Converter automatic slag splashing control system and method based on image processing

Abstract

The invention relates to a converter automatic slag splashing control system based on image processing. The converter automatic slag splashing control system comprises a mode and material quantity setting module, a slag splashing execution control module, an image collecting and processing module and an intelligent decision module. The mode and material quantity setting module is used for setting the slag splashing mode and the slag splashing material quantity of the furnace according to the production condition; the slag splashing execution control module is used for dynamically adjusting a gun for slag splashing according to the input of the slag splashing mode and slag splashing material dosage setting module; the image acquisition and processing module is used for acquiring real-time videos of the converter kennel and training a target detection model for distinguishing image states. The invention further relates to a method using the automatic converter slag splashing control system based on image processing. The method solves the problems that a traditional slag splashing control method depends on manual operation, and large errors and risks exist; and the automation level and the safety of the slag splashing process are improved.

Description

Technical Field

[0001] This invention relates to the technical field of slag splashing protection in converters during iron and steel smelting, and specifically to an automatic slag splashing control system and method for converters based on image processing. Background Technology

[0002] In converter steelmaking, slag splashing is a crucial process for furnace protection. High-pressure nitrogen is injected into the converter to splash slag onto the inner wall, forming a protective slag layer and preventing premature damage to the furnace lining. However, traditional slag splashing control methods rely on manual operation, which carries significant errors and risks. This invention introduces image processing technology to automatically detect critical moments in the slag splashing process. When the camera captures video of the converter's "doghouse" (referring to the furnace opening), the system analyzes the video image in real time, identifying the furnace opening condition. If there is severe slag discharge at the furnace opening, the system raises the slag splashing lance to prevent splashing. The precise moment of slag drying is also identified. At this point, the system automatically issues a command to raise the oxygen lance and shut off the nitrogen supply, ensuring the accuracy and safety of the slag splashing process.

[0003] Chinese patent application CN118028565A, entitled "A Method, Device, Electronic Equipment, and Storage Medium for Automatic Slag Splashing Gun Position Control in a Converter," discloses a method that includes: collecting relevant information from the current and historical furnace runs and establishing a historical furnace run database; performing analysis and segmentation to form a secondary database; simulating the slag splashing process to establish a tertiary database; first, identifying the optimal simulation curve for the tertiary database, then fitting the curve to the secondary database to obtain a library of fitting curves for the slag splashing gun position control height, and finally finding the optimal fitting curve; comparing and correcting the fitting curve and the simulation curve to obtain a corrected slag splashing gun position height control curve; adjusting the corrected slag splashing gun position height control curve in real time based on the audio signal from the audio slag treatment system; and feeding the adjusted information back to the fitting curve library for self-learning. This invention achieves scientific and precise control of the slag splashing gun position, improving the slag splashing effect.

[0004] Chinese patent application CN115896387A, entitled "A Control Method and System for Automatic Slag Splashing in a Converter," discloses a method that includes: automatically determining the slag splashing mode to be adopted in the furnace, specifically including: collecting the liquid level and comparing it with a first set value and a second set value; when the liquid level is less than the first set value, it indicates that the furnace bottom is thinning, and a furnace bottom expansion mode is adopted; when the liquid level is greater than or equal to the first set value and less than or equal to the second set value, it indicates that the furnace bottom is at a normal level, and a furnace bottom maintenance mode is adopted; when the liquid level is greater than the second set value, it indicates that the furnace bottom is thickening, and a furnace bottom lowering mode is adopted; automatically determining the amount of material to be added during slag splashing; receiving slag splashing operation instructions and controlling slag splashing according to the determined slag splashing mode and the amount of material added. This invention automatically provides the amount of material added to facilitate accurate adjustment of the slag splashing material weight, avoids the blindness of manual judgment, improves the slag splashing effect, and enables one-button slag splashing operation, reducing the labor intensity of operators.

