Angle detection method and device for continuous casting billet, electronic equipment and storage medium

Abstract

The invention provides an angle detection method and device for a continuous casting billet, electronic equipment and a storage medium, and relates to the field of automatic detection in the steel manufacturing process. Comprising the steps of obtaining a production picture of a target continuous casting billet; inputting the production picture into a trained continuous casting blank semantic segmentation model for segmentation processing, and determining a target area corresponding to the target continuous casting blank; recognizing the production picture through an image recognition algorithm, and determining a target center line of the target continuous casting billet; if an overlapping area exists between the target area and a preset area, and / or the difference value between the slope of the target center line and the preset slope is larger than a preset difference value, angle deflection of the target continuous casting billet is determined; and if the target area and the preset area do not have the overlapping area, and the difference value between the slope of the target center line and the preset slope is not larger than the preset difference value, it is determined that the angle of the target continuous casting billet does not deflect.

Description

Technical Field

[0001] The present invention relates to the field of automated detection in a steel manufacturing process, and in particular to a method, device, electronic equipment and storage medium for detecting the angle of a continuous casting billet. Background Art

[0002] In modern steel production, continuous casting technology is a critical process, and its quality directly impacts the processing and use of subsequent products. Continuous casting slab cutting is the process of cutting continuously cast long slabs into individual slabs according to specified dimensions. After cutting, the slabs are transported linearly along the roller conveyor to the next process step. However, factors such as uneven roller conveyor surfaces and incomplete slab cutting leading to slab adhesion can cause uneven force on the roller conveyor, leading to skewed movement and deviation. This deviation can lead to a series of problems, including slab jamming, roller conveyor blockage, and equipment impact, impacting production stability.

[0003] In the related art, the detection of continuous casting slab deviation relies on the operator to manually judge the monitoring screen in the control room, which is prone to the problem of low accuracy of the detection results. Summary of the Invention

[0004] The present invention provides a method, device, electronic equipment and storage medium for detecting the angle of a continuous casting billet, which are used to solve the problem in the related art that manual detection of the angle deviation of the continuous casting billet leads to low accuracy of the detection result.

[0005] In a first aspect, the present invention provides a method for detecting the angle of a continuous casting billet, the method comprising:

[0006] Acquire the production picture of the target continuous casting billet;

[0007] Inputting the production image into a trained continuous casting billet semantic segmentation model for segmentation processing to determine a target area corresponding to the target continuous casting billet;

[0008] Identifying the production picture by an image recognition algorithm to determine the target center line of the target continuous casting billet;

[0009] If there is an overlapping area between the target area and the preset area, and / or the difference between the slope of the target center line and the preset slope is greater than the preset difference, determining that the target continuous casting billet is angularly deflected;

[0010] If the target area and the preset area do not overlap, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference, it is determined that the angle of the target continuous casting billet is not skewed.

[0011] Optionally, the target continuous casting billet runs on the target roller table, and the identifying the production screen by an image recognition algorithm to determine the target center line of the target continuous casting billet includes:

[0012] Recognizing and processing the production image using the image recognition algorithm to obtain a first edge line segment and a second edge line segment, wherein the first edge line segment and the second edge line segment both represent an edge of the target continuous casting billet in the running direction of the target roller;

[0013] Obtaining coordinates of N first target points, where the first target point is any point on the first edge segment;

[0014] According to the coordinates of the N first target points, the coordinates of the N second target points on the second edge segment are obtained, and the coordinates of the second target points in the running direction perpendicular to the target roller are the same as those of the first target points.

[0015] Determining coordinates of N center points based on the N first target points and the N second target points;

[0016] The line connecting the N center points is used as the target center line.

[0017] Optionally, the preset area is a quadrilateral area adjacent to the left and right of the motion trajectory of the continuous casting billet when it is not deviated.

[0018] Optionally, the preset slope is the slope of a preset center line, and the preset center line represents the movement trajectory of the continuous casting billet when the angle is not deflected.

[0019] Optionally, the trained continuous casting slab semantic segmentation model is trained by the following steps:

[0020] Acquiring training data, wherein the training data includes production images of continuous casting billets under different conditions;

[0021] The original model is trained using the training data to obtain a trained continuous casting slab semantic segmentation model.

[0022] Optionally, the method further includes:

[0023] When it is determined that the angle of the target continuous casting billet is deviated, an alarm signal is generated.

