KR101328266B1 - Apparatus for detecting surface crack of hot slab - Google Patents

Abstract

An apparatus for detecting surface cracks of slabs is provided. In the apparatus for detecting the surface cracks of the slab, in the apparatus for detecting surface cracks in the vertical direction from the slab image, a garber filter for filtering the slab image, and generating paths by the number of each column of the filtered slab image, A candidate path selecting module that selects a path whose cost value, which is a sum of pixel values of pixels included in the corresponding path, among the generated paths is smaller than a cost value of an adjacent path, and the selected candidate path and the candidate path; A feature extraction module for extracting features for a region of interest (ROI) including a crack, and determining whether a candidate path is a surface crack by applying a SVM (Support Vector Machine) as an input vector to the extracted features. It may include a detection module.

Description

Surface crack detection device of hot slab {APPARATUS FOR DETECTING SURFACE CRACK OF HOT SLAB}

The present invention relates to an apparatus for detecting surface cracks in hot slabs.

In general, continuous casting is a process of producing slabs of a constant size by receiving molten steel produced in a steelmaking furnace and transferred to a ladle in a tundish and then supplying it to a casting furnace for a continuous casting machine.

In the slab manufactured by the above-described process, as shown in Figure 1, the surface crack (Surface Crack, SC) of the elongate shape in the conveying direction of the slab may be generated, the generated surface crack (SC) is a post-process Will cause a defect.

Therefore, the surface crack SC needs to be detected in advance so as not to go to a later process. However, in the case of a hot slab having a high temperature, the surface crack SC is not easily visually detected.

According to one Embodiment of this invention, the apparatus which can detect the surface crack which exists in a hot slab is provided.

According to the first embodiment of the present invention,

In the device for detecting the surface crack in the vertical direction from the slab image,

A garber filter for filtering the slab image;

Paths are generated as many as the number of columns of the filtered slab image, and a path whose cost value, which is a sum of pixel values of pixels included in the corresponding path among the generated paths, is smaller than a cost value of an adjacent path. A candidate path selection module for selecting as;

A feature extraction module for extracting features for the selected candidate path and a region of interest (ROI) including the candidate path; And

By applying the SVM (Support Vector Machine) to the extracted features as an input vector, it provides a surface crack detection device of the slab including a crack detection module for determining whether the candidate path is a surface crack.

According to an embodiment of the present invention, the Gabor filter,

Filtering in the vertical direction of the slab image.

According to an embodiment of the invention, the candidate path selection module,

For each column of the filtered slab image, the path may be generated by connecting a pixel having the smallest pixel value among the three pixels of the previous row adjacent to the current pixel.

According to an embodiment of the invention, the above features,

Features extracted from the histogram of the candidate path and the region of interest, and features extracted from the discrete wavelet transform of the region of interest.

According to an embodiment of the invention, the candidate path selection module,

Binarizing the Gabor filtered slab image with an average value of pixels included in the candidate path,

The binarization blobs other than the binary blobs included in the binarized candidate paths may be removed.

According to the second embodiment of the present invention, in the method for detecting the surface crack in the vertical direction from the slab image,

A filtering step of filtering the slab image using a garber filter;

Paths are generated as many as the number of columns of the filtered slab image, and a path whose cost value, which is the sum of pixel values of pixels included in the corresponding path among the generated paths, is smaller than the cost value of the adjacent path as a candidate path. Selecting a candidate path;

Extracting features of the selected candidate path and regions of interest (ROI) including the candidate path; And

The crack detection step of determining whether the candidate path is a surface crack by applying the SVM (Support Vector Machine) as the input vector to the extracted features

It provides a surface crack detection method of the slab comprising a.

According to an embodiment of the invention, the filtering step,

The method may include filtering in a vertical direction of the slab image.

According to an embodiment of the present invention, the candidate path selecting step includes:

For each column of the filtered slab image, the method may include generating the path by connecting a pixel having a smallest pixel value among three pixels of a previous row adjacent to the current pixel.

