A method for detecting hot-rolled strip steel and warning of deviation

Abstract

The present invention discloses a method for detecting hot-rolled strip steel and warning of deviation, including: collecting and constructing an image data set of hot-rolled strip steel; establishing a detection model for hot-rolled strip steel based on a support vector machine; designing a deviation warning algorithm for hot-rolled strip steel based on the steel-containing images detected by the detection model for hot-rolled strip steel; embedding the detection model for hot-rolled strip steel and the deviation warning algorithm for hot-rolled strip steel into the on-site surface detection system, and real-time feedback of strip steel warning and deviation amount to the host computer. Based on the production image data of hot-rolled strip steel, the present invention constructs a detection model for hot-rolled strip steel based on machine learning algorithms. Secondly, a deviation warning algorithm for hot-rolled strip steel is designed for the classified steel-containing images. Using this method can accurately and quickly identify the categories of steel-containing and non-steel-containing during the rolling process of different steel coils, timely warn and feedback on the deviation situation during operation, and is of great significance to the stable operation of strip steel rolling and shape control.

Description

Technical Field

[0001] The present invention belongs to the field of process detection in metallurgical rolling technology, and particularly relates to a method for detecting hot-rolled strip steel and warning of deviation from the center line. Background Art

[0002] During the hot-rolling production process, it is very easy for the center of the strip steel to deviate from the center of the roller table, which is commonly called deviation from the center line in industrial production. Many domestic and foreign scholars have conducted relevant research on the influencing factors causing deviation from the center line and the methods for rectifying it. For example: strip steel shape defects (sickle bend, single-sided wave, etc.), the influence of roller geometry, the axial movement of conveying rollers, and the process parameters of the unit. Most hot-rolling production lines are equipped with side guide plate devices in front of the coiler to clamp the strip steel and align it with the pinch roll of the coiler for rectification, but it can only rectify the strip steel entering the side guide plate part and cannot completely avoid the phenomenon of deviation from the center line. When the hot-rolled strip steel deviates from the center line before coiling, it mainly causes two types of defects. One is the strip steel surface defect: the uneven pressure distribution between the edge of the strip steel and both sides of the side guide plate caused by the deviation from the center line of the strip steel before coiling generates friction, resulting in edge peeling defects and plaque defects, as well as hole defects formed on the strip steel surface due to the strip steel sticking slag and falling off. The other is the coiling shape defect: the inner tower or outer tower defect of the steel coil caused by the excessive deviation from the center line of the strip steel head and tail or the excessive control error of the side guide plate. In addition, when the deviation from the center line is serious, the strip steel rolling at high speed will collide with the side guide plate, damaging the production equipment and causing the production line to stop production.

[0003] At present, most hot-rolling production lines lack the detection of strip steel deviation from the center line and can only rely on operators to observe the monitoring equipment and carry out manual intervention. This method is extremely vulnerable to external factors, requires operators to make correct operations based on experience in a short time, has low efficiency and cannot guarantee the stability of intervention, and also has a relatively high labor intensity requirement for operators. Therefore, designing a method for detecting hot-rolled strip steel and warning of deviation from the center line with strong versatility and fast response speed has practical significance for liberating productivity, improving the coiling shape quality of steel coils and the surface quality of strip steel. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for detecting hot-rolled strip steel and warning of deviation from the center line. This method is based on the production image data of hot-rolled strip steel, uses a support vector machine to construct a hot-rolled strip steel detection model, and then designs a hot-rolled strip steel deviation warning algorithm based on an improved probabilistic statistical Hough transform line detection algorithm for the steel-containing images classified by the hot-rolled strip steel detection model. Using this method can accurately and quickly identify the categories of the presence and absence of steel during the rolling process, and the hot-rolled strip steel deviation warning algorithm designed for the steel-containing images can give a warning feedback on the deviation from the center line of the strip steel in real time, facilitating subsequent timely rectification.