[0024] In a second aspect, the present invention provides a device for detecting the angle of a continuous casting billet, the device comprising:

[0025] An acquisition module is used to acquire the production picture of the target continuous casting billet;

[0026] An input module is used to input the production picture into a trained continuous casting billet semantic segmentation model for segmentation processing to determine the target area corresponding to the target continuous casting billet;

[0027] an identification module, configured to identify the production image through an image recognition algorithm and determine a target center line of the target continuous casting billet;

[0028] a judgment module, configured to determine that the target continuous casting billet is skewed in angle if there is an overlapping area between the target area and the preset area, and / or if the difference between the slope of the target centerline and the preset slope is greater than a preset difference;

[0029] The judgment module is also used to determine that the angle of the target continuous casting billet is not skewed if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference.

[0030] In a third aspect, the present invention provides an electronic device, comprising:

[0031] processor;

[0032] a memory for storing instructions executable by the processor;

[0033] The processor is configured to execute the instructions to implement the method as described in the first aspect.

[0034] In a fourth aspect, the present invention provides a storage medium, which, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the method described in the first aspect.

[0035] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor as described in the first aspect.

[0036] An embodiment of the present invention provides a method, device, electronic device and storage medium for detecting the angle of a continuous casting billet. Specifically, a production picture of a target continuous casting billet is obtained; the production picture is input into a trained semantic segmentation model of the continuous casting billet for segmentation processing, and a target area corresponding to the target continuous casting billet is determined; the production picture is identified by an image recognition algorithm, and a target center line of the target continuous casting billet is determined; if there is an overlapping area between the target area and a preset area, and / or the difference between the slope of the target center line and the preset slope is greater than the preset difference, it is determined that the angle of the target continuous casting billet is skewed; if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference, it is determined that the angle of the target continuous casting billet is not skewed. In this way, by utilizing the synergistic effect of the trained continuous casting billet semantic segmentation model and the image recognition algorithm, it is possible to determine whether the angle of the target continuous casting billet is skewed, that is, whether there is a deviation phenomenon, based on the target area and / or target center line. This not only realizes the automatic detection of whether the target continuous casting billet is deviated without human intervention, but also can further improve the accuracy of the detection results based on the combination of the model and the algorithm, solving the problem of low accuracy of the detection results caused by manual detection of the continuous casting billet angle in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0038] Figure 1 A flow chart of a method for detecting angles of continuous casting billets provided by an embodiment of the present invention;

[0039] Figure 2 This is a conceptual diagram of the output of a trained continuous casting slab semantic segmentation model provided by an embodiment of the present invention;

[0040] Figure 3 A conceptual diagram of a method for detecting angles of continuous casting billets provided by an embodiment of the present invention;

[0041] Figure 4 A conceptual diagram of a target centerline provided by an embodiment of the present invention;

[0042] Figure 5 A conceptual diagram of a method for detecting angles of continuous casting billets provided by an embodiment of the present invention;

[0043] Figure 6 A conceptual diagram of a preset area and a preset center line provided in an embodiment of the present invention;

[0044] Figure 7 A structural block diagram of an angle detection device for continuous casting provided by an embodiment of the present invention;

[0045] Figure 8 This is a structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings, but it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure.

[0047] As described in the background technology, continuous casting technology is a key production link in modern steel production, and its quality directly affects the processing and use of subsequent products. Continuous casting slab cutting is the process of cutting continuously cast long slabs into individual slabs according to specified dimensions. After cutting, the slabs are transported to the next process in a straight line along the roller conveyor below. However, factors such as uneven roller conveyor and incomplete slab cutting leading to slab adhesion can cause uneven force on the roller conveyor, resulting in slab deviation. This deviation can cause a series of problems, including slab jamming, roller clogging, and equipment collisions, affecting production stability.

[0048] Related technologies rely primarily on manual video monitoring to determine deviations, which is labor-intensive and prone to missed detections, making it difficult to meet the requirements for efficient and accurate detection. Therefore, an automated detection method is urgently needed to improve detection efficiency and ensure production safety.