[0005] Application number CN115323104A, entitled "A Method for Automatic Slag Splashing Protection of a Converter," discloses the following steps: Step 1: Controlling the amount of slag inside the converter after tapping; Step 2: Adjusting the internal composition of the slag inside the converter; Step 3: Initial slag splashing protection; Step 4: Secondary slag splashing protection, with the following steps: adjusting the oxygen lance height to 1000mm and performing nitrogen blowing; the nitrogen blowing time is controlled to 20s; Step 5: Tertiary slag splashing protection, with the following steps: adjusting the oxygen lance height to 800mm and performing nitrogen blowing; the nitrogen blowing time is controlled to 100s; Step 6: Final slag splashing protection, with the following steps: adjusting the oxygen lance height to 700mm and performing nitrogen blowing; the nitrogen blowing time is controlled to 100s; Step 7: Checking the slag splashing condition of the furnace lining. This automatic slag splashing protection method for converters adopts scientific slag splashing protection technology. Through multiple repeated slag splashing operations, it not only avoids slag leakage on the furnace wall but also ensures uniform slag thickness due to the gradual progression of the process.

[0006] Currently, domestic automatic slag splashing models for converters are mainly based on preset lance positions and corresponding times. Their shortcomings are: This method involves splashing slag according to a predetermined time and gun position, but it is difficult to adapt to real-time changes on-site during the slag splashing process.

[0007] Determining the endpoint of slag splashing based on time can easily lead to furnace bottom rise and lance sticking. Therefore, its applicability is somewhat limited. Summary of the Invention

[0008] To address the aforementioned problems, this invention provides an automatic slag splashing control system and method for converters based on image processing. The purpose is to solve the problems of traditional slag splashing control methods relying on manual operation, resulting in significant errors and risks. By introducing image processing technology, the system can automatically detect key moments in the slag splashing process, dynamically adjust the gun position to prevent splashing, and accurately detect the moment of drying, thereby improving the automation level and safety of the slag splashing process.

[0009] To solve the above problems, the technical solution provided by the present invention is as follows: An image processing-based automatic slag splashing control system for converters includes a mode and feed rate setting module, a slag splashing execution control module, an image acquisition and processing module, and an intelligent decision-making module, wherein: The mode and material quantity setting module is used to set the slag splashing mode and slag splashing material quantity of this furnace according to the production status; the slag splashing mode includes furnace cap maintenance mode, trunnion maintenance mode and furnace bottom maintenance mode; The slag splashing execution control module is used to dynamically adjust the slag splashing gun according to the slag splashing mode and the input of the slag splashing material dosage setting module, and then adjust the slag splashing gun according to the slag splashing time and gun position in different modes through the automatic slag splashing control unit. The image acquisition and processing module is used to acquire real-time video of the converter dog kennel, and obtain an image dataset through preprocessing and labeling. Then, the image dataset is used to train an object detection model to distinguish the image state. The intelligent decision-making module is used to generate corresponding control decision signals based on the recognition results of the target detection model, and feed them back to the automatic slag splashing control unit in real time to execute the corresponding control signals.

[0010] Preferably, the slag splashing flow rate of the slag splashing control module is 72000 Nm³. 3 / h; When the preset amount of slag splashing material is 0, the slag is splashed directly into the gun; when the preset amount of slag splashing material is not 0, the slag splashing material is added at the same time as the gun is lowered.

[0011] Preferably, the corresponding gun position for the slag splashing time in the furnace cap maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1250 when it is [40 seconds to 2 minutes 20 seconds), 1200 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1180 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1180 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing]. The corresponding gun position for the splatter time in the trunnion maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1300 when it is [40 seconds to 2 minutes 20 seconds), 1250 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1200 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1170 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of splatter time]. The corresponding gun position for the slag splashing time in the furnace bottom maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1350 when it is [40 seconds to 2 minutes 20 seconds), 1300 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1250 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1200 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing].

