A machine vision quality inspection method for galvanized steel strip production surfaces

By installing a light source and a high-definition camera in a dark room, and combining deep learning for three-channel color value and grayscale feature extraction, the problem of inaccurate galvanizing quality detection results in existing technologies has been solved, achieving a more efficient assessment of galvanizing quality on steel strip surfaces.

CN121190434BActive Publication Date: 2026-11-10TIANJIN YU RUN DE METAL PROD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202511351502.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-09-22
Publication Date
2026-11-10
Estimated Expiration
2045-09-22

AI Technical Summary

Technical Problem

Existing technologies for inspecting the quality of galvanized steel strip cannot accurately assess the quality of the galvanized layer, leading to errors in the test results. This is mainly because the preset threshold cannot adapt to the fluctuations in color values ​​in the image after galvanizing and the influence of the shooting angle.

Method used

A light source and a high-definition camera are installed in a dark room to capture images of the galvanized steel strip surface. By extracting three-channel color values ​​and grayscale features, and combining them with deep learning, a quality inspection standard is established to determine whether the quality of the galvanized layer is up to standard.

Benefits of technology

It improves the accuracy and rationality of galvanizing quality inspection, eliminates surface brightness and darkness changes caused by multiple light sources, provides more accurate pixel color value judgment, and ensures the validity of the test results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121190434B_ABST
    Figure CN121190434B_ABST
Patent Text Reader

Abstract

This invention discloses a machine vision quality inspection method for galvanized steel strip production surfaces, relating to the field of galvanized steel strip surface quality inspection technology. The method includes the following steps: setting up a black chamber; capturing images of the galvanized surface; extracting features from the galvanized surface images to obtain three-channel color value features and grayscale features; extracting the three-channel color value features and grayscale features of qualified galvanized surface images and analyzing quality inspection standards; performing quality inspection on the galvanized surface images based on the quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is qualified. This invention addresses the problems of existing galvanized steel strip surface quality inspection technologies, which suffer from over-reliance on preset thresholds and failure to accurately consider color value fluctuations in the strip steel surface images, leading to erroneous results in the galvanized steel strip surface quality inspection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of strip steel surface galvanization quality inspection technology, specifically a machine vision quality inspection method for strip steel production surface galvanization. Background Technology

[0002] The quality inspection technology for galvanized steel strip refers to the technical means of inspecting the surface of galvanized steel strip to ensure the integrity of the galvanized layer and prevent defects in the galvanized layer from affecting the normal use of the steel strip.

[0003] Existing technologies for inspecting the quality of galvanized steel strip typically employ a large number of preset threshold comparisons. However, after galvanizing, the color value of the steel strip fluctuates in the captured images. Furthermore, due to varying shooting angles, the distance between different points on the steel strip surface and the camera differs. The farther the point from the camera, the greater the color value fluctuation in the captured image. Therefore, using uniform preset thresholds cannot accurately assess the quality of galvanized steel strip, and the effectiveness of these preset thresholds cannot be guaranteed. For example, in publication number CN113281341... Patent application A discloses an "optimization method for the detection of a dual-sensor surface quality detection system for hot-dip galvanized strip steel." This method detects the galvanization quality of the strip steel surface using different preset thresholds or ranges. However, this method cannot accurately assess the galvanization quality of the strip steel surface, and the preset thresholds cannot guarantee their effectiveness. Existing strip steel surface galvanization quality detection technologies also suffer from problems such as over-reliance on preset thresholds and failure to accurately consider the fluctuations in color values ​​in the strip steel surface image, leading to errors in the galvanization quality detection results. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in the prior art. It involves constructing a black chamber, installing a light source and a high-definition camera within it, and then conveying galvanized steel strip into the black chamber via a conveyor belt to capture images of the galvanized surface. Color features are then extracted from these images to obtain three-channel color value features. Simultaneously, grayscale features are extracted to obtain grayscale features. The three-channel color value features and grayscale features of qualified galvanized surface images are then extracted and processed using deep learning to obtain the three-channel color value standards and grayscale standards for the galvanized steel strip. Finally, based on quality inspection standards, the galvanized surface images are subjected to quality inspection to determine whether the galvanized layer of the steel strip is qualified. This addresses the problem that existing galvanized steel surface quality inspection technologies suffer from over-reliance on preset thresholds and a lack of accurate consideration of color value fluctuations in the surface images, leading to erroneous results in the galvanized steel surface quality inspection.

[0005] To achieve the above objectives, in a first aspect, this application provides a machine vision quality inspection method for galvanized steel strip production surfaces, comprising the following steps: A dark room was built, and light sources and high-definition camera devices were installed inside. The galvanized steel strip is fed into a dark chamber via a conveyor belt and an image of the galvanized surface is captured. Feature extraction is performed on the image of the galvanized surface to obtain the three-channel color value features and grayscale features of the galvanized surface image; Extract the three-channel color value features and grayscale features of qualified galvanized surface images and perform deep learning to obtain quality inspection standards; The quality of the galvanized surface image is inspected based on the quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is up to standard.