According to an embodiment of the invention, the above features,

Features extracted from the histogram of the candidate path and the region of interest, and features extracted from the discrete wavelet transform of the region of interest.

According to an embodiment of the present invention, the candidate path selecting step includes:

Binarizing the Gabor-filtered slab image with an average value of pixels included in the candidate path; And

The method may further include removing the binarization blobs except for the binary blobs included in the binarized candidate path.

 According to one embodiment of the invention, by using an image processing algorithm, it is possible to perform a full inspection of the hot slab that is hard to visually check, and to significantly improve the detection capability of the surface cracks.

1 is a view showing surface cracks formed in a hot slab.
2 is a block diagram of an entire system including a surface crack detection device of a hot slab according to an embodiment of the present invention.
3A is a diagram for explaining a Gabor filter.
FIG. 3B is a diagram illustrating a slab image having surface cracks and an image after applying a Gabor filter to the slab image according to one embodiment of the present invention.
4 is a view for explaining a process of selecting a candidate path according to an embodiment of the present invention.
5 illustrates candidate paths and regions of interest selected according to an embodiment of the present invention.
6 is a view for explaining a process for removing the scale in addition to the surface cracks according to an embodiment of the present invention.
It is a figure for demonstrating the surface crack detection method of the hot slab which concerns on one Embodiment of this invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the embodiments of the present invention can be modified into various other forms, and the scope of the present invention is not limited to the embodiments described below. The shape and the size of the elements in the drawings may be exaggerated for clarity and the same elements are denoted by the same reference numerals in the drawings.

2 is a block diagram of an entire system including a surface crack detection device of a hot slab according to an embodiment of the present invention.

As shown in FIG. 2, the entire system is disposed along the circumference of the slab S in a direction perpendicular to the direction of movement of the lighting device L and the slab S, and photographs an image of the hot slab S. Surface crack detection device for detecting the surface cracks of the hot slab (S) by processing the plurality of cameras (C) for transmitting the captured image to the defect detection device 100 and the image received from the plurality of cameras (C) 200 may be included.

In addition, the entire system may further include a speedometer V for measuring the moving speed of the hot slab S and a distance meter D for measuring the moving distance.

Hereinafter, the surface crack detection apparatus 200 according to an exemplary embodiment of the present invention will be described in detail with reference to FIGS. 2 to 6. 3A is a diagram for explaining a Gabor filter, and FIG. 3B is a diagram illustrating a slab image having surface cracks and an image after applying Gabor filters to the slab images according to one embodiment of the present invention. 4 is a diagram illustrating a process of selecting candidate paths according to an embodiment of the present invention, and FIG. 5 is a view illustrating candidate paths and regions of interest selected according to an embodiment of the present invention. Finally, Figure 6 is a view for explaining a process for removing the scale in addition to the surface cracks according to an embodiment of the present invention.

First, as shown in FIG. 2, the surface crack detection apparatus 200 is a device for detecting surface cracks (Surface Crack, SC) in a vertical direction from a slab image, and a garber filter for filtering the slab image. A path is generated as many as the number of columns of the filtered slab image, and a cost value that is a sum of pixel values of pixels included in the corresponding path among the generated paths is smaller than a cost value of an adjacent path. A candidate path selection module 220 for selecting a candidate path as a candidate path, a feature extraction module 230 for extracting features of a region of interest (ROI) including the selected candidate path and the candidate path, and extracted By applying the support vector machine (SVM) as the input vector, the crack detection module 240 may determine whether the candidate path is a surface crack.

Specifically, the Gabor filter 210 is a filter capable of extracting components in a specific direction of the image. Specifically, as shown in FIG. 3A, when the input image 300 is Gabor filtered, components for various angles (θ = 0, ± 22.5, ± 45, ± 77.5, 90) from the input image 300 are obtained. Can be extracted.

According to the exemplary embodiment of the present invention, since the surface crack SC formed in the hot slab S is formed in the vertical direction of the slab image, the vertical component of the slab image may be extracted by the Gabor filter 210. .

