A method for improving the accuracy of steel sheet defect detection and the fault tolerance of a camera

Abstract

The application discloses a method for improving the accuracy of steel plate defect detection and the fault tolerance rate of a camera, comprising arranging a group of cameras above and below the steel plate, the capture ranges of the cameras overlapping; firstly, carrying out hot state defect detection, and then carrying out cold state defect detection; collecting two-dimensional image features and color information of the defects through a color camera, recording the positions of the defect targets on the plate, recording the similarity of the defects, judging whether periodicity exists, combining the upstream hot state inspection mother plate number with a downstream table inspection system, and increasing information interaction and reminding functions; since the defects and pseudo defects have different color and shape features, the defect categories can be judged through the collected defect shapes, periodicity, color and position. The application has the advantages of improving the accuracy of defect discrimination with the maximum efficiency, improving the fault tolerance rate of the camera, reducing the labor intensity of workers, reducing safety hazards, and ensuring that the data is accurate and correct.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of steel plate surface defect detection, and particularly relates to a method for improving the accuracy of steel plate defect detection and the fault tolerance rate of a camera. BACKGROUND

[0002] Domestic steel enterprises generally use manual methods for surface defect detection of medium-thick steel plates. Such detection methods often require a large amount of human resources and cannot detect both sides simultaneously. In order to improve the working and safety environment of workers, reduce the labor intensity of workers, improve the stability of the steel plate surface detection system, ensure accuracy, and promote the intelligent level of steel plate production and manufacturing, it is necessary to design a steel plate surface detection system with better functions and higher efficiency.

[0003] To solve the above problems, a steel plate surface detection system is developed for double-sided artificial intelligent detection. However, in actual production, it is found that defects or non-defects are similar in two-dimensional black and white images, and the outlines of non-defects such as water marks and oxide scales are similar to those of inclusions. Oxide scales, heavy scales, and bubble-like cracks all have jagged edges. Similar forms can easily cause confusion during classification, resulting in misidentification. Non-defects that are misidentified are called "pseudo-defects", and high misidentification rates have been a problem in the use of two-dimensional image-based surface detection methods. SUMMARY

[0004] The purpose of the present application is to solve the problem of inaccurate steel plate defect detection data and high misidentification rate. A method for improving the accuracy of steel plate defect detection and the fault tolerance rate of a camera is provided, which can maximize the accuracy of defect identification, improve the fault tolerance rate of the camera, reduce the labor intensity of workers, reduce safety hazards, and ensure accurate data.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] A method for improving the accuracy of steel plate defect detection and the fault tolerance rate of a camera, the specific steps are as follows:

[0007] (1) Equipment arrangement: a set of cameras is arranged above and below the steel plate, the lenses of the cameras are arranged facing the upper and lower surfaces of the steel plate, and every three cameras form a group. The capture ranges of the left and right cameras are L1 and L2, respectively, and the steel plate regions corresponding to L1 and L2 are connected. The capture range of the middle camera is L1+L2, and the corresponding steel plate region is exactly the sum of the steel plate regions captured by the left and right cameras;

[0008] (2) Defect hot state detection: the detection equipment acquires steel plate defect information and informs the downstream surface detection system of the information;

[0009] (3) Defect cold-state detection:

[0010] a. Collecting two-dimensional image features and color information of the defect by a color camera;

[0011] b. Recording the position of the defect target on the plate;

[0012] c. Recording the similarity of the defect to determine whether there is periodicity;

[0013] d. Combining the upstream hot-state inspection mother plate number with the downstream table inspection system to increase information interaction and reminding functions, thereby improving the classification accuracy of the defect;

[0014] (4) Defect judgment module: since the defect and the pseudo-defect have different color and shape features, the defect category can be judged by collecting the defect morphology, periodicity, color and position.

[0015] Further, in the step (2), the defect hot-state detection uses a black-and-white camera, which is pre-inspected and input into the system.

[0016] Further, in the step (3), the defect cold-state detection uses a color camera, and a white light source is used to optimize the light path.

[0017] Further, in the step (3), for periodic defects such as indentations, the length information recorded by the counter is combined to analyze the shape similarity of indentations at different positions, that is, the cosin similarity is used, the defect sub-graph obtained by target detection is expressed as a vector, and the similarity of different defect sub-graphs is represented by calculating the cosine distance between the vectors.

[0018] Further, in the step (3), for defects with position distribution characteristics such as cracks, different detection thresholds are set on different camera corresponding algorithms, that is, the longitudinal crack algorithm detection threshold of the middle camera is higher than that of the two side cameras, and the bubble-shaped crack algorithm detection threshold of the two side cameras is higher than that of the middle camera.