[0049] An embodiment of the present invention provides a method, device, electronic device and storage medium for detecting the angle of a continuous casting billet. Specifically, a production picture of a target continuous casting billet is obtained; the production picture is input into a trained semantic segmentation model of the continuous casting billet for segmentation processing, and a target area corresponding to the target continuous casting billet is determined; the production picture is identified by an image recognition algorithm, and a target center line of the target continuous casting billet is determined; if there is an overlapping area between the target area and a preset area, and / or the difference between the slope of the target center line and the preset slope is greater than the preset difference, it is determined that the angle of the target continuous casting billet is skewed; if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference, it is determined that the angle of the target continuous casting billet is not skewed. In this way, by utilizing the synergistic effect of the trained continuous casting billet semantic segmentation model and the image recognition algorithm, it is determined whether the angle of the target continuous casting billet is skewed, that is, whether there is a deviation phenomenon, based on the target area and / or target center line. This realizes the automatic detection of whether the target continuous casting billet is deviated, improves the accuracy of the detection results, and eliminates the need for human intervention, solving the problem of low accuracy of detection results caused by manual detection of continuous casting billet angle deviation in related technologies.

[0050] Furthermore, related technologies, such as traditional computer vision and deep learning-based visual algorithms, offer the potential for automated detection of cutting conditions. However, traditional computer vision suffers from insufficient robustness, poor adaptability, and the need for hand-crafted feature algorithms for detection. Furthermore, camera footage used on production sites often contains numerous interfering factors, and algorithm performance can rapidly degrade when cutting conditions or ambient lighting change. Therefore, relying solely on traditional vision technology results in poorly robust detection algorithms. Deep learning technology also requires a high level of datasets; training semantic segmentation models requires a large amount of labeled data, and the complexity of the production environment places high demands on the segmentation capabilities of deep learning models.

[0051] The angle detection method for continuous casting billets provided in the embodiment of the present invention not only realizes automated detection, but also further collaboratively applies the semantic segmentation model of continuous casting billets using deep learning technology and the image recognition algorithm using traditional computer vision technology. The dual detection method can not only overcome the technical problems of the single method in the related technology, but also further make the detection results more accurate.

[0052] It should be understood that the angle detection method for the continuous casting billet provided in the embodiment of the present invention can be executed by a target device. The target device can be a programmable logic controller (PLC) capable of driving a holding needle. The target device can be a single electronic device or multiple electronic devices that cooperate with each other to execute. The electronic device can be a server, such as an independent physical server, a server cluster composed of multiple servers, and a cloud server capable of cloud computing.

[0053] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0054] Figure 1 Flowchart of a method for detecting the angle of a continuous casting billet provided by an embodiment of the present invention. Figure 1 As shown, the angle detection method of the continuous casting billet provided by the present invention includes steps 110 to 150.

[0055] Step 110: Acquire a production picture of the target continuous casting billet.

[0056] In the embodiment of the present invention, the target continuous casting billet may be any continuous casting billet on the continuous casting billet production line (which may also be understood as moving on the continuous casting billet roller). The production image of the target continuous casting billet may be an image of the target continuous casting billet on the roller.

[0057] In an embodiment of the present invention, in order to obtain the production picture of the target continuous casting billet, a camera can be installed above the roller where the continuous casting billet moves with a bird's-eye view at the production site of the target continuous casting billet. The embodiment of the present invention does not impose specific restrictions on the installation position, number of installations and installation angle of the camera. For example, a camera can be installed in the middle above the roller, and the monitoring picture of the camera can be used as the production picture. A camera can also be installed at a certain distance above the roller for sequential detection so as to more promptly detect the deviation of the target continuous casting billet and make the detection result more accurate.

[0058] In an embodiment of the present invention, the camera's monitoring image can be used as the target continuous casting billet production image. Accordingly, the target device can obtain the target continuous casting billet production image by receiving the real-time camera image and detect the target billet's angle in real time. Furthermore, additional image processing can be performed on the camera's monitoring image to make it clearer. For example, the monitoring image can be subjected to strong light suppression processing and used as the target continuous casting billet production image after the strong light suppression processing.

[0059] In the embodiment of the present invention, after obtaining the production picture including the target continuous casting billet, the target device may respectively execute step 120 and step 130 for the production picture.

[0060] Step 120 : Input the production image into a trained continuous casting billet semantic segmentation model for segmentation processing to determine a target area corresponding to the target continuous casting billet.

[0061] In an embodiment of the present invention, the trained continuous casting billet semantic segmentation model can be any trained deep learning model that can perform segmentation processing on the production screen. After the production screen is input into the trained continuous casting billet semantic segmentation model, the trained continuous casting billet semantic segmentation model can perform pixel-level semantic segmentation processing on the production screen, identify the target continuous casting billet on the production screen, and output the annotation box, annotation and confidence corresponding to the target continuous casting billet, for example Figure 2 As shown in the figure, the large shaded area is the target continuous casting billet body identified by the trained continuous casting billet semantic segmentation model, and the box is the annotation box corresponding to the target continuous casting billet. The annotation boxes are respectively marked (such as Figure 2 The slag shown represents the continuous casting billet) and the confidence level (such as Figure 2 Figures such as 0.91 are shown).

