CN111062939B - Method for rapidly screening quality of strip steel surface and automatically extracting defect characteristics - Google Patents

Abstract

The present invention relates to the field of computer vision technology, and discloses a method for rapid quality identification and defect feature automatic extraction of steel strip surface. The steps are as follows: 1) grayscale projection is performed on the collected strip steel surface image to obtain a grayscale matrix image; 2) Find the maximum value R _Max and the minimum value R _Min of the gray projection value of each row in the grayscale matrix diagram, and the maximum value C _Max and minimum value C _Min of the gray projection value of each column, and calculate the average value R _Avg of the gray projection value of each row , the average value C _Avg of the grayscale projection of each column, and then judge whether there is a defect in the surface image of the strip according to the average value of the grayscale projection value of each row and column and the difference between the maximum value and the minimum value; 3) According to the judgment result of step 2) , crop the gray-scale matrix image, and mark the defect feature ROI area in the cropped gray-scale matrix image. The method can quickly identify the defects in the strip surface image, and the calculation speed is fast, which meets the real-time online defect detection requirements of the high-speed strip production line.

Description

A method for rapid quality screening and automatic defect feature extraction of strip steel surface

technical field

The invention belongs to the technical field of computer vision, and in particular relates to a method for rapid quality identification and defect feature automatic extraction of strip steel surfaces.

Background technique

Strip steel is one of the main products of the steel industry and an indispensable raw material for aerospace, shipbuilding, automobile, machinery manufacturing and other industries. Its quality will directly affect the quality and performance of the final product. In the process of strip steel manufacturing, due to many factors such as raw materials, rolling equipment and processing technology, different types of defects such as cracks, scars, and holes appear on the surface of the strip steel. Strip surface defects not only easily lead to serious production accidents such as strip breakage, accumulation, parking, etc., but also seriously wear the rolls, causing incalculable economic and social impacts on production enterprises.

In recent years, with the development of industrial technology, enterprises have gradually begun to use non-contact non-destructive testing technology represented by machine vision, which has high resolution, strong classification, little influence by environmental electromagnetic fields, large working distance and high measurement accuracy. and low cost advantages. In the production process, the moving speed of the steel strip can exceed 10m/s, and a large amount of image data (such as 25 frames per second) will be generated every second to be processed by the system, while the steel strip with surface defects only accounts for a small part, which is absolutely Most of the steel strips have no defects, which requires high real-time defect detection capabilities of the algorithm.

At present, the most widely used method for the rapid detection of strip surface defects is mainly the "difference method". The image of the steel surface will blur the image of the strip boundary, which will cause the strip boundary to be unable to be detected normally. When the strip steel surface image is blurred in the collected strip steel surface image, after processing the strip steel image using the canny operator edge detection, it will be found that there is an obvious slender broken line in the intersection area of the strip steel edge and the background, continue to use The area where the broken line is located will be mistakenly identified as a strip surface defect by the “difference method” for detection. Therefore, the detection of the strip surface image using the “difference method” will identify the place where the strip boundary image is blurred as "Edge defects", however, in reality there is no defect in the boundary part. A large number of edge "false defects" pose a major challenge to the rapid identification of strip surface quality and defect feature extraction. How to identify false defects must also be included in the research scope of strip defect identification. Aiming at the problems in the prior art that the defect area ratio of the strip steel is low, false defects at the edge and light interference are serious, making it difficult to effectively extract all kinds of real defects, this application is based on the "difference method" and according to the surface defects of the strip steel Based on the image characteristics, a method for rapid quality identification and defect feature automatic extraction of strip steel surface was studied.

Contents of the invention

The object of the present invention is to provide a method for rapid quality screening and defect feature automatic extraction of steel strip surface.