[0012] Preferably, the image acquisition and processing module uses FFMPEG technology to obtain real-time video of the converter slag outlet captured by the camera, and detects the sparks splashed out of the outlet during slag splashing by adjusting the video stream contrast and using image segmentation technology; the image acquisition and processing module uses OpenCV to split the converter slag outlet video stream into frame-level images and construct the image dataset; the labels of the image dataset include "slag splashing dry" and "lots of sparks".

[0013] Preferably, the labeling criteria for the image dataset are as follows: when the number of pixels in the image that meet the high brightness condition is less than the manually preset splashing threshold, the current image is labeled as "splashing"; when the number of pixels in the image that meet the high brightness condition is greater than the manually preset risk threshold, the current image is labeled as "lots of steel flowers".

[0014] Preferably, the high brightness condition is a region in the HSV color space where the value of the brightness component is in the range of 80% to 95%; the splatter threshold is 0.1% to 0.5% of the total pixels of the image; and the risk threshold is 3% to 8% of the total pixels of the image.

[0015] Preferably, the target detection model is trained using a deep learning model based on the YOLO architecture, and the training method includes the following steps: Sa1. Data Acquisition and Labeling: Collect image data of the converter slag splashing process, label the slag splashing area, spray gun position and furnace mouth status in the image with bounding boxes, and label the image data according to the labeling judgment conditions of the image dataset to obtain the training sample set; Sa2. Data Augmentation: Data augmentation processing of the training sample set, including random rotation, brightness adjustment, noise addition, and scaling transformation; Sa3. Model Construction: Using the anchor box mechanism and feature pyramid network structure of the YOLO model, a neural network suitable for multi-scale slag splash target detection is constructed; Sa4. Loss function design: CIoU loss function is used to optimize bounding box regression accuracy, and classification loss and confidence loss are combined for multi-task joint training; Sa5. Training optimization: Iterative training is performed using the Adam optimizer, the learning rate is dynamically adjusted using the cosine annealing algorithm, and an early stopping strategy is used on the validation set to prevent overfitting; Sa6. Model Deployment: Convert the trained model weights into a format that the model can call and integrate them into the image acquisition and processing module of the converter control system.

[0016] A method utilizing the aforementioned image processing-based automatic slag splashing control system for converters includes the following steps: S100. Set the slag splashing mode and the amount of slag splashing material for this furnace according to the production status; S200. With the slag splashing flow rate fixed at 72000 Nm 3 Under the condition of / h, when the set amount of splashing slag is 0, the slag is splashed directly by dropping the gun; when the set amount of splashing slag is non-zero, splashing slag is added at the same time as dropping the gun; and the gun position and time relationship in the preset mode are dynamically adjusted for splashing slag. S300. Acquire real-time video of the converter doghouse and distinguish the furnace opening from the background by brightness difference; acquire images through the image acquisition and processing module and label them to obtain the image dataset; train the target detection model based on the labels of the image dataset; S400. During the slag splashing process, if the target detection model identifies "excessive steel spatter", the height of the slag splashing oxygen lance is increased to avoid splashing; if the target detection model identifies "dry splashing", the oxygen lance is increased and the nitrogen supply is turned off to end the slag splashing.

[0017] Preferably, in step S300, the monitoring video stream of the doghouse is obtained through OpenCV, and the video stream is parsed into an RGB color gamut image; then, the RGB color gamut image is converted into an HSV color gamut image using the RGB to HSV algorithm in OpenCV; the RGB to HSV algorithm is expressed by the following formula: Wherein: R0, G0, B0 are used to represent the original values ​​of the RGB color gamut image; R, G, B are used to represent the values ​​of the RGB color gamut image after conversion to the [0, 1] interval; H is used to represent hue; S is used to represent saturation; V is used to represent brightness.

[0018] Preferably, the threshold value for determining steel flower pixels in the V channel of the HSV color gamut image is 245; when the pixel value of a pixel in the V channel is greater than 245, the current pixel is determined to be a steel flower pixel.