[0062] In the embodiment of the present invention, the target area may be the screen area occupied by the target continuous casting billet in the production screen, for example Figure 2 The shaded area within the marked box.

[0063] Step 130: Identify the production image using an image recognition algorithm to determine the target center line of the target continuous casting billet.

[0064] In an embodiment of the present invention, the target device may also input the production screen into an image recognition algorithm, which may be any conventional computer vision algorithm. The image recognition algorithm performs image preprocessing such as grayscale processing on the production screen, and further performs feature extraction on the preprocessed production screen to obtain the target center line of the target continuous casting billet. The target center line represents the movement angle of the target continuous casting billet, and accordingly, the direction of the target center line is the direction in which the target continuous casting billet moves on the roller, for example Figure 2 As shown, the target continuous casting billet moves from the top to the bottom of the image, and the target centerline runs through the top and bottom of the target continuous casting billet. The target centerline can be a line connecting the center points of two opposite sides of the target continuous casting billet, or it can be a line running through a certain section of the target continuous casting billet.

[0065] In an embodiment of the present invention, the target center line of the target continuous casting billet can be directly determined through an image recognition algorithm. The shape of the target continuous casting billet and the coordinates of all points on the target continuous casting billet and the length and width side positioning can also be determined through an image recognition algorithm, and the target center line of the target continuous casting billet can be further determined through a target device.

[0066] In the embodiment of the present invention, after obtaining the target area through step 120 and obtaining the target center line through step 130, the following can be done: Figure 3 As shown, the target area is compared with the preset area to determine whether there is overlap. The slope of the target center line can also be compared with the preset slope. Based on the two comparison results, the target device is determined to execute step 140 or step 150.

[0067] Step 140 : If the target area and the preset area have an overlapping area, and / or the difference between the slope of the target center line and the preset slope is greater than the preset difference, then determine the angular deviation of the target continuous casting billet.

[0068] In an embodiment of the present invention, the target area can be compared with the preset area to determine whether there is an overlapping phenomenon or an overlapping area. The preset area is a quadrilateral area adjacent to the left and right of the motion trajectory of the continuous casting billet when it is not deviated. Since the preset area includes the left and right sides, when the target continuous casting billet enters the preset area on either side, it can be considered that the target area and the preset area have an overlapping area. If there is an overlapping phenomenon between the target area and the preset area, it can be considered that the target continuous casting billet has not moved normally, that is, the motion angle is skewed and the target continuous casting billet has deviated. Figure 3 As shown, there is no overlap between the shadow area corresponding to the target continuous casting billet and the preset area.

[0069] In an embodiment of the present invention, the slope of the target center line can also be compared with a preset slope, where the preset slope is the slope of the preset center line, which represents the motion trajectory of the continuous casting billet when the angle is not deflected. The corresponding preset difference can represent the normal tilt range, such as 5 degrees. If the difference between the slope of the target center line and the preset slope is greater than the preset difference, it can be considered that the actual motion angle of the target continuous casting billet is deflected, and the target continuous casting billet is deviating. Figure 3 As shown, the preset slope can be 0 degrees, that is, perpendicular to Figure 3 The slope of the center line of the target roller on the surface, at this time, the difference between the slope of the target center line and the preset slope is greater than the preset difference.

[0070] In an embodiment of the present invention, when there is an overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is greater than the preset difference, it can also be considered that the target continuous casting billet is deviated.

[0071] Step 150: If the target area and the preset area do not overlap, and the difference between the slope of the target centerline and the preset slope is not greater than the preset difference, it is determined that the angle of the target continuous casting billet is not skewed.

[0072] In an embodiment of the present invention, if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference, after double judgment, if all conditions are met, it is considered that the actual motion trajectory of the target continuous casting billet is gentle with the ideal normal motion trajectory, the motion angle is not deviated, and no deviation occurs.

