JPH09184812A - Flaw detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To sufficiently suppress over-detection and to enhance detection accuracy in the automatic inspection of a surface flaw. SOLUTION: An article to be inspected fed from a feed line is irradiated with light and the surface image of the article to be inspected is obtained from the reflected light to detect the surface flaw of the article to be inspected. Two-dimensional weighting moving average operation is applied to the inputted surface image to calculate a primary processing image and weighting short-time mutual correlation operation is applied thereto to obtain a flaw emphasized image and this image is biinarized on the basis of a threshold value to detect a surface flaw.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a defect detecting method used for detecting harmful defects on the surface of a steel material, for example.

[0002]

2. Description of the Related Art Recently, hot strip charging of a continuous casting slab into a rolling mill has been carried out as a part of energy saving. In this implementation, 1) determination of whether hot strip charging is possible and 2) For the purpose of feeding back the slab surface flaw generation state to the continuous casting operation, slab surface flaws are automatically detected while hot. As this surface flaw detection method, for example, as proposed in Japanese Patent Laid-Open No. 61-34447, the surface of the object to be inspected is irradiated with illumination light and the density distribution of the reflected light is converted into an electric signal. There is a method of obtaining an emphasized image by so-called weighted moving average calculation, which averages every predetermined section and outputs a smoothed signal, and detects the surface flaw based on this. In addition, JP-A 1-2536
As proposed in Japanese Patent Laid-Open No. 39, by obtaining a differential image of the electric signal, shifting the differential signal by a predetermined amount to obtain a differential image, and obtaining an emphasized image by the sum of absolute values of the differential signal and the differential image before shifting. There is a method of detecting the surface flaw.

[0003]

However, the above-mentioned Japanese Patent Laid-Open Nos. 61-34447 and 1-253639.
In any of the publications, the property of the surface of the material to be inspected is not uniform due to, for example, wrinkles in the shrinkage portion due to width reduction, rolling pattern, scale, etc. If there is, the amount of change in the density of the normal part in the surface image may be larger than the amount of change in the density of the defective part, and it is possible to distinguish between the defective part and the non-defective part only by using the weighted moving average or the differential image. Becomes difficult. As a result, there are problems such as excessive detection and poor reliability. The present invention has an object to sufficiently suppress over-detection and improve detection accuracy in automatic inspection of surface defects.

[0004]

In order to solve the above problems, the present invention provides a defect detection for obtaining a surface image of an object to be inspected and image-processing the surface image to detect a surface defect of the object to be inspected. In the method, a plurality of short-time cross-correlation functions with different delays are obtained for two signal sequences obtained by subjecting the surface image to two types of two-dimensional weighted moving averages, and the weighted sum of these is calculated as the density of each pixel. A defect detection method in which an image configured as a value is obtained, and the image is thresholded to detect a defect. Further, for two signal sequences obtained by performing two types of two-dimensional weighted moving averages on the surface image. , Is a defect detection method in which both signals are multiplied to obtain an image in which only a defective portion is emphasized, and this is subjected to threshold processing to detect a defect.

[0005]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention has an image processing operation flow as shown in FIG. First, for the input surface image, there are two processing systems of two-dimensional weighted moving average,
In each processing system, so-called spatial filter calculation shown in the following equation (1) is performed to obtain primary processed images x ₁ and x ₂ .

[0006]

[Equation 1]

However, u (i, j): density at the (i, j) position of the input image x _k (i, j): density at the (i, j) position of the primary processed image b _k (i, j) : Two-dimensional moving average weighting factor (i,
j) Element M, N: 1/2 K of the degree of the two-dimensional weighting coefficient in the horizontal / vertical directions: Primary processing sequence number By this operation, the waveform portion similar to the pattern of the weighting coefficient b _k (i, j) is emphasized. To be done.

Next, between the two primary processed images,
The short-time cross-correlation function calculation shown in the following equation (2) is performed,

[0009]

[Equation 2]

However, x ₁ (i, j): (i, j) of the primary processed image of the first system
Density at position x ₂ (i, j): (i, j) of the primary processed image of the second system
Position Density y (i, j): Density at (i, j) Position of Defect Emphasized Image w (i, j): (i, j) Element of Short-Time Cross-Correlation Calculation Weighting Factor K, L: Short Time 1/2 of the horizontal-vertical direction order of the cross-correlation function weighting coefficient Further, the defect enhancement image y (i, j) obtained here is binarized with an appropriate threshold value to detect a defect and determine an area. To do.