[0005] In order to achieve the above purpose, a method for detecting hot-rolled strip steel and warning of deviation from the center line proposed by the present invention mainly includes the following steps:

[0006] Collect the production data of hot-rolled strip steel and perform preprocessing to construct an image dataset of hot-rolled strip steel;

[0007] Use the constructed image dataset of hot-rolled strip steel to train a support vector machine to establish a detection model for hot-rolled strip steel;

[0008] Use the detection model of hot-rolled strip steel to detect the preprocessed image data of hot-rolled strip steel, save the detected images with steel into the dataset with steel, and save the detected images without steel into the dataset without steel;

[0009] Design a deviation warning algorithm for hot-rolled strip steel based on the detected images with steel by the detection model of hot-rolled strip steel. The deviation warning algorithm for hot-rolled strip steel is to use an improved probabilistic statistical Hough transform line detection algorithm to extract the potential roller paths and strip steel edge lines in the image, and then perform deviation warning and deviation amount prediction according to the position coordinates of the strip steel edge lines screened based on the position between the strip steel and the roller path. The improved probabilistic statistical Hough transform line detection algorithm is to limit the detection angle range in the probabilistic statistical Hough transform according to the actual rolling direction and fuse the different lines detected for the same edge;

[0010] Embed the detection model of hot-rolled strip steel and the deviation warning algorithm for hot-rolled strip steel into the surface inspection system at the production site, monitor the operation of the strip steel during production, and real-time feedback the warning information and the deviation amount prediction result to the upper computer.

[0011] Furthermore, collecting the production data of hot-rolled strip steel and performing preprocessing to construct an image dataset of hot-rolled strip steel includes:

[0012] Install an image acquisition device above the roller path at the outlet of the laminar cooling section of the hot-rolling production line to collect a batch of production image data of hot-rolled strip steel;

[0013] Perform preprocessing correction for the perspective distortion existing in the image data by using perspective transformation;

[0014] Randomly select several images with steel and images without steel from the preprocessed image dataset, and randomly divide them into a training set and a test set according to a set ratio.

[0015] Furthermore, using the constructed dataset to train a support vector machine to establish a detection model for hot-rolled strip steel includes:

[0016] Design and extract the HOG feature vectors of the training set and the test set;

[0017] Use the extracted HOG feature vector set of the training set to train a support vector machine to obtain a detection model for hot-rolled strip steel, and use the extracted HOG feature vector set of the test set to verify the model accuracy.

[0018] Furthermore, design and extract the HOG feature vectors of the training set and the test set, including:

[0019] Randomly select a steel-containing image from the training set, and design and extract the HOG feature parameters according to the image and the target size, including: the sliding window size, the pixel size of each cell, and the number of cells in each block;

[0020] Convert the image into a single-channel grayscale image, and perform color space normalization on the single-channel grayscale image using the Gamma correction method, and set the Gamma coefficient;

[0021] Calculate the directional gradient, using [-1, 0, 1] and [-1, 0, 1] T operators to perform convolution calculation with the normalized image to obtain the horizontal directional gradient G x (x, y) and the vertical directional gradient G y (x, y), and calculate the gradient magnitude and gradient direction at the point (x, y) according to the directional gradient and gradient direction

[0022] Construct a gradient histogram and generate a HOG feature vector.

[0023] Furthermore, extract the potential roller path and strip edge lines based on the improved probabilistic statistical Hough transform line detection algorithm, including:

[0024] S1: Take an image from the classified steel-containing image dataset and perform grayscale processing;

[0025] S2: Perform Gaussian filtering on the grayscale image, and then perform Canny edge detection;

[0026] S3: Obtain the set P of edge points of the edge detection image;

[0027] S4: According to the duality principle between points and lines, convert the point set P from the points in the image coordinate system x-y to the lines in the parameter coordinate system α-l. Divide the parameter coordinate system into n intervals, and each interval corresponds to an accumulator. Set the interval precision α0 and l0 of the parameter coordinate division, set the range of the parameter coordinate system division as α = [a, b], l = [c, d], and set the line accumulation threshold as threshod;

[0028] S5: Randomly extract a pixel point from the point set P, calculate its α value and l value in the parameter coordinate system, find the corresponding interval accumulator and increment its count by 1. If there is no corresponding interval accumulator, delete the point from the point set P;

[0029] S6: Determine whether there is an accumulator count value greater than threshod. If so, go to step S7; otherwise, go to step S5;

[0030] S7: Complete a straight line detection, clear the votes in the accumulator, and delete all corresponding points on the straight line from the point set P;

[0031] S8: Determine whether the point set P is an empty set. If P is an empty set, go to step S9; otherwise, go to step S5;

[0032] S9: Extend each detected straight line to the upper and lower edges of the image;

[0033] S10: Use the bubble sort method to sort the straight lines in ascending order according to the abscissa value of the upper endpoint;

[0034] S11: Take two straight lines in turn according to the abscissa value, and calculate their slopes and the distance between the two straight lines at the midpoint of the ordinate;

[0035] S12: Determine whether the slopes and distances of the two straight lines meet the set thresholds, and fuse the two straight lines that meet both conditions into a new straight line;

[0036] S13: Traverse all straight lines until there are no two straight lines that meet the slope and distance thresholds;

[0037] S14: Obtain the straight line set of the strip steel and the roller path edges.

