CN112986277B - Detection method for hot-rolled strip steel finish rolling roll mark - Google Patents
Detection method for hot-rolled strip steel finish rolling roll mark Download PDFInfo
- Publication number
- CN112986277B CN112986277B CN202110365067.5A CN202110365067A CN112986277B CN 112986277 B CN112986277 B CN 112986277B CN 202110365067 A CN202110365067 A CN 202110365067A CN 112986277 B CN112986277 B CN 112986277B
- Authority
- CN
- China
- Prior art keywords
- image
- sub
- blocks
- segmentation
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention provides a method for detecting a hot-rolled strip steel finish rolling roller mark, and belongs to the technical field of intelligent detection. Splicing images acquired by a camera to obtain a spliced image covering about ten meters of strip steel, acquiring perimeter information of a roller in real time through two stages, and segmenting the spliced image by taking the perimeter of the roller as a height reference to obtain a plurality of segmented images with the same size; and then dividing each segmentation graph into a plurality of sub-blocks, carrying out preliminary filtering through image classification, carrying out feature matching on every two sub-blocks at the same position in each segmentation graph, and determining whether the finish rolling roll mark exists or not by counting the success rate of the feature matching of the sub-blocks at different positions. The method fully considers the periodicity of the finish rolling seal and the form consistency in single occurrence, can effectively avoid the missed detection of the finish rolling seal, and improves the identification rate of the defects.
Description
Technical Field
The invention relates to the technical field of intelligent detection, in particular to a method for detecting a hot-rolled strip steel finish rolling roll mark.
Background
Some defects are generated on the surface in the production process of the strip steel, and the quality of a final finished product is influenced. The most common defect is a finish rolling roller mark, and the finish rolling roller mark defect is mainly caused by roller damage or foreign objects pressed into the roller, if the detection cannot be carried out in time in the production process, mass quality accidents are easy to generate, and serious economic loss is caused. Therefore, the detection of the defects plays an important role in the detection of the defects of the strip steel, and the accurate detection can provide guidance for production, timely carry out the roll changing process and avoid more problem rolls.
The existing strip steel surface detection technology simultaneously detects the finish rolling imprints and other types of defects, and adopts a processing mode of firstly extracting interested areas and then classifying and identifying. The roll mark defects have larger form difference every time and are different from defects with standard forms such as scabs, scratches and the like, so that the intra-class feature spacing in the roll mark categories is larger, the model detection effect is not good in the classification and identification processes, and the missing rate of the roll mark is higher especially for some finish rolling roll marks with smaller sizes.
The method is used for independently processing the detection of the finish rolling mark defect, considers the periodicity of the finish rolling mark and the consistency of the single appearance form, and can detect the finish rolling mark defect through a strip steel image of dozens of meters by utilizing the two key characteristics, and meanwhile, the detection effect of the finish rolling mark with smaller size is superior to that of the traditional detection mode.
Disclosure of Invention
The invention aims to provide a detection method of a hot-rolled strip steel finish rolling mark.
Splicing high-definition strip steel surface images acquired by an industrial linear array camera, transversely segmenting the spliced images by taking the circumference of a roller as a reference to obtain n segmentation images with the height of the image being the circumference of the roller; then dividing each segmentation image into a plurality of sub-blocks in the same way, classifying each sub-block image, and removing the sub-blocks with the types as backgrounds and the sub-blocks with the types inconsistent at the same position in each segmentation image; and finally, combining and matching the subblocks reserved at the same position in every two segmentation images, if the images of the two subblocks are successfully matched, adding 1 to the confidence probability of the position with the roll mark defect, and if the confidence probability reaches a specified threshold value after all the images are matched, determining that the corresponding position has the finish rolling roll mark defect.
The method specifically comprises the following steps:
(1) splicing high-definition strip steel surface images acquired by an industrial linear array camera, transversely segmenting the spliced images by taking the circumference of a roller as a reference to obtain n segmentation images with the image height being the circumference of the roller;
(2) dividing each segmentation image into a plurality of sub-blocks in the same way, classifying each sub-block image, and removing the sub-blocks with the types as backgrounds and the sub-blocks with the types inconsistent at the same position in each segmentation image;
(3) and combining and matching the subblocks reserved at the same position in every two segmentation images, if the images of the two subblocks are successfully matched, adding 1 to the confidence probability of the position with the roll mark defect, and if the confidence probability reaches a specified confidence threshold after all the images are matched, determining that the corresponding position has the finish rolling roll mark defect.
