CN110223267A - The recognition methods of refractory brick deep defects based on height histogram divion - Google Patents

Abstract

The invention discloses a method for identifying depth defects of refractory bricks based on height histogram segmentation. Multiply the refractory brick image point cloud for plane fitting to obtain the zero plane, and obtain the height and width of the original refractory brick image, and generate the corresponding reference plane image according to the size parameters and zero plane of the original refractory brick image, for the original refractory brick image Make a difference with the reference plane image to obtain the tilt-corrected point cloud data map, filter and segment the height histogram of the tilt-corrected point cloud, and obtain the point cloud information of the set depth. The invention can analyze the image height map based on the gray moment plane fitting, and obtain the depth defect information of the refractory bricks, effectively avoid the penetration corrosion phenomenon, and prolong the service life of the refractory bricks.

Description

Recognition Method of Refractory Brick Depth Defect Based on Height Histogram Segmentation

technical field

The invention relates to a method for identifying depth defects of refractory bricks based on height histogram segmentation.

Background technique

Refractory bricks, according to international standards, refer to non-metallic materials and products that have stable chemical and physical properties and can be used normally in high temperature environments (not excluding a certain proportion of metals). Refractory bricks can withstand various mechanical actions and physical and chemical changes at high temperatures, and are widely used in metallurgy, chemical industry, petroleum, power generation and other industrial fields. In recent years, my country's refractory brick production enterprises have developed rapidly, and the market competition for refractory bricks has become increasingly fierce. Many enterprises have improved production technology and quality inspection technology, and realized production and testing automation to improve their competitiveness. On the production line of refractory bricks, before the products are off-line and packed, manual measurement with a tape measure has been used for a long time to judge the depth defects with the naked eye, such as missing corners, missing edges and pitted surfaces of refractory bricks. In the production process of refractory bricks, the vibration and noise of the press are very harmful to the physical and mental health of workers, and many defects are judged by workers based on experience, which is greatly influenced by subjective, so it is impossible to establish a unified evaluation standard. In addition, in the process of mass production, this process not only consumes a lot of labor costs, but also the repetitive and monotonous measurement and observation work can easily cause personnel fatigue and misjudgment. Serious economic losses, and even seriously affect the production of steel. Therefore, in some dangerous occasions that are not suitable for manual work, occasions where artificial vision is difficult to meet the requirements, and occasions that are highly repeatable and intelligent and cannot be continuously and stably inspected by human eyes. In the production process of refractory bricks, in addition to the geometric dimensions and depth defects, these defects have a non-negligible impact on the quality of refractory bricks and the use of the steel smelting process. Therefore, for the defects of refractory bricks, effective identification It is a vital part of the quality inspection of refractory bricks. At present, there is no measurement system or method for automatic measurement of surface defects of refractory bricks.

Contents of the invention

The purpose of the present invention is to provide an identification method capable of identifying deep defects in refractory bricks.

The technical solution adopted by the present invention to solve its technical problems is:

A method for identifying depth defects in refractory bricks based on height histogram segmentation, comprising the following steps:

Step 1. Use the structured light sensor to obtain the color point cloud data of refractory bricks. The color point cloud data is fused together from image data and 3D point cloud. The coordinate system of the color point cloud data is based on the sensor pose;

Step 2. Use the least square method to perform plane fitting on the point cloud of the refractory brick image to obtain the zero plane, and obtain the height and width of the original refractory brick image, and generate the corresponding reference plane image according to the size parameters and zero plane of the original refractory brick image ;

Step 3. Make a difference between the original refractory brick image and the reference plane image to obtain the point cloud data map after tilt correction;

Step 4. Filter and segment the height histogram of the tilt-corrected point cloud to obtain the point cloud information of the set depth, set the height value range of the height band-pass filter to filter the point cloud height histogram, within the height value range The connected domains within are regarded as deep defects.