[0006] Furthermore, constructing a darkroom and installing light sources and high-definition camera equipment within it includes the following sub-steps: Construct a black room, which is a closed space where no external light source illuminates the interior except for artificial light; Install a light fixture at the geometric center of the top of the darkroom; A high-definition camera is installed at the geometric center of any side of the black room, and the angle between the high-definition camera and the plane of the black room is a first angle.

[0007] Further, the process of feeding the galvanized steel strip into a darkroom via a conveyor belt and taking images of the galvanized surface includes the following sub-steps: The strip steel is galvanized, and after galvanizing, the strip steel is placed on a conveyor belt and sent into the black chamber. The galvanized surface is imaged by a high-definition camera in a darkroom. The length of each shot taken in the black chamber is limited. A strip of steel needs to go through a first number of shots to obtain a first number of galvanized surface images. During quality inspection, each galvanized surface image is analyzed independently.

[0008] Further, feature extraction is performed on the galvanized surface image to obtain the three-channel color value features and grayscale features of the galvanized surface image, including the following sub-steps: Color features were extracted from the image of the galvanized surface to obtain the three-channel color value features of the galvanized surface image; Gray-scale features are extracted from the image of the galvanized surface to obtain the gray-scale features of the galvanized surface image.

[0009] Further, color feature extraction is performed on the galvanized surface image to obtain the three-channel color value features of the galvanized surface image, including the following sub-steps: The contour of the strip steel in the galvanized surface image is obtained by contour extraction and named as the target contour. The pixels in the target contour are named as surface points. The surface point located in the nth row and mth column is labeled as P(n,m), where n and m are both non-zero natural numbers and (n,m) is the index of P; Obtain the color values ​​of P(n,m), which include R-pass, G-pass, and B-pass values, and use the symbol C. h Let R, G, and B represent the pass values ​​in sequence, where C1, C2, and C3 are respectively, and h is a non-zero natural number and is the index of C. With h as the X-axis, C h Establish a Cartesian coordinate system for the Y-axis, named the T-junction feature map, and set C... h Enter the three-channel feature map according to h, perform linear regression on the three-channel feature map, obtain the slope of the linear regression function, and name it as three-channel color value feature. Each P(n,m) corresponds to a three-channel color value feature.

[0010] Further, grayscale feature extraction is performed on the galvanized surface image to obtain the grayscale features of the galvanized surface image, including the following sub-steps: The image of the galvanized surface is converted to grayscale to obtain a grayscale image of the surface. P(n,m) in the grayscale image of the surface is labeled as Q(n,m), and the grayscale value of Q(n,m) is labeled as H(n,m). Calculate max(m) / 2, and mark the integer part of the result as M, where max() is the maximum value operator; Analyze each value of n independently, calculate |mM|, label the calculation result as F, integrate the two H(n,m) with equal F into a pixel group, name the H(n,m) in the pixel group where m is less than M as the left pixel gray level, represented by the symbol GS1, and name the H(n,m) in the pixel group where m is greater than M as the right pixel gray level, represented by the symbol GS2; Calculate |GS1-H(n,M)| and name the result as the left pixel difference; calculate |GS2-H(n,M)| and name the result as the right pixel difference. Pixel groups are sorted and numbered in ascending order of F, using the symbol PG. i Let PG be a non-zero natural number, where i is the index of PG, and PG is represented by the symbol R(i,j). i The left and right pixel differences are defined as follows, where j takes values ​​of 1 and 2 and (i,j) is the index of R, where R(i,1) represents PG. i The difference in pixels on the left, R(i,2) represents the difference in pixels on the left. i The pixel difference on the right side; Establish a Cartesian coordinate system with i as the horizontal axis and R(i,j) as the vertical axis, named a single-row feature map. Input R(i,j) into the single-row feature map according to i. Name the coordinate point j=1 as the first coordinate and the coordinate point j=2 as the second coordinate. Perform linear regression on the first coordinate and the second coordinate respectively to obtain the slope of the linear regression function, and obtain KL and KR. Calculate |KL-KR| and name the calculation result as gray-scale feature. Each value of n corresponds to a gray-scale feature.

[0011] Furthermore, the three-channel color value features and grayscale features of the qualified galvanized surface image are extracted and deep learning is performed to obtain the quality inspection standard, which includes the following sub-steps: The three-channel color value features of qualified galvanized surface images are extracted and deep learning is performed to obtain the three-channel color value standard of galvanized strip steel. The grayscale features of qualified galvanized surface images are extracted and deep learning is performed to obtain the grayscale standard of galvanized steel strip.

[0012] Furthermore, the three-channel color value features of qualified galvanized surface images are extracted and deep learning is performed to obtain the three-channel color value standard of galvanized strip steel, including the following sub-steps: The qualified galvanized steel strip is named as the reference sample. The three-channel color value features of all galvanized surface images of the same reference sample are obtained and named as the three-channel color value reference features. The difference between the maximum and minimum values ​​in the three-channel color value reference feature is obtained and named the three-color variability. The differences in the three colors of all reference samples are statistically analyzed, and the maximum value is obtained to obtain the three-channel color value standard.

