Slagging-off path planning method based on iron flow liquid level form recognition

Abstract

The invention relates to the technical field of automatic slagging-off of molten iron in the metallurgical industry, in particular to a slagging-off path planning method based on molten iron liquid level form recognition, which comprises the following steps: preprocessing a molten iron tank opening image and marking an internal area of a molten iron tank; the tank opening area of the hot-metal tank is extracted; identifying the form of the liquid level of the molten iron; molten iron and slag distribution conditions of the molten iron liquid level of the molten iron tank are obtained; index parameters of the slag areas are calculated, and slagging-off areas are divided; and the slagging-off path is planned, and the optimal slagging-off path is selected. The method has the advantages that the tank opening of the molten iron tank is recognized through computer vision, recognition of molten iron and slag is not interfered by other external environments, the molten iron and the slag are distinguished according to different imaging modes of the molten iron and the slag and different brightness and contrast ratios, and the distribution condition of the molten iron and the slag on the liquid level of the drossing molten iron is achieved; and a slag raking path is calculated and updated in real time according to the identified molten iron and slag distribution conditions, and a foundation is laid for automatic slag raking.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic slag skimming for hot metal in the metallurgical industry, and particularly to a slag skimming path planning method based on the recognition of the molten iron surface morphology. Background Art

[0002] The recognition of the molten iron surface morphology during slag skimming and the planning of the slag skimming path are the basis for realizing automatic slag skimming. The accuracy of the recognition of the molten iron surface morphology directly affects the subsequent planning of the slag skimming path, and thus affects the slag skimming efficiency. The traditional method is that the desulfurization worker observes with the naked eye and operates the robotic arm for slag skimming. Due to the harsh working environment, unstable factors such as strong light, thick smoke, and the worker's attention are likely to lead to misjudgment.

[0003] In the prior art, the publication number is CN117428183A, and the patent name is "Slag Skimming Control Method, System, Equipment and Medium Based on Image Recognition". "The method includes: obtaining the temperature information of the slag to be skimmed located on the surface of the metal fluid; determining the position information of the slag to be skimmed based on the fluid surface image obtained for the metal fluid; generating a slag skimming control signal based on the position information of the slag to be skimmed and the temperature information of the slag to be skimmed, where the slag skimming control signal is used to indicate skimming the slag to be skimmed from the metal fluid". Although this method combines the position information and the temperature information to generate the slag skimming control signal, the generation of the control signal depends on preset rules or algorithms, lacks adaptability, and may cause the slag skimming control signal to not accurately reflect the actual situation, thereby affecting the slag skimming effect. Summary of the Invention

[0004] The purpose of the present invention is to provide a slag skimming path planning method based on the recognition of the molten iron surface morphology, and solve the problems of potential safety hazards and low efficiency caused by the traditional method of observing with the naked eye and operating the robotic arm for slag skimming.

[0005] To achieve the above purpose, the present invention is realized through the following technical solutions:

[0006] A slag skimming path planning method based on the recognition of the molten iron surface morphology, including

[0007] S1. Obtain the image of the ladle mouth, preprocess the image of the ladle mouth and label the internal area of the ladle;

[0008] S2. Extract the ladle mouth area according to the ladle mouth instance segmentation model;

[0009] S3. Recognize the molten iron surface morphology according to the extracted ladle mouth image;

[0010] S4. Distinguish the molten iron and slag on the molten iron surface according to the brightness and contrast of the molten iron and slag on the molten iron surface, and obtain the distribution of the molten iron and slag on the molten iron surface of the ladle;

[0011] S5. Calculate the index parameters of the slag region according to the distribution of hot metal and slag, and divide the slag skimming region.

[0012] S6. Plan the slag skimming path according to the index parameters of the identified slag region, and select the optimal slag skimming path.

[0013] In S1, the preprocessing of the hot metal ladle mouth image is to adjust the contrast and brightness of the hot metal ladle mouth image and remove the noise of the hot metal ladle mouth image; the annotation of the hot metal ladle mouth image is to use labelimg to annotate the hot metal ladle mouth image, and frame the hot metal ladle mouth area with a square box.

[0014] In S2, the hot metal ladle mouth instance segmentation model includes:

[0015] S201. The steps of constructing the hot metal ladle mouth instance segmentation model include:

[0016] 1) Extract the features of the hot metal ladle mouth image.

[0017] 2) Generate candidate target boxes through the RPN network.