[0019] Compared with the prior art, the present invention has the following advantages: 1. This invention utilizes image processing technology to control the converter slag splashing process in real time, automatically adapting to complex working conditions, improving slag splashing efficiency and quality, and reducing manual intervention.

[0020] 2. This invention can detect the condition of the converter's large furnace opening in real time and accurately, and dynamically adjust the oxygen lance height to ensure the quality of slag splashing.

[0021] 3. This invention can improve detection efficiency by automating the processing of video information, thereby reducing the time and effort required for manual observation.

[0022] 4. This invention can improve detection accuracy. The detection method based on image processing technology can accurately identify the state of the converter's large furnace opening and avoid interference from human factors.

[0023] 5. This invention has broad application prospects in the field of iron and steel smelting and can significantly improve the efficiency and quality of converter slag splashing. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of a state where the number of steel flower pixels exceeds a set threshold, according to a specific embodiment of the present invention. Figure 2 This is a schematic diagram of a specific embodiment of the present invention where the number of steel flower pixels is less than a set threshold (splash drying). Figure 3 This is a schematic diagram of the process principle of a specific embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0026] This invention application claims a converter automatic slag splashing control system based on image processing, comprising a mode and material quantity setting module, a slag splashing execution control module, an image acquisition and processing module, and an intelligent decision-making module, wherein: The mode and material quantity setting module is used to set the slag splashing mode and slag material quantity of this furnace according to the production conditions; the mode and material quantity setting module allows manual setting of the slag splashing mode and slag material quantity of this furnace according to the production conditions; the slag splashing modes include furnace cap maintenance mode, trunnion maintenance mode and furnace bottom maintenance mode; The slag splashing execution control module is used to set the input of the slag splashing mode and slag splashing material amount according to the slag splashing mode and slag splashing material amount, and then dynamically adjust the slag splashing gun through the automatic slag splashing control unit according to the slag splashing time and gun position in different modes. The image acquisition and processing module is used to acquire real-time video of the converter dog kennel, and obtain an image dataset through preprocessing and labeling. Then, the image dataset is used to train an object detection model to distinguish the image state. The intelligent decision-making module is used to generate corresponding control decision signals based on the recognition results of the target detection model, and feeds them back to the automatic slag splashing control unit in real time to execute the corresponding control signals.

[0027] In this specific embodiment, the slag splashing flow rate of the slag splashing execution control module is 72000 Nm. 3 / h; When the preset amount of slag splashing material is 0, the slag is splashed directly into the gun; when the preset amount of slag splashing material is not 0, the slag splashing material is added at the same time as the gun is lowered.

[0028] In this specific embodiment, the corresponding gun position for the slag splashing time in the furnace cap maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1250 when it is [40 seconds to 2 minutes 20 seconds), 1200 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1180 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1180 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing]. The corresponding gun position for the splatter time in the trunnion maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1300 when it is [40 seconds to 2 minutes 20 seconds), 1250 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1200 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1170 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of splatter time]. The corresponding gun position for the slag splashing time in the furnace bottom maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1350 when it is [40 seconds to 2 minutes 20 seconds), 1300 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1250 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1200 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing].

[0029] It should be noted that the image acquisition and processing module uses FFMPEG technology to obtain real-time video of the converter slag outlet captured by the camera. It also uses video stream contrast adjustment and image segmentation technology to detect the sparks splashing out of the outlet during slag splashing. Image processing technology is used to reduce noise and highlight the sparks. The image acquisition and processing module uses OpenCV to split the converter slag outlet video stream into frame-level images and construct an image dataset. The image dataset is formed by extracting images at regular intervals and annotating the images. The labels of the image dataset include "slag splash dry" and "lots of sparks".

[0030] It should be further noted that in this embodiment, OpenCV is a cross-platform computer vision and machine learning software library released under the Apache 2.0 license (open source), which can run on Linux, Windows, Android and Mac OS operating systems.