[0013] Furthermore, extracting the grayscale features of qualified galvanized surface images and performing deep learning to obtain the grayscale standard of galvanized strip steel includes the following sub-steps: Extract the grayscale features of the reference sample, and label the grayscale features of the nth row as T. n Statistical analysis of T values ​​in all reference samples max(n) The range of grayscale values ​​is named the grayscale baseline standard. For any reference sample, with n as the X-axis, T n Establish a Cartesian coordinate system for the Y-axis, name it Multi-Row Feature Map, and set T... n Enter multiple rows of feature maps according to n, and name the coordinate points in the multiple rows of feature maps as multiple row feature points; For each value of n, the topmost feature point in the multi-row is named a feature vertex. Connect adjacent feature vertices with straight lines to obtain a broken line, named the multi-row feature line. Then, T is placed on the multi-row feature line. n The point corresponding to this location is marked as S. n ; Place the multi-row feature lines of each reference sample into the same new multi-row feature map, and set the S of each multi-row feature line... max(n) The overlapping is performed, and the resulting multi-row feature map is named the advanced standard map. In the advanced standard diagram, for each value of n, the topmost S... n Named the coordinate vertex, connect adjacent coordinate vertices with straight lines, and name the resulting polyline the grayscale progression standard. The grayscale basic standard and the grayscale progression standard together constitute the grayscale standard.

[0014] Furthermore, based on quality inspection standards, the images of the galvanized surface are subjected to quality inspection to determine whether the quality of the galvanized layer of the strip steel is qualified. This includes the following sub-steps: The image of the galvanized surface to be analyzed is named the image to be analyzed. The three-channel color value features and grayscale features of the image to be analyzed are extracted and named the first feature to be analyzed and the second feature to be analyzed, respectively. Analyze whether the three-color difference of the first feature to be analyzed is less than or equal to the three-channel color value standard. If yes, output a three-channel qualified signal; otherwise, output a three-channel unqualified signal. Analyze the T of the second feature to be analyzed max(n) If the grayscale standard is met, output a basic verification pass signal; otherwise, output a basic verification failure signal. If the basic verification signal is passed, then the multi-row feature map of the second feature to be analyzed is analyzed and named the feature map to be analyzed. The rightmost endpoint of the grayscale progression standard is then compared with T in the feature map to be analyzed. max(n) If the feature vertices overlap, query whether all feature points in multiple rows are below the grayscale advancement standard. If yes, output an advancement verification pass signal; otherwise, output an advancement verification failure signal. If three channels of qualified signals and advanced verification pass signals are output simultaneously, the surface quality of the galvanized steel strip is marked as qualified; otherwise, the surface quality of the galvanized steel strip is marked as unqualified.

[0015] The beneficial effects of this invention are as follows: This invention constructs a black chamber, installs a light source and a high-definition camera device in the black chamber, and then feeds the galvanized steel strip into the black chamber by a conveyor belt and captures an image of the galvanized surface. The advantage is that when constructing the black chamber, it ensures that the galvanized surface of the steel strip is illuminated by only one light source, eliminating the surface brightness variations caused by multiple light sources. At the same time, since the galvanized surface of the steel strip is reflective, when installing the high-definition camera, it is installed on the side and at a certain angle, so that no lighting lamp appears in the captured image of the galvanized surface, but only a fixed background. In this way, the color values ​​of each pixel in the obtained image of the galvanized surface can be free from external interference, making the color values ​​of different positions closer, thus improving the rationality and effectiveness of the quality inspection of the galvanized surface of the steel strip. Color feature extraction is performed on galvanized surface images to obtain three-channel color value features. Simultaneously, grayscale feature extraction is performed to obtain grayscale features. Then, the three-channel color value features and grayscale features of qualified galvanized surface images are extracted and subjected to deep learning to obtain the three-channel color value standards and grayscale standards for galvanized strip steel. Finally, quality inspection is performed on the galvanized surface images based on quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is qualified. The advantage lies in analyzing the galvanized surface images using both three-channel color value and grayscale value schemes. During deep learning, the relationship between grayscale values ​​of pixels from near to far is analyzed row by row, providing a more accurate judgment basis for more distant pixels, thus improving the accuracy and rationality of galvanized strip steel surface quality inspection. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of the method of the present invention; Figure 2 This is a schematic diagram of the three-way feature diagram of the present invention; Figure 3 This is a schematic diagram of the multi-row feature map of the present invention; Figure 4 This is a schematic diagram of the multi-row feature lines of the present invention; Figure 5 This is a schematic diagram of an advanced standard diagram of the present invention; Figure 6 This is a schematic diagram of the grayscale progression standard of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 As shown, this application provides a machine vision quality inspection method for galvanized steel strip production surfaces, comprising the following steps: Step S1: Construct a darkroom and install a light source and a high-definition camera device inside. Step S1 includes the following sub-steps: Step S101: Construct a black room. A black room is a closed space where no external light source illuminates the interior except for the light itself. Step S102: Install a lighting lamp at the geometric center of the top of the darkroom; Step S103: Install a high-definition camera at the geometric center of any side of the black room, and the angle between the high-definition camera and the plane of the black room is the first angle. In practice, the black chamber is a closed space of a cuboid, with the geometric center being the intersection of the two diagonals of the cuboid's face. The first angle is adjusted by the tester, only needing to ensure that the reflection of the lighting lamp does not appear in the image of the galvanized surface being captured.