[0019] Further, in the step (4), the defect detection uses a voting mechanism, and only when the defects detected by two adjacent cameras are defects, it is considered to be a defect, which can greatly reduce the false detection.

[0020] Further, in the step (4), the defects mainly include indentations, cracks, scratches, pressed iron scales and inclusions, the two-dimensional image features of the indentations are irregular blocks, and the color is gray, the two-dimensional image features of the cracks are crack-like and zigzag strip-like, and the color is black, the two-dimensional image features of the scratches are straight line-like and parallel line-like, and the color is metal bright color, the two-dimensional image features of the pressed iron scales are block-like and edge zigzag, and the color is black, and the two-dimensional image features of the inclusions are irregular blocks, and the color is white or yellow and red point-like.

[0021] In the technical solution of this invention, the automation and intelligent functions of the steel plate surface defect detection system are realized through machine vision technology. The accuracy of the steel plate surface defect detection system is improved by increasing defect color information, increasing the overlap area of ​​camera shooting, and determining the periodicity of defect location. This reduces the labor intensity of workers and the existing safety hazards, adapts to complex working conditions, and in particular improves the fault tolerance rate of the camera. Attached Figure Description

[0022] Figure 1 The camera arrangement optical path diagram of the present invention;

[0023] Figure 2 This is a flowchart of the method for improving the accuracy of steel plate defect detection and camera fault tolerance according to the present invention. Implementation Example

[0024] To make the present invention clearer, the following description, in conjunction with the accompanying drawings, further illustrates a method for improving the accuracy of steel plate defect detection and camera fault tolerance. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0025] A method for improving the accuracy of steel plate defect detection and camera fault tolerance includes the following steps:

[0026] (1) Equipment layout: such as Figure 1 As shown, a set of cameras is set above and below the steel plate 1. The camera lenses are set facing the upper and lower surfaces of the steel plate. Cameras A1, B1 and C1 form one group, and cameras A2, B2 and C2 form another group.

[0027] For the detection of the upper surface of the steel plate, the capture ranges of the two cameras A1 and C1 on the left and right are L1 and L2 respectively, and the steel plate areas corresponding to L1 and L2 are connected. The capture range of the middle camera B1 is L1+L2, and the steel plate area corresponding to it is exactly the sum of the steel plate areas captured by the two cameras on the left and right. The working principle of the detection camera on the lower surface of the steel plate is the same as that of the detection camera on the upper surface of the steel plate.

[0028] (2) Defect hot detection: The detection equipment acquires the defect information of the steel plate and informs the downstream surface inspection system. The defect hot detection uses a black and white camera, which is pre-inspected and entered into the system;

[0029] (3) Cold inspection of defects:

[0030] a. A color camera is used, with a white light source to optimize the optical path, to collect two-dimensional image features and color information of defects;

[0031] b. Record the location of the defect target on the board;

[0032] c. Record the similarity of the defect, determine whether there is periodicity;

[0033] d. Combine the upstream hot-state inspection motherboard number with the downstream table inspection system, increase the information interaction and prompting function, thereby improving the classification accuracy of the defect;

[0034] (4) Defect judgment module: since the defect and the pseudo-defect have different color and shape characteristics, the defect category can be judged by collecting the defect shape, periodicity, color and position.

[0035] In this embodiment, for the periodic defect of indentation, the length information recorded by the counter is combined to analyze the shape similarity of the indentation at different positions, that is, the cosine similarity is used, the defect subgraph obtained by target detection is expressed as a vector, and the cosine distance between the vectors is calculated to represent the similarity of different defect subgraphs.

[0036] In this embodiment, for the defect with position distribution characteristics such as crack, different detection thresholds are set on different camera corresponding algorithms, such as the longitudinal crack algorithm detection threshold of the middle camera is higher than that of the two side cameras, and the bubble-shaped crack algorithm detection threshold of the two side cameras is higher than that of the middle camera.

[0037] In this embodiment, the defect detection adopts a voting mechanism, only when the defects detected by two adjacent cameras are defects, which can greatly reduce the false detection.

[0038] The working principle of the present application is that: (1) the color information of the steel plate surface defect exists, as shown in the following table; (2) the defect detection accuracy of the camera observation overlapping area is higher than that of the camera observation non-overlapping area, so the camera overlapping shooting mode is adopted; (3) the detection information of hot-state table inspection and cold-state table inspection is comprehensively used for information interaction, to further ensure the defect detection accuracy.

