Image recognition method for KR stirring vortex radius

Abstract

The invention discloses a KR stirring vortex radius image recognition method, which comprises the following steps of S1, establishing an image data acquisition system, and acquiring image data; s2, preprocessing the image data according to the number of frames so as to enhance the image data; s3, carrying out threshold segmentation with correction on each frame of image data; s4, selecting an image recognition baseline position; s5, establishing a region window, and identifying a stirring radius; s6, establishing a search micro window in the area window, and determining the position of the stirring radius; and S7, carrying out filtering processing on the real-time identification data so as to smoothly process the discrete data and eliminate abnormal identification values. According to the method, the real-time vortex radius is recognized by adopting a threshold segmentation technology and combining a selective area window progressive search technology aiming at an image recognition parameter (vortex radius), and rapid calculation and image capture accuracy of vortex recognition are realized.

Description

Technical Field

[0001] The invention relates to steelmaking technology, and more specifically to an image recognition method for KR stirring vortex radius. Background Art

[0002] In the process of steelmaking continuous casting billet production, especially the molten iron pretreatment process, in order to improve the desulfurization efficiency, reduce the desulfurization cost, enhance the degree of desulfurization automation, reduce the labor intensity of operators, and achieve the goal of "intelligent steelmaking-dynamic interconnection", it is necessary to optimize the KR (Kambara Reactor) desulfurization agitator paddle height adjustment model and the agitator paddle speed selection model, that is, to realize the information transmission between equipment through the Internet of Things, guide the equipment to operate through the analysis and judgment of the transmitted information, and realize closed-loop control.

[0003] The existing vortex image recognition technology is mainly used for the recognition of vortices such as molten steel, ocean and water flow, and atmosphere, among which:

[0004] Chinese patent application No. 201210409049.3 discloses a slag detection method and device based on vortex image recognition on the surface of molten steel. A camera is installed above the ladle and connected to an image signal control unit and an industrial computer in sequence through cables. The image is then processed and extracted. The free surface image feature recognition includes vortex recognition and vortex state extraction. The center position and state feature quantity information of the vortex on the surface of the ladle of molten steel are obtained. The canny operator is used to obtain the edge of the vortex streamline. The image feature extraction is performed through direction detection and rotation center positioning.

[0005] Chinese patent application No. 202211316681.3 discloses a joint vortex detection device and method based on the physical and geometric characteristics of a two-dimensional flow field, the steps of which include: calculating the local unit surface velocity and stress vorticity difference index of the two-dimensional flow field data points; determining the potential vortex center position according to the two calculated physical parameters; screening the potential vortex center according to the closed contour line and the position of the potential vortex center; after determining the vortex center position, the outermost closed contour line around the vortex center is defined as the edge position of the vortex. This invention improves the accuracy of vortex detection by introducing two physical parameters, the local unit surface velocity and the stress vorticity difference index, and effectively combining them with traditional geometric detection methods.

[0006] Chinese patent application No. 202211185825.6 discloses an immersive visualization method for ocean flow fields based on velocity field data. The vortex detection is based on the velocity field data of at least one sampling point in the ocean flow field, and the data of the horizontal and vertical components in the velocity field data are converted into time series coordinate data. Vortex identification is performed through a multi-feature vortex identification model based on a deep neural network multi-feature perception algorithm.

[0007] Chinese patent application No. 202210888236.8 discloses a method for identifying and tracking mesoscale vortices using a Doppler meteorological radar, which solves the problem of how to effectively use Doppler meteorological radar to automatically and quickly monitor, identify and track severe convective weather systems. The invention uses the least squares derivative method to calculate the azimuth shear, which effectively alleviates the noise error of the radar radial velocity during the identification process, and uses the correlation coefficient method to comprehensively consider the splitting and merging of the convective system during the development process, and constructs a full-process method from radar data input, data processing to meteorological product generation.

