perform the nearest neighbor search, a downstream upstream within a predetermined search range L before and after the transport direction including the position O corresponding to the upstream candidate Du for one upstream upstream candidate Du The side wrinkle candidate Dd is stored in the reception buffer memory 120. Therefore, on the time axis, the downstream buffer candidate Dd is stored in the reception buffer memory 120 during the reception buffer period corresponding to the search range L including the position O serving as a search reference. Note that this reception buffer period, that is, the time interval during which the metal band F is transported by the search range L is set to at least the time interval during which the metal band F is transported by the length of the maximum wrinkle assumed as an inspection target. It is desirable. That is, the search range L is set to the length of the maximum wrinkle assumed as the inspection target.

  Correspondingly, in the delay memory 110, not only the delay period but also a period until the search is performed from the downstream detection time Td (for example, half of the reception buffer period), the upstream eyelid candidate Du is stored. . In addition, from the above, the logical length (search range L) in which the reception buffer memory 120 stores the cocoon candidate Dd is set to about the assumed maximum cocoon length, and the delay memory 110 is set to the cocoon candidate. The logical length for storing Du is preferably set to about the sum of the distance d between the imaging units 101 and 102 and a half of the reception buffer memory length.

Then, the nearest neighbor specifying unit 131 performs a nearest neighbor search as shown in FIG.
That is, the nearest neighbor specifying unit 131 uses the position O of the upstream (upper surface side) eyelid candidate Du shown in FIG. 3A as a reference and the downstream side (lower surface side) shown in FIG. The distance d to the cocoon candidates Dd1, Dd2,. At this time, since the distance d is biased in the longitudinal direction as described above, the equidistant line is extended in the longitudinal direction as shown by a broken line in FIG. Therefore, in the case of the example shown in FIG. 3, the nearest neighbor specifying unit 131 does not select the downstream eyelid candidate Dd2 adjacent in the width direction to the upstream eyelid candidate Du but the upstream eyelid candidate Dd1 adjacent in the longitudinal direction. It is possible to identify the closest candidate to be associated with each other.

  In this way, the nearest neighbor specifying unit 131 that associates the front and back eyelid candidates with each other further calculates the deviation of the width direction coordinate x and the deviation of the longitudinal direction coordinate y between the both eyelid candidates without applying a bias. To do. Then, the nearest neighbor specifying unit 131 outputs the characteristic information of the front and back both-side candidates and the deviation between the width direction coordinate x and the longitudinal direction coordinate y to the determination unit 132. However, when there is no downstream kite candidate Dd corresponding to the upstream kite candidate Du, the closest identifying unit 131 indicates that the upstream kite candidate Du feature information and the corresponding downstream kite candidate Dd do not exist. Information is output to the determination unit 132.

  The determination unit 132 is an example of a one-side determination unit, and based on not only the feature information of the upstream cocoon candidate, but also the feature information of the downstream cocoon candidate associated with the cocoon candidate, Determine at least one of the type and degree of the candy candidate.

  At this time, the determination by the determination unit 132 can be realized by a direction based on the same algorithm as the classification algorithm used in the determination by the determination unit 232. That is, for example, examples of the determination process include if the rule, neural network, decision tree, support vector machine, and the like. Of course, other classification algorithms can be used. In these classification algorithms, the type and degree of the wrinkle candidate can be output by a predetermined algorithm using the feature information of not only the upstream wrinkle candidate but also the downstream hail candidate. In order to establish this classification algorithm, the determination unit 132 receives a plurality of pieces of teacher information including feature information for a wrinkle candidate whose type and degree are known and a correct determination result for the feature information, and determines the classification. Establish an algorithm. As the teacher information, unlike the determination unit 232, the correct determination result includes not only the type and degree of the wrinkle candidate, but also the result of whether or not the corresponding wrinkle candidates on both sides are the same. Is preferably included. Then, according to the classification algorithm constructed with such teacher information, from the feature information of the double-sided wrinkle candidates (which may include the determination result of one side), whether or not the double-sided wrinkle candidates are the same, The type and degree of the wrinkle candidate can be output.

  Note that the classification algorithm establishment method, the characteristics of each algorithm, and the like are different for each algorithm, and thus detailed description thereof is omitted here. However, regardless of which classification algorithm is used, the determination unit 132 can output at least one of the types and degrees of the wrinkle candidates (hereinafter also simply referred to as “determination results”) with the feature information on the lower surface side as input. A simple classification algorithm.