Further, in step 2, the generation method of the reference plane image is: according to the fitting parameters α, β, γ, combined with the refractory brick color point cloud data Image(r, c), generate the reference plane image Image(r, c) ₀ : For a 2D continuous image f(x,y)(≥0), the p+q order moment m _pq is defined as:

Among them, p and q are non-negative integers. For discretized digital images, the above formula is:

where (r ₀ ,c ₀ ) is the coordinates of the center of mass, and

The first-order plane approximation method is described by the following formula:

Image(r,c)=α(rr ₀ )+β(cc ₀ )+γ (4-13)

Among them, r ₀ is the abscissa of the area to be fitted, c ₀ is the ordinate of the area to be fitted, γ is the average gray level of the area to be fitted, F is the area of the entire plane, and MRow is the gray area along the row direction degree moment, MCol is the gray moment along the column direction, then:

MRow=sum((rr ₀ )*(Image(r,c)-γ))/F ² (4-14)

MCol=sum((cr ₀ )*(Image(r,c)-γ))/F ² (4-15)

Further, the difference between the original refractory brick point cloud image Image(r,c) and the reference plane image Image(r,c) ₀ is obtained to obtain the tilt-corrected point cloud data image Image'(r,c).

Image'(r,c)=Image(r,c)-Image(r,c) ₀ (4-18);

Segment the tilt-corrected point cloud data map along the normal direction of the zero plane to obtain a point cloud height histogram; set the height value range of the height band-pass filter to filter the point cloud height histogram, within the range of height values Connected domains are considered as deep defects.

Further, in step 4, the connected domain is obtained according to the two-pass scanning method, including the following steps:

Step 4-1: Scan the refractory brick threshold image for the first pass, assign a label to each pixel position, assign one or more different labels to the set of pixels in the same connected domain, and merge pixels belonging to the same connected domain but with different values Label;

Step 4-2: Scan the refractory brick threshold image for the second pass, classify the pixels marked by the same label with an equal relationship into a connected domain, and assign the same label to the connected domain.

Further, step 4-1 uses the seed filling method to obtain connected domains:

(1) Scan the pixels in the refractory brick threshold map until the current pixel B(x, y)==1:

a. Use B(x, y) as a seed and give it a label, and then push all the foreground pixels adjacent to the seed into the stack;

b. Pop the top pixel of the stack, give the top pixel the same label as the seed, and then push all the foreground pixels adjacent to the top pixel into the stack;

c. Repeat step b until the stack is empty;

At this point, all pixel values with the same label form a connected domain;

Obtain any pixel outside the connected domain as a seed, repeat step (1) until the end of scanning; after scanning, all connected domains in image B can be obtained.

The advantages of the present invention are:

1. Based on gray moment plane fitting, analyze the image height map to obtain the depth defect information of refractory bricks, effectively avoid penetration corrosion and prolong the service life of refractory bricks.

2. Using the two-pass scanning method, there is no need to apply for a large amount of stack space, the speed of obtaining connected regions is fast, and multiple connected regions can be obtained, no memory leaks occur, and relatively good execution efficiency.

Description of drawings

Figure 1 is the result of tilt correction of refractory brick scanning surface.

Figure 2 is a comparison of height histograms before and after segmentation.

Figure 3 is the result of deep defect extraction.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

During the production process of refractory bricks, in addition to the surface scratches caused by contact, some deep defects will also be caused by the loss of the mold: pits, pockmarks, missing corners and edges. The image data collected by the Gocator2350 sensor is color point cloud data that is fused together with image data and 3D point cloud. In this regard, depth defects can be analyzed from the height information in the point cloud data.

The data collected by the sensor is color point cloud data, and its coordinate system is based on the sensor pose. Therefore, it is first necessary to fit the measured upper surface of the refractory brick, and perform height segmentation along its normal direction to obtain depth defect information.