For realizing the purpose of the invention, the technical scheme adopted in the present invention is as follows:

A method for rapid quality screening and defect feature automatic extraction of strip steel surface, comprising the following steps:

(1) Gray-scale projection is performed on the collected strip steel surface image in a column-down and row-to-right manner to obtain a gray-scale matrix image;

(2) Analyze the grayscale matrix image obtained in step (1), find out the maximum value R _Max , the minimum value R _Min of the grayscale projection value of each row in the grayscale matrix image, and the maximum value of the grayscale projection value of each column C _Max , minimum value C _Min , calculate the average value R _Avg of the grayscale projection of each row, the average value C _Avg of the grayscale projection of each column, and the global average value Global _Avg of the entire grayscale matrix image, and then project according to the grayscale projection of each row and column The average value of the value and the difference between the maximum value and the minimum value determine the defect area, non-defect area, boundary pseudo-defect area, transition area between the strip boundary and the background, and the background area in the strip surface image;

(3) According to the judgment result of step (2), crop and delete the boundary pseudo-defect area, conversion area and background area in the gray-scale matrix image, and then mark the defect characteristic ROI area in the gray-scale matrix image after cropping.

According to the above-mentioned method for rapid quality screening and defect feature automatic extraction of strip steel surface, preferably, in step (2), the defects in the strip steel surface image are judged according to the difference between the maximum value and the minimum value of the gray projection value of each row and column The specific operations of area, non-defect area, boundary pseudo-defect area, transition area between strip boundary and background, and background area are as follows:

A. Judge each column in the grayscale matrix:

(A1) If the difference between C _Max and C _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and C _Avg > Global _Avg /10, then there is no defect in this column;

(A2) If the difference between C _Max and C _Min and C _Avg are within [(1-µ)Global _Avg , (1+µ)Global _Avg ], then there is a defect in this column;

(A3) The difference between C _Max and C _Min is within [(1-µ)Global _Avg , (1+µ)Global _Avg ], but C _Avg <(1-2µ)Global _Avg , then this column is the strip boundary The conversion area with the background; starting from the adjacent columns of the conversion area, extending laterally toward the direction in which the gray mean value increases to obtain a boundary pseudo-defect area, wherein the gray mean value of the adjacent columns is greater than the gray value of the conversion area The average value of the degree, the width of the lateral extension is 1 to 3 times the width of the conversion area;

(A4) The difference between C _Max and C _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and C _Avg ≤ Global _Avg /10, then this column is the strip background area;

B. Judging each row in the grayscale matrix image:

(B1) The difference between R _Max and R _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg > Global _Avg /10, then there is no defect in this row;

(B2) If the difference between R _Max and R _Min and R _Avg are both within [(1-µ)Global _Avg , (1+µ)Global _Avg ], then there is a defect in this row;

(B3) The difference between R _Max and R _Min is within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg <(1-2µ)Global _Avg , then the strip boundary and The conversion area of the background; starting from the adjacent behavior of the conversion area, extending longitudinally toward the direction in which the average gray value increases to obtain a boundary pseudo-defect area, wherein the average gray value of the adjacent rows is greater than the average gray value of the converted area , the width of the longitudinal extension is 1 to 3 times the width of the conversion area;

(B4) The difference between R _Max and R _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg ≤ Global _Avg /10, then this behavior is the strip background area;

Among them, µ is the defect threshold coefficient.

Find out the column where the defect area is located according to the column judgment result, find out the row where the defect area is located according to the row judgment result, and the position where the row and column intersect is the position where the defect area is located.

According to the above-mentioned method for rapid quality screening of steel strip surface and automatic extraction of defect features, the defect threshold coefficient µ is mainly the result of pre-adjustment tests based on the lighting conditions on site and the gray value of non-defective rolled steel. The size of its value range Directly determine the frame size of the defect feature ROI area. If the value of µ is too small, the border of the ROI area will be relatively small, so that the algorithm can only frame the more obvious defect features and ignore the less obvious defect features; if the value of µ is too large, the border of the ROI area will be relatively large, and the algorithm Not only can it frame relatively obvious defect features and non-obvious defect features, but it may also frame the defect-free area around the defect.