[0018] 3) Identify the hot metal ladle mouth.

[0019] 4) Generate the bounding box of the hot metal ladle mouth through bounding box regression for wrapping the target.

[0020] S202. Divide the training set, test set and validation set according to the hot metal ladle mouth image annotated in S1, and train the hot metal ladle mouth instance segmentation model.

[0021] S203. Test and validate the trained hot metal ladle mouth instance segmentation model according to the test set and validation set divided in S202.

[0022] S204. Analyze the recognition metrics of the hot metal ladle mouth instance segmentation model, adjust the parameters of the hot metal ladle mouth instance segmentation model, and optimize the hot metal ladle mouth instance segmentation model.

[0023] In S202, when training the hot metal ladle mouth instance segmentation model, the previously annotated hot metal ladle mouth images are input into the constructed hot metal ladle mouth instance segmentation model, set the relevant training parameters such as batch size, learning rate, and epochs, and select the loss function to train the hot metal ladle mouth instance segmentation model.

[0024] In S203, the trained ladle mouth instance segmentation model is tested and verified. Before annotation, all datasets are divided into a training set, a validation set, and a test set according to a ratio. The validation set is used to verify the instance segmentation model to adjust the hyperparameters of the instance segmentation model and find the best configuration of the ladle mouth instance segmentation model; the test set is used to test the actual performance of the instance segmentation model.

[0025] In S204, the recognition metrics of the ladle mouth instance segmentation model are analyzed. The accuracy of the ladle mouth instance segmentation model is judged according to the intersection over union of the ladle mouth instance segmentation area and the manually annotated area and the accuracy of ladle mouth recognition.

[0026] Optimize the ladle mouth instance segmentation model, including hyperparameters related to the model structure and loss function. Adjust the pixel values of the input image, the number of model training times, the number of batches, and the step size according to the accuracy and speed of model training to ensure the quality of the model.

[0027] In S3, the morphology of the molten iron surface is recognized. According to the inner area of the ladle segmented in S2, the inner area and the outer area of the ladle are distinguished, and based on the molten iron surface in the distinguished inner area of the ladle, the molten iron and slag are distinguished using OpenCV image recognition technology.

[0028] In S5, the index parameters of the slag area are calculated. The index parameters include area, center point coordinates, and edge point coordinates. The formulas are as follows:

[0029]

[0030] In formula ①, Gx represents the horizontal gradient; Gy represents the vertical gradient.

[0031] M00 = ∑i∑jI(i, j) ②

[0032] M10 = ∑i∑jj·I(i, j) ③

[0033] M01 = ∑i∑ii·I(i, j) ④

[0034] In formulas ② - ④, the zero - order moment (M00) represents the area or total mass of the region; the first - order moments (M10 and M01) represent the weighted sums of the region pixel values about the x - axis and y - axis respectively; I(i, j) represents the pixel value of the image at coordinate (i, j).

[0035] cX = M10 / M00 ⑤

[0036] cY = M01 / M00 ⑥

[0037] In Formula ⑤ and Formula ⑥, cX represents the x coordinate of the center point of the region, which is the sum of the pixel values weighted by the y coordinate divided by the total number of pixels;

[0038] cY represents the y coordinate of the center point of the region, which is the sum of the pixel values weighted by the x coordinate divided by the total number of pixels;

[0039] The area of the slag region is determined according to the number of pixels in the slag region. The slag region is converted into a binary image, and then the number of pixels with a value of 1 is calculated;

[0040] Similarly, the center point coordinates of the slag region are the average values of the pixels in the slag region;

[0041] The slag removal area is divided as follows: According to the distribution of molten iron and slag in the molten iron surface of the ladle in S4, the molten iron region and the slag region are divided. The area and boundary of the slag region are calculated and sorted, and the priority of the slag region is set;

[0042] The area and boundary of the slag region are calculated and photographed, and the calculation formula is as follows:

[0043]

[0044] In Formula ⑦, ∮L represents the line integral along the closed curve L, and ∫∫D represents the double integral over the closed region D;

[0045] The priority of the slag region is to divide the slag region into three directions: 45 degrees to the left, the middle, and 45 degrees to the right. According to the number and area of the slag in each region, the priorities of the three directions are determined.