[0073] In this way, by utilizing the synergistic effect of the trained continuous casting billet semantic segmentation model and the image recognition algorithm, it is possible to determine whether the angle of the target continuous casting billet is skewed, that is, whether there is a deviation phenomenon, based on the target area and / or target center line. This not only realizes the automatic detection of whether the target continuous casting billet is deviated without human intervention, but also can further improve the accuracy of the detection results based on the combination of the model and the algorithm, solving the problem of low accuracy of the detection results caused by manual detection of the continuous casting billet angle in related technologies.

[0074] In an embodiment of the present invention, the target continuous casting billet runs on a target roller, and a specific implementation method of step 130 can refer to the following steps: the production screen is identified and processed by the image recognition algorithm to obtain a first edge line segment and a second edge line segment, the first edge line segment and the second edge line segment both represent the edge of the target continuous casting billet in the running direction of the target roller, and the coordinates of N first target points are obtained, the first target point is an arbitrary point on the first edge line segment, and according to the coordinates of the N first target points, the coordinates of N second target points on the second edge line segment are obtained, the coordinates of the second target point in the running direction perpendicular to the target roller are the same as those of the first target point, and the coordinates of N center points are determined according to the N first target points and the N second target points, and finally the line connecting the N center points is used as the target center line.

[0075] In order to better understand step 130 provided in the embodiment of the present invention, Figure 4 As shown, Figure 4 The left side can be a production screen, which shows the target continuous casting billet running on the target roller. The target roller runs in the north-south direction. In layman's terms, the center of the production screen can be used as the origin, and the running direction of the target roller is the X-axis direction. The image recognition algorithm can be used to grayscale and binarize the production screen, and use the Hough transform to identify the two edges of the target continuous casting billet, namely the first edge segment and the second edge segment. In the process of obtaining the first edge segment and the second edge segment, the image recognition algorithm further determines the coordinates of all pixel points on the production screen.

[0076] Furthermore, the target device obtains the coordinates of N first target points, and the first target point can be any point on the first edge segment. In order to make the slope of the target center line more accurate, the first target point can be the middle segment of the first edge segment (remove the upper and lower parts of the first edge segment, for example Figure 4 The coordinates of any point on the first target point can be expressed as (x1, y1). After determining N first target points, the second target point on the second edge segment can be determined. Since the second target point is perpendicular to the running direction of the target roller (such as Figure 4As shown, it can be considered that the coordinates on the east-west direction (i.e., the Y axis) are the same as those of the first target point, then the coordinates of a certain point of the second target point can be (x2, y1). Finally, the coordinates of the N center points are determined based on the N first target points and the N second target points. Using the above example, the coordinates of the center point can be ((x1+x2) / 2, y1), that is Figure 4 As shown on the right, the points on the connecting line can be understood as the center points, and the line connecting the N center points is fitted with linear regression to obtain the target center line, thereby calculating the slope of the target center line.

[0077] In an embodiment of the present invention, a certain number of sampling points can be preset in advance for the target continuous casting billet, and the horizontal coordinates of the points (which can be understood as the first target point and the second target point) where the first edge segment and the second edge segment have the same vertical coordinates as the sampling points are calculated, and then the midpoint coordinates corresponding to all the sampling points are determined to finally obtain the target center line.

[0078] In an embodiment of the present invention, before using the trained semantic segmentation model of the continuous casting billet in step 120, the original model can be trained by the following steps: obtaining training data, wherein the training data includes production images of the continuous casting billet under different conditions. The original model is trained with the training data to obtain a trained semantic segmentation model of the continuous casting billet. Specifically, the training data can be image data collected by a camera under different conditions during the movement of the continuous casting billet, and the different conditions can be different roller speeds, different ambient lighting, different billet sizes, etc. These training data are then labeled (for example, the Labelme labeling tool is used to perform pixel-level labeling on the filtered images required for the semantic segmentation algorithm). The labeled training data is used to train the original model (the original model can be YOLOv8mSeg) to achieve pixel-level detection and segmentation of the continuous casting billet.

[0079] In this embodiment of the present invention, the training data is first divided into a training set, a test set, and a validation set in a ratio of 7:2:1. An independent validation set is retained to impartially evaluate the generalization ability of the model during the experiment. The original image resolution can be scaled to obtain better training and reduce dataset consumption. After each training output, the validation set is used to evaluate model performance, monitor model accuracy and loss, adjust training parameters to avoid overfitting, and obtain the optimal original segmentation model as the trained continuous casting slab semantic segmentation model.

[0080] In an embodiment of the present invention, when the target continuous casting billet is determined to be angularly deflected, an alarm signal is generated. The alarm signal may be in the form of sound, light, text, etc., and the target device may issue a roller stop instruction to promptly correct the running direction of the target continuous casting billet.