According to the above-mentioned method for rapid quality identification and defect feature automatic extraction of strip steel surface, preferably, the defect threshold coefficient µ ranges from 20% to 30%. More preferably, the value range of the defect threshold coefficient µ is 30%.

According to the above-mentioned method for rapid quality screening and defect feature automatic extraction of strip steel surface, preferably, the specific operation of step (3) is: according to the judgment result of step (2), cut and delete the boundary pseudo-defect area in the gray matrix image , the conversion area and the background area, mark the rows and columns that are judged to contain defects in the gray scale matrix after clipping, and draw a rectangular frame to frame the defect area, and the rectangular frame area is the defect characteristic ROI area.

According to the above-mentioned method for rapid quality screening and automatic defect feature extraction of strip steel surface, preferably, in step (1), before performing grayscale projection on the strip steel surface image, the background area of the strip steel surface image is roughly cut, the specific The operation is: according to the characteristics of clear distinction between the strip steel and the background, identify the collected strip steel surface image, filter out the edge background in the strip steel surface image, and delete it. This operation can effectively filter out the interference of most edge background regions on image analysis, further reduce the calculation amount, and improve the detection efficiency.

Compared with the prior art, the positive beneficial effect that the present invention obtains is:

(1) The present invention carries out the grayscale projection of the collected strip steel surface image in a column-down and row-to-right manner to obtain a grayscale matrix image, and, according to the grayscale of each row and column in the grayscale matrix image The average value of the projection value and the difference between the maximum value and the minimum value can effectively filter out the non-defective surface, false defects on the edge and light interference, and quickly detect the defects in the image (especially for the longitudinal and transverse grains on the strip surface. and large-area scale defects), and output the characteristic ROI region of the defect in the strip surface image; this method greatly reduces the amount of calculation, the calculation speed is fast, the efficiency is high, and the accuracy of the detection result is high. Defects in steel surface images meet the real-time online defect detection requirements of high-speed strip steel production lines, and can accurately and automatically detect and extract defect feature ROI regions in images, providing high-quality labeling data for strip steel defects, saving a lot of manual data labeling data problem, practicality is strong.

(2) The method of the present invention has low requirements on the hardware performance of the equipment, and fully meets the needs of deployment in industrial sites.

Description of drawings

Fig. 1 is the detection result diagram of the strip image with longitudinal inclusion defects on the surface;

Fig. 2 is the partial diagram of the column analysis result that the gray scale matrix figure of original image A in Fig. 1 is analyzed by column;

Fig. 3 is the partial diagram of the row analysis result that the grayscale matrix image of original image A in Fig. 1 is analyzed by row;

Figure 4 is an enlarged view of area a in Figure 3;

Fig. 5 is the detection result diagram of the strip steel image with longitudinal scratch defect on the surface;

Fig. 6 is the partial diagram of the column analysis result that the grayscale matrix image of original image A in Fig. 5 is analyzed by column;

Fig. 7 is the partial diagram of the row analysis result that the grayscale matrix image of original image A in Fig. 5 is analyzed by row;

Figure 8 is a diagram of the image detection results of the steel strip covered with scale.

Detailed ways

The present invention will be described in further detail below through specific examples, but the scope of the present invention is not limited.

Example 1:

A method for rapid quality screening and defect feature automatic extraction of strip steel surface, comprising the following steps:

(1) Gray-scale projection is performed on the collected strip steel surface image in a column-down and row-to-right manner to obtain a gray-scale matrix image;

(2) Analyze the grayscale matrix image obtained in step (1), find out the maximum value R _Max , the minimum value R _Min of the grayscale projection value of each row in the grayscale matrix image, and the maximum value of the grayscale projection value of each column C _Max , minimum value C _Min , calculate the average value R _Avg of the grayscale projection of each row, the average value C _Avg of the grayscale projection of each column, and the global average value Global _Avg of the entire grayscale matrix image, and then project according to the grayscale projection of each row and column The average value of the value and the difference between the maximum value and the minimum value can be used to judge the defect area, non-defect area, boundary pseudo-defect area, transition area between the strip boundary and the background, and the background area in the strip surface image.