[0008] Chinese Patent Application No. 202111675313.3 discloses a vortex identification method and device based on width learning, which includes: obtaining sea surface flow field data of the target area, and determining the flow field characteristics of the sea-air interface of the target area based on the sea surface flow field data; judging the vortex distribution of the target area and determining the vortex distribution result of the target area; after fitting the vortex distribution result through the quadratic surface equation, determining the three-dimensional structure type of the vortex; after learning the characteristics of the identified vortex using the width learning method, predicting the near-shore sub-mesoscale vortex according to the flow field characteristics in the target area, and obtaining the vortex identification result.

[0009] Chinese Patent Application No. 202110796512.3 discloses a method for identifying mesoscale eddies based on sea surface height anomalies and rotation speed, including 1) judging possible cyclonic vortices, and for sea surface height anomaly data with a spatial resolution of 0.25°, by finding the local minimum or maximum value of the sea surface height anomaly within a 1°×1° longitude and latitude moving window, the center of the possible cyclonic vortex or anticyclonic vortex is judged; 2) determining the closed contour line of the sea surface height anomaly, by looking for the closed contour line of the sea surface height anomaly from the inside of each possible cyclonic vortex or anticyclonic vortex center with an increase or decrease of 1 cm outward, and determining the closed contour line of the sea surface height anomaly that meets the conditions, thereby effectively identifying the mesoscale vortex in the ocean.

[0010] Chinese patent application No. 202110525651.2 discloses a vortex identification method, device, storage medium and electronic device. The invention selects valid grid points that may be the vortex center based on wind field data, divides the area around the valid grid points into multiple quadrants, and determines the vortex center based on the effective vorticity when each mode satisfies the counterclockwise rotation condition. This invention can be applied to the identification of mesoscale vortices in complex wind fields.

[0011] Chinese patent application No. 202110441779.0 discloses a method for identifying vortices on the water surface in front of a dam based on deep learning, which includes the following steps: camera installation, normal water surface shooting, water surface vortex simulation, vortex image shooting, data enhancement, transfer learning, identification model distribution and vortex identification.

[0012] Chinese patent application No. 202010594330.3 discloses a mesoscale vortex recognition method based on deep learning. It uses an improved PSPNet (Pyramid Scene Parsing Network) to analyze and recognize two-dimensional images with sea surface height data as the content, thereby realizing the recognition of ocean mesoscale vortices. The ResNet network with dilated convolution is used as the feature extraction network, and a point-by-point spatial attention mechanism neural network module is used to improve the recognition performance of mesoscale vortices in the pyramid scene parsing network model.

[0013] Chinese patent application No. 202010026457.5 discloses a correlated vortex identification device based on an equivalent tensor reconstruction method, which uses a deep neural network to identify vortices;

[0014] Chinese patent application No. 201610714546.2 discloses a global ocean mesoscale eddy fast identification algorithm based on parallelism. The invention first performs high-pass filtering on the data based on the sky-averaged ocean dynamic field sea level anomaly (SLA) satellite altimeter data, and then divides the world into several regions, each of which has an overlapping area with the surrounding area. Identify the vortex seed points in each region and generate SLA contour lines, iterate the SLA contour lines, and determine whether the vortex boundary conditions are met for vortex identification.

[0015] Chinese patent application No. 200480025668.6 (Detection of wake vortices, etc. in the lower atmosphere) measures and identifies aircraft wake atmospheric vortices through acoustic echo analysis technology using multiple receivers.

[0016] In summary, the existing technologies have rich research results on vortex identification. However, these methods cannot be fully used to monitor the real-time stirring scene of KR impellers for the following reasons:

[0017] 1) During the real-time stirring process of the KR stirring paddle, the stirring shaft partially blocks the vortex, and the entire vortex cannot be observed. Therefore, the canny operator streamline edge cannot be effectively implemented; the altimeter cannot measure the full depth data of the vortex;

[0018] 2) The stirring of the KR stirring paddle is under high temperature and severe dust environment. Most detection equipment does not meet the installation and use requirements. The stirring vortex is irregular and highly random. Affected by dust, the grayscale change of the KR stirring image fluctuates greatly. Summary of the invention

[0019] In view of the defects existing in the prior art, the purpose of the present invention is to provide an image recognition method for the KR stirring vortex radius. For the image recognition parameter (vortex radius), the threshold segmentation technology is adopted and combined with the selection window progressive search technology to identify the real-time vortex radius, thereby realizing the rapid calculation of vortex recognition and the image capture accuracy.