[0019] Step S2 involves feeding the galvanized steel strip into a darkroom via a conveyor belt and capturing an image of the galvanized surface. Step S2 includes the following sub-steps: Step S201: The strip steel is galvanized. After galvanizing, the strip steel is placed on a conveyor belt and sent into a black chamber. Step S202: The surface of the strip steel is photographed by a high-definition camera in a dark room to obtain an image of the galvanized surface; Step S203: The length of each shot in the black room is limited. A strip of steel needs to go through the first number of shots to obtain the first number of galvanized surface images. During quality inspection, each galvanized surface image is analyzed independently. In practice, for example, the darkroom can only photograph 5m of strip steel at a time, while the current strip steel is 20m long. Theoretically, it needs to be photographed 4 times. However, usually, the theoretical optimal number of photographs is rounded up, that is, 5 photographs are taken. This is to ensure that there is a certain overlap between two adjacent galvanized surface images during subsequent analysis, to prevent missing the boundary area. Also, when calculating M and pixel groups in subsequent analysis, there may be cases where there is only one H(n,m) in a pixel group. In this case, the corresponding pixel group needs to be removed, but it must be located at the intersection of two adjacent galvanized surface images, so it can still be analyzed in the next galvanized surface image.

[0020] Step S3 involves extracting features from the galvanized surface image to obtain the three-channel color value features and grayscale features of the galvanized surface image. Step S3 includes the following sub-steps: Step S301: Extract color features from the galvanized surface image to obtain the three-channel color value features of the galvanized surface image; Step S301 includes the following sub-steps: Step S301.1: Obtain the outline of the strip steel in the galvanized surface image by contour extraction, name it as the target outline, and name the pixels in the target outline as surface points. Step S301.2: Mark the surface point in the nth row and mth column as P(n,m), where n and m are both non-zero natural numbers and (n,m) is the index of P; Step S301.3: Obtain the color values ​​of P(n,m), including R-pass, G-pass, and B-pass values, using the symbol C. hLet R, G, and B represent the pass values ​​in sequence, where C1, C2, and C3 are respectively, and h is a non-zero natural number and is the index of C. Please see Figure 2 As shown, in step S301.4, with h as the X-axis, C h Establish a Cartesian coordinate system for the Y-axis, named the T-junction feature map, and set C... h Enter the three-channel feature map according to h, perform linear regression on the three-channel feature map, obtain the slope of the linear regression function, and name it as the three-channel color value feature. Each P(n,m) corresponds to a three-channel color value feature. In practice, contour extraction is used to limit the pixels on the strip steel surface, preventing environmental factors from affecting the detection results of the galvanized surface quality of the strip steel. Under the shooting conditions in this embodiment, the colors of the acquired surface points are usually very similar, i.e., they have more stable color values. Since different R-pass, G-pass, and B-pass values ​​can form different colors, and when the colors are similar, the differences between the R-pass, G-pass, and B-pass values ​​are relatively stable and will not have large differences. The three-channel color value features extracted in this embodiment reflect the differences between the R-pass, G-pass, and B-pass values. For example, the R-pass, G-pass, and B-pass values ​​of a certain surface point are 209, 213, and 222, respectively. The three-channel feature map is constructed as follows: Figure 2 As shown, the extracted three-channel color value feature is 6.5. Each surface point has a three-channel color value feature. When analyzing the three-channel color value feature, regression analysis can also be performed directly through the three coordinate points without constructing a three-channel feature map. In this embodiment, the three-channel feature map is constructed only to represent the differences in magnitude between the R-channel value, G-channel value, and B-channel value in the form of an image.