[0046] In S6, the slag removal path is planned as follows: According to the index parameters of the slag region calculated in S5, the path planning is called to calculate multiple planned slag removal paths, and the characteristic values of each slag removal path are calculated. The slag removal path with the highest priority is selected as the slag removal path for this time, which specifically includes:

[0047] S601: According to the identified slag region, by integrating and calling the Dijkstra algorithm and the GA algorithm, multiple slag removal paths are calculated, and the formula is as follows:

[0048] dist[v] = min(dist[v], dist[u] + weight(u, v)) ⑧

[0049] In Formula ⑧, dist[v] represents the current shortest distance from the source vertex s to vertex v; dist[u] represents the current shortest distance from the source vertex s to vertex u; weight(u, v) represents the weight of the edge from vertex u to vertex v;

[0050] S602: Calculate the eigenvalue and weight of each slag removal path based on the distance between adjacent slag regions, the deviation of the center points of adjacent slags, and the boundary of the slag region calculated in S5;

[0051] Calculate the slag removal path according to the area and center point coordinates of the slag region. The formula is as follows:

[0052]

[0053] In formula ⑨, xi represents the value of the i-th data point, wi represents the weight associated with the i-th data point, and n represents the total number of data points;

[0054] S603: Select the optimal slag removal path as the path for this slag removal.

[0055] A computer device includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a slag removal path planning method based on the recognition of the molten iron surface morphology.

[0056] A computer-readable storage medium stores computer instructions for causing a computer to execute a slag removal path planning method based on the recognition of the molten iron surface morphology.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. Identify the ladle mouth of the hot metal ladle through computer vision to ensure that the recognition of hot metal and slag is not interfered by other external environments. According to the different imaging methods of hot metal and slag, distinguish hot metal and slag according to different brightness and contrast, realize the distribution of hot metal and slag on the molten iron surface during slag removal, and calculate and update the slag removal path in real time according to the recognized distribution of hot metal and slag, laying a foundation for realizing automatic slag removal;

[0059] 2. Update the slag removal path in real time according to the dynamic changes of the molten iron surface morphology of the hot metal ladle;

[0060] 3. Can well distinguish the hot metal region and the slag region in the internal area of the hot metal ladle, ensuring the accuracy of slag removal. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a flowchart of a slag removal path planning method based on the recognition of the molten iron surface morphology. DETAILED DESCRIPTION OF THE INVENTION

[0062] The present invention will be described in detail below with reference to the accompanying drawings of the specification, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0063] The following embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given. However, the protection scope of the present invention is not limited to the following embodiments. The methods used in the following embodiments are all conventional methods unless otherwise specified.

[0064] Embodiment 1

[0065] See Figure 1 , a slag skimming path planning method based on the recognition of the molten iron liquid surface morphology, including:

[0066] S1. Obtain the image of the molten iron ladle mouth through a camera to identify the molten iron liquid surface morphology and analyze the change of the molten iron liquid surface before and after slag skimming, and preprocess the image of the molten iron ladle mouth and label the internal area of the molten iron ladle;

[0067] Preprocess the image of the molten iron ladle mouth: To ensure the clarity of the molten iron ladle mouth and lay a foundation for subsequent recognition of the molten iron ladle mouth, adjust the contrast and brightness of the image of the molten iron ladle mouth and remove the noise in the image of the molten iron ladle mouth;

[0068] Label the image of the molten iron ladle mouth: Use labelimg to label the image of the molten iron ladle mouth, and frame the area of the molten iron ladle mouth with a square frame.

[0069] S2. Extract the area of the molten iron ladle mouth according to the instance segmentation model of the molten iron ladle mouth;

[0070] The instance segmentation model of the molten iron ladle mouth includes:

[0071] S201. Build the instance segmentation model of the molten iron ladle mouth, and the steps include:

[0072] 1) Extract the features of the image of the molten iron ladle mouth;

[0073] 2) Generate candidate target boxes through the RPN network;

[0074] 3) Identify the molten iron ladle mouth;

[0075] 4) Generate the bounding box of the molten iron ladle mouth through bounding box regression and more accurately wrap the target;

[0076] S202. Divide the labeled image of the molten iron ladle mouth into a training set, a test set and a validation set, and train the instance segmentation model of the molten iron ladle mouth;

[0077] To train the ladle mouth instance segmentation model, the pre-annotated ladle mouth images are input into the constructed ladle mouth instance segmentation model. Set the training parameters such as batch size, learning rate, and epochs, and select a loss function to train the ladle mouth instance segmentation model;