  As described above, unlike the determination unit 232, the determination unit 132 determines an upstream wrinkle candidate based on the feature information of the corresponding front and back wrinkle candidates. In other words, the determination unit 232 determines the wrinkle candidate based only on the feature information of the downstream wrinkle candidate, whereas the determination unit 132 determines whether the front and back surface wrinkle candidates are likely to be the same wrinkle. Both feature information is used to determine the upstream side candidate. Therefore, the number of input variables for the classification algorithm increases compared to the case where the feature information of only one side is targeted, and the accuracy of the classification result by the classification algorithm can be improved. In this sense, it can be said that the determination unit 132 performs determination based on the feature information on both sides, and the determination unit 232 performs determination based on the feature information on one side.

  Note that the determination unit 132 uses, as the input variable, feature information of upstream and downstream eyelid candidates, at least the amount of deviation between the positions of both eyelid candidates (deviations Δx and Δy of the width direction coordinate x and the longitudinal direction coordinate y). It is desirable to make a determination using Furthermore, as the feature information, the feature amount information of the upstream and downstream eyelid candidates may also be used for the determination. Furthermore, for the determination, it is also possible to use the determination result of the single downstream side candidate by the determination unit 232 as input information. An example of a set of front and back feature information used for this classification algorithm is shown in Table 1. The unit is mm.

  The above <1. As in the case of the surface defect inspection apparatus according to the prior art as described in the section “Surface defect inspection apparatus according to the prior art”, the front and back flaw candidates are simply determined independently, and further determination is performed based on the determination result. In this case, since the determination result used for the final determination is the type and degree of the wrinkle candidate, it takes a discrete value and may reduce the accuracy of the classification logic. On the other hand, the deviation amount used by the determination unit 132 in the present embodiment takes a continuous value. Therefore, the accuracy of the determination result by the classification algorithm can be improved.

  Similar to the determination result by the determination unit 232, the determination result by the determination unit 132 is recorded in the determination result storage device 300 in association with the feature information of at least one of the upstream and downstream eyelid candidates. In the determination result storage device 300, the first image and the second image before being processed by the image processing units 102 and 202 may be recorded together with the determination result as necessary. And the display apparatus 400 can display the determination result currently recorded on this determination result memory | storage device 300 on a display screen, and can provide it to users, such as an inspector.

(2-2. Operation of surface defect inspection apparatus)
The configuration of the surface defect inspection apparatus 10 according to the first embodiment of the present invention has been described above. Next, the operation of the surface defect inspection apparatus 10 according to the first embodiment of the present invention will be described with reference to FIG.

  FIG. 4 is an explanatory diagram for explaining the operation of the surface defect inspection apparatus according to the present embodiment. In a state where the metal band F to be inspected is conveyed in the longitudinal direction, the upstream inspection apparatus 100 first processes step S101 (an example of a first imaging step) to image the upper surface of the metal band F. Then, the process proceeds to step S103.

  In step S103, the image processing unit 102 of the upstream inspection apparatus 100 performs predetermined image processing to extract a wrinkle candidate. Then, the process proceeds to step S105.

  In step S105 (an example of a first extraction step), the feature extraction unit 103 of the upstream inspection device 100 extracts feature information for each of the wrinkle candidates extracted in step S103. This feature information is stored in the delay memory 110 in step S107.

  On the other hand, in parallel with this, the downstream inspection apparatus 200 performs step S111 (an example of a second imaging step) when the part of the metal band F imaged in step S101 reaches the imaging unit 201 within the imaging range. The imaging unit 201 of the downstream inspection apparatus 200 images the lower surface of the metal band F. Then, the process proceeds to step S113.

  In step S113, the image processing unit 202 of the downstream inspection apparatus 200 performs predetermined image processing to extract a wrinkle candidate. Then, the process proceeds to step S115.

  In step S115 (an example of a second extraction step), the feature extraction unit 203 of the downstream inspection device 200 extracts feature information for each of the wrinkle candidates extracted in step S113. This feature information is stored in the reception buffer memory 120 in step S121. Then, the process proceeds to step S117.

  In step S117, the determination unit 232 of the downstream inspection device 200 determines at least one of the type and degree of the wrinkle candidate based on the feature information of the downstream wrinkle candidate. The determination result is recorded in the reception buffer memory 120 in step S121 and in the determination result storage device 300 in association with the feature information of the wrinkle candidate in step S141.

  On the other hand, in the upstream inspection apparatus 100, after the process of step S123 (after the elapse of the reception buffer period), step S131 (an example of the nearest neighbor specifying step) is processed, and the closest neighbor specifying unit 131 is assigned to each upstream side candidate. On the other hand, the closest downstream side candidate is identified, and the two are associated with each other. Then, the process proceeds to step S133.

  In step S133 (an example of a one-side determination step), the determination unit 132 of the upstream inspection device 100 performs upstream processing based on the feature information of the upstream side candidate and the downstream side candidate and the determination result of the downstream side candidate. Determine at least one of the type and degree of the candy candidate. Then, the determination result is recorded in the determination result storage device 300 in step S143. After the processing of step S143, step S145 is processed, and the determination result in steps S117 and S133 is displayed on the display screen by the display device 400.