[0020] To achieve the above object, the present invention adopts the following technical solution:

[0021] A method for image recognition of KR stirring vortex radius comprises the following steps:

[0022] S1. Establish an image data acquisition system to collect image data;

[0023] S2, preprocessing the image data according to the number of frames to enhance the image data;

[0024] S3, performing threshold segmentation with correction on each frame of the image data;

[0025] S4, selecting the image recognition baseline position;

[0026] S5, establishing a regional window and identifying the stirring radius;

[0027] S6, establishing a search micro-window within the area window to determine the position of the stirring radius;

[0028] S7, real-time recognition data filtering processing to smooth discrete data and eliminate abnormal recognition values.

[0029] Preferably, in step S1, the image data acquisition system includes a visible light camera system and a control server connected to the visible light camera system through a network.

[0030] Preferably, in step S2, digital image edge enhancement processing technology is used to amplify feature position information in the image data.

[0031] Preferably, the threshold segmentation process with correction in step S3 is a binary basic threshold plus a dynamic incremental threshold, and the dynamic incremental threshold adopts the image gray background mean B gray Multiply by the adjustment coefficient α, that is, threshold Threshold = T base +B gray *α;

[0032] Where, T base is the basic threshold of binarization, ranging from 80 to 250; B gray is the gray background mean of the image; α is the adjustment coefficient, which is between 0.01 and 0.3.

[0033] Preferably, identifying the stirring radius in step S5 specifically includes:

[0034] Image enhancement processing is performed within the area window, and the vortex radius is identified by using a selected window progressive search technique based on the vortex image features.

[0035] Preferably, the micro-window judgment threshold in the selected window progressive search technology used to identify the vortex radius is between 10% and 90%.

[0036] Preferably, determining the position of the stirring radius in step S6 specifically includes:

[0037] The slag distribution area in the search micro-window statistics window is used, and when the slag distribution area is greater than a threshold, the position of the stirring radius is determined.

[0038] Preferably, in step S7, a real-time discrete data filtering method is used to perform real-time identification data filtering processing.

[0039] Preferably, the real-time discrete data filtering method adopts a limiting average filtering method, that is, a limiting filtering method plus a recursive average filtering method.

[0040] The present invention provides an image recognition method for the KR stirring vortex radius, that is, for an image recognition parameter (vortex radius), a threshold segmentation technology is used in combination with a selection window progressive search technology to identify the real-time vortex radius, thereby realizing rapid calculation of vortex recognition and achieving higher image capture accuracy; the image recognition processing and calculation of the vortex radius of the KR stirring process is realized, and the working condition simulation of the image processing system can be performed at any time, which effectively reduces the development and debugging time, reduces the on-site debugging time and intensity, and achieves the purpose of reducing costs and increasing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the flow of the image recognition method of the present invention;

[0042] Figure 2 It is a schematic diagram of the KR stirring vortex and slag area recognition in the image recognition method of the present invention, wherein the left side is a schematic diagram of the KR stirring vortex and the right side is a schematic diagram of the slag area. DETAILED DESCRIPTION

[0043] In order to better understand the above technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0044] Combination Figure 1 As shown, the image recognition method of a KR stirring vortex radius provided by the present invention comprises the following steps:

[0045] S1. Establish an image data acquisition system, use a conventional visible light camera system to capture KR mixing process images in real time, and transmit them to the control server in real time via Gigabit Ethernet;

[0046] S2, pre-processing the collected real-time video according to the frame number, applying digital image edge enhancement processing technology to amplify the feature position information to improve recognition accuracy and enhance image data;