[0031] It needs to be further explained that, such as Figure 1 , Figure 2 As shown, the labeling criteria for the image dataset are as follows: when the number of pixels in the image that meet the high brightness condition is less than the manually preset splashing threshold, the furnace opening brightness is low and the steel sparks are sparse, and the current image is labeled as "splashing"; when the number of pixels in the image that meet the high brightness condition is greater than the manually preset risk threshold, the steel sparks are active and there is a risk of splashing, and the current image is labeled as "lots of steel sparks".

[0032] In this specific embodiment, the high brightness condition refers to the region where the brightness component in the HSV color space ranges from 80% to 95%, preferably ≥85%; the splatter threshold is 0.1% to 0.5% of the total image pixels; and the risk threshold is 3% to 8% of the total image pixels. High-brightness pixels represent high-temperature steel splinters, and their quantity directly reflects the severity of splattering. The specific threshold values ​​are determined through statistical analysis of historical data from the field and are manually calibrated and optimized during the system debugging phase.

[0033] It should be noted that the object detection model is trained using a deep learning model based on the YOLO architecture, and its training method includes the following steps: Sa1. Data Acquisition and Labeling: Collect image data of the converter slag splashing process, label the slag splashing area, spray gun position and furnace mouth status in the image with bounding boxes, and label the image data according to the label labeling judgment conditions of the image dataset to obtain the training sample set; Sa2. Data Augmentation: Data augmentation processing of the training sample set, including random rotation, brightness adjustment, noise addition, and scaling transformation; Sa3. Model Construction: Using the anchor box mechanism and feature pyramid network structure of the YOLO model, a neural network suitable for multi-scale slag splash target detection is constructed; Sa4. Loss function design: CIoU loss function is used to optimize bounding box regression accuracy, and classification loss and confidence loss are combined for multi-task joint training; Sa5. Training optimization: Iterative training is performed using the Adam optimizer, the learning rate is dynamically adjusted using the cosine annealing algorithm, and an early stopping strategy is used on the validation set to prevent overfitting; Sa6. Model Deployment: Convert the trained model weights into a format that the model can call and integrate them into the image acquisition and processing module of the converter control system.

[0034] A method for an automatic slag splashing control system for converters based on image processing, such as... Figure 3 As shown, it includes the following steps: S100. Set the slag splashing mode and slag material usage for this furnace according to the production status; S200. With the slag splashing flow rate fixed at 72000 Nm 3 Under the condition of / h, when the set amount of splashing material is 0, the slag is directly splashed by dropping the gun; when the set amount of splashing material is non-zero, the splashing material is added at the same time as dropping the gun; and the gun position and time relationship in the preset mode are dynamically adjusted for splashing. S300. Acquire real-time video of the converter doghouse and distinguish the furnace opening from the background by brightness difference; acquire images through the image acquisition and processing module and label them to obtain an image dataset; train a target detection model based on the labels of the image dataset; S400. During the slag splashing process, if the target detection model identifies "excessive steel spatter," the height of the oxygen lance is increased to avoid splashing; if the target detection model identifies "dry splashing," the oxygen lance is increased and the nitrogen supply is turned off to end the slag splashing.

[0035] It should be noted that in step S300, the monitoring video stream of the doghouse is obtained through OpenCV, and the video stream is parsed into an RGB color gamut image; then, the RGB color gamut image is converted into an HSV color gamut image using the RGB to HSV algorithm in OpenCV; the RGB to HSV algorithm is expressed as Equation 1: (1) Wherein: R0, G0, B0 are used to represent the original values ​​of the RGB color gamut image; R, G, B are used to represent the values ​​of the RGB color gamut image after conversion to the [0, 1] interval; H is used to represent hue; S is used to represent saturation; V is used to represent brightness.