[0078] S203. Test and validate the trained ladle mouth instance segmentation model according to the test set and validation set divided in S202;

[0079] To test and validate the trained ladle mouth instance segmentation model, before annotation, all datasets are divided into a training set, a validation set, and a test set according to a ratio. The purpose of the validation set to validate the instance segmentation model is to adjust the hyperparameters of the instance segmentation model and find the best configuration of the ladle mouth instance segmentation model; the test set tests the actual performance of the instance segmentation model. The test set is a dataset that the model has never used during training and adjustment. Using the test set can ensure the fairness and accuracy of the ladle mouth instance segmentation results;

[0080] S204. Analyze the recognition metrics of the ladle mouth instance segmentation model, adjust the parameters of the ladle mouth instance segmentation model, and optimize the ladle mouth instance segmentation model;

[0081] To analyze the recognition metrics of the ladle mouth instance segmentation model, it is to judge the accuracy of the ladle mouth instance segmentation model according to the intersection over union of the ladle mouth instance segmentation area and the manually annotated area and the accuracy of ladle mouth recognition;

[0082] To optimize the ladle mouth instance segmentation model, it includes not only the relevant parameters of the input data, but also the hyperparameters such as the model structure and the loss function. Adjust the pixel value of the input image, the number of training times of the model, the number of batch processes, and the step size according to the accuracy and speed of model training, so as to ensure the quality of the model.

[0083] S3. Identify the molten iron surface shape according to the extracted ladle mouth images;

[0084] To identify the molten iron surface shape, according to the inner area of the ladle segmented in S2, distinguish the inner area and the outer area of the ladle, and based on the molten iron surface in the distinguished inner area of the ladle, use OpenCV image recognition techniques such as grayscale, binarization, and edge recognition methods to identify the molten iron surface shape of the ladle, and distinguish the molten iron and slag with different colors.

[0085] S4. Distinguish the molten iron and slag on the molten iron surface through the brightness and contrast of the molten iron and slag on the molten iron surface, and obtain the distribution of the molten iron and slag on the molten iron surface of the ladle;

[0086] S5. Calculate the index parameters of the slag region according to the distribution of hot metal and slag, and divide the slag skimming region.

[0087] Calculate the index parameters of the slag region. The index parameters include area, center point coordinates, and edge point coordinates. The formulas are as follows:

[0088]

[0089] In formula ①, Gx represents the horizontal direction gradient; Gy represents the vertical direction gradient.

[0090] M00 = ∑i∑jI(i, j) ②

[0091] M10 = ∑i∑jj·I(i, j) ③

[0092] M01 = ∑i∑ii·I(i, j) ④

[0093] In formulas ② - ④, the zero - order moment (M00) represents the area or total mass of the region; the first - order moments (M10 and M01) represent the weighted sums of the region pixel values about the x - axis and y - axis respectively; I(i, j) represents the pixel value of the image at the coordinate (i, j).

[0094] cX = M10 / M00 ⑤

[0095] cY = M01 / M00 ⑥

[0096] In formulas ⑤ and ⑥, cX represents the x - coordinate of the center point of the region, which is the sum of the pixel values weighted by the y - coordinate divided by the total pixel value.

[0097] cY represents the y - coordinate of the center point of the region, which is the sum of the pixel values weighted by the x - coordinate divided by the total pixel value.

[0098] The area of the slag region is determined according to the number of pixel points in the slag region. Convert the slag region into a binary image, and then calculate the number of pixel points with a value of 1.

[0099] Similarly, the center point coordinates of the slag region are the average values of the pixel points in the slag region.

[0100] The slag skimming region is divided as follows: According to the distribution of hot metal and slag on the hot metal surface in the ladle in S4, divide the hot metal region and the slag region, calculate and sort the area and boundary of the slag region, and set the priority of the slag region.

[0101] Calculate and photograph the area and boundary of the slag region. The calculation formulas are as follows:

[0102]

[0103] In formula ⑦, ∮L represents the line integral along the closed curve L, and ∫∫D represents the double integral over the closed region D;

[0104] The priority of the slag region is to divide the slag region into three directions: 45 degrees to the left, the middle, and 45 degrees to the right. According to the number and area of the slag in each region, the priority of the three directions is confirmed.