[0081] In order to better understand the angle detection method of the continuous casting billet provided by the embodiment of the present invention, an example is given below. In the embodiment of the present invention, the angle detection method of the continuous casting billet mainly includes steps S1 to S7:

[0082] Step S1: Data collection and model training.

[0083] S101 camera installation: At the continuous casting production site, a high-resolution camera is installed facing the roller conveyor where the continuous casting billets are moving, with a bird's-eye view. The camera is set to strong light suppression mode.

[0084] S102 Image acquisition: The image data of the continuous casting billet during its movement are collected by a camera (i.e., the training data described above). These image data should cover different conditions (including different roller speeds, ambient lighting, billet sizes, etc.).

[0085] S103 Data Labeling: Use the Labelme labeling tool to perform pixel-level labeling on the filtered images required for the semantic segmentation algorithm. The labeled dataset contains 1,500 images, and each image has a corresponding labeled label file.

[0086] Step S2:

[0087] The labeled dataset is used to train a deep learning model (i.e., the original model described above) to achieve pixel-level detection and segmentation of continuous casting billets.

[0088] S201 Dataset Partitioning: The dataset is divided into training set, test set, and validation set in a ratio of 7:2:1. An independent validation set is retained to fairly evaluate the generalization ability of the model during the experiment.

[0089] S202 model training: Using the NVIDIA RTX 4090 training platform and the Pytorch deep learning framework, the original image resolution of 2560*1440 was scaled to 640*640. YOLOv8mSeg was used as the base model to balance speed and accuracy. The batch size was set to 16, the initial learning rate was set to 0.001, and the cosine annealing strategy was used to adjust the learning rate. The model was trained for 500 rounds.

[0090] S203 Validation Model: Use the validation set to evaluate model performance after each epoch, monitor the accuracy and loss of the model, adjust training parameters to avoid overfitting, and obtain the semantic segmentation model with the best performance for continuous casting billets.

[0091] Through steps S1 and S2, the trained continuous casting slab semantic segmentation model is obtained, and then reasoning and application are performed, such as Figure 5 As shown, the following steps are included:

[0092] Step S3:

[0093] like Figure 6 As shown, the two side areas ROIL and ROIR where the continuous casting billet deviates, as well as the center line Line1 of the normal movement of the continuous casting billet are set.

[0094] S301 Delineation of out-of-bounds areas: By analyzing the motion trajectory of the continuous casting billet when it is in normal motion and not deviating, two quadrilateral areas are found on the left and right sides of the trajectory (i.e., the preset areas, such as Figure 6 The left and right areas are ROIL and ROIR respectively, and ROIL and ROIR are considered to be out-of-bounds areas.

[0095] S302 center line delineation: find the straight line of the motion trajectory (i.e., the preset center line, such as Figure 6 The middle line shown in the figure is used as the center line Line1 of the normal movement of the continuous casting billet, and the slope of the preset center line is used as the preset slope.

[0096] Step S4:

[0097] In step S1, the real-time video stream collected by the camera at the continuous casting production site is subjected to strong light suppression processing and then used as the input of steps 5 and 6, that is, the production picture of the target continuous casting billet mentioned above.

[0098] Step S5:

[0099] The continuous casting billet semantic segmentation model is used to perform pixel-level segmentation on the continuous casting billet in the production screen, and to detect whether there is any out-of-bounds situation in real time.

[0100] S501 segments the continuous casting billet: uses the trained continuous casting billet semantic segmentation model to perform high-precision pixel-level segmentation on the continuous casting billet in the input image, obtains the segmented mask, and determines the target area corresponding to the target continuous casting billet.

[0101] S502 Out-of-bounds detection: The segmented continuous casting billet mask (target area) is compared with the ROIL and ROIR. If the continuous casting billet enters the inside of the ROIL or ROIR, or enters the inside of both at the same time, it is considered that the continuous casting billet has deviated.

[0102] Step S6:

[0103] Use traditional computer vision algorithms to find the center line of the continuous casting billet and determine whether it is deviating.

[0104] S601 Line Detection: The collected production images are grayscaled and binarized, and then the Hough transform is used to find the edge lines on the left and right sides of the continuous casting billet in the image (the first edge segment and the second edge segment), which are recorded as L1 and L2.