[0036] In this specific embodiment, image processing techniques are used to reduce image noise and highlight the characteristics of splashed steel sparks. By comparing the brightness values ​​of numerous previous video images of slag splashing processes, the threshold for identifying steel spark pixels in the V channel of the HSV color gamut image is determined to be 245. This threshold indicates that the image is bright enough to clearly distinguish the steel sparks from the background. The V channel matrix of the detection area in the HSV color gamut image is extracted. When the pixel value of a pixel in the V channel is greater than 245, the current pixel is determined to be a steel spark pixel. The number of these molten steel pixels is recorded. In the HSV color gamut image, the V channel represents the brightness of the image. When steel sparks appear, there will be a noticeable brightening area on the screen, hence its larger V value.

[0037] It should be further explained that the trained target detection model is used to distinguish between two situations: excessive slag and dry splashing. After confirming that the target detection model meets the accuracy requirements, real-time data is imported into the model. Based on the trained target detection model, it is determined whether the phenomena of excessive slag and dry splashing occur. The comparison result of the number of slag pixels with the set threshold is used as the input feature, and the output control command controls the oxygen lance to rise or stop the automatic slag splashing.

[0038] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.

[0039] The disclosed embodiments have been described above to enable any person skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments without departing from the spirit and scope of this disclosure. Therefore, this disclosure is not limited to the embodiments given herein, but is consistent with the broadest scope of the principles and novel features disclosed in this application.

[0040] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."

[0041] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic slag splashing control system for converters based on image processing, characterized in that: It includes a mode and material quantity setting module, a slag splashing execution control module, an image acquisition and processing module, and an intelligent decision-making module, among which: The mode and material quantity setting module is used to set the slag splashing mode and slag splashing material quantity of this furnace according to the production status; the slag splashing mode includes furnace cap maintenance mode, trunnion maintenance mode and furnace bottom maintenance mode; The slag splashing execution control module is used to dynamically adjust the slag splashing gun according to the slag splashing mode and the input of the slag splashing material dosage setting module, and then adjust the slag splashing gun according to the slag splashing time and gun position in different modes through the automatic slag splashing control unit. The image acquisition and processing module is used to acquire real-time video of the converter dog kennel, and obtain an image dataset through preprocessing and labeling. Then, the image dataset is used to train an object detection model to distinguish the image state. The intelligent decision-making module is used to generate corresponding control decision signals based on the recognition results of the target detection model, and feed them back to the automatic slag splashing control unit in real time to execute the corresponding control signals.

2. The automatic slag splashing control system for converters based on image processing according to claim 1, characterized in that: The slag splashing flow rate of the slag splashing control module is 72000 Nm. 3 / h; When the preset amount of slag splashing material is 0, the slag is splashed directly into the gun; when the preset amount of slag splashing material is not 0, the slag splashing material is added at the same time as the gun is lowered.

3. The automatic slag splashing control system for converters based on image processing according to claim 2, characterized in that: The corresponding gun position for the slag splashing time in the furnace cap maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1250 when it is [40 seconds to 2 minutes 20 seconds), 1200 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1180 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1180 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing]. The corresponding gun position for the splatter time in the trunnion maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1300 when it is [40 seconds to 2 minutes 20 seconds), 1250 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1200 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1170 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of splatter time]. The corresponding gun position for the slag splashing time in the furnace bottom maintenance mode is 1180 when it is [0 seconds to 40 seconds), 1350 when it is [40 seconds to 2 minutes 20 seconds), 1300 when it is [2 minutes 20 seconds to 2 minutes 30 seconds], 1250 when it is [2 minutes 30 seconds to 2 minutes 40 seconds], 1200 when it is [2 minutes 40 seconds to 2 minutes 50 seconds], and 1180 when it is [2 minutes 50 seconds to the end of slag splashing].

4. The automatic slag splashing control system for converters based on image processing according to claim 3, characterized in that: The image acquisition and processing module uses FFMPEG technology to obtain real-time video of the converter slag inlet captured by the camera, and detects the sparks splashing out of the slag inlet by adjusting the video stream contrast and image segmentation technology. The image acquisition and processing module uses OpenCV to split the converter slag inlet video stream into frame-level images and construct the image dataset. The image dataset is labeled with "slag splash" and "lots of sparks".