A method for identifying depth defects in refractory bricks based on height histogram segmentation, comprising the following steps:

Step 1. Use the structured light sensor to obtain the color point cloud data of refractory bricks. The color point cloud data is fused together from image data and 3D point cloud. The coordinate system of the color point cloud data is based on the sensor pose;

Step 2. Use the least square method to perform plane fitting on the point cloud of the refractory brick image to obtain the zero plane, and obtain the height and width of the original refractory brick image, and generate the corresponding reference plane image according to the size parameters and zero plane of the original refractory brick image ;

Step 3. Make a difference between the original refractory brick image and the reference plane image to obtain the point cloud data map after tilt correction;

Step 4. Filter and segment the height histogram of the tilt-corrected point cloud to obtain the point cloud information of the set depth, set the height value range of the height band-pass filter to filter the point cloud height histogram, within the height value range The connected domains within are regarded as deep defects.

Plane Fitting Based on Calculation of Gray Value Moment and First Order Plane Approximation Method

Moments are operators that describe image features. Nowadays, image moment technology has been widely used in image retrieval and recognition, image matching, image reconstruction, digital compression, digital watermarking and moving image sequence analysis and other fields.

The image can be regarded as a flat object, and the value of each pixel is regarded as the density there. The expectation of a point is the moment of the image at that point. For a 2D continuous image f(x,y)(≥0), the p+q order moment m _pq is defined as:

Among them, p and q are non-negative integers. For discretized digital images, the above formula is:

where (r ₀ ,c ₀ ) is the coordinates of the center of mass, and

The first-order plane approximation method is realized by minimizing the distance between the gray value and the plane ^[61] , which can be described by the following formula:

Image(r,c)=α(rr ₀ )+β(cc ₀ )+γ (4-13)

Among them, r ₀ and c ₀ are the horizontal and vertical coordinates of the area to be fitted, and γ is the average gray level of the area to be fitted.

Let F be the area of the entire plane, and MRow and MCol be the moments along the row and column directions respectively, then:

MRow=sum((rr ₀ )*(Image(r,c)-γ))/F ² (4-14)

MCol=sum((cr ₀ )*(Image(r,c)-γ))/F ² (4-15)

Height histogram threshold segmentation based on tilt correction to extract depth defects

According to the fitting parameters α, β, γ obtained by the above formula, combined with the size of the original image, an image can be generated as a reference. At this point, the point cloud data map after tilt correction can be obtained by making a difference between the original image and the reference image. The tilt correction situation is shown in Figure 1 below.

At this time, by segmenting the tilt-corrected height map, the effect of segmenting along the normal direction of the fitting plane can be obtained. The histogram shown in Figure 2 can be obtained by probabilistic analysis of the tilt-corrected height data.

Set the height band-pass filter, the height value range is -2.5-0mm, and filter and segment, you can get all the depth defects in this range, as shown in Figure 3 below, that is, pits and pockmarks (left) and missing corners and edges (right).

All the segmented depth defects that meet the height range are marked as 1, and the rest are marked as 0. According to the connected domain algorithm, the connected domain is extracted, and the area with a depth defect area above ^40mm2 is extracted as follows:

Dimpled surface: [98, 122, 178, 59, 281, 117, 107, 120, 116, 55, 290, 50, 103, 139]

Missing corners and edges: [92, 120, 97, 165, 63, 160]

Multiple measurement experiments were carried out on different types of refractory bricks. In repeated experiments, the defect test data of refractory bricks are shown in the table below. The defect values in the table are in accordance with the agreed indicators, and the upper limit of pits and pockmarks is 1.5mm ,, where OK means qualified and NG means unqualified. It can be seen from the measurement results that the method of quantifying defects in the connected domain has been effectively verified, that is, the measurement system can meet the requirements of defect identification.

Table defect measurement experiment results

Data analysis shows that the defect recognition of the system meets the accuracy requirements, thus proving that the stability and repeatability accuracy of this set of visual measurement methods meet the requirements.

The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments. Equivalent technical means that a person can think of based on the concept of the present invention.