[0038] Further, perform deviation warning and deviation amount prediction based on the position coordinates of the strip steel edge straight lines screened according to the position between the strip steel and the roller path, including:

[0039] S21: Obtain the pixel width W of the image, the actual width W mm_strip of the strip steel, and the actual width W mm_roll of the roller path, and calculate the pixel width of the strip steel in the image Obtain the straight line set of the strip steel and the roller path edges;

[0040] S22: Take a straight line, calculate the average value of the x-axis coordinates of its two endpoints, and determine whether it is greater than W / 2. If it meets the condition, save it as the right straight line set; otherwise, save it as the left straight line set, and then delete the straight line from the straight line set of the strip steel and the roller path edges;

[0041] S23: Determine whether the straight line set of the strip steel and the roller path edges is an empty set. If it is an empty set, go to step S24; otherwise, go to step S22;

[0042] S24: Take the last straight line in the left straight line set and the first straight line in the right straight line set to obtain the strip steel edge straight line set;

[0043] S25: Set the left and right deviation warning position coordinates of the strip steel (xleft_warning , x right_warning ), calculate the proportional relationship between the actual size and the image pixel size

[0044] S26: Calculate the average value of the x-axis coordinates of the two endpoints of the left and right edge lines in the strip edge straight line set and denote it as x left_edge and x right_edge ;

[0045] S27: If x left_edge ≤ x left_warning and x right_edge ≤ x left_warning + W pixel_strip , output a warning for the left deviation of the strip;

[0046] If x left_warning < x left_edge < x right_warning - W pixel_strip and x left_warning + W pixel_strip < x right_edge < x right_warning , output that the strip deviation is normal;

[0047] If x left_edge ≥ x right_warning - W strip and x right_edge ≥ x right_warning , output a warning for the right deviation of the strip;

[0048] S28: Calculate the predicted value of the strip deviation,

[0049] The above technical solution has the following beneficial effects:[[ID=]]

[0050] The present invention provides a method for detecting hot-rolled strip steel and warning of deviation. Image data is collected based on machine vision technology. Feature engineering design is carried out on the production images of hot-rolled strip steel to extract the HOG feature vectors of the images, and a detection model for hot-rolled strip steel is constructed based on the HOG feature vector set. This model can automatically and accurately identify the categories of the presence or absence of steel in the images during the rolling process. Secondly, the present invention designs a deviation warning algorithm for the images with steel. This algorithm is based on an improved probabilistic statistical Hough transform line detection algorithm. Compared with the original algorithm, the improved algorithm limits the detection angle range in the probabilistic statistical Hough transform according to the actual rolling angle of the strip steel, improves the detection speed of the algorithm, and eliminates the redundant lines detected by the useless background edges. Secondly, the different lines detected by the same edge are fused to eliminate the overlapping lines detected by environmental influences such as water mist. This deviation warning algorithm for hot-rolled strip steel can accurately and quickly detect the edge lines of the strip steel for deviation warning. The method of the present invention has low cost, high precision, and does not require complex device mechanisms, and has practical significance for liberating productivity, improving the coil shape quality of steel coils and the surface quality of strip steel. Description of the Drawings

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0052] Figure 1 It is a flowchart of the method for detecting hot-rolled strip steel and warning of deviation in the embodiments of the present invention;

[0053] Figure 2 It is the original image and the image after data preprocessing in the embodiments of the present invention;

[0054] Figure 3 It is the HOG feature map for extracting the images with steel and without steel in the embodiments of the present invention;

[0055] Figure 4 It is the straight line detection process of the deviation warning algorithm for hot-rolled strip steel in the embodiments of the present invention and the comparison before and after the algorithm improvement;

[0056] Figure 5 It is a flowchart of the improved probabilistic statistical Hough transform line detection algorithm in the embodiments of the present invention;

[0057] Figure 6 It is a flowchart of the strip steel deviation warning and deviation amount prediction algorithm in the embodiments of the present invention. Detailed Embodiments

[0058] To enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solution in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0059] In the embodiments of the present invention, a method for detecting and warning the deviation of hot-rolled strip steel is provided, including the following implementation steps:

[0060] S1: Collect the production data of hot-rolled strip steel and perform preprocessing to construct an image data set of hot-rolled strip steel;

[0061] In specific implementation, the specific steps include:

[0062] S11: Install an image acquisition device above the roller path at the outlet of the laminar cooling in the hot-rolling production line to collect a batch of rolling image data of hot-rolled strip steel. A total of n1 = 19916 on-site image data are collected and saved as the original image data set Imgset1.