[0047] S3, performing threshold segmentation and binarization processing on each frame of image data to highlight the molten iron and slag surface areas;

[0048] S4, selecting the image recognition baseline position;

[0049] S5, establish a regional window, perform image enhancement processing (image contrast processing) in the window, and identify the vortex radius by selecting a window progressive search technology according to the vortex image characteristics;

[0050] S6. Since the vortex space is a bright area and the slag area distribution area is a dark area, a search is performed from the center of the stirring blade to the edge of the vortex in the area window to establish a search micro-window. The micro-window is used to count the slag distribution area in the window. When the slag area is greater than a certain threshold, it is identified as the vortex radius position.

[0051] S7, real-time recognition data filtering processing to smooth discrete data and eliminate abnormal recognition values.

[0052] In step S1, the visible light camera system requires multiple high temperature and dust protection measures such as water cooling, air cooling, glass baffles and air sweeping dust removal.

[0053] The corrected threshold segmentation process in step S3 is a binary basic threshold plus a dynamic incremental threshold, and the dynamic incremental threshold adopts the image gray background mean value B gray Multiply by the adjustment coefficient α, that is, threshold Threshold = T base +B gray *α;

[0054] Where, T base is the basic threshold of binarization, ranging from 80 to 250; B gray is the gray background mean of the image; α is the adjustment coefficient, which is between 0.01 and 0.3.

[0055] In step S5, the micro-window judgment threshold in the window selection progressive search technology used to identify the vortex radius is between 10% and 90%.

[0056] In step S7, a real-time discrete data filtering method is used to identify data filtering in real time. The real-time discrete data filtering method adopts a limited average filtering method, that is, a limited filtering method + a recursive average filtering method. The principle is that each new data sampled is first subjected to a limited processing, and then sent to a queue for recursive average filtering; and the recursive average filtering method (sliding average filtering method) regards N consecutive sampled values ​​as a queue, and the length of the queue is fixed to N. Each time a new data is sampled, it is put into the end of the queue, and the original data at the head of the queue is discarded (first-in-first-out principle), and the N data in the queue are averaged to obtain a new filtering result.

[0057] Combination Figure 2 As shown, in the KR stirring vortex schematic diagram on the left, the circle represents the center of the vortex radius, and the dotted line represents the vortex radius schematic curve. When the vortex radius changes, the dotted line will expand and contract accordingly with the recognition result.

[0058] The slag area diagram on the right is a binary image based on the threshold segmentation technology, wherein the threshold segmentation can be dynamically adjusted according to the background grayscale mean of the image. It can be seen from the figure that selecting a suitable threshold can display the slag distribution on the surface of molten iron. The present invention adopts the selected area progressive search technology to effectively identify the radius of the stirring vortex. The principle is based on the distribution near the baseline in the figure and close to the center of the vortex. The white-bright area is related. The larger the vortex radius, the larger the average white-bright area. A sub-window area is first selected, starting from the edge of the stirring paddle, and searching one area at a time from the center to the edge of the vortex. When the ratio of the slag area in the sub-window area to the molten iron is greater than a certain threshold, it is considered that the edge of the stirring radius has been searched and identified as the recognition result. At the same time, in order to reduce the random volatility of the numerical value, the technology of the present invention also uses the limited mean filtering technology for data processing. The final result is shown in the KR stirring vortex diagram on the left.

[0059] The present invention indirectly characterizes the desulfurizer dispersion state by real-time identification of the vortex distribution during the KR stirring process, thereby controlling the height and rotation speed of the stirring paddle to achieve the optimal stirring state, thereby improving the desulfurization efficiency. It is an intelligent recognition system technology based on digital image processing technology and regional space search technology.