(2-3. Arrangement example of surface defect inspection device)
Note that the surface defect inspection apparatus 10 according to this embodiment may be disposed at a position where the metal strip F is sandwiched and offset in the transport direction at a position where the surface defect transport apparatus 10 is linearly transported as shown in FIG. It is desirable to arrange as shown in FIG.

  FIG. 5 is an explanatory diagram for explaining an arrangement example of the surface defect inspection apparatus according to the present embodiment. Usually, the metal strip F is not always transported in a straight line, and its transport direction may be changed by a roll R or the like. In such a case, it is desirable that the surface defect inspection apparatus 10 according to the present embodiment is arranged so as to image the metal band F at a position wound around the roll R. In many cases, the metal band F vibrates vertically and horizontally during the conveyance process. Thus, by imaging the metal band F with the imaging units 101 and 201 at the position wound around the roll R in this way, such fluttering and vibration of the metal band F are reduced, and the detection accuracy of the wrinkle candidate is improved. be able to.

(2-4. Inspection example of surface defect inspection device)
An example of inspection by the surface defect inspection apparatus 10 according to the present embodiment configured as described above will be described with reference to FIG. FIG. 6 is an explanatory diagram for explaining an inspection example of the surface defect inspection apparatus according to the present embodiment. In addition, since it is as above-mentioned that a bald wrinkle, an uneven | corrugated wrinkle, etc. can be detected appropriately, the case where a wrinkle is a pushing rod and a hole is demonstrated as another example of an inspection here.

(2-4-1. Push rod)
The push rod is generated when the metal band F is plastically deformed by a hard foreign material adhering to the transport roll R or the like. In this case, as shown in FIG. 6, half of the region where the eyelid candidate is imaged is whiter than the surroundings depending on the direction of light applied to the metal band F and the direction imaged by the imaging units 101 and 201 (brightness) The value is high) and the other half is black (the luminance value is low).

  In such a case, if the determination is made for each side, it is usually determined whether or not the image is a push-down on the condition that black and white exist adjacent to each other in the transport direction and the circumscribed rectangle is close to a square. And when using a line camera as the image pick-up part 101,201 and picking up the metal strip F at the part to be rubbed against the roll R as shown in FIG. Even if there is a foreign object, an image like a push stick may be taken.

  At this time, in the case of a soft foreign matter, the metal strip F is often deformed in the range of elastic deformation, and returns to the original flat shape after passing through the scoring roll R. In such a case, a determination on only one side results in false detection. On the other hand, since the upstream inspection apparatus 100 according to the present embodiment performs both-side determination by the determination unit 132, it is possible to reduce the possibility that a wrinkle candidate due to elastic deformation is erroneously detected as a pressed wrinkle. .

  On the other hand, as for the push rod actually plastically deformed, whether the push rod is concave or convex when viewed from the imaging units 101 and 201 is determined as shown in FIGS. 6 (A) and 6 (B). It is possible to judge whether the top or bottom is white. Therefore, the feature extraction units 103 and 203 calculate (luminance average value of the upstream half of the kite) − (luminance average value of the downstream half of the kite) as the feature amount. Then, the determination unit 232 can determine that the feature amount is positive when the feature amount is positive, and that the feature amount is negative when the feature amount is negative. Therefore, in the determination unit 132 according to the present embodiment, the determination result by the determination unit 232 or the calculated feature amount is used for determination together with the feature information of the upstream side candidate, so that the upstream side is the upstream side and the downstream side. If it is negative or vice versa, it can be determined that there is the same pushing rod on both the upper and lower surfaces.

  In addition, when the degree of the push tack is relatively light, the shading becomes light, so that only one of adjacent white and black may be recognized as a wrinkle candidate. In this case, there is a possibility that the determination on only one side may be determined as dirt or the like. Therefore, by adding the position deviation on the upstream side and the downstream side to the determination, the determination unit 132 according to the present embodiment determines that the position is a slight push if the position deviation is as large as the heel. And the determination accuracy can be improved.

(2-4-1. Hole)
For example, when the foreign matter in the metal band F falls off after being rolled, such as steel, the metal band F may have a hole. Since this is a very harmful defect, it is desirable to detect it with high accuracy.

  In this case, the roll R is imaged through the hole in the hitting portion as shown in FIG. However, when the scoring roll R is a metal, the appearance looks similar to the surface of the metal band F, and it may be difficult to distinguish. More specifically, only the edge of the hole appears black or white, and the inside of the hole has an image (luminance value) similar to that of the plate surface.