Among them, according to the difference between the maximum value and the minimum value of the gray projection value of each row and column, judge the defect area, non-defect area, boundary pseudo-defect area, transition area between the strip boundary and the background, and the background area in the strip surface image The specific operation is as follows:

A. Judge each column in the grayscale matrix:

(A1) If the difference between C _Max and C _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and C _Avg > Global _Avg /10, then there is no defect in this column;

(A2) If the difference between C _Max and C _Min and C _Avg are within [(1-µ)Global _Avg , (1+µ)Global _Avg ], then there is a defect in this column;

(A3) The difference between C _Max and C _Min is within [(1-µ)Global _Avg , (1+µ)Global _Avg ], but C _Avg <(1-2µ)Global _Avg , then this column is the strip boundary The conversion area with the background; starting from the adjacent columns of the conversion area, extending laterally toward the direction in which the gray mean value increases to obtain a boundary pseudo-defect area, wherein the gray mean value of the adjacent columns is greater than the gray value of the conversion area degree mean value, the width of the horizontal extension is twice the width of the conversion area;

(A4) The difference between C _Max and C _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and C _Avg ≤ Global _Avg /10, then this column is the strip background area.

Among them, µ is the defect threshold coefficient, and its value is 30%.

B. Judging each row in the grayscale matrix image:

(B1) The difference between R _Max and R _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg > Global _Avg /10, then there is no defect in this row;

(B2) If the difference between R _Max and R _Min and R _Avg are both within [(1-µ)Global _Avg , (1+µ)Global _Avg ], then there is a defect in this line;

(B3) The difference between R _Max and R _Min is within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg <(1-2µ)Global _Avg , then the strip boundary and The conversion area of the background; starting from the adjacent behavior of the conversion area, extending longitudinally toward the direction in which the average gray value increases to obtain a boundary pseudo-defect area, wherein the average gray value of the adjacent rows is greater than the average gray value of the converted area , the width of the vertical extension is twice the width of the conversion area;

(B4) The difference between R _Max and R _Min is not within [(1-µ)Global _Avg , (1+µ)Global _Avg ], and R _Avg ≤ Global _Avg /10, then this behavior is the strip background area;

Among them, µ is the defect threshold coefficient, and its value is 30%.

(3) According to the judgment result of step (2), cut out and delete the boundary pseudo-defect area, conversion area, and background area in the gray-scale matrix image, and mark the rows and columns that are judged to contain defects in the gray-scale matrix image after clipping, At the same time, a rectangular frame is drawn to frame the defect area, and the rectangular frame area is the defect characteristic ROI area.

Example 2: Validation experiment of the present invention's strip steel surface rapid quality screening and defect feature automatic extraction method

1. Detect the strip steel image with longitudinal inclusion defects on the surface

Adopt the method for the embodiment of the present invention 1 to detect the steel strip image that surface has longitudinal grain defect, wherein the original image of the steel strip surface collected is as shown in A in Fig. 1, as can be seen from A in Fig. 1, the strip steel image to be detected There are longitudinal grain defects, the background area of the strip shooting, and the transition area between the strip boundary and the background. The detection result is shown in B in Figure 1 (B is a partial image of the detection result), and the defect area in Figure B has been framed by a box, that is, the area framed by the box is the detected defect characteristic ROI area, and Figure B has The background area, transition area and border pseudo-defect area in the original image A have been deleted, thus illustrating that the method described in Embodiment 1 of the present invention can identify the defect area, background area, transition area and border in the original image of the strip surface Pseudo-defect area.