Claims (5)

1. The recognition method of the depth defect of refractory brick based on height histogram segmentation, comprises the following steps: Step 1. Use the structured light sensor to obtain the color point cloud data of refractory bricks. The color point cloud data is fused together from image data and 3D point cloud. The coordinate system of the color point cloud data is based on the sensor pose; Step 2. Use the least square method to perform plane fitting on the point cloud of the refractory brick image to obtain the zero plane, and obtain the height and width of the original refractory brick image, and generate the corresponding reference plane image according to the size parameters and zero plane of the original refractory brick image ; Step 3. Make a difference between the original refractory brick image and the reference plane image to obtain the point cloud data map after tilt correction; Step 4. Filter and segment the height histogram of the tilt-corrected point cloud to obtain the point cloud information of the set depth, set the height value range of the height band-pass filter to filter the point cloud height histogram, within the height value range The connected domains within are regarded as deep defects. 2. The recognition method of the refractory brick depth defect based on the height histogram segmentation as claimed in claim 1, characterized in that, in step 2, the generation method of the reference plane image is: according to the fitting parameters α, β, γ, combined with Refractory brick color point cloud data Image(r,c), generate reference plane image Image(r,c) ₀ : For a 2D continuous image f(x,y)(≥0), p+q order moment m _pq is defined for: Among them, p and q are non-negative integers. For discretized digital images, the above formula is: where (r ₀ ,c ₀ ) is the coordinates of the center of mass, and The first-order plane approximation method is described by the following formula: Image(r,c)=α(rr ₀ )+β(cc ₀ )+γ (4-13) Among them, r ₀ is the abscissa of the area to be fitted, c ₀ is the ordinate of the area to be fitted, γ is the average gray level of the area to be fitted, F is the area of the entire plane, and MRow is the gray area along the row direction degree moment, MCol is the gray moment along the column direction, then: MRow=sum((rr ₀ )*(Image(r,c)-γ))/F ² (4-14) MCol=sum((cr ₀ )*(Image(r,c)-γ))/F ² (4-15) 3. the recognition method of the refractory brick depth defect based on height histogram segmentation as claimed in claim 2, is characterized in that, to original refractory brick point cloud image Image (r, c) and reference plane image Image (r, c) ₀ Make a difference to obtain the tilt-corrected point cloud data image Image'(r,c). Image'(r,c)=Image(r,c)-Image(r,c) ₀ (4-18); Segment the tilt-corrected point cloud data map along the normal direction of the zero plane to obtain a point cloud height histogram; set the height value range of the height band-pass filter to filter the point cloud height histogram, within the range of height values Connected domains are considered as deep defects. 4. the identification method of the refractory brick depth defect based on height histogram segmentation as claimed in claim 3, is characterized in that, in step 4, obtains connected domain according to two-pass scanning method, comprises the following steps: Step 4-1: Scan the refractory brick threshold image for the first pass, assign a label to each pixel position, assign one or more different labels to the set of pixels in the same connected domain, and merge pixels belonging to the same connected domain but with different values Label; Step 4-2: Scan the refractory brick threshold image for the second pass, classify the pixels marked by the same label with an equal relationship into a connected domain, and assign the same label to the connected domain. 5. the identification method of the refractory brick depth defect based on height histogram segmentation as claimed in claim 4, is characterized in that, step 4-1 uses seed filling method to carry out connected domain acquisition: (1) Scan the pixels in the refractory brick threshold map until the current pixel B(x, y)==1: a. Use B(x, y) as a seed and give it a label, and then push all the foreground pixels adjacent to the seed into the stack; b. Pop the top pixel of the stack, give the top pixel the same label as the seed, and then push all the foreground pixels adjacent to the top pixel into the stack; c. Repeat step b until the stack is empty; At this point, all pixel values with the same label form a connected domain; Obtain any pixel outside the connected domain as a seed, repeat step (1) until the end of scanning; after scanning, all connected domains in image B can be obtained.