Since the hole has an irregular shape, it is difficult to distinguish it from an irregular defect such as dirt with a feature value on only one side. In such a case, the feature extraction units 103 and 203 have some feature quantities representing the shape of the hole (for example, the aspect ratio of the circumscribed rectangle, the roundness of the wrinkle candidate (= 4 × π × area / (perimeter length) ² ). ) Etc.). Then, the determination unit 132 adds a condition that the feature information is approximated in addition to the positional information on both sides to be approximated to the classification algorithm, so that the defect having the same shape on both sides is obtained. It can be determined that there is a hole, and the hole can be reliably determined. The degree of roundness of the heel candidate is closer to 1 and closer to 1 as the heel candidate is closer to the circle, and is smaller to 1 and lower as it is different from the circle.
(2-5. Examples of effects according to first embodiment)
The surface defect inspection apparatus 10 according to the first embodiment of the present invention has been described above.
According to this surface defect inspection apparatus 10, front and back wrinkle candidates that are highly likely to be the same wrinkle can be associated by the closest identifying unit 131 based on the distance d between the wrinkle candidates. Therefore, it is possible to increase a possibility that the front and back wrinkle candidates associated with each other are the same wrinkle. At this time, the nearest neighbor specifying unit 131 uses a distance biased to the longitudinal coordinate y as the distance d between the eyelid candidates when performing the nearest neighbor search. Therefore, for example, even when the same wrinkles are front and back and are greatly displaced in the longitudinal direction than the width direction of the metal band F, and the deviation Δy is larger than the deviation Δx, the both wrinkle candidates can be accurately associated. Is possible.

  Then, the determination unit 132 can determine at least one of the type and degree of the wrinkle candidate based on the feature information of the double-sided wrinkle candidates that are associated with high accuracy. Therefore, it is possible to more accurately determine the type of soot or the like as compared with the case where the determination is performed using the feature information of any one of the eyelid candidates. In particular, wrinkles with a high degree of harmfulness often appear as wrinkle candidates on both sides, and the detection accuracy of such wrinkles with a high degree of harmfulness can be dramatically improved. In addition, in the case of a point-shaped push rod, the determination unit 132 includes the feature information of the double-sided eyelid candidates, such as a condition that “they are the same eyelids only when the deviations Δx and Δy are approximately equal to the width and length of the eyelids” A condition for the type or degree of the wrinkle candidate based on the above can be added to the classification algorithm, and it is possible to accurately detect not only high-poisoning wrinkles but also low-poisoning wrinkles.

<3. Second Embodiment>
Next, the surface defect inspection apparatus 20 according to the second embodiment of the present invention will be described.
In the surface defect inspection apparatus 10 according to the first embodiment, the case has been described in which the determination unit 132 that performs determination based on the feature information of the double-sided wrinkle candidates is arranged on the upstream inspection apparatus 100 side. However, the determination unit 132 that performs such double-sided determination may be disposed on the downstream inspection apparatus 200 side. Accordingly, the closest specifying unit 131 may also be disposed on the downstream inspection device 200 side, and conversely, the determination unit 232 that performs single-side determination may be disposed on the upstream inspection device 100 side. The surface defect inspection apparatus 20 according to the second embodiment will be described as an example in the case where both-side determination is performed on the downstream inspection apparatus 200 side in this way. However, since the basic configuration and the like have many portions that overlap with the first embodiment, the following description will focus on differences from the first embodiment.

(3-1. Configuration of surface defect inspection apparatus)
FIG. 7 is an explanatory diagram for explaining the configuration of the surface defect inspection apparatus according to the second embodiment of the present invention.

  As shown in FIG. 7, the surface defect inspection apparatus 20 according to the present embodiment is basically formed in the same manner as the surface defect inspection apparatus 10 according to the first embodiment. However, the closest specifying unit 131 and the determination unit 132 that are arranged in the upstream inspection device 100 are arranged in the downstream inspection device 200, while the determination unit 232 that is arranged in the downstream inspection device 200 is an upstream inspection. Arranged in the apparatus 100. Furthermore, instead of the two memories, the delay memory 110 and the reception buffer memory 120, which are arranged in the upstream inspection device 100 in the first embodiment, in the present embodiment, one memory and a delay memory 310 are provided in the downstream inspection device 200. Be placed.

  In the present embodiment, the determination unit 232, the nearest neighbor specifying unit 131, and the determination unit 132 whose arrangement positions have been changed basically perform the same operations as in the first embodiment.

  That is, the determination unit 232 is an example of the other-side determination unit, and determines at least one of the type and the degree of the wrinkle candidate on the upper surface side based on the feature information on the upper surface side (an example of the other surface in the present embodiment). To do. The determination result is stored in the determination result storage device 300 and the delay memory 310.

