Feeding control method for refining furnace and related equipment

Abstract

The invention discloses a refining furnace feeding control method and related equipment, and relates to the technical field of iron and steel smelting, and the method comprises the steps that a material adding instruction and a material feature image for a target material are obtained; based on the material feature image, determining an evaluation type, an evaluation physical quantity and a type confidence coefficient of the target material; the preset type is compared with the evaluation type, the preset physical quantity is compared with the evaluation physical quantity, and a material verification result is determined; and when the material verification result indicates that the material verification result is abnormal or the type confidence coefficient is smaller than a preset confidence coefficient threshold value, performing material flow intervention operation. According to the invention, accurate verification of the materials before entering the furnace is realized through image identification and instruction comparison, feeding errors and identification errors are effectively prevented, and the feeding accuracy and the downstream product quality stability under complex working conditions are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of steel smelting, and more particularly, to a refining furnace feeding control method and related equipment. BACKGROUND

[0002] With the continuous improvement of the intelligent level of steel manufacturing, the refining furnace feeding link, as an important process affecting the smelting quality, efficiency and safety, is increasingly concerned by automatic control and intelligent recognition technology. Especially in the RH vacuum refining and other complex processes, the types of alloy materials are various, the addition amount precision is high, and the feeding sequence is sensitive. Once the material type is wrongly added, the quantity is deviated, or the sequence is wrong, it is easy to cause the composition fluctuation of molten steel, the unstable product performance, and even cause serious consequences such as continuous casting blockage and rework. Therefore, how to realize real-time verification and error intervention of the material type and quantity in the feeding process has become a key technical problem in the refining process control.

[0003] In the related art, the material feeding process still mainly relies on manual identification and operation, and part of the feeding amount verification is assisted by a weighing system. However, the manual method has problems such as strong experience dependence, high risk of misoperation, and low standardization, which is difficult to meet the requirements of high-quality refining operation. The weighing system can only verify the total weight of the material, and cannot judge the actual type or sequence, lacks image recognition and intelligent comparison means in the process, and is difficult to realize real whole-process closed-loop control. Especially in typical metallurgical environments such as high temperature, dust and strong interference, the traditional method is difficult to balance the recognition accuracy and response speed, resulting in that the feeding error is not discovered or intervened in time. That is, the related art has the technical problems of large deviation of molten steel composition, low product quality and production safety hazards. SUMMARY

[0004] A series of simplified concepts are introduced in the summary part of the present application, which will be further described in detail in the specific embodiment part. The summary part of the present application does not mean to try to limit the key features and necessary technical features of the claimed technical solution, and even less means to determine the protection scope of the claimed technical solution.

[0005] The refining furnace feeding control method and related equipment provided by the present application can realize accurate verification of the material before entering the furnace through image recognition and instruction comparison, effectively prevent feeding errors and recognition mistakes, and improve the feeding accuracy and downstream product quality stability under complex working conditions.

[0006] In a first aspect, this application provides a refining furnace charging control method, comprising: acquiring a material adding instruction and a material feature image for a target material, wherein the material adding instruction includes a preset type and a preset physical quantity; determining an evaluation type, an evaluation physical quantity, and a type confidence level for the target material based on the material feature image; comparing the preset type with the evaluation type and the preset physical quantity with the evaluation physical quantity to determine a material verification result; and performing a material flow intervention operation when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold.

[0007] In some implementations, determining the evaluation type, evaluation physical quantity, and type confidence level of the target material based on the material feature image includes: determining the material outline in the material feature image using a preset target detection algorithm; determining spatial volume parameters based on the material outline and generating the evaluation physical quantity by combining it with a preset density mapping table; extracting spectral feature vectors and visual texture feature vectors from the material feature image; performing similarity matching between the spectral feature vectors and a pre-constructed material spectral library to obtain a first type probability distribution; inputting the visual texture feature vectors into a preset classification model to obtain a second type probability distribution; and determining the evaluation type and the type confidence level based on the first type probability distribution and the second type probability distribution.

[0008] In some embodiments, extracting the spectral feature vector and visual texture feature vector of the material feature image includes: extracting the packaging area in the material feature image; parsing the text information of the packaging area using an optical character recognition engine; when parsing is successful, determining the material type corresponding to the text information as the evaluation type, and determining a preset confidence threshold as the type confidence; when parsing fails, extracting the spectral feature vector and visual texture feature vector of the material feature image.

[0009] In some embodiments, acquiring the material addition instruction and material feature image for the target material includes: in response to a material conveyor belt start signal, acquiring the material addition instruction from the production execution system via an industrial communication protocol; acquiring an original image of the target material using a multispectral imaging unit deployed above the conveyor belt area; determining the image sharpness evaluation value of the original image and the sharpness deviation value between it and a preset sharpness threshold; when the sharpness deviation value is not within the preset deviation range, increasing the light intensity and increasing the lens cleaning frequency; and when the sharpness deviation value is within the preset deviation range, determining the original image as the material feature image.

[0010] In some embodiments, before acquiring the original image of the target material by a multispectral imaging unit deployed above the belt conveyor area, the refining furnace feeding control method further includes: determining the reflective properties of the target material according to the preset type in the material feeding instruction; and switching the illumination mode for the target material according to the reflective properties, wherein the illumination mode is at least one of coaxial illumination mode, bright field illumination mode, dark field illumination mode, or diffuse reflection illumination mode.

[0011] In some implementations, the step of intervening in the material flow when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold includes: when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold, determining an anomaly severity assessment value for the target material based on the material verification result and the type confidence level, wherein the anomaly level is calculated and generated based on a comprehensive calculation of the type deviation severity, the physical quantity deviation ratio, and the confidence level deviation magnitude; in response to the anomaly severity assessment value being greater than a first preset risk threshold, generating a belt deceleration command and triggering an audible and visual warning signal; in response to the anomaly severity assessment value being greater than a second preset risk threshold, generating an emergency stop command and locking the associated silo discharge valve, while simultaneously pushing the material feature image to the central control terminal.

[0012] In some implementations, determining the anomaly assessment value of the target material based on the material verification result and the type confidence level includes: determining the type deviation severity based on the type matching status in the material verification result, wherein the type deviation severity is negatively correlated with the metallurgical compatibility between associated materials in a preset material grade library; determining the physical quantity deviation ratio based on the proportion of the absolute deviation between the preset physical quantity and the assessed physical quantity to the preset physical quantity; determining the confidence level deviation magnitude based on the absolute value of the difference between the preset confidence threshold and the type confidence level; and weighting the type deviation severity, the physical quantity deviation ratio, and the confidence level deviation magnitude according to preset weighting coefficients to obtain the anomaly assessment value.