Specifically, referring to FIG. 3B, it can be seen that by filtering Gabor filtering the slab image having the surface crack (see (a)), the filtered slab image as shown in (b) of FIG. 3B can be obtained. . The filtered slab image may be delivered to the candidate path selection module 220.

The candidate path selection module 220 selects a candidate path including the surface crack SC from the Gabor filtered slab image. In order to select candidate paths, embodiments of the present invention use a dynamic programming method. Dynamic programming is an algorithm used to gather partial solutions to find the total solution.

Specifically, as shown in FIG. 4, the candidate path selection module 220 moves pixels downward of each pixel in a vertical direction (row direction) with respect to each column of the Gabor filtered image (see (a)). The paths are generated for each column by comparing the values and then connecting the pixels with the smallest pixel values. For example, for the current pixel ((x, y)), three adjacent pixels in the previous row ((x, y-1) (x-1, y-1), (x + 1, y-1) ), The pixel having the smallest pixel value is selected. By this means, as many paths as there are columns can be generated.

Next, the candidate path selection module 220 obtains a cost value that is a sum of pixel values of pixels included in each of the paths. The cost value for each route thus obtained is as shown in FIG.

Subsequently, as illustrated in FIG. 4B, the candidate path selection module 220 may find portions 410, 420, and 430 in which the cost values of the respective paths are smaller than the cost values of the adjacent paths.

The candidate paths CP1, CP2, and CP3 selected for the small portions 410, 420, and 430 described above are as shown in FIG. 4C. The candidate paths CP1, CP2, CP3 may be one or more. The reason why the path having a small cost value is selected as the candidate path is that the characteristic of the surface crack SC is stronger than the surroundings and thus has a minimum value. The selected candidate paths CP1, CP2, and CP3 may be delivered to the feature extraction module 230.

Thereafter, the feature extraction module 230 may extract features of a region of interest (ROI) including the selected candidate paths CP1, CP2, and CP3 and the candidate paths.

In detail, the feature extraction module 230 may extract five features using histograms for the candidate paths CP1, CP2, and CP3. The five features may be mean, standard deviation, smoothness, third moment, and uniformity.

In addition, the feature extraction module 230 may extract the above-described five features for the ROI including the candidate paths CP1, CP2, and CP3. Here, the ROI may be an area including the candidate path CP, as shown in FIG. 5.

In addition, the feature extraction module 230 may extract a feature for frequency information using Discrete Wavelet Transformation (DCT) for the ROI including the candidate paths CP1, CP2, and CP3. have. Discrete wavelet transform can be performed up to 5 levels and then extract 5 components for vertical, horizontal, and diagonal.

Meanwhile, in order to distinguish between the narrow surface crack SC and the path formed along the boundary of the wide scale, the feature extraction module 230 may further extract the following features. That is, the feature extraction module 230, as shown in (a) of FIG. 6, the image filtered by the average value of the pixels included in the candidate alert (refer to (c), CP1, CP2, CP3 of FIG. 4). Binarize Then, the feature extraction module 230 removes the remaining binary blobs except for the binary blobs included in the candidate paths CP1, CP2, and CP3, as shown in FIG. 6B. The path formed by can be excluded. Thereafter, five features using the histogram as described above may be extracted with respect to the binarized image from which the path formed by the scale is removed.

As described above, the features based on the histogram extracted by the feature extraction module 230 and the features extracted based on the discrete wavelet transform may be transferred to the crack detection module 240.

Finally, the crack detection module 240 may determine whether the candidate path is a surface crack by applying a SVM (Support Vector Machine) to the extracted features as an input vector. Here, SVM (Support Vector Machine) is to convert the data into a space of a dimension higher than a given space, and then to obtain a linear hyperplane that distinguishes each class, and the candidate path selected from the features extracted by learning is a surface Whether or not it is a crack SC can be determined. Since the above-described SVM is a known algorithm, detailed description is omitted for simplicity of the invention.