[0063] S12: Perform preprocessing correction of the perspective distortion on the perspective distortion existing in the image data by using perspective transformation; Take an image from the original image data set Imgset1 and denote it as I org And perform perspective transformation processing: Determine the size of the image I after perspective transformation transform Is 600×280; Select the original coordinate points of the image I org At the edge positions of the middle three roller paths, denoted as (788, 818), (1195, 848), (718, 985), (1223, 1020), and select the target coordinate points of the image I after perspective transformation transform Are denoted as (0, 0), (600, 0), (0, 280), (600, 280). The actual production data and the perspective transformation image are as Figure 2 Shown. The perspective transformation process is as follows: First, convert the original coordinate points O i (x, y) (i = 1 to 4) and the target coordinate points T i (x′, y′) (i = 1 to 4) in the Cartesian coordinate system to the homogeneous coordinate system and denote them as (X, Y, Z) and (X′, Y′, Z′), and the conversion relationship is denoted as Generally, let Z = 1, then the original coordinate points in the homogeneous coordinate system can become (x, y, 1); Then, according to the perspective transformation formula (M transform Is the perspective transformation matrix, And generally let a 33 = 1) can be obtained According to the formula we can obtain that is Eight equations can be obtained according to the given 4 original coordinate points and 4 target coordinate points Then the perspective transformation matrix M transf o rm The eight position parameters in can be obtained, and then according to M transf o rm Perform perspective transformation on all points of the image Io rg to obtain I transf o rm and save it to the image dataset Imgset2

[0064] S13: Randomly select 1000 steel images and 1000 non-steel images from the preprocessed image dataset Imgset2, denoted as GS_Imgset1 and NS_Imgset1 respectively, and then randomly divide them into a training set and a test set according to the ratio of 8:2, denoted as Trainset{GS_Imgset, NS_Imgset} and Testset{GS_Imgset, NS_Imgset}

[0065] S2: Use the constructed hot-rolled strip steel image dataset to train a support vector machine to establish a hot-rolled strip steel detection model

[0066] In the specific implementation, the specific steps include

[0067] S21: Design and extract the HOG feature vectors of the training set and the test set, which specifically include the following steps

[0068] S211: Randomly select a steel image from the training set and denote it as I train and design and extract the HOG feature parameters according to the image and the target size: I train The width and height dimensions are 600×280. In order to reduce the complexity of the extracted features, the image is scaled to 1 / 4 of the original size before feature extraction, and the scaled width and height are 150×70; set the pixel size of each cell to 10×10, and the entire image can be divided into 15×7 cells; set each block to contain 4 cells, that is, the pixel size of each block is 20×20

[0069] S212: In order to reduce the computational complexity, convert the input image I train into a single-channel grayscale image I gray , and use the Gamma correction method to perform color space normalization on the single-channel image I gray , set the Gamma coefficient γ = 0.5, and the normalization process is: I gamma (x,y)=Igray (x, y) γ . Among them, I gray (x, y) is the gray value at (x, y) in a single-channel image, and I gamma (x, y) is the gray value at (x, y) in the image after Gamma normalization, and the Gamma-corrected image I is obtained gamma ;;

[0070] S213: Calculate the directional gradient: Use [-1, 0, 1] and [-1, 0, 1] respectively T operators to convolve with the image I gamma to obtain the horizontal directional gradient G gamma (x, y) at the point (x, y) and the vertical directional gradient G x (x, y). Calculate the gradient magnitude and gradient direction at the point (x, y) according to the directional gradient y (x, y). Calculate the gradient magnitude and gradient direction at the point (x, y) according to the directional gradient and the gradient direction