[0021] Step S302: Extract grayscale features from the galvanized surface image to obtain the grayscale features of the galvanized surface image; Step S302 includes the following sub-steps: Step S302.1: Perform grayscale processing on the galvanized surface image to obtain a surface grayscale image. Mark P(n,m) in the surface grayscale image as Q(n,m) and mark the grayscale value of Q(n,m) as H(n,m). Step S302.2: Calculate max(m) / 2, and mark the integer part of the calculation result as M, where max() is the maximum value operator; Step S302.3: Analyze each value of n independently, calculate |mM|, mark the calculation result as F, integrate the two H(n,m) with equal F into a pixel group, name the H(n,m) in the pixel group where m is less than M as the left pixel gray level, represented by the symbol GS1, and name the H(n,m) in the pixel group where m is greater than M as the right pixel gray level, represented by the symbol GS2; In practice, slight color shifts may occur during shooting, which is normal. In this embodiment, analyzing H(n,M) and pixel groups is essentially analyzing the difference in grayscale values ​​between the pixel group and H(n,M) in each row of surface points. H(n,M) represents the middle column of pixels in the surface grayscale image. The distance between any two surface points in a pixel group and H(n,M) is the same, and their grayscale value deviations are usually very similar. H(n,M) is therefore m =M of H(n,m), for example, in this embodiment, max(m) is obtained as 1024, and M is calculated as 512, resulting in 511 pixel groups. The surface point in column 1024 is not included in the reference range because there is no H(n,m) with the same F. For example, H(2,510) and H(2,514) both have F of 2, so they are integrated into one pixel group, where H(2,510) is GS1 and H(2,514) is GS2.

[0022] Step S302.4: Calculate |GS1-H(n,M)| and name the calculation result as the left pixel difference; calculate |GS2-H(n,M)| and name the calculation result as the right pixel difference. Step S302.5: Sort and number the pixel groups according to F in ascending order, using the symbol PG. i Let PG be a non-zero natural number, where i is the index of PG, and PG is represented by the symbol R(i,j). i The left and right pixel differences are defined as follows, where j takes values ​​of 1 and 2 and (i,j) is the index of R, where R(i,1) represents PG. i The difference in pixels on the left, R(i,2) represents the difference in pixels on the left. i The pixel difference on the right side; Step S302.6: Establish a Cartesian coordinate system with i as the horizontal axis and R(i,j) as the vertical axis, and name it a single-row feature map. Enter R(i,j) into the single-row feature map according to i. Name the coordinate point j=1 as the first coordinate and the coordinate point j=2 as the second coordinate. Perform linear regression on the first coordinate and the second coordinate respectively to obtain the slope of the linear regression function, obtain KL and KR, calculate |KL-KR|, and name the calculation result as gray-scale feature. Each value of n corresponds to a gray-scale feature. In this specific implementation, H(2,510) is 230, H(2,514) is 231, and H(2,512) is 230. The calculated difference between the left and right pixels is 0, and the difference between the right and right pixels is 1. The F of the pixel group listed here is 2, and it is sorted in ascending order as PG2, i=2. Thus, R(2,1) is 0 and R(2,2) is 1. A single-row feature map is constructed. At this time, the data in the single-row feature map all belong to H(2,m). When the galvanized surface of the strip steel is qualified, KL and KR are very close, with only a small difference. The grayscale feature is the difference between KL and KR. The grayscale feature when n is 2 is 0.0003.

[0023] Step S4 involves extracting the three-channel color value features and grayscale features of the qualified galvanized surface image and performing deep learning to obtain the quality inspection standard. Step S4 includes the following sub-steps: Step S401: Extract the three-channel color value features of the qualified galvanized surface image and perform deep learning to obtain the three-channel color value standard of galvanized strip steel. Step S401 includes the following sub-steps: Step S401.1: Name the qualified galvanized steel strip as a reference sample, and obtain the three-channel color value features of all galvanized surface images of the same reference sample, and name them as three-channel color value reference features. Step S401.2: Obtain the difference between the maximum and minimum values ​​in the three-channel color value reference features, and name it as the three-color difference. Step S401.3: Statistically analyze the three-color differences of all reference samples, obtain the maximum value, and obtain the three-channel color value standard; In practice, there is a certain fluctuation range between different three-channel color value reference features. The three-color difference reveals the fluctuation range of the three-channel color value reference features. By obtaining the maximum value, the fluctuation range of the three-channel color value features under normal circumstances can be obtained, thereby determining whether the difference between the color values ​​of the pixels on the galvanized steel strip surface is normal.