[0105] S602 Slope Calculation: At a predetermined number of longitudinal sampling points, calculate the horizontal coordinates of lines L1 and L2 at those sampling points, denoted as x1 and x2. The midpoint horizontal coordinate x_c = (x1 + x2) / 2. Perform a linear regression fit on the midpoint coordinates (x_c, y) of all sampling points to generate the target centerline Line0 of the real-time continuous casting slab. Calculate the slope K0 of Line0.

[0106] S603 Slope comparison: Compare the slope K0 of the center line of the continuous casting billet with the slope K1 of the center line Line1 of the normal movement of the continuous casting billet. If the difference between the two is greater than 5° (preset difference), it is considered that deviation has occurred.

[0107] Finally, step S7 is executed: the developed algorithm is deployed to the server at the continuous casting production site, and a real-time video monitoring system is established. When the algorithm determines that deviation has occurred, an audible and visual alarm is triggered to indicate the deviation.

[0108] Step S701: Data access: directly connect the camera at the production site to the on-site monitoring server to obtain real-time data stream data at the site.

[0109] Step S702: Algorithm deployment: deploy the complete continuous casting slab deviation detection algorithm to the on-site monitoring server to form a real-time video monitoring system.

[0110] Step S703: Real-time detection and control: When the algorithm detects deviation, the system automatically triggers an audible and visual alarm and links the on-site PLC to issue a roller stop instruction.

[0111] The angle detection method of the continuous casting billet provided by the embodiment of the present invention has the following effects:

[0112] 1. Construction of a Semantic Segmentation Dataset: By systematically collecting, filtering, and labeling continuous casting billet image data at the continuous casting production site, we developed a high-quality, large-scale semantic segmentation dataset for continuous casting billets. Using this dataset, we trained a semantic segmentation model for continuous casting billets based on YOLOv8mSeg, laying the foundation for subsequent real-time detection.

[0113] 2. Collaborative Application of Deep Learning and Traditional Vision Technologies: Continuous casting slab deviation detection is performed using both the YOLOv8mSeg semantic segmentation model and traditional machine vision. This dual detection mechanism complements the technologies and achieves highly accurate and robust detection. During the detection process, the YOLOv8mSeg model accurately locates the continuous casting slab's position and determines whether it has crossed the boundary. Simultaneously, traditional image processing techniques locate the edges of the continuous casting slab, ensuring accurate and real-time determination of slab deviation. This dual detection approach overcomes the shortcomings of a single method. When relying solely on the semantic segmentation algorithm, narrow slabs, due to their narrow size, will not quickly enter the out-of-bounds ROI even if they slightly deviate. However, traditional vision methods can detect deviation based on their skew angle. When relying solely on traditional vision methods, wide slabs may not have deviated by more than 5°, but their large size may indicate they are about to impact the equipment. In this case, the semantic segmentation-based out-of-bounds detection method can detect deviation.

[0114] 3. Integrated deployment of real-time monitoring and alarm systems: The introduction of an automated and intelligent monitoring system reduces reliance on operators and the risk of human misjudgment, ensuring the continuity and stability of the production line and significantly improving the efficiency and safety of the production process. Real-time monitoring and audio and visual alarm functions enhance the safety and timeliness of the production process and ensure the stable operation of the continuous production line. The developed detection algorithm was deployed on a server at the continuous casting production site, and a real-time video monitoring system was established. When deviation is detected, the system automatically triggers audio and visual alarms and simultaneously activates the on-site PLC to issue a roller stop instruction, effectively ensuring production safety and reducing potential safety hazards and the risk of production interruptions.

[0115] Figure 7 This is a structural block diagram of an angle detection device for continuous casting provided by an embodiment of the present invention. Figure 7 As shown, the angle detection device for a continuous casting billet provided by an embodiment of the present invention includes an acquisition module 710 , an input module 720 , an identification module 730 and a judgment module 740 .

[0116] An acquisition module 710 is used to acquire a production picture of a target continuous casting billet;

[0117] An input module 720 is configured to input the production image into a trained continuous casting slab semantic segmentation model for segmentation processing to determine a target area corresponding to the target continuous casting slab;

[0118] The recognition module 730 is used to recognize the production picture through an image recognition algorithm to determine the target center line of the target continuous casting billet;

[0119] A judgment module 740 is configured to determine that the target continuous casting slab is skewed if there is an overlap between the target area and the preset area, and / or if the difference between the slope of the target centerline and the preset slope is greater than a preset difference;

[0120] The judgment module 740 is further used to determine that the angle of the target continuous casting billet is not skewed if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference.