5. The automatic slag splashing control system for converters based on image processing according to claim 4, characterized in that: The labeling criteria for the image dataset are as follows: when the number of pixels in the image that meet the high brightness condition is less than the manually preset splashing threshold, the current image is labeled as "splashing"; when the number of pixels in the image that meet the high brightness condition is greater than the manually preset risk threshold, the current image is labeled as "lots of steel flowers".

6. The automatic slag splashing control system for converters based on image processing according to claim 5, characterized in that: The high brightness condition is the region in the HSV color space where the value of the brightness component is in the range of 80% to 95%; the splatter threshold is 0.1% to 0.5% of the total pixels of the image; and the risk threshold is 3% to 8% of the total pixels of the image.

7. The automatic slag splashing control system for converters based on image processing according to claim 6, characterized in that: The object detection model is trained using a deep learning model based on the YOLO architecture, and its training method includes the following steps: Sa1. Data Acquisition and Labeling: Collect image data of the converter slag splashing process, label the slag splashing area, spray gun position and furnace mouth status in the image with bounding boxes, and label the image data according to the labeling judgment conditions of the image dataset to obtain the training sample set; Sa2. Data Augmentation: Data augmentation processing of the training sample set, including random rotation, brightness adjustment, noise addition, and scaling transformation; Sa3. Model Construction: Using the anchor box mechanism and feature pyramid network structure of the YOLO model, a neural network suitable for multi-scale slag splash target detection is constructed; Sa4. Loss function design: CIoU loss function is used to optimize bounding box regression accuracy, and classification loss and confidence loss are combined for multi-task joint training; Sa5. Training optimization: Iterative training is performed using the Adam optimizer, the learning rate is dynamically adjusted using the cosine annealing algorithm, and an early stopping strategy is used on the validation set to prevent overfitting; Sa6. Model Deployment: Convert the trained model weights into a format that the model can call and integrate them into the image acquisition and processing module of the converter control system.

8. A method for an automatic slag splashing control system for a converter based on image processing as described in any one of claims 1 to 7, characterized in that: Includes the following steps: S100. Set the slag splashing mode and the amount of slag splashing material for this furnace according to the production status; S200. With the slag splashing flow rate fixed at 72000 Nm 3 Under the condition of / h, when the set amount of splashing slag is 0, the slag is splashed directly by dropping the gun; when the set amount of splashing slag is non-zero, splashing slag is added at the same time as dropping the gun; and the gun position and time relationship in the preset mode are dynamically adjusted for splashing slag. S300. Acquire real-time video of the converter doghouse and distinguish the furnace opening from the background by brightness difference; acquire images through the image acquisition and processing module and label them to obtain the image dataset; train the target detection model based on the labels of the image dataset; S400. During the slag splashing process, if the target detection model identifies "many steel sputtering" as the result, the height of the slag splashing oxygen lance is increased to avoid splashing. If the target detection model identifies "splashing dry", then raise the oxygen lance and shut off the nitrogen supply to end the slag splashing.

9. The method according to claim 8, characterized in that: In step S300, the monitoring video stream of the doghouse is obtained through OpenCV, and the video stream is parsed into an RGB color gamut image; then, the RGB color gamut image is converted into an HSV color gamut image using the RGB to HSV algorithm in OpenCV; the RGB to HSV algorithm is expressed by the following formula: Wherein: R0, G0, B0 are used to represent the original values ​​of the RGB color gamut image; R, G, B are used to represent the values ​​of the RGB color gamut image after conversion to the [0, 1] interval; H is used to represent hue; S is used to represent saturation; V is used to represent brightness.

10. The method according to claim 9, characterized in that: The threshold for determining steel flower pixels in the V channel of the HSV color gamut image is 245; when the pixel value of a pixel in the V channel is greater than 245, the current pixel is determined to be a steel flower pixel.

Images