[0060] Example 1

[0061] S1. Use a pinhole camera to capture real-time images of KR stirring, obtain the background image through threshold segmentation, and use a selection window progressive search and recognition technology to identify information such as the vortex radius; first establish an image data acquisition system, use a conventional pinhole camera to capture real-time images of the KR stirring process, and transmit them to the control server in real time through Gigabit Ethernet; the camera system requires multiple high temperature and dust protection measures such as water cooling, air cooling, glass baffles and air sweeping dust removal;

[0062] S2, pre-processing the collected real-time video according to the frame image, applying the digital image edge enhancement processing technology to amplify the feature position information to improve the recognition accuracy. This embodiment 1 adopts the corrosion expansion operation method;

[0063] S3, for a frame of image, threshold segmentation, binarization processing, highlighting the molten iron and slag surface areas; in the threshold segmentation processing, the binarization basic threshold T base The dynamic increment threshold is taken as 150, and the image gray background mean value B gray (taken as 150) multiplied by the adjustment coefficient, that is:

[0064] Threshold=T base +B gray *α=150+150*0.2=180;

[0065] Among them, α is 0.2; (α is the adjustment coefficient)

[0066] S4. Select the image recognition baseline position, such as Figure 2 As shown in the slag area diagram on the right;

[0067] S5, establish a regional window, perform image enhancement processing (image contrast processing) in the window, and identify the vortex radius by selecting a window progressive search technology according to the vortex image characteristics;

[0068] S6. Since the vortex space is a bright area and the slag area distribution area is a dark area, a search is performed from the center of the stirring blade to the edge of the vortex in the area window to establish a search micro-window. The micro-window is used to count the slag distribution area in the window. When the slag area is greater than a certain threshold (the micro-window judgment threshold value in the selection window progressive search technology used in the embodiment 1 to identify the vortex radius is 90), it is identified as the vortex radius position;

[0069] S7, real-time identification data filtering processing to smooth discrete data and eliminate abnormal identification values; the amount of filtered data is 125 data points.

[0070] Example 2

[0071] S1. Use a pinhole camera to capture real-time images of KR stirring, obtain the background image through threshold segmentation, and use a selection window progressive search and recognition technology to identify information such as the vortex radius; first establish an image data acquisition system, use a conventional pinhole camera to capture real-time images of the KR stirring process, and transmit them to the control server in real time through Gigabit Ethernet; the camera system requires multiple high temperature and dust protection measures such as water cooling, air cooling, glass baffles and air sweeping dust removal;

[0072] S2, pre-processing the collected real-time video according to the frame image, applying the digital image edge enhancement processing technology to amplify the feature position information to improve the recognition accuracy. This embodiment 2 adopts the Scharr+Laplace method;

[0073] S3, for a frame of image, threshold segmentation, binarization processing, highlighting the molten iron and slag surface areas; in the threshold segmentation processing, the binarization basic threshold T base The dynamic increment threshold is taken as 160, and the image gray background mean value B gray (taken as 90) multiplied by the adjustment coefficient, that is:

[0074] Threshold=T base +B gray *α=160+90*0.1=169;

[0075] Among them, α is 0.1; (α is the adjustment coefficient)

[0076] S4. Select the image recognition baseline position, such as Figure 2 As shown in the slag area diagram on the right;

[0077] S5, establish a regional window, perform image enhancement processing (image contrast processing) in the window, and identify the vortex radius by selecting a window progressive search technology according to the vortex image characteristics;

[0078] S6. Since the vortex space is a bright area and the slag area distribution area is a dark area, a search is performed from the center of the stirring blade to the edge of the vortex in the area window to establish a search micro-window. The micro-window is used to count the slag distribution area in the window. When the slag area is greater than a certain threshold (the micro-window judgment threshold value in the selection window progressive search technology used in the embodiment 2 to identify the vortex radius is 85), it is identified as the vortex radius position;

[0079] S7, real-time identification data filtering processing to smooth discrete data and eliminate abnormal identification values; the amount of filtered data is 120 data points.