[0105] S6. Plan the slag removal path based on the index parameters of the identified slag region, and select the optimal slag removal path;

[0106] Plan the slag removal path: According to the index parameters of the slag region calculated in S5, call the path planning to calculate multiple planned slag removal paths, and calculate the eigenvalue of each slag removal path. Select the slag removal path with the highest priority as the slag removal path for this time, specifically including:

[0107] S601: According to the identified slag region, combined with the integrated call of the Dijkstra algorithm and the GA algorithm, calculate multiple slag removal paths. The formula is as follows:

[0108] dist[v] = min(dist[v], dist[u] + weight(u, v)) ②

[0109] In formula ②, dist[v] is the current shortest distance from the source vertex s to vertex v; dist[u] is the current shortest distance from the source vertex s to vertex u; weight(u, v) is the weight of the edge from vertex u to vertex v;

[0110] S602: Calculate the eigenvalue and weight of each slag removal path according to the index parameters such as the distance between adjacent slag regions, the deviation of the center points of adjacent slag, and the boundary of the slag region calculated in S5;

[0111] Calculate the slag removal path according to the area and center point coordinates of the slag region. The formula is as follows:

[0112] ⑨

[0113] In formula ⑨, xi represents the value of the i-th data point, wi represents the weight associated with the i-th data point, and n represents the total number of data points;

[0114] S603: Select the optimal slag removal path as the slag removal path for this time.

[0115] The present invention identifies the ladle mouth of the hot metal ladle through computer vision, ensuring that the identification of hot metal and slag is not interfered by other external environments. According to the different imaging methods of hot metal and slag, hot metal and slag are distinguished based on different brightness and contrast, realizing the distribution of hot metal and slag on the surface of the hot metal during slag skimming. Based on the identified distribution of hot metal and slag, the slag skimming path is calculated and updated in real time, laying a foundation for realizing automatic slag skimming; according to the dynamic changes in the shape of the hot metal surface in the hot metal ladle, the slag skimming path is updated in real time; it can well distinguish the hot metal area and the slag area in the internal area of the hot metal ladle, ensuring the accuracy of slag skimming.

Claims

1. A method for slag removal path planning based on iron flow liquid surface morphology recognition, characterized in that: include S1. Obtain an image of the ladle mouth, pre-process the image of the ladle mouth, and mark the internal area of ​​the ladle; S2, extracting the ladle mouth area according to the ladle mouth instance segmentation model; S3, identifying the molten iron liquid surface morphology according to the extracted molten iron tank mouth image; S4, distinguishing the molten iron and slag on the molten iron surface according to the brightness and contrast of the molten iron and slag on the molten iron surface, and obtaining the distribution of the molten iron and slag on the molten iron surface of the molten iron tank; S5. Calculate the index parameters of the slag area and divide the slag removal area according to the distribution of molten iron and slag; S6. According to the index parameters of the identified slag area, the slag removal path is planned and the optimal slag removal path is selected.

2. A method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 1, characterized in that: In S1, the ladle mouth image is preprocessed by adjusting the contrast and brightness of the ladle mouth image to remove noise from the ladle mouth image; the ladle mouth image is labeled by labelimg to frame the ladle mouth area with a square frame.

3. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 1 is characterized in that: In S2, the iron ladle mouth instance segmentation model includes: S201, constructing an instance segmentation model for the molten iron ladle mouth, the steps comprising: 1) Extract the features of the ladle mouth image; 2) Generate candidate target boxes through the RPN network; 3) Identify the mouth of the molten iron tank; 4) Generate a bounding box of the ladle mouth through bounding box regression to wrap the target; S202, dividing the ladle mouth images into training sets, test sets, and validation sets according to the ladle mouth images annotated in S1, and training the ladle mouth instance segmentation model; S203, testing and verifying the trained molten iron ladle mouth instance segmentation model according to the test set and verification set divided in S202; S204, analyzing the recognition index of the molten iron ladle mouth instance segmentation model, adjusting the parameters of the molten iron ladle mouth instance segmentation model, and optimizing the molten iron ladle mouth instance segmentation model.

4. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 3 is characterized in that: In S202, the training of the molten iron ladle mouth instance segmentation model is to input the previously labeled molten iron ladle mouth image into the constructed molten iron ladle mouth instance segmentation model, set the batch size, learning rate, epochs and related training parameters, and select the loss function to train the molten iron ladle mouth instance segmentation model.

5. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 3 is characterized in that: In S203, the testing and verification of the trained molten iron ladle mouth instance segmentation model is to divide all data sets into training set, validation set and test set in proportion before labeling, the validation set verifies the instance segmentation model for adjusting the hyperparameters of the instance segmentation model and finding the optimal configuration of the molten iron ladle mouth instance segmentation model; the test set tests the actual performance of the instance segmentation model.

6. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 3 is characterized in that: In S204, the recognition index of the ladle mouth instance segmentation model is analyzed, and the accuracy of the ladle mouth instance segmentation model is determined according to the intersection-over-union ratio of the ladle mouth instance segmentation area and the manually labeled area and the ladle mouth recognition accuracy. Optimize the instance segmentation model of the molten iron ladle mouth, including the hyperparameters related to the model structure and loss function. Adjust the pixel value of the input image, the number of model training times, the number of batches, and the step size according to the accuracy and speed of model training to ensure the quality of the model.

7. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 1 is characterized in that: In S3, the molten iron liquid surface morphology is identified, and the internal area and the external area of ​​the molten iron tank are distinguished according to the internal area of ​​the molten iron tank segmented by S2, and the molten iron and slag are distinguished based on the OpenCV image recognition technology according to the molten iron liquid surface of the distinguished internal area of ​​the molten iron tank.

8. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 1 is characterized in that: In S5, the index parameters of the slag area are calculated. The index parameters include area, center point coordinates, and edge point coordinates. The formula is as follows: In formula ①, Gx represents the horizontal gradient; Gy represents the vertical gradient; M00=i∑j∑I(i,j)② M10=i∑j∑j·I(i,j) ③ M01=i∑j∑i·I(i,j)④ In formulas ②-④, the zero-order moment (M00) represents the area or total mass of the region; the first-order moment (M10 and M01) represents the weighted sum of the pixel values ​​of the region about the x-axis and y-axis respectively; I(i,j) represents the pixel value of the image at the coordinate (i,j); cX=M10 / M00 ⑤ cY=M01 / M00 ⑥ In formula ⑤ and formula ⑥, cX represents the x-coordinate of the center point of the region, which is the sum of the weighted pixel values ​​of the y-coordinate divided by the total pixel value; cY represents the y coordinate of the center point of the region, which is the sum of the weighted pixel values ​​of the x coordinates divided by the total pixel value; The area of ​​the slag region is determined according to the number of pixels in the slag region. The slag region is converted into a binary image, and then the number of pixels that are 1 is calculated. Similarly, the coordinates of the center point of the slag area are taken as the average value of the pixel points in the slag area; The slag removal area is divided as follows: according to the distribution of molten iron and slag on the molten iron level of the molten iron ladle in S4, the molten iron area and the slag area are divided, the area and boundary of the slag area are calculated and sorted, and the priority of the slag area is set; The area and boundary of the slag area are calculated and photographed. The calculation formula is as follows: In formula ⑦, ∮L represents the curve integral along the closed curve L, and ∫∫D represents the double integral over the closed area D; The priority of the slag area is to divide the slag area into three directions: 45 degrees on the left, the middle, and 45 degrees on the right. The priorities of the three directions are determined according to the number and area size of the slag in each area.

9. The method for slag removal path planning based on iron flow liquid surface morphology recognition according to claim 1 is characterized in that: In S6, the slag removal path is planned: according to the index parameters of the slag area calculated in S5, the path planning is called to calculate multiple planned slag removal paths, and the characteristic values ​​of each slag removal path are calculated, and the slag removal path with the highest priority is selected as the slag removal path for this time, which specifically includes: S601: Based on the identified slag area, multiple slag removal paths are calculated by combining the integrated call of the Dijkstra algorithm and the GA algorithm. The formula is as follows: dist[v]=min(dist[v],dist[u]+weight(u,v)) ⑧ In formula ⑧, dist[v] represents the current shortest distance from source vertex s to vertex v; dist[u] represents the current shortest distance from source vertex s to vertex u; weight(u,v) represents the weight of the edge from vertex u to vertex v; S602: Calculate the characteristic value and weight of each slag removal path according to the distance between adjacent slag areas, the deviation between adjacent slag center points, and the boundary of the slag area and other index parameters calculated in S5; According to the area of ​​the slag area and the coordinates of the center point, the slag removal path is calculated using the following formula: In formula 9, xi represents the value of the i-th data point, wi represents the weight associated with the i-th data point, and n represents the total number of data points; S603: Selecting the best slag removal path as the path for this slag removal.

Images