  Further, the nearest neighbor specifying unit 131 acquires the feature information of the upstream eyelid candidate stored in the delay memory 310 and the feature information of the downstream eyelid candidate extracted by the feature extracting unit 203. Then, the nearest neighbor specifying unit 131 performs a nearest neighbor search based on both feature information, and specifies the closest other eyelid candidate to the one side eyelid candidate. At this time, unlike the first embodiment, the nearest neighbor specifying unit 131 is a downstream side candidate, that is, a lower side candidate (an example of a second side candidate, and one side candidate) As an example, the upstream cocoon candidate closest to the cocoon candidate, that is, the top cocoon candidate (an example of the other cocoon candidate) is specified.

  The determination unit 132 is an example of a one-side determination unit, and based on not only the feature information of the downstream cocoon candidate but also the upstream cocoon candidate feature information associated with the cocoon candidate. At least one of the type and the degree of the side wrinkle candidate is determined. In other words, the determination unit 132 according to the first embodiment determines the upstream wrinkle candidate based on the feature information of the front and back surface wrinkle candidates, but the determination unit 132 according to the present embodiment similarly applies to the front and back surface wrinkle candidates. On the basis of the feature information of the nodule candidate, the next nodule candidate is determined.

  On the other hand, in such a configuration, the surface defect inspection apparatus 20 according to the present embodiment includes a delay memory 310 in the downstream inspection apparatus 200 instead of the delay memory 110 and the reception buffer memory 120.

  The delay memory 310 serves not only as the delay memory 110 of the first embodiment but also as the reception buffer memory 120. That is, the delay memory 310 holds the upstream cocoon candidate feature information during the reception buffer period until the upstream cocoon candidate is associated with the downstream cocoon candidate by the nearest neighbor specifying unit 131. It will be.

  The operation of the delay memory 310 as a reception buffer memory will be conceptually described with reference to FIG. FIG. 8 is an explanatory diagram for explaining the nearest neighbor search process by the surface defect inspection apparatus according to the present embodiment.

  The feature information (position information) of the wrinkle candidate extracted from the feature extraction unit 103 of the upstream inspection device 100 is stored in the delay memory 310 as in the first embodiment. The concept of the stored upstream cocoon candidate Du is shown in FIG. On the other hand, the imaging unit 201 of the downstream inspection apparatus 200 is arranged offset with respect to the imaging unit 101 by an interval ΔL in the transport direction. Accordingly, when the time point at which the candy candidate Du is stored is “upstream detection time Tu”, the time point “downstream detection time Td” at which the feature information of the downstream candy candidate Dd corresponding to the point is extracted is the interval ΔL. Delay. Therefore, the delay memory 310 holds the upstream wing candidate Du for a delay period corresponding to at least the interval ΔL. An area of the delay memory 310 corresponding to this delay period is shown as a delay area in FIG. On the other hand, in this embodiment, unlike the first embodiment, the downstream eyelid candidate Dd is used as a reference for performing the nearest neighbor search. Therefore, the delay memory 310 also holds the upstream eyelid candidate Du within the predetermined search range L before and after the transport direction including the position O corresponding to the eyelid candidate Dd with respect to the downstream eyelid candidate Dd. . Therefore, a reception buffer area is secured in the delay memory 310 in addition to the delay area. In terms of the time axis, this reception buffer area corresponds to a reception buffer period corresponding to the search range L including the position O serving as a search reference. Therefore, the delay memory 310 holds the feature information of the upstream cocoon candidate Du not only in the delay period corresponding to the interval ΔL but also in the reception buffer period corresponding to the search range L.

  Then, when the feature extraction unit 203 extracts the feature information of the downstream side candidate Dd at the downstream detection time Td, the closest search unit 131 stores the feature information (position information) in the search range L of the delay memory 310. Based on the feature information of the upper wing candidate Du on the upper side, the upstream wing candidate Du closest to the downstream wing candidate Dd is specified.

(3-2. Operation of surface defect inspection apparatus)
The configuration of the surface defect inspection apparatus 20 according to the second embodiment of the present invention has been described above. Next, the operation of the surface defect inspection apparatus 20 according to the second embodiment of the present invention will be described with reference to FIG. FIG. 9 is an explanatory diagram for explaining the operation of the surface defect inspection apparatus according to the present embodiment.

  As shown in FIG. 9, the upstream inspection apparatus 100 first processes step S <b> 101 in the state where the metal band F to be inspected is conveyed in the longitudinal direction, and images the upper surface of the metal band F. In step S103, the image processing unit 102 performs predetermined image processing to extract a wrinkle candidate, and in step S105, the feature extraction unit 103 extracts feature information. In step S107, the feature information is recorded in the delay memory 310 in the downstream inspection apparatus 200, unlike the first embodiment. Further, unlike the first embodiment, the single-sided determination in step S117 is processed on the upstream inspection apparatus 100 side, and the determination result is stored in the delay memory 310 in step S223 and determined in step S141. It is stored in the result storage device 300.