[0024] Step S402: Extract the grayscale features of the qualified galvanized surface image and perform deep learning to obtain the grayscale standard of galvanized strip steel. Step S402 includes the following sub-steps: Step S402.1: Extract the grayscale features of the reference sample, and label the grayscale features of the nth row as T. n Statistical analysis of T values ​​in all reference samples max(n) The range of grayscale values ​​is named the grayscale baseline standard. In specific implementation, T max(n)This refers to the grayscale features of the row of surface points closest to the camera, which exhibits the least fluctuation in grayscale features. However, as the distance from the camera increases, the fluctuation in grayscale features changes to some extent. Statistical T max(n) The range is used to obtain the grayscale baseline standard, which is used to determine whether the grayscale features of the surface points in the row closest to the camera are normal; Please see Figure 3 As shown, in step S402.2, for any reference sample, with n as the X-axis, T n Establish a Cartesian coordinate system for the Y-axis, name it Multi-Row Feature Map, and set T... n Enter multiple rows of feature maps according to n, and name the coordinate points in the multiple rows of feature maps as multiple row feature points; Please see Figure 4 As shown in step S402.3, the topmost feature point in each value of n is named a feature vertex. Adjacent feature vertices are connected by straight lines to form a broken line, named the multi-row feature line. The T-line on the multi-row feature line... n The point corresponding to this location is marked as S. n ; In practice, a multi-row feature map of a reference sample is constructed as follows: Figure 3 As shown, due to the excessive and dense amount of data, it is not convenient to observe in this embodiment, therefore... Figure 3 The image only shows a portion of the data to clearly illustrate the feature vertices, connecting them to form multiple rows of feature lines, such as... Figure 4 As shown, each feature vertex corresponds to an S. n It is important to note that Figure 3 A large amount of data has been omitted, and this is not the multi-row feature map obtained in the final implementation. Please see Figure 5 As shown, in step S402.4, the multi-row feature lines of each reference sample are placed in the same new multi-row feature map, and the S of each multi-row feature line is... max(n) The overlapping is performed, and the resulting multi-row feature map is named the advanced standard map. Please see Figure 6 As shown, in step S402.5, for each value of n in the advanced standard diagram, the topmost S... n Named the coordinate vertex, connect adjacent coordinate vertices with straight lines, and name the resulting polyline the grayscale progression standard. The grayscale basic standard and the grayscale progression standard together constitute the grayscale standard. In practice, assuming there are three reference samples, the multi-row feature lines of the three reference samples are placed on the same new multi-row feature map to obtain the advanced standard map, as shown below. Figure 5 As shown, the rightmost endpoints of each multi-row feature line are aligned. Figure 5In this process, only multiple feature lines need to be analyzed; the X and Y axes are not meaningful. The final grayscale progression standard is obtained as follows: Figure 6 As shown, Figure 6 The grayscale progression standard shown ensures a higher degree of error reception for pixel rows far from the camera, and in all cases, all grayscale features are below the grayscale progression standard.

[0025] Step S5 involves performing a quality inspection on the galvanized surface image based on quality inspection standards to determine whether the quality of the galvanized layer on the strip steel is qualified. Step S5 includes the following sub-steps: Step S501: Name the image of the galvanized surface to be analyzed as the image to be analyzed, and extract the three-channel color value features and grayscale features of the image to be analyzed, and name them as the first feature to be analyzed and the second feature to be analyzed, respectively. Step S502: Analyze whether the three-color difference of the first feature to be analyzed is less than or equal to the three-channel color value standard. If yes, output a three-channel qualified signal; otherwise, output a three-channel unqualified signal. In practice, if the three-color difference of the first feature is less than or equal to the three-channel color value standard, it means that the difference between the R-pass value, G-pass value and B-pass value in the color value of each pixel in the image to be analyzed is within the normal range. Therefore, the three-channel qualified signal is output, indicating that there is no obvious quality defect in color in the image to be analyzed. Step S503, analyze the T of the second feature to be analyzed. max(n) If the grayscale standard is met, output a basic verification pass signal; otherwise, output a basic verification failure signal. Step S504: If the basic verification signal is passed, analyze the multi-row feature map of the second feature to be analyzed, name it the feature map to be analyzed, and connect the rightmost endpoint of the grayscale progression standard to the T in the feature map to be analyzed. max(n) If the feature vertices overlap, query whether all feature points in multiple rows are below the grayscale advancement standard. If yes, output an advancement verification pass signal; otherwise, output an advancement verification failure signal. Step S505: If three-channel qualified signals and advanced verification pass signals are output simultaneously, the surface quality of the galvanized strip steel is marked as qualified; otherwise, the surface quality of the galvanized strip steel is marked as unqualified. In specific implementation, the second feature T to be analyzed max(n) The closest distance to the camera results in the lowest difference in grayscale values. If T max(n) If it is at the grayscale baseline standard, then it represents T. max(n) No abnormal offset was found, so the basic verification signal was passed. At this point, advanced analysis is performed, specifically verifying the T of the second feature to be analyzed. nTo determine whether the distribution conforms to the grayscale progression standard, the rightmost endpoint of the grayscale progression standard is overlapped with the rightmost endpoint of the multi-line feature line. If the multi-line feature line has no part larger than the grayscale progression standard, the grayscale features of the image to be analyzed are considered normal. If there are defects on the galvanized steel strip surface in the image to be analyzed, the balance of grayscale features will be broken, and the multi-line feature line will definitely have a part larger than the grayscale progression standard.

[0026] Example 2: This application provides an electronic device, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, and the processor can call these instructions. When the processor executes a computer-readable instruction, it performs steps such as those in a machine vision quality inspection method for galvanized steel strip production, to achieve the following functions: setting up a darkroom; capturing an image of the galvanized surface; extracting features from the galvanized surface image to obtain its three-channel color value features and grayscale features; extracting the three-channel color value features and grayscale features of a qualified galvanized surface image and analyzing the quality inspection standards; and performing quality inspection on the galvanized surface image based on the quality inspection standards to determine whether the quality of the galvanized layer on the strip is qualified.