In the present invention, the weighting factor b _k (i, j) in the two-dimensional moving average calculation is of the linear phase type in order to make the group delay of the signal constant, and the defect to be inspected. Smoothing is performed in the direction parallel to the representative direction, and incomplete differentiation is performed in the vertical direction. For example, for the detection of the defect whose density distribution of the surface image is shown in FIG. 2, if the weighting coefficient b _k (i, j) as shown in FIG. 3 is used, the result of the two-dimensional weighted moving average calculation is the primary output image. The concentration distribution of is as shown in FIG.

If one line perpendicular to the defect within the defect range is taken out from FIG. 4, the waveform becomes as shown in FIG. 5, and the sudden change portion of the density gradient around the defect appears as positive and negative peaks in the primary processed image. The positional deviation between the positive and negative peaks is measured, and the weight w of the short-time autocorrelation function is measured as shown in FIG.
By setting (i, j) to have a negative peak at the position shift amount position and obtaining the short-time autocorrelation in FIG. 4, a large positive value is obtained only at the defect position as shown in FIG. It is possible to generate an image in which only the signal, that is, the defective portion is emphasized.

Therefore, a short-time autocorrelation of the same type as that obtained by multiplying the histogram of the positional deviation amount of the positive and negative peaks of the defect portion by -1 by collecting surface images of many defects and analyzing each primary processed image By obtaining the short-time autocorrelation function using the function weight w (i, j), it becomes possible to obtain a defect-enhanced image in which all the defects can be detected. Moreover,
This defect-enhanced image has a large positive value only in the area where the image density gradient changes abruptly at a specified distance, so only minute areas such as defects are emphasized and accurate without being affected by density fluctuations over a wide area such as the background. It is possible to detect only the target defect.

When two types of primary processed image calculation means are provided, the weighting coefficient b _{1 in} each two-dimensional moving average calculation
When (i, j) and b ₂ (i, j) are detected, for example, in the defect of FIG. 2, b _b of FIG.
As shown in (i), a form aiming at smoothing, and as a vertical direction component, as shown in b _a1 (i) and b _a2 (i) in FIG. When set to a format shifted by the defect width, the waveforms of one line extracted from the primary output image are as shown in FIG. 10, and the weight w for calculating the short-time cross-correlation function of both is calculated. (I, j) is the same as 0 when j ≠ 0, and when j = 0, the horizontal axis of the positive / negative peak position deviation amount histogram of the defect portion is shifted in the negative direction by a typical defect width. If set to, one line of the defect-emphasized image is obtained as shown in FIG. 11, and it is possible to obtain the same result as when the short-time autocorrelation function is used.

Further, in the case of detecting only defects in which the size of the defect in the width direction is substantially uniform and the distribution of the positive and negative peak position deviation amounts of the defect portion is concentrated in a narrow range of 1 pixel or less. Is

[0016]

(Equation 3)

It is possible to create a defect-enhanced image by simply setting the following, that is, performing the multiplication between the images of the two primary output images.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be specifically described below with reference to the drawings showing the embodiments. In this embodiment, a vertical crack defect inspection is performed on the surface of the hot slab during transportation. FIG. 12 is a schematic diagram showing an implementation state of the defect detection method according to the present embodiment. The slab 1, which is the object to be inspected for detecting surface defects,
The transport table 2 transports the sheet from left to right toward the paper surface. An illumination light source 3 and a line sensor camera 4 are installed above the slab 1, and a surface image of the slab 1 can be obtained by the camera 4 by illumination from the light source 3. In addition, the camera 4 and the pulse signal of the encoder 6 mechanically connected to the driving device 5 of the transport table 2,
It is connected to an image processing device 7 that detects surface defects on the slab.

Next, the operation will be described. The surface image of one line of the slab 1 obtained by the camera 4 is sequentially sent to the image processing device 7 for each scan cycle, and the signal of the encoder 6 is used to store the two-dimensional image of the slab 1 in the memory of the image processing device 7. The surface image is developed, and image processing is performed by the procedure described below to detect the surface defect.