  On the other hand, on the downstream inspection device 200 side, when the part of the metal band F imaged in step S101 reaches the imaging unit 201 within the imaging range, the imaging unit 201 is in metal in step S111, as in the first embodiment. The lower surface of the band F is imaged. In step S113, the image processing unit 202 performs predetermined image processing to extract a wrinkle candidate. In step S115, the feature extraction unit 203 extracts feature information. Here, step S121 (recording in the reception buffer memory) processed in the first embodiment is not processed, and steps S131 and S133 are processed on the downstream inspection apparatus 200 side. In this step S131, the nearest neighbor specifying unit 131 of the downstream inspection device 200 specifies the upstream eyelid candidate closest to the downstream eyelid candidate and associates them, and in step S133, the determination of the downstream inspection device 200 is performed. The unit 132 determines at least one of the type and the degree of the downstream kite candidate based on the feature information of the downstream kite candidate and the upstream kite candidate and the determination result of the upstream kite candidate. Hereinafter, as in the first embodiment, the determination result is recorded in the determination result storage device 300 in step S143, and the determination result in steps S117 and S133 is displayed on the display screen by the display device 400 in step S145. The

(3-3. Examples of effects according to second embodiment)
The surface defect inspection apparatus 20 according to the second embodiment of the present invention has been described above.
According to the surface defect inspection apparatus 20, it is possible not only to achieve the same effects as the surface defect inspection apparatus 10 according to the first embodiment, but also to omit the reception buffer memory 120. Can do. Therefore, the device configuration can be facilitated and the manufacturing cost can be reduced.

<4. Third Embodiment>
Next, a surface defect inspection apparatus 30 according to the third embodiment of the present invention will be described.
In the surface defect inspection apparatus 10 according to the first embodiment described above, a case will be described in which the upstream (that is, the upper surface side) defect candidate is determined from the feature information on both sides, and the surface defect inspection apparatus 20 according to the second embodiment described above. In the above description, the case where the downstream (that is, the lower surface) eyelid candidate is determined from the feature information on both sides has been described. Note that in both embodiments, the face wrinkle candidate for which the double-sided determination is not performed is determined based only on the single-side feature information. However, it is also possible to make determination on both sides using the feature information on both sides. Therefore, as an example of this case, a surface defect inspection apparatus 30 according to the third embodiment will be described. However, since the basic configuration and the like have many portions that overlap with the first embodiment and the second embodiment, the following description will focus on differences from the first embodiment and the second embodiment.

(4-1. Configuration of surface defect inspection apparatus)
FIG. 10 is an explanatory diagram for explaining the configuration of the surface defect inspection apparatus according to the third embodiment of the present invention.

  As shown in FIG. 10, the surface defect inspection apparatus 30 according to the present embodiment basically includes the configuration of the surface defect inspection apparatus 10 according to the first embodiment and the configuration of the surface defect inspection apparatus 20 according to the second embodiment. It has a structure that is added together. However, since both sides are determined based on the feature information on both sides, the determination unit 232 that performs determination on one side is omitted. In addition, since the determination based on the feature information on both sides is performed in parallel, the transfer of the determination result between the upstream inspection apparatus 100 and the downstream inspection apparatus 200 is omitted for each determination. However, when providing a time difference between the determination on the upstream inspection device 100 side and the determination on the downstream inspection device 200 side, it is of course possible to transfer the determination result of the preceding determination to the processing side with a delay.

  In this embodiment, since the determination result is omitted, the determination by the determination units 132A and 132B is performed based on the feature information on both sides, and the determination result on the other side is not used in this determination. Therefore, the classification algorithm of the determination units 132A and 132B is constructed based on the feature information on both sides. The determination units 132A and 132B correspond to the determination unit 132 of the first embodiment and the determination unit 132 of the second embodiment, respectively, and perform the same operation except that the determination result on the other side is not used. Do. In the present embodiment, the determination units 132A and 132B are examples of one-side determination units or other-side determination units, respectively. Similarly, the nearest neighbor specifying units 131A and 131B respectively correspond to the nearest neighbor specifying unit 131 of the first embodiment and the nearest neighbor specifying unit 131 of the second embodiment, and perform the same operation.

(4-2. Operation of surface defect inspection apparatus)
The configuration and the like of the surface defect inspection apparatus 30 according to the third embodiment of the present invention have been described above. Next, the operation of the surface defect inspection apparatus 30 according to the third embodiment of the present invention will be described with reference to FIG. FIG. 11 is an explanatory diagram for explaining the operation of the surface defect inspection apparatus according to the present embodiment.