Hereinafter, with reference to FIGS. 2-7, the surface crack detection method of the hot slab which concerns on one Embodiment of this invention is demonstrated. However, for simplicity of the invention, descriptions duplicated with those described in FIGS. 2 to 6 will be omitted.

2 to 7, the Gabor filter 210 may Gabor filter the slab image (S700). The filtered slab image may be delivered to the candidate path selection module 220. By Gabor filtering, vertical components of the slab image may be extracted.

Next, as shown in FIG. 4, the candidate path selection module 220 moves the pixels of adjacent pixels downward by one row in the vertical direction (row direction) with respect to each column of the Gabor filtered image (see (a)). The paths are generated for each column by comparing the values and then connecting the pixels with the smallest pixel values.

Next, the candidate path selection module 220 obtains a cost value that is a sum of pixel values of pixels included in each of the paths. The cost value for each route thus obtained is as shown in FIG.

Subsequently, the candidate path selection module 220 finds portions 410, 420, and 430 in which the cost value of each path is smaller than the cost value of the adjacent path, as shown in FIG. The paths of the portions 410, 420, and 430 having a small cost value may be selected as candidate paths CP1, CP2, and CP3 (S701). The selected candidate paths CP1, CP2, and CP3 are the feature extraction module 230. Can be delivered.

Thereafter, the feature extraction module 230 may extract features of a region of interest (ROI) including the selected candidate paths CP1, CP2, and CP3 and the candidate paths (S702). Here, the features may include features extracted from histograms of the candidate paths CP1, CP2, CP3 and the region of interest ROI, and features extracted from the discrete wavelet transform of the region of interest ROI. Additionally, the features may further include five features using the histogram as described above for the binarized image from which the path formed by the scale has been removed, as described above. The extracted features may be transferred to the crack detection module 240.

Finally, the crack detection module 240 may determine whether the candidate path is a surface crack by applying the SVM (Support Vector Machine) to the extracted features as an input vector (S703).

As described above, according to one embodiment of the invention, by using the image processing algorithm, it is possible to perform a full inspection of the hot slab which is hard to visually check, and there is a technical effect that can significantly improve the detection capability of the surface cracks.

The present invention is not limited by the above-described embodiment and the accompanying drawings. It is intended to limit the scope of the claims by the appended claims, and that various forms of substitution, modification and change can be made without departing from the spirit of the present invention as set forth in the claims to those skilled in the art. Will be self explanatory.

200: surface crack detection device 210: Gabor filter
220: candidate path selection module 230: feature extraction module
240: crack detection module CP, CP1, CP2, CP3: candidate path
ROI: area of interest

Claims (5)

In the device for detecting the surface crack in the vertical direction from the slab image,
A garber filter for filtering the slab image;
Paths are generated as many as the number of columns of the filtered slab image, and a path whose cost value, which is a sum of pixel values of pixels included in the corresponding path among the generated paths, is smaller than a cost value of an adjacent path. A candidate path selection module for selecting as;
A feature extraction module for extracting features for the selected candidate path and a region of interest (ROI) including the candidate path; And
A crack detection module for determining whether the candidate path is a surface crack by applying a SVM (Support Vector Machine) to the extracted features as an input vector
Surface crack detection device of the slab comprising a.
The method of claim 1,
The Gabor filter,
Surface crack detection device of the slab for filtering in the vertical direction of the slab image.
The method of claim 1,
The candidate path selection module,
And generating a path for each column of the filtered slab image by connecting the pixel having the smallest pixel value among the three pixels of the previous row adjacent to the current pixel.
The method of claim 1,
The above features,
And a feature extracted from the histogram of the candidate path and the region of interest, and a feature extracted from the discrete wavelet transform of the region of interest.
The method of claim 1,
The candidate path selection module,
Binarizing the Gabor filtered slab image with an average value of pixels included in the candidate path,
Surface crack detection device of the slab to remove the binarization blobs other than the binary blob (binary blob) included in the binarized candidate path.

Images