Figure 2 is a partial image of the column analysis result of the gray matrix image of the original image A in Figure 1 analyzed by column (because the column analysis result image is relatively large, only the partial image is shown here). It can be seen from Figure 2 that the Global _Avg of the grayscale matrix image is 111.3083, and the value range of (1±µ)Global _Avg is [77.9, 144.7]. In Figure 2, C _Max - C _Min = 41, C _Avg = 2 in the column where the yellow area is located, so this column is the strip background area; C _Max - C _Min = 101, C _Avg = 28.8 in the column where the green area is located, Therefore, this column is the transition area between the strip border and the background; starting from the adjacent columns of the transition area (the gray mean value of the adjacent columns is greater than the gray level mean value of the transition area), the direction of the gray mean value increases laterally Extending, the width of the lateral extension is twice the width of the conversion area, and the border pseudo-defect area is obtained, which is the blue area in the figure; the difference between C _Max and C _Min of the three columns corresponding to the red area is 119, 126, and 111 in sequence , the C _Avg values corresponding to the three columns are 103, 116, and 105 respectively, therefore, there are defects in the three columns corresponding to the red area. Therefore, it can also be seen from FIG. 2 that the detection method of the present invention can clearly detect the background area in the strip surface image, the transition area between the strip boundary and the background, the boundary pseudo-defect area and the defect area.

Fig. 3 is a partial image of the row analysis result of the gray matrix image of the original image A in Fig. 1 (because the row analysis result image is relatively large, only a partial image is shown here). The behavioral defect area corresponding to the red area in Figure 3. It can be seen from Fig. 2 and Fig. 3 that the position of the defect area can be accurately found by combining the column detection results with the row detection results, that is, the area enclosed by the box in B in Fig. 1 . In order to clearly present the row analysis results, taking the area a in Figure 3 as an example, the method was carried out on the area a, and its enlarged view is shown in Figure 4.

, Detection of strip steel images with longitudinal scratch defects on the surface

Adopt the method of embodiment 1 of the present invention to detect the strip steel image that the surface has longitudinal grain defect, wherein the strip steel surface raw image of acquisition is as shown in A in Fig. 5, as can be known from Fig. 5 A, the strip steel image to be detected There are longitudinal grain defects and background areas in strip shots. The detection result is shown in B in Figure 5 (B is a partial image of the detection result), and the defect area in Figure B has been framed by a box, that is, the area framed by the box is the detected defect feature ROI area, and it is consistent with the original image A In comparison, the background area in the original image A has been deleted in Figure B, which shows that the defect area in the original image of the strip surface can be identified by using the method described in Embodiment 1 of the present invention.

Figure 6 is a partial image of the column analysis results of the gray matrix image of the original image A in Figure 5 analyzed by column (because the image of the column analysis results is relatively large, only a partial image is shown here). The column corresponding to the red area in Figure 6 is the defect area. Figure 7 is a partial view of the row analysis result of the gray matrix image of the original image A in Figure 5 analyzed by row (because the row analysis result image is too large, only the reduced image is shown here). The behavioral defect area corresponding to the red area in Figure 7. It can be seen from Fig. 6 and Fig. 7 that the position of the defect region can be accurately found by combining the column detection result with the row detection result, that is, the region surrounded by the box in B in Fig. 5 .

、Detect the strip steel image covered with scale on the surface

The method described in Embodiment 1 of the present invention is used to detect a strip steel image whose surface is covered with scale, wherein the original image of the strip steel surface collected is as shown in A in Figure 8, as can be seen from A in Figure 8, to be detected The strip image is full of scale defects. The detection result is shown in B in Figure 8. The defect area in Figure B has been framed by a box, that is, the area framed by the box is the detected defect characteristic ROI area, which shows that the method described in Embodiment 1 of the present invention can Identify defect areas in the raw image of the strip surface.

The above description is only a preferred embodiment of the present invention, but not limited to the above examples, and any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention within.