  As shown in FIG. 11, in the surface defect inspection apparatus 30 according to this embodiment, the operation in the first embodiment shown in FIG. 4 and the operation in the second embodiment shown in FIG. 9 are combined. I do. However, in this case, step S117 for performing single-sided determination and steps S123 and S223 for recording the determination result by single-sided determination on the other side are omitted. Further, step S131 for performing the nearest neighbor search process, step S133 for performing double-sided determination, and step S143 for recording the determination result are performed for both the upstream inspection apparatus 100 and the downstream inspection apparatus 200, respectively. Are described separately from steps S131A, S131B, steps S132A, S132B, and steps S143A, S143B. The specific contents of each process have been described in detail in the first embodiment and the second embodiment, and will be omitted.

(4-3. Examples of effects according to third embodiment)
The surface defect inspection apparatus 30 according to the third embodiment of the present invention has been described above.
According to the surface defect inspection apparatus 30, in addition to being able to achieve the same operational effects and the like that the surface defect inspection apparatus 10 according to the first embodiment has, it is possible to provide either the upstream or the downstream side. Not only can the wrinkle candidate be determined based on the feature information on both sides, but also the wrinkle candidate on the other side can be determined based on the feature information on both sides.

  As mentioned above, although preferred embodiment of this invention was described in detail, referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to this example. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

  For example, in each of the embodiments described above, the case where the imaging unit 201 of the downstream inspection device 200 is disposed downstream by the interval ΔL from the imaging unit 101 of the upstream inspection device 100 has been described. However, it is also possible to place both the imaging units 101 and 201 at the same position in the transport direction and simultaneously capture images. In this case, the delay memory and the reception buffer memory can be appropriately increased / decreased to achieve the same operational effects as the first to third embodiments. However, as shown in FIG. 5, in order to reduce the flapping or vibration of the metal band F and increase the detection accuracy, it is desirable to take an image at a position where the metal band F is in contact with the roll R. For example, as described in the first to third embodiments, it is desirable to perform both-side imaging that is offset in the conveyance direction.

  In each of the above embodiments, the case where a predetermined classification algorithm is used for the determination by the determination unit 132 or the like has been described. This classification algorithm has a learning function based on teacher information, and the classification algorithm is automatically constructed as a learning result. However, of course, users such as inspectors may use a classification algorithm according to rules that are artificially determined based on experience and experimental results, or may combine this artificial algorithm with the above learning result algorithm. Is possible.

  In each of the above embodiments, for convenience of explanation, each configuration has been described separately for the upstream inspection device 100 and the downstream inspection device 200, but it is needless to say that this division is not necessarily required.

  In addition, the series of processes described in the above embodiments may be executed by dedicated hardware, but may be executed by software. When the series of processes is performed by software, the above series of processes can be realized by causing a general-purpose or dedicated computer to execute the program. The computer includes a CPU (Central Processing Unit), a recording device such as a HDD (Hard Disk Drive), a ROM (Read Only Memory), a RAM (Random Access Memory), a LAN (Local Area Network), etc. A communication device, an input device such as an imaging device / mouse / keyboard, a magnetic disk such as a flexible disk, an optical disk such as various CDs (Compact Discs), MO (Magneto Optical) disks, DVDs (Digital Versatile Discs), etc. Drives that read and write removable storage media such as semiconductor memory, and display devices such as monitors, speakers, and headphones And an output device such as an audio output device. The computer may execute the above-described series of processes by executing a program recorded in a recording device / removable storage medium, a program acquired via a network, or the like.

  In this specification, the steps described in the flowcharts are executed in parallel or individually even if they are not necessarily processed in time series, as well as processes performed in time series in the described order. Including processing to be performed. Further, it goes without saying that the order can be appropriately changed even in the steps processed in time series.

10, 20, 30 Surface defect inspection device 100 Upstream inspection device 200 Downstream inspection device 300 Determination result storage device 400 Display device 101 Imaging unit 102 Image processing unit 103 Feature extraction unit 110, 310 Delay memory 120 Reception buffer memory 131, 131A, 131B Closest-identification part 132,132A, 132B determination part 201 imaging part 202 image processing part 203 feature extraction part 232 determination part Ar same range Du, Dd, Dd1, Dd2 疵 candidates F metal band x width direction coordinate y longitudinal direction coordinate Δx, Δy deviation d distance ΔL interval

Claims (11)