[0080] Example 3

[0081] S1. Use a pinhole camera to capture real-time images of KR stirring, obtain the background image through threshold segmentation, and use a selection window progressive search and recognition technology to identify information such as the vortex radius; first establish an image data acquisition system, use a conventional pinhole camera to capture real-time images of the KR stirring process, and transmit them to the control server in real time through Gigabit Ethernet; the camera system requires multiple high temperature and dust protection measures such as water cooling, air cooling, glass baffles and air sweeping dust removal;

[0082] S2, pre-processing the collected real-time video according to the frame image, applying the digital image edge enhancement processing technology to amplify the feature position information to improve the recognition accuracy. This embodiment 3 adopts the corrosion enhancement method;

[0083] S3, for a frame of image, threshold segmentation, binarization processing, highlighting the molten iron and slag surface areas; in the threshold segmentation processing, the binarization basic threshold T base The dynamic increment threshold is taken as 130, and the image gray background mean value B gray (taken as 200) multiplied by the adjustment coefficient, that is:

[0084] Threshold=T base +B gray *α=130+200*0.15=160;

[0085] Among them, α is 0.15; (α is the adjustment coefficient)

[0086] S4. Select the image recognition baseline position, such as Figure 2 As shown in the slag area diagram on the right;

[0087] S5, establish a regional window, perform image enhancement processing (image contrast processing) in the window, and identify the vortex radius by selecting a window progressive search technology according to the vortex image characteristics;

[0088] S6. Since the vortex space is a bright area and the slag area distribution area is a dark area, a search is performed from the center of the stirring blade to the edge of the vortex in the area window to establish a search micro-window. The micro-window is used to count the slag distribution area in the window. When the slag area is greater than a certain threshold (the micro-window judgment threshold value in the selection window progressive search technology used in the vortex radius identification of this embodiment 3 is 100), it is identified as the vortex radius position;

[0089] S7, real-time identification data filtering processing, to smooth discrete data, eliminate abnormal identification values; the amount of filtered data is 130 data points.

[0090] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A method for image recognition of KR stirring vortex radius, characterized in that: The following steps are involved: S1. Establish an image data acquisition system to collect image data; S2, preprocessing the image data according to the number of frames to enhance the image data; S3, performing threshold segmentation with correction on each frame of the image data; S4, selecting the image recognition baseline position; S5, establishing a regional window and identifying the stirring radius; S6, establishing a search micro-window within the area window to determine the position of the stirring radius; S7, real-time recognition data filtering processing to smooth discrete data and eliminate abnormal recognition values.

2. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: In the step S1, the image data acquisition system includes a visible light camera system and a control server connected to the visible light camera system through a network.

3. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: In the step S2, digital image edge enhancement processing technology is used to amplify the feature position information in the image data.

4. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: The threshold segmentation process with correction in step S3 is a binary basic threshold plus a dynamic incremental threshold, and the dynamic incremental threshold adopts the image gray background mean value B gray Multiply by the adjustment factor α, That is, threshold Threshold = T base +B gray *α; Where, T base is the basic threshold of binarization, ranging from 80 to 250; B gray is the gray background mean of the image; α is the adjustment coefficient, which is between 0.01 and 0.

3.

5. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: The identification of the stirring radius in step S5 specifically includes: Image enhancement processing is performed within the area window, and the vortex radius is identified by using a selected window progressive search technique based on the vortex image features.

6. The image recognition method of KR stirring vortex radius according to claim 5, characterized in that: The micro-window judgment threshold in the selection window progressive search technology used to identify the vortex radius is between 10% and 90%.

7. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: Determining the position of the stirring radius in step S6 specifically includes: The slag distribution area in the search micro-window statistics window is used, and when the slag distribution area is greater than a threshold, the position of the stirring radius is determined.

8. The image recognition method of KR stirring vortex radius according to claim 1, characterized in that: In step S7, a real-time discrete data filtering method is used to perform real-time identification data filtering processing.

9. The image recognition method of KR stirring vortex radius according to claim 8, characterized in that: The real-time discrete data filtering method adopts a limiting average filtering method, that is, a limiting filtering method plus a recursive average filtering method.

Images