Since the slab to be inspected has been descaled in advance for the purpose of detecting surface defects, the surface defects existing there are almost visually observable in the surface image developed in the image processing apparatus 7, but the scale remains. , And the rolling pattern associated with the width rolling in the previous step, wrinkles in the shrinkage area near the center in the width direction, and oscillation marks at the time of casting by the continuous casting machine appear on the surface image. Therefore, the surface image developed in the image processing 7 has the same pixel density amplitude and density gradient as the defect portion not only in the defect but also in the disturbance such as scale remaining. However, it can be confirmed that the portion where the vertical crack defect exists has continuity of a valley waveform in the direction parallel to the defect and the width of the valley on the waveform is narrower than that of the other disturbance portion.

Each line of the surface image is input from the original signal input unit 11 of the image processing circuit shown in FIG. In the figure, 12 and 16 are FIR filter elements for performing a weighted moving average in the camera scanning direction, 13 and 17 are rotate memories for holding the scanning direction moving average results, and 14 and 1.
Reference numeral 8 is an FIR filter element for performing a weighted moving average in the slab length direction on the image stored in the rotating memory, and 15 and 19 are for making the sum of squares for each area of the calculation result image constant, that is,

[0022]

(Equation 4)

Where u (i, j): the density of the (i, j) element of the input image σ _u : a normalization processing unit for calculating the square root of the root mean square value of the input image, There are two systems in which these are connected in cascade, and each operates in parallel to output two primary processed images.

Next, for the two primary processed images obtained as a result of the above-mentioned calculation, both signals are calculated in the inter-image multiplication section 20, and a short-time cross-correlation function in the camera scanning direction is calculated using the FIR filter element 21. Calculation is performed, and this value is used as the density of each pixel to form a defect-emphasized image. Furthermore, this defect emphasized image is
The defect determination section 23 determines a defect based on the dimension information obtained by binarizing the area with a constant threshold and inputting the binarized area dimension to the measuring section 22. FI for performing one-dimensional weighted moving average calculation in the camera scanning direction and slab length direction
Cascade connection of the R filter elements 12 and 14 (or 16 and 18) via the rotation memory 13 (or 17) means that the moving average weighting factors in the camera scanning direction and the slab length direction are b _a ( i) and b _b (i),

[0025]

(Equation 5)

It is equivalent to performing a two-dimensional weighted moving average using the weighting factor of. Further, the normalization operation is performed by the normalization operation processing units 15 and 19 so that the surface texture of the slab is generally for each material (for example, the difference in compression ratio of width rolling) or for each place (for example, the thinning portion after width rolling). This is for suppressing that the absolute value of the two-dimensional weighted moving average calculation result fluctuates.

In the case of the vertical crack defect detection of this example, the moving average weighting coefficient in the camera scanning direction is b _a1 (i), b in FIG. 9 for the FIR filter elements 12, 16 in FIG. 13, respectively.
_a2 (i) is set, and the moving average weighting coefficient in the slab length direction is set to _bb (i) in FIG. 8 for both the FIR filter elements 14 and 18 in FIG. Two
One line around the defect in each of the primary processed images becomes x ₁ and x _{2 in} FIG. 10 (i), and both have a relatively large density value that is positive at the defect position. Next, the primary processed image x
_After multiplying the density values at the same pixel positions in ₁ and x _{2 for} all pixels in the image, the short-time cross-correlation function calculation weight w (i, j) is

[0028]

(Equation 6)

FIG. 11 is obtained when one line of the defect-emphasized image obtained by performing the arithmetic processing with the FIR filter 21 set to is obtained, and it is possible to obtain a large positive value only for the defective portion as compared with the other. Become.

Next, regarding a slab cast by a continuous casting machine and rough-rolled by a width rolling mill called a sizing mill and having surface defects (vertical cracks), this embodiment and the above-mentioned Japanese Patent Laid-Open No. 1-253639. The results of the surface defect investigation by and are shown in FIGS. 14 (a) and 14 (b). FIG. 14 (a) is the method proposed in Japanese Patent Application Laid-Open No. 1-253639 (conventional method), and FIG. 14 (b) is the method investigated in the present embodiment. As can be seen from the above, at the inspection level with an overdetection rate of 0%, the defect detection rate when the conventional method was used was about 20%, but in the present embodiment, it was improved to about 95%.

At the inspection level with a detection rate of 95%, the overdetection rate when the conventional method was used was 100% (that is, all images were judged to be defective).
In this example, it was possible to suppress the amount to 0%. Although only the vertical crack defect detection on the upper surface side of the slab is shown in this example, another illumination and camera are installed on the lower surface side of the slab, and the same signal processing is performed on the signal obtained here. In this case, it is possible to inspect both sides at the same time, and for detecting defects such as edge cracks and lateral cracks that occur in the slab length direction and the vertical direction, a weighting coefficient for moving average in the camera scanning direction is used. the b _b in FIG. 8 (i), respectively the moving average for the weighting coefficients of the slab longitudinal direction in each line, b _a1 (i) of FIG. 9, of course it can be coped with by a b _a2 (i) Is.