Claims (4)

1. A method for fast quality screening and defect characteristic automatic extraction of a strip steel surface is characterized by comprising the following steps:

(1) Carrying out gray projection on the acquired strip steel surface image in a mode of downward column by column and rightward row by row to obtain a gray matrix diagram;

(2) For the gray scale obtained in step (1)Analyzing the matrix diagram to find out the maximum value R of each row of gray projection values in the gray matrix diagram _Max Minimum value R _Min Maximum value C of gray projection value of each column _Max Minimum value C _Min Calculating the gray projection mean value R of each row _Avg Mean value C of gray projection of each column _Avg And Global gray average Global for the entire gray matrix map _Avg Then judging a defect area, a defect-free area, a boundary pseudo defect area, a conversion area of a band steel boundary and a background area in the band steel surface image according to the average value of gray projection values of each row and each column and the difference value of the maximum value and the minimum value;

(3) Cutting and deleting boundary pseudo defect areas, conversion areas and background areas in the gray matrix diagram according to the judging result in the step (2), and marking defect feature ROI areas in the cut gray matrix diagram.

2. The method for fast quality screening and defect feature extraction on a strip steel surface according to claim 1, wherein in the step (2), the specific operations of determining a defect area, a defect-free area, a boundary pseudo defect area, a conversion area between a strip steel boundary and a background, and a background area in the strip steel surface image according to the difference between the maximum value and the minimum value of each row and each column of gray projection values are as follows:

A. judging each column in the gray matrix diagram:

（A1）C _Max and C _Min The difference value of (2) is not in [ (1- [ mu ] O) Global _Avg ，(1+µ)Global _Avg ]And C in _Avg ＞Global _Avg 10, the column is defect-free;

（A2）C _Max and C _Min Difference of C _Avg Are all in [ (1-mu) Global _Avg ，(1+µ)Global _Avg ]If there is a defect in the column;

（A3）C _Max and C _Min The difference value between [ (1-mu) Global _Avg ，(1+µ)Global _Avg ]In, but C _Avg ＜(1-2µ) Global _Avg The column is a conversion area of the boundary of the strip steel and the background; with adjacent columns of conversion regionsStarting, transversely extending in the direction of increasing the gray average value to obtain a boundary pseudo-defect region, wherein the gray average value of the adjacent columns is larger than that of the conversion region, and the transversely extending width is 1-3 times of the width of the conversion region;

（A4）C _Max and C _Min The difference value of (2) is not in [ (1- [ mu ] O) Global _Avg ，(1+µ)Global _Avg ]And C in _Avg ≤Global _Avg And/10, the column is a strip steel background area;

B. judging each row in the gray matrix diagram:

（B1）R _Max and R is R _Min The difference value of (2) is not in [ (1- [ mu ] O) Global _Avg ，(1+µ)Global _Avg ]And R is _Avg ＞Global _Avg 10, then the row is defect free;

（B2）R _Max and R is R _Min Is the difference of R _Avg Are all in [ (1-mu) Global _Avg ，(1+µ)Global _Avg ]If there is a defect in the row;

（B3）R _Max and R is R _Min The difference value between [ (1-mu) Global _Avg ，(1+µ)Global _Avg ]And R is _Avg ＜(1-2µ) Global _Avg The transition area of the boundary and the background of the behavior strip steel; starting with adjacent rows of the conversion area, and longitudinally extending in the direction of increasing the gray average value to obtain a boundary pseudo-defect area, wherein the gray average value of the adjacent rows is larger than the gray average value of the conversion area, and the longitudinal extending width is 1-3 times of the width of the conversion area;

（B4）R _Max and R is R _Min The difference value of (2) is not in [ (1- [ mu ] O) Global _Avg ，(1+µ)Global _Avg ]And R is _Avg ≤Global _Avg And/10, the behavior strip steel background area;

wherein [ mu ] is a defect threshold coefficient.

3. The method for rapidly screening the quality and automatically extracting the defect characteristics of the surface of the strip steel according to claim 2, wherein the value range of the mu is 20% -30%.

4. The method for rapid quality screening and automatic defect feature extraction on a strip steel surface according to claim 3, wherein the specific operation of step (3) is as follows: and (3) cutting and deleting a boundary pseudo defect region, a conversion region and a background region in the gray matrix diagram according to the judgment result in the step (2), marking rows and columns judged to contain defects in the cut gray matrix diagram, and simultaneously drawing a rectangular frame to form a defect region, wherein the rectangular frame region is a defect characteristic ROI region.