And (2) respectively arranging two industrial linear array cameras on the upper surface and the lower surface of the strip steel in the step (1), and setting the transverse resolution and the longitudinal resolution of the surface of the shot steel plate to be 0.2mm in order to ensure that the concerned minimum roll mark defect can be detected.
Actual length represented by each image in step (1):
l＝h*fy
wherein l is the actual length represented by each image, h is the pixel height of the image, fyIs the image longitudinal resolution;
and sequentially splicing a plurality of continuous images, and stopping splicing after the length L of the accumulated spliced image is more than 10m to obtain a spliced image.
In the step (1), the roller perimeter is obtained in two stages in real time, and segmentation and rounding are carried out on the spliced image in the height direction according to the roller perimeter to obtain the number of segmentation maps:
wherein L is the accumulated length of the spliced image, and C is the circumference of the roller.
And (3) dividing the segmentation graph in the step (2) into subblocks with the same size by adopting 4 number division modes (32 × 32, 16 × 16, 8 × 8 and 4 × 4) so as to adapt to roll mark defects with different sizes, and respectively carrying out subsequent classification, elimination and matching processes after each division.
In the step (2), the classification process of the sub-block images can adopt, but is not limited to, ResNet101 and Inception V3 network models, the output types of the models mainly comprise backgrounds, scabs, iron scales, heavy skins, scratches, speckles and foreign matter pressing, blocks with the types as the backgrounds are obtained through a classifier and are removed, the classification types of the sub-blocks at the same position in each segmentation image are inconsistent, and all the sub-blocks at the position are removed.
The method for matching the combinations of the sub-blocks in the step (3) includes, but is not limited to, corner matching and sift matching.
And (4) in the step (3), the confidence level threshold is specified according to the combination number of the same position and the actual field situation.
The technical scheme of the invention has the following beneficial effects:
in the scheme, the detection of the finish rolling mark defect is processed independently, the finish rolling mark defect can be detected through a strip steel image of dozens of meters, and the real-time performance is high; the defect detection is carried out in a mode of taking a period as a drive, so that the precision rate of the finish rolling mark detection is greatly improved, a detection algorithm is applied to an actual field, the number of times of detecting the finish rolling mark is averagely 3-5 times every day, and the frequency of manually unwinding the roll every 5 times can reduce about 20 rolls of the problem roll of the finish rolling mark, so that the qualification rate of strip steel products is greatly improved, and unnecessary loss and quality objections are reduced.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a flow chart of segmentation and filtering for sub-blocks of a stitched image according to the present invention;
fig. 3 is a diagram illustrating the effect of feature matching of the sub-block image according to the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
The invention provides a method for detecting a hot-rolled strip steel finish rolling mark.
The method comprises the steps of firstly splicing high-definition strip steel surface images acquired by an industrial linear array camera, transversely segmenting the spliced images by taking the circumference of a roller as a reference to obtain n segmentation images with the height of the image being the circumference of the roller; then dividing each segmentation image into a plurality of sub-blocks in the same way, classifying each sub-block image, and removing the sub-blocks with the types as backgrounds and the sub-blocks with the types inconsistent at the same position in each segmentation image; and finally, combining and matching the subblocks reserved at the same position in every two segmentation images, if the images of the two subblocks are successfully matched, adding 1 to the confidence probability of the position with the roll mark defect, and if the confidence probability reaches a specified threshold value after all the images are matched, determining that the corresponding position has the finish rolling roll mark defect.
As shown in fig. 1, the method specifically comprises the following steps:
(1) splicing high-definition strip steel surface images acquired by an industrial linear array camera, transversely segmenting the spliced images by taking the circumference of a roller as a reference to obtain n segmentation images with the image height being the circumference of the roller;
(2) dividing each segmentation image into a plurality of sub-blocks in the same way, classifying each sub-block image, and removing the sub-blocks with the types as backgrounds and the sub-blocks with the types inconsistent at the same position in each segmentation image;
(3) and combining and matching the subblocks reserved at the same position in every two segmentation images, if the images of the two subblocks are successfully matched, adding 1 to the confidence probability of the position with the roll mark defect, and if the confidence probability reaches a specified confidence threshold after all the images are matched, determining that the corresponding position has the finish rolling roll mark defect.
The following description is given with reference to specific examples.