[0027] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0028] Example 3: This application also provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the machine vision quality inspection method for galvanized steel strip production provided by the above methods. The method includes: setting up a dark room; capturing an image of the galvanized surface; extracting features from the galvanized surface image to obtain the three-channel color value features and grayscale features of the galvanized surface image; extracting the three-channel color value features and grayscale features of the galvanized surface image that meets the quality standards and analyzing the quality inspection standards; and performing quality inspection on the galvanized surface image based on the quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is qualified.

[0029] Example 4: This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it performs the steps of the machine vision quality inspection method for galvanized steel strip production surface as described above, to achieve the following functions: setting up a darkroom; capturing an image of the galvanized surface; extracting features from the galvanized surface image to obtain the three-channel color value features and grayscale features of the galvanized surface image; extracting the three-channel color value features and grayscale features of the galvanized surface image that meets the quality standards and analyzing the quality inspection standards; and performing quality inspection on the galvanized surface image based on the quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is qualified.

[0030] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0031] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A machine vision quality inspection method for galvanized steel strip production surfaces, characterized in that, Includes the following steps: A dark room was built, and light sources and high-definition camera devices were installed inside. The galvanized steel strip is fed into a dark chamber via a conveyor belt and an image of the galvanized surface is captured. Feature extraction is performed on the image of the galvanized surface to obtain the three-channel color value features and grayscale features of the galvanized surface image; Extract the three-channel color value features and grayscale features of the qualified galvanized surface image and process them to obtain the quality inspection standard; The quality of the galvanized surface image is inspected based on the quality inspection standards to determine whether the quality of the galvanized layer of the strip steel is qualified. Feature extraction of galvanized surface images, obtaining the three-channel color value features and grayscale features of the galvanized surface images, includes the following sub-steps: Color features were extracted from the image of the galvanized surface to obtain the three-channel color value features of the galvanized surface image; Gray-scale features are extracted from the image of the galvanized surface to obtain the gray-scale features of the galvanized surface image; Extracting color features from an image of a galvanized surface to obtain its three-channel color value features includes the following sub-steps: The contour of the strip steel in the galvanized surface image is obtained by contour extraction and named as the target contour. The pixels in the target contour are named as surface points. The surface point located in the nth row and mth column is labeled as P(n,m), where n and m are both non-zero natural numbers and (n,m) is the index of P; Obtain the color values ​​of P(n,m), which include R-pass, G-pass, and B-pass values, and use the symbol C. h Let R represent the pass value, G represent the pass value, and B represent the pass value, where R represent the pass value, G represent the pass value, and B represent the pass value, C1, C2, and C3 respectively, and h is a non-zero natural number and h is the index of C. With h as the X-axis, C h Establish a Cartesian coordinate system for the Y-axis, named the T-junction feature map, and set C... h Enter the three-channel feature map according to h, perform linear regression on the three-channel feature map, obtain the slope of the linear regression function, and name it as three-channel color value feature. Each P(n,m) corresponds to a three-channel color value feature. Extracting grayscale features from an image of a galvanized surface involves the following sub-steps: The image of the galvanized surface is converted to grayscale to obtain a grayscale image of the surface. P(n,m) in the grayscale image of the surface is labeled as Q(n,m), and the grayscale value of Q(n,m) is labeled as H(n,m). Calculate max(m) / 2, and mark the integer part of the result as M, where max() is the maximum value operator; Analyze each value of n independently, calculate |mM|, label the calculation result as F, integrate the two H(n,m) with equal F into a pixel group, name the H(n,m) in the pixel group where m is less than M as the left pixel gray level, represented by the symbol GS1, and name the H(n,m) in the pixel group where m is greater than M as the right pixel gray level, represented by the symbol GS2; Calculate |GS1-H(n,M)| and name the result as the left pixel difference; calculate |GS2-H(n,M)| and name the result as the right pixel difference. Pixel groups are sorted and numbered in ascending order of F, using the symbol PG. i Let PG be a non-zero natural number, where i is the index of PG, and PG is represented by the symbol R(i,j). i The left and right pixel differences are defined as follows, where j takes values ​​of 1 and 2 and (i,j) is the index of R, where R(i,1) represents PG. i The difference in pixels on the left, R(i,2) represents the difference in pixels on the left. i The difference in pixels on the right side; Establish a Cartesian coordinate system with i as the horizontal axis and R(i,j) as the vertical axis, named a single-row feature map. Input R(i,j) into the single-row feature map according to i. Name the coordinate point j=1 as the first coordinate and the coordinate point j=2 as the second coordinate. Perform linear regression on the first coordinate and the second coordinate respectively to obtain the slope of the linear regression function, and obtain KL and KR. Calculate |KL-KR| and name the calculation result as gray-scale feature. Each value of n corresponds to a gray-scale feature.