[0071] S214: Construct a gradient histogram and generate a HOG feature vector: Construct a gradient histogram in units of cells. First, divide the gradient range. Select the unsigned gradient direction (i.e., from 0 - 180°) and divide it into 9 regions, with each region being 20° wide. Then, count the gradient magnitude G(x, y) of each pixel in a cell into the respective regions according to its gradient direction θ(x, y). Each cell will generate a gradient histogram with 9 directions. Since each block has 4 cells, each block will generate 36 feature vectors. Set the block step size to 10. Then, 14 blocks will be obtained in the horizontal direction and 6 blocks will be obtained in the vertical direction. The dimension of the HOG feature vector extracted from an image is 3024. The visualized results of the HOG feature vectors with and without steel are as Figure 2 shown

[0072] S22: Extract the HOG feature vectors of the training set and the test set according to step S21 and save them as datasets, denoted as Trainset{GS_HOGset, NS_HOGset} and Testset{GS_HOGset, NS_HOGset}

[0073] S23: Establish a hot-rolled strip detection model based on the support vector machine. Use Trainset {GS_HOGset, NS_HOGset} to train the support vector machine to obtain the hot-rolled strip detection model denoted as HSD_M. Then use Testset {GS_HOGset, NS_HOGset} to verify the accuracy of model HSD_M. The classification accuracy of model HSD_M on the test set Testset {GS_HOGset, NS_HOGset} reaches 100%.

[0074] S3: Use model HSD_M to detect strips in the image dataset Imgset2. Save the classified steel-containing images into the steel-containing dataset GS_Imgset2, with a total of 10,256 images, and save the classified steel-free images into the steel-free dataset NS_Imgset2, with a total of 9,660 images.

[0075] S4: Design a hot-rolled strip deviation warning algorithm based on the steel-containing images detected by the hot-rolled strip detection model;

[0076] In the specific implementation, the specific steps include:

[0077] Take an image data from the steel-containing image dataset GS_Imgset2 and denote it as I GS , Given I GS The actual width of the strip in it is denoted as W mm_strip = 1000 (mm), and the actual width of the on-site roller table is denoted as W mm_roll = 1380 (mm). For image I GS Design a hot-rolled strip deviation warning algorithm. The comparison of the strip edge straight line detection process and the algorithm before and after improvement is as Figure 4 shown. The specific implementation steps include:

[0078] S41: Extract the potential straight lines of the roller table and the strip edge in image I GS based on the improved probabilistic statistical Hough transform straight line detection algorithm. The algorithm flow is as Figure 5 shown. The specific implementation steps include:

[0079] (101) Grayscale the steel-containing image I GS to convert it into a grayscale image I gray , and create an empty set Line_1;

[0080] (102) Perform Gaussian filtering on image I gray . Set the Gaussian filter kernel size to (3, 3) to obtain the filtered image I Gauss ;

[0081] (103) For image I GaussPerform Canny edge detection with the first detection threshold set to 15 and the second detection threshold set to 50 to obtain the edge image I Canny ;

[0082] (104) Obtain the edge point set of image I Canny , denoted as set P;

[0083] (105) According to the duality principle between points and lines, convert the point set P from points in the image coordinate system x - y to lines in the parametric coordinate system α - l. Set the interval precision of the parametric coordinates α0 to π / 360 and l0 to 1. Set the range of the parametric coordinate system division intervals as α = [87, 93] and l = [0, H]. Set the line accumulation threshold threshod to 38. Divide the parametric coordinate system into n = 1960 intervals, and each interval corresponds to an accumulator;

[0084] (106) Randomly extract a pixel point from the point set P, calculate its α value and l value in the parametric coordinate system, and determine whether there is an accumulator in the corresponding interval. If there is an accumulator in the corresponding interval, increment the vote count of that accumulator by 1. If there is no accumulator in the corresponding interval, delete that point from the point set P;

[0085] (107) Determine whether there is an accumulator count value greater than threshod. If so, go to step (108); otherwise, go to step (106);

[0086] (108) Complete one line detection, clear the vote count in this accumulator and delete all the corresponding points on this line from the point set P. The detected line is denoted as l1 i (x [i][1] , y [i][1] , x [i][2] , y [i][2] ). Each line has 4 parameters (the x and y coordinates of two points of the line in the image), and save them to the set Line_1{l1 i}(i = 1 ~ N1);

[0087] (109) Determine whether the point set P is an empty set. If P is an empty set, go to step (110); otherwise, go to step (106);

[0088] (110) Obtain the strip steel and roller path edge line set Line_1{l1 i}(i = 1 ~ N1), and set the temporary variable o = 1;

[0089] (111) Take the line l1o from the set Line_1 and calculate the slope of the line l1o