Example 1
The application steps of the method are described in the following by the practical example applied in a steel mill in China.
Firstly, a linear array CCD camera is adopted to acquire images on the surface of the strip steel, the size of the images is 4096 x 1024 pixels, the longitudinal resolution of the images is preset before acquisition, and the default is 0.2 mm.
Wherein each image represents an actual length:
l＝h*fy
where h is the pixel height of the image, fyIs the image longitudinal resolution;
the actual strip steel length represented by each image is calculated by the formula to be 204 mm.
And sequentially splicing the continuously acquired images, stopping splicing when the cumulative length of the images exceeds 10m to obtain a complete spliced image, and splicing 50 images in the actual process to obtain the spliced image, wherein the total length of the spliced image is 10.2 m.
The roller perimeter is obtained in two stages in real time, and segmentation and rounding are carried out on the spliced image in the height direction according to the roller perimeter to obtain the number of segmentation images:
wherein L is the length of the spliced image, and C is the circumference of the roller.
Taking the 2200mm of the rolling mark period of the common finish rolling of a steel mill as an example, obtaining 4 segmentation maps in total by calculation, dividing each segmentation map into a plurality of sub-blocks, uniformly dividing the possible rolling mark defects with different sizes by the number of 32 × 32, 16 × 16, 8 × 8 and 4 × 4 respectively, and performing subsequent classification, elimination and matching processes after each division.
The sub-block classification process can adopt, but is not limited to, ResNet101 and Inception V3 network models, the output types of the models mainly comprise background, scab, iron scale, heavy skin, scratch, speckles and foreign matter pressing, sub-blocks with the types as the background are obtained through a classifier and are removed, the classification types of the sub-blocks at the same position in each cutting map are inconsistent, and all the sub-blocks at the position are removed, as shown in FIG. 2.
And combining the reserved subblocks at the same position on the multiple segmentation images in pairs, performing feature matching on the two subblock images in each pair of combinations, wherein the matching method can adopt but not limited to corner point matching and sift matching, and if the two subblock images are successfully matched, the confidence probability of the position with the roll mark defect is increased by 1, as shown in fig. 3.
And if the confidence probability reaches a specified threshold value after all matching is finished, determining that the corresponding position has the defect of finish rolling and roll mark, wherein 4 segmentation graphs are applied, and every two segmentation graphs are matched for 6 combinations, so that the threshold value is set to be 4 appropriately.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (8)
1. A detection method of a hot-rolled strip steel finish rolling roll mark is characterized by comprising the following steps: the method comprises the following steps:
(1) splicing high-definition strip steel surface images acquired by an industrial linear array camera, transversely segmenting the spliced images by taking the circumference of a roller as a reference to obtain n segmentation images with the image height being the circumference of the roller;
(2) dividing each segmentation image into a plurality of sub-blocks in the same way, classifying each sub-block image, and removing the sub-blocks with the types as backgrounds and the sub-blocks with the types inconsistent at the same position in each segmentation image;
(3) and combining and matching the subblocks reserved at the same position in every two segmentation images, if the images of the two subblocks are successfully matched, adding 1 to the confidence probability of the position with the roll mark defect, and if the confidence probability reaches a specified confidence threshold after all the images are matched, determining that the corresponding position has the finish rolling roll mark defect.
2. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: in the step (1), two industrial linear array cameras are respectively arranged on the upper surface and the lower surface of the strip steel, and in order to ensure that the concerned minimum roll mark defect can be detected, the transverse resolution and the longitudinal resolution of the surface of the shot steel plate are both set to be 0.2 mm.
3. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: the actual length represented by each image in the step (1):
l＝h*fy
wherein l is the actual length represented by each image, h is the pixel height of the image, fyIs the image longitudinal resolution;
and sequentially splicing a plurality of continuous images, and stopping splicing after the length L of the accumulated spliced image is more than 10m to obtain a spliced image.
4. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: in the step (1), the roller perimeter is obtained in two stages in real time, and segmentation and rounding are performed on the spliced image in the height direction according to the roller perimeter to obtain the number of segmentation maps:
wherein L is the accumulated length of the spliced image, and C is the circumference of the roller.
5. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: the segmentation chart in the step (2) adopts 4 number division modes: 32, 16, 8, 4 are divided into sub-blocks of the same size to accommodate different sizes of roll mark defects.
6. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: the sub-block image classification process in the step (2) comprises a ResNet101 and an inclusion V3 network model, the output types of the model comprise a background, a scab, an iron scale, a heavy skin, a scratch, a spot and a foreign matter press-in, sub-blocks with the types as the background are obtained through a classifier and are removed, the classification types of the sub-blocks at the same position in each segmentation image are inconsistent, and all the sub-blocks at the position are removed.
7. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: the method for matching the combinations of the sub-blocks in the step (3) comprises corner point matching and sift matching.
8. The method of detecting a hot rolled strip finishing rolling mark as claimed in claim 1, wherein: and (4) in the step (3), the confidence level threshold is specified according to the combination number of the same position and the actual field situation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110365067.5A CN112986277B (en) | 2021-04-01 | 2021-04-01 | Detection method for hot-rolled strip steel finish rolling roll mark |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110365067.5A CN112986277B (en) | 2021-04-01 | 2021-04-01 | Detection method for hot-rolled strip steel finish rolling roll mark |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112986277A CN112986277A (en) | 2021-06-18 |
| CN112986277B true CN112986277B (en) | 2022-04-19 |
Family
ID=76339272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110365067.5A Active CN112986277B (en) | 2021-04-01 | 2021-04-01 | Detection method for hot-rolled strip steel finish rolling roll mark |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112986277B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114522985B (en) * | 2022-04-13 | 2022-10-14 | 东莞海裕百特智能装备有限公司 | Roller rolling control method, system and computer readable storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104914111B (en) * | 2015-05-18 | 2018-08-03 | 北京华检智研软件技术有限责任公司 | A kind of steel strip surface defect online intelligent recognition detecting system and its detection method |
| CN105447817B (en) * | 2015-11-09 | 2018-11-09 | 北京科技大学 | A kind of hot-rolled intermediate billet flat image joining method based on edge contour |
| CN107576664A (en) * | 2017-09-28 | 2018-01-12 | 清华大学 | A kind of roll dressing surface defect Machine Vision Inspecting System |
| CN212228765U (en) * | 2020-01-14 | 2020-12-25 | 杭州谱诚泰迪实业有限公司 | Cold-state heavy rail surface defect detection system for deep learning |
-
2021
- 2021-04-01 CN CN202110365067.5A patent/CN112986277B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112986277A (en) | 2021-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110175982B (en) | Defect detection method based on target detection | |
| CN110009618B (en) | Shaft part surface quality detection method and device | |
| CN112712512A (en) | Hot-rolled strip steel scab defect detection method and system based on artificial intelligence | |
| CN115345885A (en) | Method for detecting appearance quality of metal fitness equipment | |
| CN113628189B (en) | Rapid strip steel scratch defect detection method based on image recognition | |
| CN109215009B (en) | Continuous casting billet surface image defect detection method based on deep convolution neural network | |
| CN115953398B (en) | Defect identification method for strip steel surface | |
| CN108596872B (en) | Steel rail disease detection method based on Gabor wavelet and SVM | |
| CN113034442B (en) | Hot rolled strip steel surface quality grading method based on defect distribution map | |
| WO2021135675A1 (en) | Automatic bar counting method and apparatus | |
| CN115035106B (en) | Strip steel defect intelligent detection method | |
| CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
| CN112986277B (en) | Detection method for hot-rolled strip steel finish rolling roll mark | |
| CN108020554A (en) | A kind of steel strip surface defect recognition detection method | |
| CN114494226B (en) | Method for detecting greasy dirt defect of spinning cake based on graph centroid tracking algorithm | |
| CN115049645B (en) | Solar cell panel surface defect detection method | |
| CN113935953A (en) | Steel coil defect detection method based on image processing | |
| CN112669265A (en) | Method for realizing surface defect detection based on Fourier transform and image gradient characteristics | |
| CN107240086A (en) | A kind of fabric defects detection method based on integration nomography | |
| CN112767359A (en) | Steel plate corner detection method and system under complex background | |
| CN115294089A (en) | Steel surface defect detection method based on improved YOLOv5 | |
| CN112668725A (en) | Metal hand basin defect target training method based on improved features | |
| CN104240252A (en) | Detecting Algorithm for cracks of surface of high-temperature billet of machine vision bar | |
| CN115511775A (en) | Light-weight ceramic tile surface defect detection method based on semantic segmentation | |
| CN113436102A (en) | Method for detecting surface defects of machined part |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |