A method and system for controlling a scraper based on image recognition

Abstract

The present invention discloses a method for controlling a slag skimmer based on image recognition. The method comprises the following steps: using a first image acquisition device to acquire an image of a molten iron ladle surface, performing boundary recognition to obtain the distribution of molten iron and slag, as well as the ladle boundary, and performing regional segmentation on the ladle surface image to obtain a plurality of subregions; calculating the slag area of ​​each subregion within the ladle boundary; constructing and training an improved deep residual contraction network to input the slag area of ​​each subregion into the improved deep residual contraction network, output an optimal slag skimming path, and determine the slag skimming arm swing angle and slag skimming head extension and contraction amount; using a second image acquisition device to acquire images of the molten iron level and the bottom boundary of the slag skimming head, and obtaining the slag skimming arm lifting amount and slag skimming arm inclination angle based on the images; and controlling the slag skimmer operation based on the slag skimming arm swing angle, slag skimming head extension and contraction amount, slag skimming arm lifting amount, and slag skimming arm inclination angle. Furthermore, the present invention discloses a slag skimmer control system for implementing the above method.

Description

Technical Field

[0001] The present invention relates to a molten iron pretreatment method and system, and in particular to a molten iron pretreatment slagging method and system. Background Art

[0002] In the steelmaking field of the metallurgical industry, the KR mechanical stirring method of the molten iron ladle is the preferred process technology for the desulfurization process of the existing blast furnace molten iron pretreatment.

[0003] During the desulfurization process, the sulfur-containing slag produced by KR mechanical stirring can be removed by slagging. Currently, the advanced KR slagging systems at home and abroad generally use manual remote slagging, that is, the operator performs remote manual slagging in the remote control room.

[0004] In existing technologies, remote manual slag removal can achieve slag removal while protecting the operator's safety. However, this process relies on manual labor, resulting in high production costs and low efficiency. It also requires high operator skills, and the slag removal time, effectiveness, and quality vary significantly between operators. This makes remote manual slag removal significantly affected by human factors. Therefore, users and the market urgently need an automated slag removal method to achieve automatic slag removal.

[0005] At present, some researchers have developed an automatic slag removal system based on image recognition. The system first reads the real-time image of the camera through the image recognition processing module to obtain information such as the constraint range of the molten iron ladle wall, slag distribution and molten iron distribution; secondly, the path planning module is applied. Under certain slag removal strategy constraints, the optimization algorithm is used to obtain a slag removal path with a large slag removal amount, short slag removal time and collision avoidance constraints between the slag removal head and the molten iron ladle. The path is further converted into the coordinate data of the slag removal head and then transmitted to the PLC for real-time control; finally, when the slag amount is less than a certain threshold, the slag removal end point is reached and the slag removal is stopped.

[0006] The key to this automatic scraping technology based on image recognition is the automated and intelligent determination of the scraper's four control parameters: the scraper arm's swing angle, scraper head extension, scraper arm vertical displacement, and scraper arm inclination. However, existing technologies have not been able to achieve effective scraping results, as they lack the ability to accurately and automatically measure these four control parameters in real time.

[0007] Based on this, and in response to the defects and shortcomings of existing slag removal processes, and to further reduce costs and increase efficiency, the present invention aims to provide a slag removal machine control method and system based on image recognition. This slag removal machine control method and system is easy to operate and has a simple process. It can realize automatic slag removal based on image recognition, and its image recognition is highly accurate and can accurately obtain the slag removal machine's operating control parameters. Using this slag removal machine control method and system, intelligent slag removal can be completed without human intervention, which not only effectively improves production efficiency, but also helps reduce production costs and protects the physical and mental health of on-site operators. Summary of the Invention

[0008] One of the objectives of the present invention is to provide a method for controlling a scraper based on image recognition. This method is easy to operate and has a simple process. It can automatically remove scraper material based on image recognition, with high image recognition accuracy and the ability to accurately obtain the scraper's operating control parameters. This method allows for intelligent scraping without human intervention, effectively improving production efficiency while also reducing production costs and protecting the physical and mental health of on-site operators.

[0009] In order to achieve the above object, the present invention proposes a method for controlling a scraper based on image recognition, which comprises the following steps:

[0010] (1) using a first image acquisition device located above the ladle to acquire an image of the ladle liquid surface;

[0011] (2) performing boundary recognition on the ladle surface image to obtain the distribution of molten iron and slag and the ladle boundary; and performing region segmentation on the ladle surface image to obtain a plurality of sub-regions;

[0012] (3) Calculate the slag area of ​​each sub-region within the ladle boundary;

[0013] (4) constructing and training an improved deep residual contraction network to input the slag area of ​​each sub-region into the improved deep residual contraction network, and the improved deep residual contraction network outputs the optimal slag removal path;

[0014] (5) determining the swing angle of the scraper arm and the extension and contraction amount of the scraper head based on the optimal scraper path;

[0015] (6) A second image acquisition device located on the side of the ladle is used to capture images of the molten iron level and the bottom boundary of the slag skimmer head, and the lifting amount and inclination angle of the slag skimmer arm are obtained based on the images of the molten iron level and the bottom boundary of the slag skimmer head;

[0016] (7) The slag scraping action of the slag scraper is controlled based on the swing angle of the slag scraper arm, the extension and contraction amount of the slag scraper head, the lifting amount of the slag scraper arm and the inclination angle of the slag scraper arm.

[0017] In the technical solution described in the present invention, the key to the scraper control method described in the present invention is to determine the automation and intelligence of the four control parameters of the scraper (including the scraper arm swing angle, the scraper head extension, the scraper arm up and down displacement and the scraper arm inclination angle).

[0018] In the present invention, the slag scraping arm's swing angle and the extension of the slag scraping head can be identified and measured by a first image acquisition device above the ladle. By performing boundary recognition and slag volume calculation on the ladle surface image captured by the first image acquisition device, the optimal slag scraping path within the horizontal plane can be calculated, thereby determining the slag scraping arm's swing angle and the extension and contraction of the slag scraping head at each point on the optimal scraping path. The slag scraping arm's lifting amount and the slag scraping arm's inclination angle can be identified and measured by a second image acquisition device on the side of the ladle, obtained by capturing images of the molten iron level and the bottom boundary of the slag scraping head.

[0019] It should be noted that in this embodiment, a modified deep residual shrinkage network (MDRSN) is constructed and trained. The slag area of ​​each sub-region can be used to output the optimal slag removal path based on the modified deep residual shrinkage network. The MDRSN is an integration of a deep residual network, an attention mechanism, and an improved soft threshold function.

[0020] The MDRSN eliminates redundant features based on the characteristics of each image recognition sample using a special soft thresholding method. The MDRSN uses an attention mechanism to identify noisy features and sets them to zero using a soft thresholding function. Furthermore, the attention mechanism retains important features, thereby enhancing the deep neural network's ability to extract useful features from noisy signals. When classifying slag and molten iron in the scraping area, image samples inevitably contain some noise, including dust, Gaussian noise, and abnormal information such as scraper failures. This noise can adversely affect classification results.

[0021] In the present invention, the improved deep residual shrinkage network (MDRSN) can use a special soft thresholding model to perform noise reduction processing; that is, from the perspective of deep learning, the features corresponding to these noises should be deleted inside the deep neural network to avoid affecting the effect of image recognition; even for the same image sample set, the amount of noise in each sample is often different. The location of the molten iron slag in each picture may be different, and the attention mechanism can identify the location of the slag for each picture; in addition, when training the molten iron and slag classifier, for images labeled "slag", they may be interfered by the slag scraper arm and smoke, resulting in a decrease in classification accuracy. Therefore, in the improved deep residual shrinkage network (MDRSN) described in the present invention, an attention mechanism is used to identify the interference of smoke and the shadow area of ​​the slag scraper arm, thereby effectively improving the accuracy of the slag and molten iron classifier.

[0022] Furthermore, in the slag skimmer control method based on image recognition described in the present invention, in step (2), image threshold processing and image enhancement are used to perform boundary recognition on the ladle liquid surface image.

[0023] In the above technical solution of the present invention, traditional image processing methods, including image threshold processing, image enhancement, etc., are used to perform boundary recognition on the ladle liquid surface image in order to obtain the distribution of molten iron and slag and the boundary with the ladle.

[0024] Accordingly, in step (2) of the slag skimmer control method of the present invention, the ladle liquid surface image can also be automatically processed for smoke recognition, and the region can be segmented to obtain a number of sub-regions.

[0025] It should be noted that in some embodiments, the region can be divided into a uniform fine grid of length (50-1920) and width (10-1280), the purpose of which is to prepare data for the deep learning algorithm. Each feature channel in the deep learning is a segmented sub-region (the minimum is a single pixel). Then, in the subsequent step (3), the slag area of ​​each sub-region within the ladle boundary can be calculated and corrected by the slag color. For example, for a grayscale image, the slag color can be analogous to the thickness of the slag. Then, the total slag amount and the slag removal path are labeled separately by industry experts. Among them, the total slag amount is used for endpoint slag judgment, that is, when the real-time surface residual slag amount is less than the set endpoint slag amount, the automatic slag removal process is terminated; and the slag removal path is a number of pre-set optimized paths (derived from the experience of industry experts), which are identified by industry experts based on the actual slag and molten iron distribution, and the optimal path is automatically trained by the deep learning network. Each path consists of 3-10 points, each point corresponds to a pixel point on the image, and uses image calibration or reference comparison methods to correspond one-to-one with the horizontal and vertical coordinates of the actual slag removal head. Finally, in step (4), the image recognition technology of the improved deep residual shrinkage network (MDRSN) can output the optimal slag removal path for the slag area of ​​each sub-region. Among them, the application of the deep residual network is to solve the problem of deep network degradation and avoid the gradient disappearance or gradient explosion phenomenon. It is used for training and learning to determine the corresponding total surface slag amount and optimal path number under the slag area distribution and color distribution. The attention mechanism automatically learns a set of thresholds by adding a small network to the deep residual network. The small network obtains a threshold for each feature channel. The threshold is equivalent to the weight of each feature. The size of the weight determines whether the feature is valued, retained or deleted. The present invention adopts the attention mechanism to weaken the adverse effects of abnormal noise such as dust environment noise and the shadow area of ​​the slag removal arm on slag amount recognition and path recognition. In addition, the soft threshold function processing method is improved to shield abnormal noise.

[0026] Furthermore, in the scraper control method based on image recognition described in the present invention, in step (4), the improved soft threshold function S(x, T) of the improved deep residual shrinkage network is:

[0027]

[0028] Where x represents the slag area of ​​the sub-region; T represents the threshold automatically learned through the attention mechanism; α represents the correction factor.

[0029] It should be noted that the present invention can correct the threshold value of the conventional soft threshold function through actual field experience fitting data to improve the classification accuracy of slag amount recognition and path recognition under a certain environment.

[0030] Furthermore, in the control method of the scraper based on image recognition according to the present invention, the correction factor is obtained based on the following formula:

[0031] α=β0+β1×X+β2×X 2 +β3×X 3

[0032] Where X represents the average grayscale value of the sub-region; β0, β1, β2, and β3 all represent fitting coefficients, where 0<β0≤10, 0<β1≤5, 0<β2≤3, and 0<β3≤3.

[0033] Furthermore, in the image recognition-based slag scraper control method described in the present invention, the improved deep residual shrinkage network outputs the optimal slag scraping path in the form of a label index number.

[0034] Furthermore, in the scraper control method based on image recognition described in the present invention, in step (5), an image calibration method or a reference comparison method is used to correspond the horizontal and vertical coordinates of the points on the optimal scraping path to the swing angle of the scraper arm and the telescopic amount of the scraper head, respectively, to determine the swing angle of the scraper arm and the telescopic amount of the scraper head.

[0035] Furthermore, in the image recognition-based slag skimmer control method described in the present invention, in step (6), obtaining the lifting amount of the slag skimmer arm based on the molten iron level and the bottom boundary image of the slag skimmer head includes the following steps: preprocessing the molten iron level and the bottom boundary image of the slag skimmer head; segmenting the preprocessed image to obtain recognition sub-regions; using different recognition thresholds to identify the lower boundary of the slag skimmer head and the molten iron liquid level position in each recognition sub-region; and obtaining the lifting amount of the slag skimmer arm using an image calibration method.

[0036] In the technical solution described in the present invention, since the lifting amount of the slag scraper arm controls the insertion depth of the slag scraper head in the molten iron, if the insertion depth is too deep, it is easy to cause increased slag scraping iron loss, while if the insertion depth is too shallow, the removal efficiency is too low. Therefore, it is necessary to accurately determine the vertical displacement of the slag scraper arm. At the same time, since this parameter is closely related to parameters such as the slag height of the slag scraper plate and the liquid level of the molten iron ladle, specifically, the slag height of the slag scraper plate is a random value. With different working conditions, the melting loss and slag state of the slag scraper plate vary greatly, and the slag scraper plate height can increase or decrease. Therefore, it is difficult to accurately calculate the slag height of the slag scraper plate by direct calculation alone. The present invention adopts a machine vision system to measure the vertical displacement of the slag scraper arm by image recognition. That is, through digital image processing technology, including binarization, image enhancement, morphological processing and other methods, using regional recognition technology, the image is divided into different recognition areas, and different recognition thresholds are used in different sub-areas to respectively identify the lower boundary of the slag scraper head and the molten iron liquid level position. After image calibration, the slag scraper arm lifting data is obtained to achieve the purpose of intelligent control.

[0037] Furthermore, in the image recognition-based slag skimmer control method described in the present invention, in step (6), obtaining the inclination angle of the slag skimmer arm based on the molten iron level and the bottom boundary image of the slag skimmer head includes the following steps: obtaining the height difference between the slag skimmer head at the near end and the far end of the standby position based on the molten iron level and the bottom boundary image of the slag skimmer head; obtaining the center line of the slag skimmer arm based on the height difference; and obtaining the inclination angle of the slag skimmer arm based on the angle between the center line of the slag skimmer arm and the horizontal line.

[0038] In the technical solution described in the present invention, the preferred solution for the inclination angle of the slag scraper arm is to set the slag scraper arm to remain horizontal during the slag scraping operation, and the insertion depth of the slag scraper head is controlled only by adjusting the up and down displacement of the slag scraper arm. Since the slag scraper machine itself cannot perform feedback control of the inclination angle of the slag scraper arm, the present invention adopts an image recognition method for calibration control, that is, the height information of the slag scraper arm on the image is determined by a machine vision recognition algorithm. If the height is consistent, it is considered to be kept horizontal. If the height is inconsistent, the horizontal value of the inclination angle of the slag scraper arm is adjusted and set according to the comparison between the image height difference and the actual height difference.

[0039] Correspondingly, another object of the present invention is to provide a slag scraper control system based on image recognition. When slag scraping is performed using this slag scraper control system, intelligent slag scraping can be completed without human intervention. It can not only effectively replace the manual operation in the current slag scraping process, but also greatly shorten the slag scraping time, improve production efficiency, and help reduce production costs.

[0040] In order to achieve the above objectives, the present invention proposes a scraper control system based on image recognition, which includes:

[0041] A first image acquisition device is provided above the molten iron ladle, and the first image acquisition device acquires an image of the liquid surface of the molten iron ladle;

[0042] A second image acquisition device is provided on the side of the molten iron ladle, and the second image acquisition device acquires images of the molten iron level and the bottom boundary of the slag skimmer;

[0043] The control module performs the following steps:

[0044] Performing boundary recognition on the ladle surface image to obtain the distribution of molten iron and slag as well as the ladle boundary, and performing region segmentation on the ladle surface image to obtain a plurality of sub-regions;

[0045] Calculate the slag area of ​​each sub-region within the ladle boundary;

[0046] Constructing and training an improved deep residual contraction network to input the slag area of ​​each sub-region into the improved deep residual contraction network, wherein the improved deep residual contraction network outputs an optimal slag removal path;

[0047] Determine the slag scraping arm swing angle and the slag scraping head extension and contraction amount based on the optimal slag scraping path;

[0048] The lifting amount and inclination angle of the slag scraper arm are obtained based on the molten iron level and the bottom boundary image of the slag scraper head;

[0049] The slag scraping action of the slag scraper is controlled based on the slag scraper arm swing angle, the slag scraper head extension and contraction amount, the slag scraper arm lifting amount and the slag scraper arm inclination angle.

[0050] Furthermore, in the scraper control system of the present invention, the improved soft threshold function S(x, T) of the improved deep residual shrinkage network is:

[0051]

[0052] Where x represents the slag area of ​​the sub-region; T represents the threshold automatically learned through the attention mechanism; α represents the correction factor.

[0053] Furthermore, in the control system of the scraper according to the present invention, the correction factor is obtained based on the following formula:

[0054] α=β0+β1×X+β2×X 2 +β3×X 3

[0055] Where X represents the average grayscale value of the sub-region; β0, β1, β2, and β3 all represent fitting coefficients, where 0<β0≤10, 0<β1≤5, 0<β2≤3, and 0<β3≤3.

[0056] Compared with the prior art, the scraper control method and system based on image recognition according to the present invention has the following advantages and beneficial effects:

[0057] (1) The scraper control method based on image recognition described in the present invention adopts a multi-machine vision integration method to obtain all scraper control parameters in real time, thereby effectively improving the level of intelligent control.

[0058] (2) The image recognition-based slag scraper control method described in the present invention adopts an image recognition technology based on an improved deep residual shrinkage network (MDRSN) for slag amount recognition and optimal slag scraping path recognition, which can effectively improve the recognition and classification accuracy in actual metallurgical high-temperature dust environments.

[0059] (3) The image recognition-based method for controlling a slag scraper according to the present invention can determine the distance parameter between the slag scraper arm and the slag scraper plate by identifying the position of the horizontal line of the slag scraper arm in the image and the position of the bottom edge line of the slag scraper plate in the image, thereby calculating the slag height of the slag scraper plate; the image recognition method for determining the liquid level height of the molten iron ladle is to determine the liquid level height by identifying the position of the liquid level of the molten iron ladle in the image and comparing it with a standard height reference value; the distance between the slag scraper plate and the liquid level of the molten iron ladle is to determine the distance between the two by identifying the position of the bottom of the slag scraper plate in the image and the position of the molten iron liquid surface in the image.

[0060] In summary, the image recognition-based control method for a slag scraper according to the present invention is easy to operate and has a simple process. It can achieve automatic slag removal based on image recognition, with high image recognition accuracy and the ability to accurately obtain slag scraper operating control parameters. This slag scraper control method enables intelligent slag removal without human intervention, effectively improving production efficiency while also reducing production costs and protecting the physical and mental health of on-site operators.

[0061] Accordingly, the scraper control system based on image recognition described in the present invention can be used to implement the above-mentioned scraper control method, which also has the above-mentioned advantages and beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The process flow chart of the scraper control method based on image recognition according to the present invention is schematically shown in one embodiment. DETAILED DESCRIPTION

[0063] The image recognition-based scraper control method and system of the present invention will be further described below with reference to specific embodiments of the present invention and the accompanying drawings. However, this description does not constitute an improper limitation to the present invention.

[0064] In the present invention, the image recognition-based slag skimmer control system can include: a first image acquisition device, a second image acquisition device, and a control device. The first image acquisition device is located above the ladle and can capture images of the ladle's liquid surface; the second image acquisition device is located on the side of the ladle and can capture images of the molten iron level and the bottom boundary of the slag skimmer.

[0065] It should be noted that in the system described in the present invention, the control module can perform boundary recognition on the ladle liquid surface image to obtain the distribution of molten iron and slag and the ladle boundary, and perform regional segmentation on the ladle liquid surface image to obtain several sub-regions; then calculate the slag area of ​​each sub-region within the ladle boundary; construct and train an improved deep residual shrinkage network (MDRSN) to input the slag area of ​​each sub-region into the improved deep residual shrinkage network (MDRSN), and the improved deep residual shrinkage network (MDRSN) outputs the optimal slag scraping path; then, based on the optimal slag scraping path, the slag scraping arm swing angle and the slag scraping head extension and contraction amount can be determined, and the slag scraping arm lifting amount and the slag scraping arm inclination angle can be obtained based on the molten iron level and the slag scraping head bottom boundary image; based on the slag scraping arm swing angle, the slag scraping head extension and contraction amount, the slag scraping arm lifting amount and the slag scraping arm inclination angle can be controlled.

[0066] In the present invention, the scraper control system based on image recognition described in the present invention can be used to implement the scraper control method based on image recognition described in the present invention.

[0067] In order to further illustrate the implementation of the scraper control system of the present invention, the present invention adopts a specific embodiment for illustration. In this embodiment, the process flow of the scraper control method implemented by the system can be as follows: Figure 1 shown.

[0068] Figure 1 The process flow chart of the scraper control method based on image recognition according to the present invention is schematically shown in one embodiment.

[0069] It should be noted that, in this embodiment, the first image acquisition device in the system for executing the slag skimmer control method described in the present invention can be set at a position 30° obliquely above the molten iron ladle and 5 meters away, and it can capture the liquid surface image of the molten iron ladle, and perform real-time measurement and identification on the liquid surface image of the molten iron ladle captured here by using image boundary recognition and slag amount recognition methods.

[0070] Correspondingly, the second image acquisition device is arranged at a lateral position between the ladle and the slag skimmer, and can acquire images of the molten iron level and the bottom boundary of the slag skimmer head. In this embodiment, the first image acquisition device and the second image acquisition device can be selected as cameras to acquire images.

[0071] like Figure 1 As shown, in this embodiment, the control method of the scraper implemented by the system of the present invention may include the following steps:

[0072] (1) A first image acquisition device located above the ladle is used to acquire an image of the ladle liquid surface.

[0073] (2) performing boundary recognition on the ladle surface image to obtain the distribution of molten iron and slag as well as the ladle boundary; and performing region segmentation on the ladle surface image to obtain several sub-regions.

[0074] In the above step (2) of the present invention, image threshold processing and image enhancement can be used to perform boundary recognition on the ladle surface image.

[0075] (3) Calculate the slag area of ​​each sub-region within the ladle boundary.

[0076] (4) Constructing and training an improved deep residual shrinkage network to input the slag area of ​​each sub-region into the improved deep residual shrinkage network, and the improved deep residual shrinkage network outputs the optimal slag removal path.

[0077] In the above step (4) of the present invention, in the improved deep residual shrinkage network (MDRSN), the improved soft threshold function S(x, T) can be expressed as:

[0078]

[0079] Where x represents the slag area of ​​the sub-region; T represents the threshold automatically learned by the attention mechanism; α represents the correction factor. Accordingly, the correction factor α can be further obtained based on the following formula:

[0080] α=β0+β1×X+β2×X 2 +β3×X 3

[0081] In the above formula, X represents the average grayscale value of the sub-region; β0, β1, β2, and β3 all represent fitting coefficients, where 0<β0≤10, 0<β1≤5, 0<β2≤3, and 0<β3≤3.

[0082] It should be noted that, in this embodiment, the present invention can correct the threshold size of the improved soft threshold function by fitting data based on actual field experience, thereby effectively improving the classification accuracy of the improved deep residual shrinkage network (MDRSN) for optimal slag path identification and slag amount identification under a certain environment. The improved deep residual shrinkage network (MDRSN) can output the optimal slag path in the form of a label index number.

[0083] (5) Determine the swing angle of the scraper arm and the extension and contraction amount of the scraper head based on the optimal scraper path.

[0084] In the above step (5) of the present invention, an image calibration method or a reference comparison method can be further used to correspond the horizontal and vertical coordinates of the points on the optimal scraping path to the swing angle of the scraping arm and the telescopic amount of the scraping head, so as to determine the swing angle of the scraping arm and the telescopic amount of the scraping head.

[0085] (6) A second image acquisition device located on the side of the ladle is used to capture images of the molten iron level and the bottom boundary of the slag skimmer head, and the lifting amount and inclination angle of the slag skimmer arm are obtained based on the images of the molten iron level and the bottom boundary of the slag skimmer head.

[0086] In the above step (6) of the present invention, the process of obtaining the lifting amount of the slag skimmer arm based on the molten iron level and the bottom boundary image of the slag skimmer head may include the following steps: preprocessing the molten iron level and the bottom boundary image of the slag skimmer head; performing region segmentation on the preprocessed image to obtain identification sub-regions; using different identification thresholds to identify the lower boundary of the slag skimmer head and the molten iron liquid surface position in each identification sub-region; and using image calibration method to obtain the lifting amount of the slag skimmer arm.

[0087] Accordingly, the process of obtaining the inclination angle of the slag scraper arm based on the molten iron level and the bottom boundary image of the slag scraper head may include the following steps: obtaining the inclination angle of the slag scraper arm based on the molten iron level and the bottom boundary image of the slag scraper head includes the steps of: obtaining the height difference of the slag scraper head at the near end and the far end of the standby position based on the molten iron level and the bottom boundary image of the slag scraper head; obtaining the center line of the slag scraper arm based on the height difference; obtaining the inclination angle of the slag scraper arm based on the angle between the center line of the slag scraper arm and the horizontal line.

[0088] (7) The slag scraping action of the slag scraper is controlled based on the swing angle of the slag scraper arm, the extension and contraction amount of the slag scraper head, the lifting amount of the slag scraper arm and the inclination angle of the slag scraper arm.

[0089] It should be noted that when adopting the real-time slag skimmer control method steps of the slag skimmer control system described in the present invention, in this embodiment, when performing regional segmentation on the ladle liquid surface image in the above step (2), the image grid can be divided into 100 lengths and 50 widths, obtaining a total of 5000 sub-regions, 8000 training images, and 2500 test images.

[0090] Accordingly, in this embodiment, the above step (4) needs to construct and train an improved deep residual shrinkage network (MDRSN), wherein the fitting coefficient β0 of the improved soft threshold function can be 0.15, the fitting coefficient β1 can be 1.2, the fitting coefficient β2 can be 0.3, and the fitting coefficient β3 can be 0.1.

[0091] In this embodiment, after determining the positions of the first image acquisition device and the second image acquisition device and completing the construction and training of the improved deep residual shrinkage network (MDRSN), the scraper control system described in this embodiment can effectively control the scraper to remove slag, and its real-time control success rate reaches more than 95%.

[0092] In summary, the image recognition-based control method for a slag scraper according to the present invention is easy to operate and has a simple process. It can achieve automatic slag removal based on image recognition, with high image recognition accuracy and the ability to accurately obtain slag scraper operating control parameters. This slag scraper control method enables intelligent slag removal without human intervention, effectively improving production efficiency while also reducing production costs and protecting the physical and mental health of on-site operators.

[0093] Accordingly, the scraper control system based on image recognition described in the present invention can be used to implement the above-mentioned scraper control method, which also has the above-mentioned advantages and beneficial effects.

[0094] It should be noted that the prior art within the scope of protection of the present invention is not limited to the embodiments given in this application document. All prior art that does not contradict the solutions of the present invention, including but not limited to prior patent documents, prior publications, prior public uses, etc., can be included in the scope of protection of the present invention.

[0095] In addition, the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0096] It should also be noted that the embodiments listed above are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or modifications made therefrom that can be directly derived from or easily conceived by those skilled in the art based on the disclosure of the present invention are intended to fall within the scope of protection of the present invention.

Claims

1. A method for controlling a scraper based on image recognition, characterized in that: Including steps: (1) using a first image acquisition device located above the ladle to acquire an image of the ladle liquid surface; (2) performing boundary recognition on the ladle surface image to obtain the distribution of molten iron and slag and the ladle boundary; and performing region segmentation on the ladle surface image to obtain a plurality of sub-regions; (3) Calculate the slag area of ​​each sub-region within the ladle boundary; (4) Construct and train an improved deep residual contraction network to input the slag area of ​​each sub-region into the improved deep residual contraction network, and the improved deep residual contraction network outputs the optimal slag removal path, wherein the improved soft threshold function S(x,T) of the improved deep residual contraction network is: Where x represents the slag area of ​​the sub-region; T represents the threshold automatically learned through the attention mechanism; α represents the correction factor; (5) determining the swing angle of the scraper arm and the extension and contraction amount of the scraper head based on the optimal scraper path; (6) A second image acquisition device located on the side of the ladle is used to capture images of the molten iron level and the bottom boundary of the slag skimmer head, and the lifting amount and inclination angle of the slag skimmer arm are obtained based on the images of the molten iron level and the bottom boundary of the slag skimmer head; (7) The slag scraping action of the slag scraper is controlled based on the swing angle of the slag scraper arm, the extension and contraction amount of the slag scraper head, the lifting amount of the slag scraper arm and the inclination angle of the slag scraper arm.

2. The method for controlling a scraper based on image recognition according to claim 1, wherein: In step (2), image threshold processing and image enhancement are used to perform boundary recognition on the ladle surface image.

3. The method for controlling a scraper based on image recognition according to claim 1, wherein: The correction factor is obtained based on the following formula: α=β0+β1×X+β2×X 2 +β3×X 3 Where X represents the average grayscale value of the sub-region; β0, β1, β2, and β3 all represent fitting coefficients, where 0<β0≤10, 0<β1≤5, 0<β2≤3, and 0<β3≤3.

4. The method for controlling a scraper based on image recognition according to claim 1, wherein: The improved deep residual contraction network outputs the optimal scraping path in the form of label index numbers.

5. The method for controlling a scraper based on image recognition according to claim 1, wherein: In step (5), the horizontal and vertical coordinates of the points on the optimal scraping path are respectively matched with the scraping arm swing angle and the scraping head extension amount by an image calibration method or a reference comparison method to determine the scraping arm swing angle and the scraping head extension amount.

6. The method for controlling a scraper based on image recognition according to claim 1, wherein: In step (6), obtaining the lifting amount of the slag skimmer arm based on the molten iron level and the bottom boundary image of the slag skimmer head includes the following steps: preprocessing the molten iron level and the bottom boundary image of the slag skimmer head; segmenting the preprocessed image to obtain recognition sub-regions; using different recognition thresholds to identify the lower boundary of the slag skimmer head and the molten iron liquid surface position in each recognition sub-region; and using image calibration method to obtain the lifting amount of the slag skimmer arm.

7. The method for controlling a scraper based on image recognition according to claim 1, wherein: In step (6), obtaining the inclination angle of the slag scraper arm based on the molten iron level and the bottom boundary image of the slag scraper head includes the following steps: obtaining the height difference between the slag scraper head at the near end of the standby position and the far end of the standby position based on the molten iron level and the bottom boundary image of the slag scraper head; obtaining the center line of the slag scraper arm based on the height difference; and obtaining the inclination angle of the slag scraper arm based on the angle between the center line of the slag scraper arm and the horizontal line.

8. A scraper control system based on image recognition, characterized in that: include: A first image acquisition device is provided above the molten iron ladle, and the first image acquisition device acquires an image of the liquid surface of the molten iron ladle; A second image acquisition device is provided on the side of the molten iron ladle, and the second image acquisition device acquires images of the molten iron level and the bottom boundary of the slag skimmer; The control module performs the following steps: Performing boundary recognition on the ladle surface image to obtain the distribution of molten iron and slag as well as the ladle boundary, and performing region segmentation on the ladle surface image to obtain a plurality of sub-regions; Calculate the slag area of ​​each sub-region within the ladle boundary; An improved deep residual contraction network is constructed and trained to input the slag area of ​​each sub-region into the improved deep residual contraction network, which outputs the optimal slag removal path; wherein the improved soft threshold function S(x,T) of the improved deep residual contraction network is: Where x represents the slag area of ​​the sub-region; T represents the threshold automatically learned through the attention mechanism; α represents the correction factor; Determine the slag scraping arm swing angle and the slag scraping head extension and contraction amount based on the optimal slag scraping path; The lifting amount and inclination angle of the slag scraper arm are obtained based on the molten iron level and the bottom boundary image of the slag scraper head; The slag scraping action of the slag scraper is controlled based on the slag scraper arm swing angle, the slag scraper head extension and contraction amount, the slag scraper arm lifting amount and the slag scraper arm inclination angle.

9. The control system for a scraper according to claim 8, characterized in that: The correction factor is obtained based on the following formula: α=β0+β1×X+β2×X 2 +β3×X 3 Where X represents the average grayscale value of the sub-region; β0, β1, β2, and β3 all represent fitting coefficients, where 0<β0≤10, 0<β1≤5, 0<β2≤3, and 0<β3≤3.

Images