[0090] (112) Determine whether it satisfies y[o][1] > y [o][2] , if it satisfies making x' [o][1] = x [o][2] , y' [o][1] = y [o][2] , x' [o][2] = x [o][1] , y' [o][2] = y [o][1] , and assign the value of (x' [o][1] , y' [o][1] , x' [o][2] , y' [o][2] ) to the straight line l1 o (x [o][1] , y [o][1] , x [o][2] , y [o][2] ), otherwise go to step (113);

[0091] (113) Judge whether it satisfies y [o][1] ≠ 0, if it satisfies, make and assign the value of (x new_[o][1] , 0, x [o][2] , y [o][2] ) to the straight line l1o(x [o][1] , y [o][1] , x [o][2] , y [ o ][2] ), otherwise go to step (114);

[0092] (114) Judge whether it satisfies y [o][2] ≠ H, if it satisfies, make and assign the value of (x [o][1] , y [o][1] , x new_[o][2] , H) to the straight line l1o(x [o][1] , y [o][1] , x [o][2] , y [o][2] ), otherwise go to step (115);

[0093] (115) Save the straight line l1o to the set Line_2, and make o = o + 1;

[0094] (116) Judge whether it satisfies o > N1, if it satisfies, go to step (117), otherwise go to step (111);

[0095] (117) Obtain the straight line set Line_2{l2 i}(i = 1 ~ N2), use the bubble sort method to sort the straight lines in Line_2 in ascending order according to the x [i][1] value, set the temporary variable r = 1, and set the temporary straight line variable (x[temp][1] ,y [temp][1] ,x [temp][2] ,y [temp][2] );

[0096] (118) Take the line l2 r (x [r][1] ,y [r][1] ,x [r][2] ,y [r][2] ) and l2 r+1 (x [r+1][1] ,y [r+1][1] ,x [r+1][2] ,y [r+1][2] ), determine whether x is satisfied [r][1] >x [r+1][1] , if satisfied, go to step (119), otherwise go to step (120);

[0097] (119) Let the straight line l2 r (x [r][1] ,y [r][1] ,x [r][2] ,y [r][2] ) is assigned to the temporary linear variable (x [temp][1] ,y [temp][1] ,x [temp][2] ,y [temp][2] ), the straight line l2 r+1 (x [r+1][1] ,y [r+1][1] ,x [r+1][2] ,y [r+1][2] ) is assigned to l2 r (x [r][1] ,y [r][1] ,x [r][2] ,y [r][2] );

[0098] (120) The temporary linear variable (x [temp][1] ,y [temp][1] ,x [temp][2] ,y [temp][2] ) is assigned to the line l2 r+1 (x [r+1][1] ,y [r+1][1] ,x [r+1][2] ,y [r+1][2] );

[0099] (121) Let r = r + 1, determine whether r = N2. If so, go to step (122); otherwise, go to step (118).

[0100] (122) Set the fusion line angle threshold θ0 = 1 (°), the fusion line distance threshold d0 = 2 (pixel), and set the temporary variable j = 1;

[0101] Take the straight line l2 from the set Line_2 j and l2 j+1 , calculate the slopes of the two straight lines respectively and denote them as k j and k j+1 , and denote the distance between the two straight lines at y = H / 2 as

[0102] (124) Judge whether it satisfies If it satisfies, fuse the straight line l2 j and l2 j+1 to become a new straight line and save it to the straight line set Line_3, otherwise go to step (121);

[0103] (125) Let j = j + 1;

[0104] (126) Judge whether it satisfies j > N2, if it satisfies, go to step (127), otherwise go to step (123);

[0105] (127) Obtain the fused strip steel and the straight line set Line_3{l3 i}(i = 4 ~ N3);

[0106] S42: The position of the roller table is fixed, and the strip steel moves left and right inside the roller table. According to the position relationship between the strip steel and the roller table, the straight lines at the edges of the strip steel are screened out, and the strip steel deviation warning and deviation amount prediction are carried out. The algorithm flow chart is as Figure 6 shown. According to the detected straight line set Line_3 at the edges of the strip steel and the roller table, the strip steel deviation warning and deviation amount detection are carried out. The specific implementation steps include:

[0107] (201) Denote the image pixel width as the variable W = 600 (pixel), and calculate the pixel width of the strip steel in the image Set the temporary variable p = 1, create empty sets Line_LE and Line_RE for saving the left and right edges of the strip steel and an empty set Line_4 for the screened straight lines;