[0013] In some implementations, the step of extracting the spectral feature vector and visual texture feature vector of the material feature image when parsing fails includes: when parsing fails, calculating the morphological entropy feature of the packaging area and the packaging similarity of a preset material packaging template; when the packaging similarity is greater than a preset similarity threshold, taking the material type associated with the preset material packaging template as the evaluation type, and determining the type confidence level based on the packaging similarity; when the packaging similarity is less than or equal to the preset similarity threshold, extracting the spectral feature vector and visual texture feature vector of the material feature image.

[0014] Secondly, this application also provides a refining furnace charging control device, comprising: an image acquisition unit for acquiring a material adding instruction and a material feature image for a target material, wherein the material adding instruction includes a preset type and a preset physical quantity; an image evaluation unit for determining the evaluation type, evaluation physical quantity, and type confidence level of the target material based on the material feature image; a verification determination unit for comparing the preset type with the evaluation type and the preset physical quantity with the evaluation physical quantity to determine the material verification result; and a process intervention unit for performing material process intervention operations when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold.

[0015] Thirdly, this application also provides an electronic device, including: a memory and a processor, wherein the processor is configured to execute a computer program stored in the memory to implement the steps of the refining furnace charging control method described in the first aspect.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the refining furnace charging control method described in the first aspect.

[0017] Fifthly, this application also provides a computer program product, including a computer program or computer-executable instructions, which, when executed by a processor, implement the refining furnace feeding control method provided in the embodiments of this application.

[0018] In summary, this application, through the recognition of material characteristic images and comparison with material addition instructions, can accurately verify materials before they actually enter the furnace. This solves the problems of experience dependence and human error commonly found in traditional manual feeding, significantly reducing the risk of incorrect, missed, or excessive feeding. When the recognition reliability is low, it also triggers warnings / interventions to avoid the risk of incorrect recognition but still allowing the material to pass. It is more adaptable to complex operating conditions and avoids the addition of incorrect alloys, thereby preventing problems such as steel composition deviation, decreased mechanical properties, and excessive harmful impurities, ensuring the quality and consistency of downstream products, and reducing scrap and rework rates. In conclusion, the refining furnace feeding control method provided by this application achieves accurate verification of materials before they enter the furnace through image recognition and instruction comparison, effectively preventing feeding errors and recognition mistakes, and improving the accuracy of feeding and the stability of downstream product quality under complex operating conditions. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0020] Figure 1 A flowchart illustrating a refining furnace charging control method provided in an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of the composition of a refining furnace charging control device provided in an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] The terms used in the specification, claims, and drawings of this application, such as "first," "second," "third," "fourth," etc. (if any), are used to distinguish similar objects and not to describe a specific order or sequence. Therefore, it is to be understood that these terms can be used interchangeably where appropriate, allowing the described embodiments to be used in different orders, unless specifically required by the illustrations or description. Furthermore, the terms "is" and "has," and any variations thereof, are intended to cover, non-exclusively, all possible constituent elements. For example, a process, method, system, product, or apparatus comprising several steps or units is not necessarily limited to the steps or units explicitly listed, but may also include other steps or units not explicitly listed, or steps or units inherent to the process, method, product, or apparatus.

[0024] In this application, a "module" or "unit" refers to a computer program or part of a computer program that has a specific function and works in conjunction with other related parts to achieve a predetermined goal. These modules or units can be implemented by software, hardware (e.g., processing circuitry or memory), or a combination of both. One or more processors or memories can implement one or more modules or units. Furthermore, each module or unit can also be part of a larger module or unit.

[0025] The technical solutions of this application will be described in detail below with reference to the accompanying drawings of the embodiments. It should be noted that the described embodiments are only a part of this application, and not all embodiments. In the following description, the "some embodiments" mentioned are only a subset of all possible embodiments, which may be the same or different subsets, and different embodiments can be combined with each other without conflict.

[0026] Figure 1 This is a schematic flowchart illustrating a refining furnace charging control method provided in an embodiment of this application. For example, see [link to example]. Figure 1 The refining furnace charging control method provided in this application embodiment may include the following steps 101 to 104:

[0027] Step 101: Obtain the material addition instruction and material feature image for the target material, wherein the material addition instruction may include a preset type and a preset physical quantity;

[0028] In some examples, the target material refers to the specific material currently being fed into the refining furnace during the feeding process; for example, the material being conveyed on the belt is a batch of "FeSi 75" ferrosilicon alloy, which is the current target material. The material addition instruction is a task parameter from the Manufacturing Execution System (MES) or the smelting automatic control system, indicating the type and quantity of material required for this feeding operation. It can be retrieved in real-time from the MES via industrial communication protocols for the current furnace feeding task; for example, the current furnace instruction is "add 30kg ferrosilicon manganese alloy," which is the material addition instruction for that material. The preset type is the standard material type specified in the material addition instruction, serving as the basis for subsequent identification, evaluation, and comparison. It can be preset by process engineers in the MES formula or automatically generated based on the current steel grade or process; for example, if it is the deoxidation stage of a low-alloy steel, the preset type could be "ferrosilicon (FeSi75)". The preset physical quantity refers to the quantity or mass of the target material to be added this time, which can be expressed in "kg"; for example, if 25kg of FeSi 75 alloy needs to be added in the current furnace, then 25kg is the preset physical quantity.

[0029] For example, when the belt conveyor system is detected to be started, the feeding formula instruction for the current furnace is automatically obtained from the production execution system, including the type of alloy material to be added next (preset type) and the corresponding addition mass (preset physical quantity); at the same time, the imaging unit located above the belt begins to collect images of the material being conveyed, forming a corresponding material feature image; so that the image can be identified and processed in the future, and compared with the material addition instruction content to ensure that the feeding is accurate.

[0030] By implementing step 101, the production execution system automatically acquires task instructions and combines multispectral imaging to collect images of actual materials, providing a key data foundation for subsequent identification and verification. This achieves dual-source input of material "should be added" and "actually added" data, which helps to break away from manual reliance and build an automated and digital pre-identification mechanism for materials.

[0031] Step 102: Based on the material feature image, determine the evaluation type, evaluation physical quantity, and type confidence level of the target material;

[0032] In some examples, the evaluation type is the material category determined by intelligent analysis of the material feature image, reflecting the actual material's type and attribute, and is used for comparison and verification with a preset type; for example, if the image analysis determines the material to be "silicon-manganese alloy (SiMn)," this is the evaluation type. The evaluation physical quantity is the material mass or quantity estimated based on the material's spatial volume information in the material feature image combined with its density, and is used for comparison with preset physical quantities. Image processing algorithms can be used to identify the material outline, calculate the three-dimensional spatial projected volume, and then convert the mass value according to a table based on the density of different materials; for example, image analysis might determine the batch of material to have a volume of 7.5 liters, and the density to be found to be 2.5 g / cm³. 3 If the weight is 18.75 kg, then the assessed weight is 18.75 kg. Type confidence is a reliability score for the currently identified assessment type, which can be represented by a probability value between 0 and 1. It is used to determine whether the identification is sufficiently reliable. It can be the probability distribution value output by an image recognition model (such as a neural network classifier), or a weighted confidence value calculated based on the fusion of multiple models. For example, if the probability of identifying the current material as "FeSi" type is 92%, then the type confidence is 0.92.

[0033] For example, after acquiring the material feature image, the color distribution, texture structure and edge morphology of the image are first analyzed using multispectral imaging, and the corresponding feature vectors are extracted. Then, the most likely material category is determined by model comparison as the evaluation type, and the evaluation physical quantity is calculated by combining contour analysis and density mapping. At the same time, the type confidence is calculated based on the output confidence of the image recognition model or the matching degree of multi-source fusion, which serves as the basis for subsequent comparison and process intervention.

[0034] By implementing step 102, multi-dimensional information such as image visual texture, spectral features, and volume estimation can be integrated to accurately determine the actual type and weight of materials. At the same time, the confidence level output by the intelligent model can be used to quantify the reliability of recognition, thereby effectively addressing the problems of diverse material types, complex packaging, and high image noise in metallurgical sites, and improving recognition accuracy and robustness.

[0035] Step 103: Compare the preset type with the evaluation type, and the preset physical quantity with the evaluation physical quantity to determine the material verification result;

[0036] In some examples, the material verification result is a judgment of the feeding accuracy obtained by comparing preset values ​​(preset type and preset physical quantity) with evaluated values ​​(evaluated type and evaluated physical quantity), used to determine whether the current material meets the feeding requirements. Judgments can be made based on preset comparison logic, such as whether the type matches, whether the difference in physical quantity is within the tolerance range, and finally generating a verification conclusion, such as "normal," "type abnormal," or "physical quantity abnormal." For example, if the preset type is "SiMn" and the evaluated type is "FeMn," then the types are inconsistent, and the material verification result is "type abnormal." Comparing the preset type and the evaluated type refers to checking whether the preset type and the identified evaluated type are the same material. This can be done using string exact matching, or mismatch matching of materials within the same family. For example, if "FeMn" is allowed to replace "SiMn," a compatibility rule must be attached. For example, preset type = "SiMn"; evaluated type = "FeMn"; if set to be non-substitutable, it is determined to be a mismatch. Comparing the preset physical quantity with the evaluated physical quantity involves comparing the deviation between the preset physical quantity and the evaluated physical quantity obtained from image analysis to determine whether it is within the preset allowable range. For example, if the preset physical quantity is 20kg and the evaluated physical quantity is 18.5kg, the deviation is 7.5%. If the preset allowable range is 10%, it is considered qualified.

[0037] For example, after obtaining the evaluation type and evaluation physical quantity of the material, it can be automatically compared with the preset type and preset physical quantity in the corresponding material addition instruction; first, type matching is performed to determine whether it is the same material or an acceptable substitute; then, physical quantity deviation is calculated to determine whether it exceeds the set error range; finally, the two comparison results are combined into a comprehensive judgment and the "material verification result" is output for the decision-making basis of whether to intervene in the subsequent operation process.

[0038] By implementing step 103, the set "material to be added" is compared with the "actual material added" obtained by image recognition. This allows for the automatic determination of whether there are any errors or deviations in material addition, thus realizing an intelligent verification mechanism before material addition. This effectively avoids operational errors such as incorrect material addition, leakage, and over-addition, ensuring the quality of steelmaking from the source.

[0039] Step 104: When the material verification result indicates an anomaly or the type confidence level is less than the preset confidence threshold, perform material flow intervention.

[0040] In some examples, an anomaly in the material verification result indicates a significant deviation or mismatch between the preset value and the evaluated value, suggesting a risk of adding the target material. Anomalies include, but are not limited to, type mismatch (e.g., requesting "lime" but identifying "fluorite") and excessive physical quantity deviation (e.g., a planned addition of 10 kg but an identified result of 14 kg, exceeding the preset allowable range). For example, an evaluated type of "FeSi" and a preset type of "SiMn," or a physical quantity deviation exceeding ±10% of the preset allowable range, can both be considered "abnormal." The preset confidence threshold is the minimum acceptable recognition confidence standard used to determine the reliability of target material identification. It can be set by engineers based on historical recognition accuracy experience or production risk levels, such as 85% or 90%. For example, if the confidence threshold is set to 0.88, and the output confidence score for the material type after image recognition is 0.76, it indicates insufficient recognition confidence. Material flow intervention operations are a series of preventative control actions performed when an anomaly is identified or the confidence level is insufficient, to prevent erroneous materials from entering the refining furnace; for example, slowing down the conveyor belt and issuing audible and visual warnings when the anomaly is minor, and immediately stopping the belt, locking the discharge valve, and sending central control alarm information when the anomaly is severe.

[0041] For example, upon detecting an anomaly in material identification or an identification confidence level below a set threshold, the level of the anomaly will be immediately assessed, and specific intervention measures will be determined accordingly. For instance, if it is only a slight deviation in physical quantity, the material can be slowed down and awaited manual confirmation. If it is a type error or an extremely blurry image, the system will automatically stop and lock the relevant silos to prevent incorrect materials from being mixed into the refining furnace.

[0042] By implementing step 104, audible and visual alarms and central control prompts can be linked, proactively intervening before feeding errors may occur, and minimizing the risk of quality accidents and equipment damage caused by feeding errors in the metallurgical process.

[0043] In summary, the embodiments of this application, by recognizing material feature images and comparing them with material addition instructions, can accurately verify materials before they actually enter the furnace. This solves the problems of experience dependence and human error commonly found in traditional manual feeding, significantly reducing the risk of incorrect, missed, or excessive feeding. When the recognition reliability is low, it also triggers warnings / interventions to avoid the risk of incorrect recognition but still allowing the material to pass. It is more adaptable to complex operating conditions and avoids the addition of incorrect alloys, thereby preventing problems such as steel composition deviation, decreased mechanical properties, and excessive harmful impurities, ensuring the quality and consistency of downstream products, and reducing scrap and rework rates. In conclusion, the refining furnace feeding control method provided by the embodiments of this application achieves accurate verification of materials before they enter the furnace through image recognition and instruction comparison, effectively preventing feeding errors and recognition mistakes, and improving the accuracy of feeding and the stability of downstream product quality under complex operating conditions.

[0044] In some embodiments, step 102 may include: determining the material contour in the material feature image using a preset target detection algorithm; determining spatial volume parameters based on the material contour and generating an evaluation physical quantity by combining a preset density mapping table; extracting the spectral feature vector and visual texture feature vector of the material feature image; performing similarity matching between the spectral feature vector and a pre-constructed material spectral library to obtain a first type probability distribution; inputting the visual texture feature vector into a preset classification model to obtain a second type probability distribution; and determining the evaluation type and type confidence based on the first type probability distribution and the second type probability distribution.

[0045] In some examples, the pre-defined object detection algorithm is used to identify material targets in a material feature image. It can be implemented based on deep learning models such as YOLO, Faster R-CNN, and Mask R-CNN. This pre-defined algorithm is trained on labeled samples and can automatically locate the bounding box or contour of the material in the material feature image. For example, the YOLOv5 model can be used to detect the main body of the material in a material feature image and exclude background or clutter interference. The material contour is the boundary shape of the target material identified in the material feature image, which can be represented by pixel coordinates and can be further used to generate a mask image. The bounding box output by the object detection algorithm can be further finely segmented to extract the contour, such as using GrabCut or a semantic segmentation network like DeepLabV3. The spatial volume parameter is used to estimate the actual spatial volume of the material, such as length, width, height, or three-dimensional volume (m), based on the material contour in the material feature image and known camera intrinsic parameters (such as focal length and distance). 3 For example, by combining the pixel area of ​​the contour in the image and the distance from the camera to the material, the volume can be calculated through back projection. A preset density mapping table is a table that establishes a mapping relationship between material types and their standard densities, used to convert volume to mass; for example, if the identified material type is limestone, its density is 2.5 g / cm³. 3Therefore, the estimated mass can be calculated as volume × density ≈ mass. Spectral feature vectors are sequences of reflectance or spectral response values ​​obtained through multispectral imaging systems, reflecting the optical properties of materials in different wavelength bands; for example, the spectral reflectance of a target material in the 400nm to 1000nm wavelength band constitutes its spectral feature vector. Visual texture feature vectors are features describing the surface texture of materials (such as roughness, particle shape, and structural repeatability) extracted using image processing or convolutional neural networks. Features can be extracted through the intermediate layers of pre-trained CNNs (such as ResNet and EfficientNet). The similarity matching process compares the spectral feature vector of the target material with the standard feature vectors in the material spectral library, outputting the degree of matching between different types of materials. Cosine similarity or Euclidean distance can be used to measure the difference, outputting a matching score with each standard material. The first type probability distribution is a material category probability distribution generated based on the spectral similarity results, reflecting the probability that the target material belongs to each type under the spectral information; for example: {ferroalloy: 0.70, limestone: 0.20, scrap steel: 0.10}. The preset classification model is an image classification model trained based on visual texture features. It can be a traditional SVM, XGBoost, or a deep network such as MobileNet or Vision Transformer. It can perform multi-class classification on the input visual features and output the possible material type distribution. The second type of probability distribution is a probability distribution generated after the visual texture features are input into the classification model, reflecting the degree of class matching of the material in the texture dimension; for example: {ferroalloy: 0.60, limestone: 0.30, scrap steel: 0.10}. The first and second type probability distributions can be fused, such as by weighted averaging or Bayesian combination, to output the final evaluation type (the category corresponding to the highest probability) and its confidence score (the fused highest probability value); for example, after fusion, {ferroalloy: 0.68, limestone: 0.25, scrap steel: 0.07}, the evaluation type is ferroalloy, and the confidence score is 0.68.

[0046] For example, when the target material is conveyed to the vision acquisition area by a belt, images can be automatically acquired and a detection model can be run to identify the material outline. Then, based on the image resolution and camera parameters, the volume is estimated and converted into an estimated weight using a density table. In parallel, the spectral and texture features of the image are also extracted, and the probability distributions of the two categories are obtained through a dual recognition mechanism. The final type and confidence level are then fused to provide a reliable basis for subsequent comparison and intervention.

[0047] By implementing the above embodiments, integrating target detection algorithms, spectral feature analysis, and texture feature recognition, material types and physical quantities can be identified in a refined manner at the visual level. This enables accurate discrimination under multi-dimensional feature fusion, effectively improving the accuracy and stability of material identification. Furthermore, the fusion of two types of probability distributions enhances the robustness of the evaluation results, enabling it to adapt to the identification needs under various material states and complex image conditions, thereby improving the reliability and generalization ability of intelligent recognition.

[0048] In some embodiments, the aforementioned extraction of the spectral feature vector and visual texture feature vector of the material feature image may include: extracting the packaging area in the material feature image; parsing the text information of the packaging area using an optical character recognition engine; when parsing is successful, determining the material type corresponding to the text information as the evaluation type and determining a preset confidence threshold as the type confidence; when parsing fails, extracting the spectral feature vector and visual texture feature vector of the material feature image.

[0049] In some examples, the packaging area is a prominent part of the material's outer packaging in the material feature image. It may contain labels, printed text, barcodes, or other identification information. Object detection or image segmentation algorithms such as YOLO and Mask R-CNN can be used to identify printed or regular contour areas in the material feature image; for example, identifying a rectangular area printed with "FeSi75" indicates that this is the packaging area. An Optical Character Recognition (OCR) engine is used to convert printed or handwritten text in the image into recognizable machine text. It can call text recognition engines such as Tesseract, PaddleOCR, Baidu OCR, and Google Vision API. The process of parsing the text information in the packaging area involves structuring the text content extracted by the OCR engine to identify information such as material names, component indicators, or model numbers. Natural language processing and regular expressions can be combined to identify keywords or standard naming conventions; for example, "FeMn 65 / 15" can be parsed as high-carbon ferromanganese, with the model number 65 / 15. By matching the results of the optical character recognition engine with the built-in material knowledge base, the material type referred to by the text can be determined. For example, if the parsed "FeSi75" matches the knowledge base, the evaluation type is determined to be "ferrosilicon (containing 75% silicon)". When text recognition is successful and the match is clear and unambiguous, a high confidence value, such as 1.0 or 0.95, can be directly assigned as the type confidence. If the optical character recognition engine cannot recognize the text, or the recognition result is blurry or invalid, image processing and multispectral analysis are then performed to identify the material type.

[0050] For example, on the feeding line of a steel refinery, a package of material is photographed. First, the label area with "CaF2≥90%" is extracted through image recognition. OCR successfully recognizes and parses the text, thus determining it to be "fluorite" and directly setting its type confidence level to 0.98. If the image is blurry or the label is damaged, causing recognition failure, it automatically switches to using multispectral imaging equipment and image analysis algorithms to perform in-depth analysis of the material's appearance and spectral characteristics, thereby completing the assessment of type and confidence level inference, ensuring that the feeding recognition process has a high degree of robustness and intelligent redundancy mechanism.

[0051] Through the implementation of the above embodiments, the text information in the material packaging area is identified first to achieve rapid positioning and accurate identification of the material type. When text recognition fails, it returns to the image feature extraction path, effectively constructing a dual-layer recognition mechanism of text and image. This not only improves the overall recognition efficiency but also reduces the dependence on a single path of image algorithm, enhancing recognition stability. It is particularly suitable for complex situations such as dust obstruction or packaging wear common in smelting environments.

[0052] In some embodiments, the aforementioned step 101 may include: in response to a material conveyor belt start signal, obtaining a material addition instruction from a production execution system via an industrial communication protocol; acquiring an original image of the target material using a multispectral imaging unit deployed above the conveyor belt area; determining the image sharpness evaluation value of the original image and the sharpness deviation value between the original image and a preset sharpness threshold; when the sharpness deviation value is not within the preset deviation range, increasing the light intensity and increasing the lens cleaning frequency; when the sharpness deviation value is within the preset deviation range, determining the original image as a material feature image.

[0053] In some examples, the material conveyor belt start signal is an electrical or logical signal that initiates the operation of the belt conveyor system. It triggers subsequent image acquisition and information processing flows and can be read via a PLC or I / O interface from the motor start status, position sensor signals, or start command signals. For example, a high-level signal is generated when the belt motor starts; receiving this signal initiates the image acquisition program. The Manufacturing Execution System (MES) is an intermediate scheduling system connecting the underlying equipment with the upper-level Enterprise Resource Planning (ERP) system. It is responsible for issuing process tasks and material formulas. Information such as material type and proportion corresponding to the current feeding task can be obtained from the MES through industrial communication protocols such as Modbus, OPC UA, and REST API. For example, the MES issues a task: "Refining Furnace 2, add 5kg of CaF2," recording "CaF2" as the preset type and 5kg as the preset physical quantity. A multispectral imaging unit deployed above the belt conveyor area is an industrial camera with multi-band imaging capabilities. Installed above the conveyor belt, it can acquire material images under different spectra, such as visible light, infrared, and ultraviolet, to improve recognition accuracy. The production execution system can be an industrial camera integrating red, green, blue, and near-infrared bands for identifying color, reflectivity, and material particle size. The raw image is unprocessed image data acquired by the multispectral imaging unit at a specific time, serving as the basic input for subsequent sharpness assessment and material identification. The image sharpness evaluation value is an indicator that quantitatively evaluates the image quality of the raw image. Methods that can be used include image gradient entropy, Laplacian variance, and frequency energy. For example, a Laplacian variance of 180 and a sharpness threshold of 200 indicate that the raw image is slightly blurry. The preset sharpness threshold is a predefined minimum acceptable standard for image sharpness; images below this standard are considered unacceptable. For example, with a sharpness threshold of 200, an image evaluation value of 180 is considered blurry. The sharpness deviation value is the numerical difference between the image sharpness evaluation value and the preset sharpness threshold, reflecting the degree of deviation between the current image and the ideal sharpness state. For example, if the image sharpness evaluation value is 180 and the preset sharpness threshold is 200, the deviation is -20. The preset deviation range is the allowable sharpness deviation interval. If the deviation exceeds this range, the current image quality is considered insufficient. For example, if the preset deviation range is ±15, the deviation is -20, therefore the image needs to be re-acquired or intervened. When the image is not sharp, the light source brightness can be automatically adjusted, and the lens maintenance frequency can be increased, such as triggering automatic lens cleaning or manual inspection; for example, the LED lighting brightness is increased from 80% to 100%, and compressed air is triggered to clean the lens once. If the sharpness meets the requirements, the current image quality is considered sufficient and can be used as a valid input image for subsequent image analysis and feature extraction.

[0054] For example, on an actual refining feeding line, when the conveyor belt starts, the system immediately retrieves a material addition task from the production execution system upon receiving the belt start signal, such as "add 8kg SiMn". Subsequently, the multispectral industrial camera above starts capturing images. The system performs a sharpness test on the acquired image and finds that the sharpness value is 195, falling within the set ±10 deviation range, indicating that the image quality is qualified. Therefore, this image is identified as a material characteristic image and enters the subsequent identification and comparison process. If the image is not clear, the system will automatically adjust the light brightness and issue a maintenance command to clean the lens, ensuring that the quality of subsequent image acquisitions is consistent and reliable.

[0055] Through the implementation of the above embodiments, the production execution system obtains feeding instructions and, combined with multispectral image acquisition and dynamic image quality evaluation mechanisms, can realize intelligent and adaptive control of image acquisition in the refining furnace feeding system. When the image quality is substandard, the lighting and cleaning strategies can be actively adjusted, effectively ensuring the image input quality, thereby providing a solid foundation for subsequent recognition and judgment, and avoiding recognition errors caused by image blurring or contamination.

[0056] In some embodiments, before acquiring the original image of the target material by a multispectral imaging unit deployed above the belt conveyor area, the refining furnace feeding control method may further include: determining the reflective properties of the target material according to a preset type in the material feeding instruction; and switching the illumination mode for the target material according to the reflective properties, wherein the illumination mode is at least one of a coaxial illumination mode, a bright field illumination mode, a dark field illumination mode, or a diffuse reflection illumination mode.

[0057] In some examples, reflectivity refers to the light reflection behavior of a material surface, which can include high reflectivity (such as metal surfaces), medium reflectivity (such as plastics), and low reflectivity (such as powders and rough ores). This can be obtained from an empirical database based on historical test samples or by measuring the material's reflectivity using sensors. For example, aluminum granules are highly reflective, lime powder is low reflective, and silicon carbide is medium reflective. The reflectivity of the target material can be automatically inferred by consulting a built-in material reflectivity table and using the type field in the "Material Addition Instruction." Based on the reflectivity of the target material, the most suitable lighting method for capturing clear images can be selected; for example, dark field lighting is used for highly reflective materials, while bright field or diffused lighting is used for powdery materials. Lighting modes are industrial lighting configurations used for material imaging. Different lighting modes affect the recognizability of features such as edges, textures, and shapes in the image. Coaxial Illumination uses a semi-transparent lens to project light perpendicularly onto the object's surface, suitable for flat and mirror-like objects, and applicable to smooth, regularly reflective metallic materials. Bright Field Illumination illuminates the object's surface directly from multiple directions, highlighting its bright and reflective areas, suitable for objects with moderate surface roughness, and applicable to granular, moderately reflective materials. Dark Field Illumination illuminates from the side at an angle, with only edges or protrusions reflecting light into the lens, suitable for highly reflective objects or objects with prominent edge features, and applicable to highly reflective materials such as metal powder and aluminum granules. Diffuse Illumination creates uniform, soft lighting through a semi-transparent or diffuser, used for photographing objects with low reflectivity or complex textures, and applicable to powder and oxide materials.

[0058] For example, when the refining furnace receives an addition instruction from the production execution system, such as "add 3kg of metallic silicon (FeSi)," it first retrieves the reflectivity of metallic silicon from the database as "high reflectivity." Then, it automatically switches the illumination mode of the multispectral imaging device to "dark field illumination" to avoid image overexposure or edge blurring caused by strong reflection. After completing the illumination mode adjustment, the image acquisition process is started to ensure that the acquired image has clear outlines and effective features, providing a high-quality data foundation for subsequent identification and comparison.

[0059] By implementing the above embodiments, the appropriate lighting mode can be dynamically switched according to the reflective characteristics of the material before image acquisition, such as coaxial illumination, bright field illumination, dark field illumination, or diffuse reflection illumination. This can significantly improve the clarity and contrast of multispectral images and enhance the feature extraction effect. In particular, on the packaging surface of highly reflective or complex materials, it can effectively suppress light spots and shadow interference, improve recognition accuracy and stability, and enhance the practicality and environmental adaptability of the method in the harsh on-site environment of steel smelting.

[0060] In some embodiments, step 104 may include: when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold, determining the anomaly assessment value of the target material based on the material verification result and the type confidence level, wherein the anomaly level is calculated and generated based on the severity of the type deviation, the proportion of the physical quantity deviation, and the confidence level deviation; in response to the anomaly assessment value being greater than a first preset risk threshold, generating a belt deceleration command and triggering an audible and visual warning signal; in response to the anomaly assessment value being greater than a second preset risk threshold, generating an emergency stop command and locking the discharge valve of the associated silo, while simultaneously pushing a material feature image to the central control terminal.

[0061] In some examples, the anomaly assessment value refers to a comprehensive quantitative score reflecting the severity of the deviation between the current target material and the preset material instruction. It is used to determine whether intervention is necessary and can be calculated using a weighted model based on multiple parameters such as type deviation, physical quantity deviation, and confidence deviation. Type deviation severity indicates the degree of difference between the actual identified type of the target material and the preset type in the instruction. This can be calculated through the difference in material classification confidence probabilities. For example, after outputting the type probability distribution from an image recognition classification model, the difference index such as Euclidean distance or KL divergence can be calculated between the model and the preset type. For example, if the preset type is "Fe" and the identified type is "FeSi", the deviation is low; if the preset type is "Fe" and the identified type is "SiO2", the deviation is severe. Physical quantity deviation ratio represents the relative difference between the estimated volume or weight of the target material and the preset physical quantity in the addition instruction. For example, if the preset physical quantity is 3 kg and the estimated physical quantity is 2.5 kg, the deviation ratio is (3-2.5) / 3 = 16.7%. The confidence level deviation represents the difference between the type confidence level of the identification model output and the set minimum reliability standard; for example, if the identification confidence level is 0.60 and the threshold is 0.85, the deviation is 0.25. The first preset risk threshold is a scoring threshold for judging "minor anomalies." Exceeding this value requires slowing down operations rather than immediately stopping the system. This threshold can be determined through historical parameter tuning of the risk assessment model, and can be set to 50 points. When a minor anomaly is detected, a belt deceleration command is actively generated to slow the conveyor rhythm, and an audible and visual warning signal is triggered to alert the operator; for example, the belt speed decreases from 1.0 m / s to 0.5 m / s, while a red light flashes and a buzzer sounds. The second preset risk threshold is the scoring boundary for judging "serious anomalies." Exceeding this threshold requires immediate shutdown measures; when a serious anomaly is detected, an emergency shutdown command is generated and the associated hopper discharge valve is locked, while a material characteristic image is pushed to the central control terminal to prevent unqualified materials from entering the refining furnace.

[0062] For example, upon receiving the addition instruction: type "SiC", preset weight "4kg"; the image recognition result is "FeSi", with a confidence level of only 0.62, and the image weight estimate is 3.0kg. The combined deviation of the three items results in an abnormality score of 82. This value exceeds the second preset threshold (80), and a stop instruction is immediately issued, the SiC hopper discharge valve is closed, and the image is pushed to the central control interface for manual review, ensuring that the raw material feeding process is safe, accurate, and controllable.

[0063] Through the implementation of the above embodiments, an anomaly assessment mechanism is introduced. After an anomaly is identified, not only is a static judgment made, but the risk level can also be dynamically calculated and a graded response can be implemented, including intervention measures such as deceleration warning and shutdown material locking. It has a high degree of safety tolerance. Especially in cases where the misaddition of critical materials may seriously affect the quality of molten steel, it can achieve rapid handling and visual feedback push, effectively ensuring the stability of the smelting process and the safety of personnel and equipment.

[0064] In some embodiments, the aforementioned determination of the anomaly assessment value of the target material based on the material verification results and type confidence level may include: determining the severity of type deviation based on the type matching status in the material verification results, wherein the severity of type deviation is negatively correlated with the metallurgical compatibility between associated materials in the preset material grade library; determining the physical quantity deviation ratio based on the ratio of the absolute deviation between the preset physical quantity and the evaluated physical quantity to the preset physical quantity; determining the confidence level deviation magnitude based on the absolute value of the difference between the preset confidence threshold and the type confidence level; and weighting the severity of type deviation, the physical quantity deviation ratio, and the confidence level deviation magnitude according to preset weighting coefficients to obtain the anomaly assessment value.

[0065] In some examples, the severity of type deviation is determined based on the type matching status in the material verification results. This is done by quantifying the severity of the deviation based on whether the assessed type matches the preset type and the metallurgical differences between the two. For example, if the types match perfectly, the severity is 0; if they belong to the same compatible material class (e.g., "Fe" and "FeSi"), a lower deviation score, such as 20, is assigned; if they are incompatible (e.g., "Al" and "SiC"), a higher score, such as 80, is assigned. The higher the metallurgical compatibility between related materials in the preset material grade library, the lower the severity of the type deviation; the worse the metallurgical compatibility between related materials in the preset material grade library, the higher the severity of the type deviation. The physical quantity deviation ratio can be determined based on the ratio of the absolute deviation between the preset physical quantity and the assessed physical quantity to the preset physical quantity. For example, if the preset physical quantity is 5 kg and the assessed physical quantity is 4.2 kg, then the physical quantity deviation ratio = |4.2 - 5| / 5 = 16%. The confidence deviation magnitude can be determined based on the absolute value of the difference between the preset confidence threshold and the type confidence level. For example, if the preset confidence threshold is 0.85 and the type confidence level is 0.62, then the confidence deviation magnitude is 0.23. Weighting coefficients can be set according to the contribution of factors such as the severity of type deviation, the proportion of physical quantity deviation, and the confidence deviation magnitude to the anomaly, and a comprehensive score, i.e., the anomaly assessment value, can be calculated.

[0066] Through the implementation of the above embodiments, the severity of type deviation, the proportion of physical quantity deviation, and the magnitude of confidence deviation are quantified and weighted and fused to form an anomaly assessment model with strong interpretability and responsiveness. This model can more accurately assess the severity of feeding deviation, which helps to realize a closed-loop logic from anomaly detection to risk quantification, providing a more decision-support basis for feeding control, and thus improving the self-diagnosis and risk perception capabilities of the method.

[0067] In some embodiments, the aforementioned extraction of the spectral feature vector and visual texture feature vector of the material feature image when parsing fails may include: when parsing fails, calculating the morphological entropy feature of the packaging area and the packaging similarity of a preset material packaging template; when the packaging similarity is greater than a preset similarity threshold, using the material type associated with the preset material packaging template as the evaluation type, and determining the type confidence based on the packaging similarity; when the packaging similarity is less than or equal to the preset similarity threshold, extracting the spectral feature vector and visual texture feature vector of the material feature image.

[0068] In some examples, morphological entropy measures the complexity of the shape and structure distribution of the packaging area in a material feature image, reflecting the orderliness of information such as packaging outline, label arrangement, and graphic texture. Packaging similarity refers to the similarity matching score between the current packaging form and the "standard material packaging template" in the database. Image processing algorithms can be used to extract morphological information such as edges, outlines, and label distribution, and then morphological entropy can be calculated using methods such as gray-level co-occurrence matrix and local binary pattern. Comparison with a preset template can be performed using cosine similarity, SSIM, or distance metrics based on depth features to calculate similarity. The preset similarity threshold is the minimum similarity requirement set to determine whether "the current packaging is sufficiently similar to the template." Precision and recall can be evaluated through historical sample testing, and can be set to, for example, 0.85 or 0.90. Different thresholds can also be configured for different material types; for example, the similarity threshold for granular materials is 0.88, and for powder packaging it is 0.92. If the current image is highly similar to the template, the material type corresponding to the template is directly used as the evaluation type. The type confidence score is then mapped from the packaging similarity. For example, a confidence score can be generated using a linear or piecewise function, such as Type Confidence = Similarity × Mapping Coefficient. If the similarity is 0.95, the type confidence score could be 0.95 or 0.95 × 0.98 ≈ 0.93. When the packaging similarity is less than or equal to a preset similarity threshold, the packaging similarity is insufficient to make a reliable judgment, and a more in-depth image analysis method will be used to extract the spectral feature vector and visual texture feature vector of the material feature image.

[0069] For example, on the production line, the packaging area is first located, and an attempt is made to identify the label information using OCR. If this fails, morphological entropy features are further extracted and compared with the "FeSi standard packaging template." If the packaging similarity reaches 0.94 (exceeding the threshold of 0.90), the material is immediately identified as "FeSi," with a type confidence level of 0.94. This eliminates the need for complex spectral and texture analysis processes, thus accelerating the identification response and improving efficiency. When the similarity is only 0.72, the process automatically switches to a deeper analysis process to ensure identification accuracy. This process fully embodies an intelligent hierarchical strategy that prioritizes simple paths and, when necessary, backs down to complex identification methods.

[0070] Through the implementation of the above embodiments, when text information cannot be parsed, type identification is achieved by comparing the package shape features with the template library, providing a reliable alternative path for material identification under complex or non-standard packaging. Especially in the material feeding environment of old, damaged or unlabeled packaging, the identification failure rate can be significantly reduced, the method's adaptability and fault tolerance can be improved, the integrity and continuity of the identification logic can be ensured, and the risk of misjudgment can be further reduced.

[0071] Furthermore, as an implementation of the aforementioned method embodiments, this application also provides a refining furnace charging control device for implementing the aforementioned method embodiments. This device embodiment corresponds to the aforementioned method embodiments. For ease of reading, this refining furnace charging control device embodiment will not repeat the details of the aforementioned method embodiments one by one, but it should be understood that the device in this application embodiment can correspondingly implement all the contents of the aforementioned method embodiments. For example... Figure 2 As shown, the refining furnace charging control device 20 includes: an image acquisition unit 201, an image evaluation unit 202, a verification and determination unit 203, and a process intervention unit 204. The image acquisition unit 201 acquires material addition instructions and material characteristic images for the target material. The material addition instructions may include a preset type and a preset physical quantity. The image evaluation unit 202 determines the evaluation type, evaluation physical quantity, and type confidence level of the target material based on the material characteristic image. The verification and determination unit 203 compares the preset type with the evaluation type and the preset physical quantity with the evaluation physical quantity to determine the material verification result. The process intervention unit 204 performs material process intervention operations when the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold.

[0072] In some implementations, the image evaluation unit 202 is further configured to determine the material contour in the material feature image using a preset target detection algorithm; determine the spatial volume parameters based on the material contour and generate the evaluation physical quantity by combining a preset density mapping table; extract the spectral feature vector and visual texture feature vector of the material feature image; perform similarity matching between the spectral feature vector and a pre-constructed material spectral library to obtain a first type probability distribution; input the visual texture feature vector into a preset classification model to obtain a second type probability distribution; and determine the evaluation type and type confidence based on the first type probability distribution and the second type probability distribution.

[0073] In some implementations, the image evaluation unit 202 is also used to extract the packaging area in the material feature image; parse the text information of the packaging area using an optical character recognition engine; when the parsing is successful, determine the material type corresponding to the text information as the evaluation type and determine the preset confidence threshold as the type confidence; when the parsing fails, extract the spectral feature vector and visual texture feature vector of the material feature image.

[0074] In some embodiments, the image acquisition unit 201 is further configured to, in response to a material conveyor belt start signal, acquire a material addition instruction from the production execution system via an industrial communication protocol; acquire an original image of the target material using a multispectral imaging unit deployed above the conveyor belt area; determine the image sharpness evaluation value of the original image and the sharpness deviation value between the original image and a preset sharpness threshold; when the sharpness deviation value is not within the preset deviation range, increase the light intensity and increase the lens cleaning frequency; and when the sharpness deviation value is within the preset deviation range, determine the original image as a material feature image.

[0075] In some embodiments, the refining furnace charging control device 20 further includes a lighting control unit for determining the reflective characteristics of the target material according to the preset type in the material adding instruction; and switching the lighting mode for the target material according to the reflective characteristics, wherein the lighting mode is at least one of coaxial lighting mode, bright field lighting mode, dark field lighting mode or diffuse reflection lighting mode.

[0076] In some implementations, the process intervention unit 204 is also used to determine the anomaly assessment value of the target material based on the material verification result and the type confidence value when the material verification result indicates an anomaly or the type confidence value is less than a preset confidence threshold. The anomaly level is calculated and generated based on the severity of the type deviation, the proportion of the physical quantity deviation, and the confidence deviation magnitude. In response to the anomaly assessment value being greater than a first preset risk threshold, a belt deceleration command is generated and an audible and visual warning signal is triggered. In response to the anomaly assessment value being greater than a second preset risk threshold, an emergency stop command is generated and the discharge valve of the associated silo is locked. At the same time, a material feature image is pushed to the central control terminal.

[0077] In some implementations, the process intervention unit 204 is also used to determine the severity of type deviation based on the type matching status in the material verification results, wherein the severity of type deviation is negatively correlated with the metallurgical compatibility between related materials in the preset material grade library; to determine the physical quantity deviation ratio based on the ratio of the absolute deviation between the preset physical quantity and the evaluated physical quantity to the preset physical quantity; to determine the confidence level deviation magnitude based on the absolute value of the difference between the preset confidence threshold and the type confidence level; and to obtain an anomaly assessment value by weighting the severity of type deviation, the physical quantity deviation ratio, and the confidence level deviation magnitude according to the preset weight coefficient.

[0078] In some implementations, the image evaluation unit 202 is also used to calculate the morphological entropy features of the packaging area and the packaging similarity of the preset material packaging template when parsing fails; when the packaging similarity is greater than the preset similarity threshold, the material type associated with the preset material packaging template is used as the evaluation type, and the type confidence is determined based on the packaging similarity; when the packaging similarity is less than or equal to the preset similarity threshold, the spectral feature vector and visual texture feature vector of the material feature image are extracted.

[0079] This application also provides a computer-readable storage medium storing computer-executable instructions or computer programs, which, when executed by a processor, will cause the processor to perform any step of the refining furnace charging control method provided in this application.

[0080] In some embodiments, the computer-readable storage medium may be a random access memory (RAM), a read-only memory (ROM), flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); or it may be a variety of devices that include one or any combination of the above-mentioned memories.

[0081] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0082] In some embodiments, computer-executable instructions may, but do not necessarily, correspond to files in a file system, and may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0083] In some embodiments, computer-executable instructions may be deployed to execute on an electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0084] like Figure 3 As shown, this application also provides an electronic device 30, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements any step of the above-described refining furnace feeding control method.

[0085] This application also provides a computer program product comprising a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer program or computer-executable instructions from the computer-readable storage medium and executes the computer program or computer-executable instructions, causing the electronic device to perform any step of the refining furnace charging control method described above.

[0086] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for controlling the charging of a refining furnace, characterized in that, include: Acquire material addition instructions and material feature images for the target material, wherein the material addition instructions include a preset type and a preset physical quantity; Based on the material feature image, determine the evaluation type, evaluation physical quantity, and type confidence level of the target material; The preset type is compared with the evaluation type, and the preset physical quantity is compared with the evaluation physical quantity to determine the material verification result; When the material verification result indicates an anomaly or the confidence level of the type is less than the preset confidence threshold, a material flow intervention operation is performed.

2. The refining furnace charging control method according to claim 1, characterized in that, The step of determining the evaluation type, evaluation physical quantity, and type confidence level of the target material based on the material feature image includes: The material outline in the material feature image is determined by a preset target detection algorithm; Based on the material profile, spatial volume parameters are determined, and the evaluation physical quantity is generated by combining the preset density mapping table. Extract the spectral feature vector and visual texture feature vector from the material feature image; The spectral feature vectors are matched with a pre-constructed material spectral library for similarity to obtain the first type of probability distribution; The visual texture feature vector is input into a preset classification model to obtain the second type probability distribution; The evaluation type and the type confidence level are determined based on the first type probability distribution and the second type probability distribution.

3. The refining furnace charging control method according to claim 2, characterized in that, The extraction of the spectral feature vector and visual texture feature vector of the material feature image includes: Extract the packaging area from the material feature image; The text information in the packaging area is parsed using an optical character recognition engine; When the parsing is successful, the material type corresponding to the text information is determined as the evaluation type, and the preset confidence threshold is determined as the confidence level of the type; When parsing fails, the spectral feature vector and visual texture feature vector of the material feature image are extracted.

4. The refining furnace charging control method according to claim 1, characterized in that, The process of acquiring the material addition instruction and material feature image for the target material includes: In response to the material conveyor belt start signal, the material addition instruction is obtained from the production execution system via an industrial communication protocol; The original images of the target material are acquired by a multispectral imaging unit deployed above the belt conveyor area; Determine the image sharpness evaluation value of the original image and the sharpness deviation value of the preset sharpness threshold; When the sharpness deviation value is not within the preset deviation range, the light intensity is increased and the lens cleaning frequency is increased; When the sharpness deviation value is within the preset deviation range, the original image is determined as the material feature image.

5. The refining furnace charging control method according to claim 4, characterized in that, Before acquiring raw images of the target material using a multispectral imaging unit deployed above the belt conveyor area, the refining furnace charging control method further includes: The reflective properties of the target material are determined according to the preset type in the material addition instruction; Based on the reflective properties, the illumination mode for the target material is switched, wherein the illumination mode is at least one of coaxial illumination mode, bright field illumination mode, dark field illumination mode, or diffuse reflection illumination mode.

6. The refining furnace charging control method according to claim 1, characterized in that, When the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold, the material flow intervention operation is performed, including: When the material verification result indicates an anomaly or the type confidence level is less than a preset confidence threshold, the anomaly assessment value of the target material is determined based on the material verification result and the type confidence level. The anomaly level is calculated and generated based on the severity of the type deviation, the proportion of the physical quantity deviation, and the confidence level deviation. In response to the anomaly assessment value being greater than a first preset risk threshold, a belt deceleration command is generated and an audible and visual warning signal is triggered. In response to the anomaly assessment value being greater than the second preset risk threshold, an emergency shutdown command is generated and the discharge valve of the associated silo is locked, while the material feature image is pushed to the central control terminal.

7. The refining furnace charging control method according to claim 1, characterized in that, The step of determining the anomaly assessment value of the target material based on the material verification results and the type confidence level includes: Based on the type matching status in the material verification results, the severity of the type deviation is determined, wherein the severity of the type deviation is negatively correlated with the metallurgical compatibility between associated materials in the preset material grade library; The deviation ratio of the physical quantity is determined based on the ratio of the absolute deviation between the preset physical quantity and the evaluated physical quantity to the preset physical quantity. The confidence deviation magnitude is determined based on the absolute value of the difference between the preset confidence threshold and the type confidence. The severity of the type deviation, the proportion of the physical quantity deviation, and the magnitude of the confidence deviation are weighted and combined according to preset weighting coefficients to obtain the anomaly assessment value.

8. A refining furnace charging control device, characterized in that, include: The image acquisition unit is used to acquire material addition instructions and material feature images for the target material, wherein the material addition instructions include a preset type and a preset physical quantity; An image evaluation unit is used to determine the evaluation type, evaluation physical quantity, and type confidence level of the target material based on the material feature image. The verification and determination unit is used to compare the preset type with the evaluation type, and the preset physical quantity with the evaluation physical quantity, to determine the material verification result; The process intervention unit is used to perform material process intervention operations when the material verification result indicates an abnormality or the confidence level of the type is less than a preset confidence threshold.

9. An electronic device, comprising: The memory and processor are characterized in that the processor, when executing a computer program stored in the memory, implements the steps of the refining furnace charging control method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the refining furnace charging control method as described in any one of claims 1-7.

Images