2. The machine vision quality inspection method for galvanized steel strip production surface according to claim 1, characterized in that, Setting up a dark room and installing light sources and high-definition camera equipment within it includes the following sub-steps: Construct a black room, which is a closed space where no external light source illuminates the interior except for artificial light; Install a light fixture at the geometric center of the top of the darkroom; A high-definition camera is installed at the geometric center of any side of the black room, and the angle between the high-definition camera and the plane of the black room is a first angle.

3. The machine vision quality inspection method for galvanized steel strip production surface according to claim 2, characterized in that, The process of feeding the galvanized steel strip into a darkroom via a conveyor belt and taking images of the galvanized surface includes the following sub-steps: The strip steel is galvanized, and after galvanizing, the strip steel is placed on a conveyor belt and sent into the black chamber. The galvanized surface is imaged by a high-definition camera in a darkroom. The length of each shot taken in the black chamber is limited. A strip of steel needs to go through a first number of shots to obtain a first number of galvanized surface images. During quality inspection, each galvanized surface image is analyzed independently.

4. The machine vision quality inspection method for galvanized steel strip production surface according to claim 3, characterized in that, Extracting and processing the three-channel color value features and grayscale features of the qualified galvanized surface image to obtain the quality inspection standard includes the following sub-steps: Extract the three-channel color value features from the qualified galvanized surface image and process them to obtain the three-channel color value standard of galvanized strip steel; The grayscale features of qualified galvanized surface images are extracted and processed to obtain the grayscale standard of galvanized steel strip.

5. The machine vision quality inspection method for galvanized steel strip production surface according to claim 4, characterized in that, Extracting and processing the three-channel color value features of qualified galvanized surface images to obtain the three-channel color value standard for galvanized steel strip includes the following sub-steps: The qualified galvanized steel strip is named as the reference sample. The three-channel color value features of all galvanized surface images of the same reference sample are obtained and named as the three-channel color value reference features. The difference between the maximum and minimum values ​​in the three-channel color value reference feature is obtained and named the three-color variability. The differences in the three colors of all reference samples are statistically analyzed, and the maximum value is obtained to obtain the three-channel color value standard.

6. The machine vision quality inspection method for galvanized steel strip production surface according to claim 5, characterized in that, Extracting and processing the grayscale features of qualified galvanized surface images to obtain the grayscale standard of galvanized steel strip includes the following sub-steps: Extract the grayscale features of the reference sample, and label the grayscale features of the nth row as T. n Statistical analysis of T values ​​in all reference samples max(n) The range of grayscale values ​​is named the grayscale baseline standard. For any reference sample, with n as the X-axis, T n Establish a Cartesian coordinate system for the Y-axis, named "Multi-row Feature Map," and set T... n Enter multiple rows of feature maps according to n, and name the coordinate points in the multiple rows of feature maps as multiple row feature points; For each value of n, the topmost feature point in the multi-row is named a feature vertex. Connect adjacent feature vertices with straight lines to obtain a broken line, named the multi-row feature line. Then, T is placed on the multi-row feature line. n The point corresponding to this location is marked as S. n ; Place the multi-row feature lines of each reference sample into the same new multi-row feature map, and set the S of each multi-row feature line... max(n) The overlapping is performed, and the resulting multi-row feature map is named the advanced standard map. In the advanced standard diagram, for each value of n, the topmost S... n Named the coordinate vertex, connect adjacent coordinate vertices with straight lines, and name the resulting polyline the grayscale progression standard. The grayscale basic standard and the grayscale progression standard together constitute the grayscale standard.

7. A machine vision quality inspection method for galvanized steel strip production surface according to claim 6, characterized in that, Quality inspection of galvanized surface images based on quality inspection standards to determine whether the quality of the galvanized layer on the strip steel is qualified includes the following sub-steps: The image of the galvanized surface to be analyzed is named the image to be analyzed. The three-channel color value features and grayscale features of the image to be analyzed are extracted and named the first feature to be analyzed and the second feature to be analyzed, respectively. Analyze whether the three-color difference of the first feature to be analyzed is less than or equal to the three-channel color value standard. If yes, output a three-channel qualified signal; otherwise, output a three-channel unqualified signal. Analyze the T of the second feature to be analyzed max(n) If the grayscale standard is met, output a basic verification pass signal; otherwise, output a basic verification failure signal. If the basic verification signal is passed, then the multi-row feature map of the second feature to be analyzed is analyzed and named the feature map to be analyzed. The rightmost endpoint of the grayscale progression standard is then compared with T in the feature map to be analyzed. max(n) If the feature vertices overlap, query whether all feature points in multiple rows are below the grayscale advancement standard. If yes, output an advancement verification pass signal; otherwise, output an advancement verification failure signal. If three channels of qualified signals and advanced verification pass signals are output simultaneously, the surface quality of the galvanized steel strip is marked as qualified; otherwise, the surface quality of the galvanized steel strip is marked as unqualified.

Citation Information

Patent Citations

  • Detection optimization method of double-sensor surface quality detection system for hot-dip galvanized strip steel

    CN113281341A