[0108] (202) Take the straight line l3 from the set Line_3 p , and calculate the average value of the x-axis coordinates of the two endpoints of this straight line

[0109] (203) Judge whether it satisfies If it satisfies, save the straight line l3 p to the set Line_LE{lL i}(i = 2 ~ N L ), let p = p + 1, otherwise save the straight line l3 pSave to the set Line_RE{lR i}(i = 2 to N R ), and let p = p + 1;

[0110] (204) Determine whether p > N3 is satisfied. If it is satisfied, take the last straight line in the set Line_LE and the first straight line in the set Line_RE Save to the set Line_4, otherwise go to step (202);

[0111] (205) Obtain the set of strip edge straight lines Line_4{l4 i}(i = 2);

[0112] (206) Set the warning position coordinates (x left_warning , x right_warning ) of the strip left and right deviation to (59, 551), set the variables x left_edge and x right_edge to save the x-axis coordinates of the detected strip edge straight lines, let x left_edge = 0, x right_edge = 0, set the proportional relationship between the actual size and the image pixel size

[0113] (207) Calculate the average value of the x-axis coordinates of the two straight lines in the set Line_4, and record them as and

[0114] (208) Judge the value range of x left_edge and x right_edge :

[0115] If x left_edge ≤ x left_warning and x right_edge ≤ x left_warning + W pixel_strip Go to step (209);

[0116] If x left_warning < x left_edge < x right_warning - W pixel_strip and x left_warning + W pixel_strip < x right_edge < x right_warning Go to step (210);

[0117] If x left_edge ≥ x right_warning - W strip and x right_edge ≥ x right_warningGo to step (211);

[0118] (209) Output the warning of the left deviation of the strip steel, and calculate the predicted value of the strip steel deviation

[0119] (210) Output that the strip steel runs normally, and calculate the predicted value of the strip steel deviation

[0120] (211) Output the warning of the right deviation of the strip steel, and calculate the predicted value of the strip steel deviation

[0121] S5: Embed the hot-rolled strip steel detection model HSD_M and the hot-rolled strip steel deviation warning algorithm into the on-site surface detection system, monitor the strip steel operation in real time during production, and feedback the strip steel warning and deviation amount prediction results to the host computer in real time.

[0122] Use the steel dataset GS_Imgset2 with a total of 10,256 images to verify the hot-rolled strip steel deviation warning algorithm, and manually calibrate the classification results, as shown in Table 1. It can be seen that the accuracy of each category reaches more than 90%, and the overall accuracy reaches 99.30%, which can meet the production requirements.

[0123] Table 1

[0124]

[0125]

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting hot-rolled strip steel and warning of deviation, characterized in that, Including: Collecting hot-rolled strip production data and performing preprocessing to construct a hot-rolled strip image dataset; Using the constructed hot-rolled strip image dataset to train a support vector machine to establish a hot-rolled strip detection model; Using the hot-rolled strip detection model to detect the preprocessed hot-rolled strip image data, saving the detected steel-containing images to the steel-containing dataset, and saving the detected steel-free images to the steel-free dataset; Designing a hot-rolled strip deviation warning algorithm based on the steel-containing images detected by the hot-rolled strip detection model. The hot-rolled strip deviation warning algorithm is to use an improved probabilistic statistical Hough transform line detection algorithm to extract the potential roller paths and strip edge lines in the image, and then perform deviation warning and deviation amount prediction according to the position coordinates of the strip edge lines screened out between the strip and the roller path. The improved probabilistic statistical Hough transform line detection algorithm is to limit the detection angle range in the probabilistic statistical Hough transform according to the actual rolling direction, and fuse the different lines detected for the same edge; Embedding the hot-rolled strip detection model and the hot-rolled strip deviation warning algorithm into the surface inspection system at the production site to monitor the strip operation during production, and real-time feedback the warning information and the deviation amount prediction results to the upper computer; Among them, the improved probabilistic statistical Hough transform line detection algorithm includes: S1: Taking an image from the classified steel-containing image dataset for grayscale processing; S2: Performing Gaussian filtering on the grayscale image, and then performing Canny edge detection; S3: Obtaining the edge point set P of the edge detection image; S4: According to the duality principle between points and lines, convert the point set P from the points in the image coordinate system x - y to lines in the parametric coordinate system -l . Divide the parametric coordinate system into n intervals, with each interval corresponding to an accumulator. Set the precision and of the interval division in the parametric coordinate system, set the range of the interval division in the parametric coordinate system, , and set the line accumulation threshold to threshod; S5: Randomly extract a pixel point from the point set P, and calculate its value and l value, increment the accumulator count of the corresponding interval by 1. If there is no accumulator for the corresponding interval, delete the point from the point set P; S6: Judging whether there is an accumulator count value greater than threshod. If so, go to step S7, otherwise, go to step S5; S7: Completing a line detection, clearing the votes in the accumulator and deleting all the points corresponding to the line from the point set P; S8: Judging whether the point set P is an empty set. If P is an empty set, go to step S9, otherwise, go to step S5; S9: Extending each detected line to the upper and lower edges of the image; S10: Using the bubble sort method to sort the lines in ascending order of the abscissa value of the upper endpoint; S11: Taking two lines in turn according to the abscissa value, calculating their slopes and the distance between the two lines at the midpoint of the ordinate; S12: Judging whether the slopes and distances of the two lines meet the set thresholds, and fusing the two lines that meet both conditions into a new line; S13: Traversing all the lines until there are no two lines that meet the slope and distance thresholds; S14: Obtaining the strip and roller path edge line set.