[0039]

[0040] As can be seen from the above table, the defects mainly include indentation, crack, scratch, pressed iron oxide scale and inclusion, the two-dimensional image feature of indentation is irregular block, the color is gray, the two-dimensional image feature of crack is crack and zigzag strip, the color is black, the two-dimensional image feature of scratch is straight line and parallel line, the color is metal bright color, the two-dimensional image feature of pressed iron oxide scale is block and edge zigzag, the color is black, and the two-dimensional image feature of inclusion is irregular block, the color is white or yellow and red point. It can be seen that the two-dimensional image features and colors corresponding to the defect and the pseudo-defect are different, and the type of the defect can be accurately determined by comparing the defect shape and color of the two.

[0041] The present application can greatly improve the accuracy of defect detection by finding defects through upstream hot state inspection, interacting defect data with downstream, and facilitating workers' daily operation. Since the same defect is distributed on two adjacent pictures due to being at the top or bottom edge of the picture when the same linear array camera constructs the image, the defect data set can be expanded by using the method of cutting the head and tail of the complete defect picture. The present application can realize the automation and intelligent function of the steel plate surface detection system through machine vision technology on the basis of double-sided artificial intelligence detection, ensure the accuracy of defect identification, reduce the labor intensity and safety hazards of workers, optimize the camera shooting scheme, maximize the efficiency of improving the accuracy of defect identification, and improve the fault tolerance of the camera.

[0042] In addition to the above embodiments, the present application can have other implementation manners. Any technical solution formed by equivalent replacement or equivalent transformation falls within the protection scope required by the present application.

Claims

1. A method for improving the accuracy of steel plate defect detection and camera fault tolerance, comprising the following steps, characterized in that: (1) Equipment layout: A set of cameras is set above and below the steel plate. The camera lens is set facing the upper and lower surfaces of the steel plate. Every three cameras are set together. The capture ranges of the left and right cameras are L1 and L2 respectively. The steel plate areas corresponding to L1 and L2 are connected. The capture range of the middle camera is L1+L2. The steel plate area corresponding to it is exactly the sum of the steel plate areas captured by the left and right cameras. (2) Defect hot detection: Defect hot detection uses a black and white camera, which is pre-inspected and entered into the system. The detection equipment acquires the defect information of the steel plate and informs the downstream surface inspection system. (3) Cold inspection of defects: Cold inspection of defects uses a color camera and a white light source to optimize the optical path, specifically: a. Collect two-dimensional image features and color information of defects using a color camera; b. Record the location of the defective target on the board; c. Record the similarity of defects to determine if there is a periodicity; d. Integrate the upstream hot-state inspection motherboard number with the downstream surface inspection system to enhance information interaction and reminder functions; For crack defects with location distribution characteristics, different detection thresholds are set on the corresponding algorithms of different cameras to solve the problem. That is, the detection threshold of the longitudinal crack algorithm of the middle camera is higher than that of the two side cameras, and the detection threshold of the bubble crack algorithm of the two side cameras is higher than that of the middle camera. (4) Defect judgment module: Since defects and pseudo-defects have different color and shape characteristics, the defect category can be determined by collecting the defect morphology, periodicity, color and location.

2. The method for improving the accuracy of steel plate defect detection and camera fault tolerance rate according to claim 1, characterized in that: In step (3), for periodic indentation defects, the similarity of indentation shapes at different locations is analyzed by combining the length information recorded by the counter. That is, the defect sub-maps obtained by target detection are represented as a vector by using cosine similarity, and the similarity of different defect sub-maps is characterized by calculating the cosine distance between the vectors.

3. The method for improving the accuracy of steel plate defect detection and camera fault tolerance according to claim 1 or 2, characterized in that: In step (4), defect detection adopts a voting mechanism. Only when two adjacent cameras detect a defect is it considered a defect, thereby reducing false detections.

4. The method for improving the accuracy of steel plate defect detection and camera fault tolerance according to claim 1 or 2, characterized in that: In step (4), the defects include indentations, cracks, scratches, indented iron oxide scale, and inclusions. The two-dimensional image features of indentations are irregular blocks and gray in color. The two-dimensional image features of cracks are chapped and tortuous stripes and black in color. The two-dimensional image features of scratches are straight lines and parallel lines and bright metallic in color. The two-dimensional image features of indented iron oxide scale are blocks and tortuous edges and black in color. The two-dimensional image features of inclusions are irregular blocks and white, yellow, or red dots.

Images