[0121] It should be noted that the embodiment of the angle detection device for continuous casting billets in this specification and the embodiment of the angle detection method for continuous casting billets in this specification are based on the same inventive concept. Therefore, the specific implementation method of this embodiment can refer to the corresponding embodiment of the angle detection method for continuous casting billets in the previous text, and the repeated parts will not be repeated.

[0122] Figure 8 A structural block diagram of an electronic device provided for the implementation of the present invention. Figure 8 As shown, the electronic device provided by the embodiment of the present invention includes a processor 810 and a memory 820, wherein the memory is used to store instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the above method.

[0123] In an exemplary embodiment, the electronic device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above methods.

[0124] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions, such as a memory including instructions, is also provided. The instructions are executable by a processor of a device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, or an optical data storage device. When the instructions in the non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the above-described method.

[0125] The present invention also provides a computer program product, comprising a computer program, which implements the above method when executed by a processor.

[0126] While the above description does not provide detailed technical details regarding the patterning of each layer, those skilled in the art will appreciate that various technical means can be employed to form layers, regions, and the like in desired shapes. Furthermore, those skilled in the art may devise methods that differ from those described above to achieve the same structure. Furthermore, while each embodiment has been described separately, this does not mean that the measures in each embodiment cannot be advantageously combined.

[0127] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0128] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for detecting the angle of a continuous casting billet, characterized in that: The method comprises: Acquire the production picture of the target continuous casting billet; Inputting the production image into a trained continuous casting billet semantic segmentation model for segmentation processing to determine a target area corresponding to the target continuous casting billet; Identifying the production picture by an image recognition algorithm to determine the target center line of the target continuous casting billet; If there is an overlapping area between the target area and the preset area, and / or the difference between the slope of the target center line and the preset slope is greater than the preset difference, determining that the target continuous casting billet is angularly deflected; If the target area and the preset area do not overlap, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference, it is determined that the angle of the target continuous casting billet is not skewed.

2. The method according to claim 1, characterized in that The target continuous casting billet runs on a target roller table, and the production screen is identified by an image recognition algorithm to determine a target center line of the target continuous casting billet, including: Recognizing and processing the production image using the image recognition algorithm to obtain a first edge line segment and a second edge line segment, wherein the first edge line segment and the second edge line segment both represent an edge of the target continuous casting billet in the running direction of the target roller; Obtaining coordinates of N first target points, where the first target point is any point on the first edge segment; According to the coordinates of the N first target points, the coordinates of the N second target points on the second edge segment are obtained, and the coordinates of the second target points in the running direction perpendicular to the target roller are the same as those of the first target points. Determining coordinates of N center points based on the N first target points and the N second target points; The line connecting the N center points is used as the target center line.

3. The method according to claim 1, characterized in that The preset area is a quadrilateral area adjacent to the left and right of the motion trajectory of the continuous casting billet when it is not deviated.

4. The method according to claim 1, wherein The preset slope is the slope of a preset center line, and the preset center line represents the movement trajectory of the continuous casting billet when the angle is not deflected.

5. The method according to claim 1, wherein The trained continuous casting slab semantic segmentation model is trained by the following steps: Acquiring training data, wherein the training data includes production images of continuous casting billets under different conditions; The original model is trained using the training data to obtain a trained continuous casting slab semantic segmentation model.

6. The method according to claim 1, characterized in that The method further comprises: When it is determined that the angle of the target continuous casting billet is deviated, an alarm signal is generated.

7. An angle detection device for continuous casting billet, characterized in that: The device comprises: An acquisition module is used to acquire the production picture of the target continuous casting billet; An input module is used to input the production picture into a trained continuous casting billet semantic segmentation model for segmentation processing to determine the target area corresponding to the target continuous casting billet; an identification module, configured to identify the production image through an image recognition algorithm and determine a target center line of the target continuous casting billet; a judgment module, configured to determine that the target continuous casting billet is skewed in angle if there is an overlapping area between the target area and the preset area, and / or if the difference between the slope of the target centerline and the preset slope is greater than a preset difference; The judgment module is also used to determine that the angle of the target continuous casting billet is not skewed if there is no overlapping area between the target area and the preset area, and the difference between the slope of the target center line and the preset slope is not greater than the preset difference.

8. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 6. 9 . A storage medium, wherein when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method according to claim 1 .

10. A computer program product comprising a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 6 by a processor.

Images