2. The hot-rolled strip steel detection and deviation warning method according to claim 1, wherein Collecting hot-rolled strip production data and performing preprocessing to construct a hot-rolled strip image dataset, including: Installing an image acquisition device above the roller path at the laminar cooling outlet of the hot-rolled production line to collect a batch of hot-rolled strip production image data; Performing preprocessing correction of the distortion on the image data with perspective distortion by using perspective transformation; Randomly selecting several steel-containing images and steel-free images from the preprocessed image dataset, and randomly dividing them into a training set and a test set according to a set ratio.

3. The hot-rolled strip detection and deviation warning method according to claim 1, characterized in that Using the constructed dataset to train a support vector machine to establish a hot-rolled strip detection model, including: Design and extract the HOG feature vectors of the training set and the test set; Use the extracted HOG feature vector set of the training set to train a support vector machine to obtain a hot-rolled strip detection model, and use the extracted HOG feature vector set of the test set to verify the model accuracy.

4. The hot-rolled strip detection and deviation warning method according to claim 3, wherein, Extract the HOG feature vectors of the training set and the test set, including: Randomly select a steel-containing image from the training set, and design and extract HOG feature parameters according to the image size and strip size, including: sliding window size, pixel size of each cell, and number of cells in each block; Convert the image into a single-channel grayscale image, perform color space normalization on the single-channel grayscale image using the Gamma correction method, and set the Gamma coefficient; Calculate the directional gradient using [-1, 0, 1] and [-1, 0, 1] T operators to perform convolution calculations with the normalized image to obtain the horizontal directional gradient of the image at the point ([[]] x , y ) and the vertical directional gradient , and calculate the gradient magnitude x , y ) and the gradient direction at the point ([[]] ; Construct a gradient histogram and generate HOG feature vectors.

5. The hot-rolled strip steel detection and deviation warning method according to claim 1, wherein Perform deviation warning and deviation amount prediction based on the position coordinates of the strip edge straight line screened according to the position between the strip and the roller table, including: S21: Obtain the pixel width of the image W , the actual width of the strip W mm_strip , the actual width of the roller table W mm_roll , calculate the pixel width of the strip in the image , obtain the set of straight lines at the edges of the strip and the roller table; S22: Take a straight line and calculate the average value of the axis coordinates of its two endpoints x and determine whether it is greater than W / 2. If it is satisfied, save it as the set of right-side straight lines; otherwise, save it as the set of left-side straight lines. Then delete this straight line from the set of straight lines at the strip steel and roller path edges; S23: Determine whether the set of straight lines of the strip and the roller table edge is an empty set. If it is an empty set, go to step S24; otherwise, go to step S22; S24: Take the last straight line in the left straight line set and the first straight line in the right straight line set to obtain the strip edge straight line set; S25: Set the warning position coordinates for the left and right deviation of the strip steel , calculate the proportional relationship between the actual size and the image pixel size ; S26: Calculate the average value of the axial coordinates of the two end points of the left and right edge straight lines in the strip edge straight line set x and denote it as and ; S27: If and are satisfied, output a warning for the left deviation of the strip steel. If satisfied and , the output strip runs normally without deviation; If and are satisfied, output a warning for the right deviation of the strip steel. S28: Calculate the predicted value of strip deviation .

Images