[0032]

According to the present invention, a two-dimensional weighted moving average, that is, a so-called spatial filter is applied to the original image, and then the weighted autocorrelation or cross-correlation is obtained using the information of the surface of the steel sheet to be detected. Since the defect-enhanced image is obtained, it is possible to detect only the defect to be detected with high accuracy even if there is a sudden change in background density such as uneven illumination, scale, or oil drop.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an image processing procedure for obtaining a defect-enhanced image according to the present invention.

FIG. 2 is a three-dimensional display diagram of the density distribution of the surface image, showing the transition of the image along with the processing flow from the surface image to the defect emphasized image in the present invention, using an actual image example of the vertical crack defect.

FIG. 3 shows a transition of images according to a processing flow from a surface image to a defect-enhanced image in the present invention by using an example of an actual image of a vertical crack defect, and a moving average weight used for obtaining a primary processed image. The figure which shows a coefficient.

FIG. 4 is a three-dimensional display diagram of a primary output image, showing a transition of an image along with a processing flow from a surface image to a defect-enhanced image in the present invention, using an actual image example of a vertical crack defect.

FIG. 5 shows an image transition along with the processing flow from the surface image to the defect-enhanced image in the present invention by using an example of an actual image of a vertical crack defect, and one line perpendicular to the defect of the primary processed image is extracted. Fig.

FIG. 6 shows a transition of images according to a processing flow from a surface image to a defect-enhanced image according to the present invention by using an example of an actual image of a vertical crack defect, and shows a short-time self-time when a primary processed image of one system is used. The figure which shows the weighting coefficient of a correlation function.

FIG. 7 is a three-dimensional display diagram of defect-enhanced image densities showing transitions of images according to a flow of processing from a surface image to a defect-enhanced image in the present invention, using an example of an actual image of a vertical crack defect.

FIG. 8 shows a transition of images according to a flow of processing from a surface image to a defect-enhanced image in the present invention by using an actual image example of a vertical crack defect, and shows a component parallel to a flaw of a moving average weighting coefficient, Of various components.

FIG. 9 shows an image transition accompanying a processing flow from a surface image to a defect-enhanced image according to the present invention, using an example of an actual image of a vertical crack defect, showing a component parallel to a flaw of a moving average weighting coefficient Of various components.

FIG. 10 shows a transition of images according to the flow of processing from the surface image to the defect-enhanced image in the present invention by using an example of an actual image of a vertical crack defect, and shows one line perpendicular to the defect from the primary processed image of two systems. The figure which took out.

FIG. 11 shows a transition of an image according to a processing flow from a surface image to a defect-enhanced image according to the present invention by using an actual image example of a vertical crack defect, and one line perpendicular to the defect in the defect-enhanced image is extracted. The figure.

FIG. 12 is a schematic diagram showing the equipment arrangement of the present embodiment.

FIG. 13 is a flowchart showing an image processing procedure in this embodiment.

FIG. 14 is a graph showing a non-detection rate and an over-detection rate according to the conventional and defect detection methods of the present invention.

[Explanation of symbols]

 1 Slab 2 Transport table 3 Illumination light source 4 Line sensor camera 5 Driving device 6 Encoder 7 Image processing device

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] January 12, 1996

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction contents]

[0016]

(Equation 3)

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0028

[Correction method] Change

[Correction contents]

[0028]

(Equation 6)

Claims (2)

[Claims] 1. A defect detection method for obtaining a surface image of an object to be inspected and image-processing the surface image to detect a surface defect of the object to be inspected, wherein two kinds of two-dimensional weighted movement are performed on the surface image. A defect detection method, characterized in that a plurality of short-time cross-correlation functions with different delays are obtained for two averaged signal sequences, and an image in which the weighted sum of these is configured as the density value of each pixel is obtained. . 2. A defect detection method for obtaining a surface image of an object to be inspected and image-processing the surface image to detect surface defects of the object to be inspected, wherein two kinds of two-dimensional weighted movements are applied to the surface image. A defect detection method, wherein an image in which only a defective portion is emphasized is obtained by multiplying both signals by averaging two signal sequences.