A surface defect inspection apparatus for inspecting wrinkles formed on the surface of a metal strip to be conveyed,
A first imaging unit that is disposed on the first surface side, which is one of both surfaces of the metal band, and images the first surface;
A first extraction unit that extracts feature information including at least position information indicating a longitudinal position and a width direction position of the eyelid candidate on the first surface side based on a first image captured by the first imaging unit;
A second imaging unit that is disposed on the second surface side, which is the other surface of both surfaces of the metal band, and images the second surface;
A second extraction unit that extracts feature information including at least position information indicating the position of the eyelid candidate on the second surface side based on the second image captured by the second imaging unit;
Based on the position information of the wrinkle candidates on the first surface side and the second surface side, the wrinkle on the other side closest to the wrinkle candidate on either the first surface side or the second surface side A closest identification unit that identifies candidates and associates the candidates with each other as closest candidates;
Based on the feature information of the one-side eyelid candidate and the feature information of the other-side eyelid candidate associated with the one-side eyelid candidate, at least one of the type and degree of the one-side eyelid candidate is determined. A one-side determination unit that
A surface defect inspection apparatus characterized by comprising:   The one-side determination unit, as the feature information of each of the one-side and other-side eyelid candidates associated with each other as the nearest neighbor candidate, includes at least a longitudinal position shift amount and a width direction position of the eyelid candidates. The surface defect inspection apparatus according to claim 1, wherein at least one of the type and degree of the one-side wrinkle candidate is determined based on a deviation amount. The feature information of the wrinkle candidate includes feature amount information indicating the feature amount of the wrinkle candidate that can be extracted from the first image or the second image.
The one-side determining unit, as the feature information of each of the one-side and other-side cocoon candidates associated with each other as the nearest neighbor candidate, based on at least the feature amount information of each of the heel candidates, The surface defect inspection apparatus according to claim 1, wherein at least one of the type and the degree of the wrinkle candidate is determined. The metal strip is conveyed in the longitudinal direction,
The closest identification unit calculates a distance between the one side candidate and the plurality of other side candidates on a space where the unit distance in the longitudinal direction is longer than the unit distance in the width direction. The surface defect inspection apparatus according to any one of claims 1 to 3, wherein the other wrinkle candidate having the shortest distance is specified and the wrinkle candidates are associated with each other as the closest wrinkle candidates. .   5. The apparatus according to claim 1, further comprising an other side determination unit that determines at least one of a type and a degree of the other side cocoon candidate based on the feature information of the other side cocoon candidate. The surface defect inspection apparatus described in 1.   The one-side determination unit further uses the determination result for the other-side cocoon candidate by the other-side determination unit, and associates the one-side cocoon candidate with the other-side cocoon candidate as the nearest-neighbor candidate The surface defect inspection apparatus according to claim 5, wherein at least one of the type and the degree of the determination is determined.   In addition to the feature information of the other-side cocoon candidate, the other-side determination unit is further based on the feature information of the one-side cocoon candidate associated with the other-side cocoon candidate as the closest candidate The surface defect inspection apparatus according to claim 5, wherein at least one of a type and a degree of the other side defect candidate is determined. The second imaging unit images the second surface on the downstream side in the transport direction from the imaging position of the first imaging unit,
The surface defect inspection apparatus includes at least the first extraction during a delay period that is a time interval required when the metal strip is transported from the imaging range of the first imaging unit to the imaging range of the second imaging unit. The surface defect inspection apparatus according to claim 1, further comprising a delay memory capable of recording feature information of the first surface side defect candidate extracted by a section.   The surface defect inspection apparatus has at least a maximum wrinkle length assumed as an inspection target, and a reception buffer period that is a time interval required for transporting the metal band, for the one side wrinkle candidate. The surface defect according to claim 8, further comprising: a reception buffer memory capable of recording feature information as a search target of the nearest neighbor candidate that the nearest neighbor specifying unit associates with the other side candidate. Inspection device. The nearest neighbor specifying unit specifies the closest eyelid candidate on the first surface side relative to the second surface side candidate, and associates the candidate candidates with each other as the closest candidate,
The delay memory also serves as the reception buffer memory, and not only during the delay period but also at least during the reception buffer period, the nearest neighbor specifying unit displays the feature information of the heel candidate on the first surface side. The surface defect inspection apparatus according to claim 9, wherein the surface defect inspection apparatus can be recorded as a search target of the nearest neighbor candidate associated with the second face side candidate. A surface defect inspection method for inspecting wrinkles formed on the surface of a metal strip to be conveyed,
A first imaging step in which a first imaging unit arranged on the first surface side, which is one of both surfaces of the metal band, images the first surface;
A first extraction step of extracting feature information including at least position information indicating a longitudinal position and a width direction position of the eyelid candidate on the first surface side based on the first image captured in the first imaging step;
A second imaging step in which a second imaging unit arranged on the second surface side, which is the other surface of both sides of the metal strip, images the second surface;
A second extraction step of extracting feature information including at least position information representing the position of the eyelid candidate on the second surface side based on the second image imaged in the second imaging step;
Based on the position information of the wrinkle candidates on the first surface side and the second surface side, the wrinkle on the other side closest to the wrinkle candidate on either the first surface side or the second surface side A closest identification step of identifying candidates and associating the candidate candidates with each other as closest candidate candidates;
Based on the feature information of the one-side eyelid candidate and the feature information of the other-side eyelid candidate associated with the one-side eyelid candidate, at least one of the type and degree of the one-side eyelid candidate is determined. A one-side determination step,
A surface defect inspection method characterized by comprising:

Images