Method and device for identifying iron ladle number based on multi-algorithm cooperation, equipment and medium

Abstract

The application discloses a molten iron tank number recognition method and device based on multi-algorithm cooperation, equipment and medium, relates to the technical field of image recognition, and comprises the following steps: collecting a molten iron tank video stream, and recognizing a tank number digit category, an end identifier and a recognition box coordinate by using a target detection model. In combination with width, height threshold value and effective recognition area judgment, an initial tank number is generated and stored in a temporary list. A multi-frame voting decision algorithm is used to optimize the recognition result, and a period termination algorithm is used for deduplication processing. Finally, the recognition record is packaged in combination with a time stamp. The method does not need to rely on RFID hardware, is suitable for complex industrial environments, improves the tank number recognition accuracy and reliability of the molten iron tank, reduces the maintenance cost, and improves the production efficiency.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, apparatus, equipment and medium for identifying the number of molten iron ladles based on multi-algorithm collaboration. Background Technology

[0002] Currently, RFID technology is mainly used for identifying the ladle numbers in molten iron. This technology identifies the ladle number by welding RFID tags onto it and placing readers at key points along the transfer track. In addition, computer vision technology has also been attempted for ladle number identification, but most of these use single-frame recognition models and lack dedicated processing algorithms designed for complex environments.

[0003] However, RFID technology has many drawbacks: the harsh working environment of molten iron ladles, with high temperatures, dust, and vibrations causing RFID tags to age and fail easily, reader antennas to become clogged, and the welding points between the tags and the ladle body to break and detach. This results in a short lifespan for RFID sensors, high annual maintenance costs, and the need for manual intervention to verify ladle number information after a failure, which is inefficient and prone to errors. While computer vision technology can achieve contactless identification, existing solutions have low accuracy in complex scenarios such as changes in lighting, stains, and ladle vibration, failing to meet the industrial-grade reliability requirements of steel companies.

[0004] Therefore, how to achieve high-precision and high-reliability identification of molten iron ladle numbers in complex industrial environments has become an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, equipment, and medium for identifying molten iron ladle numbers based on multi-algorithm collaboration, aiming to solve the technical problem of how to improve the accuracy and reliability of molten iron ladle number identification.

[0006] To achieve the above objectives, this application proposes a method for identifying ladle numbers based on multi-algorithm collaboration, including:

[0007] Acquire video streams from molten iron ladles to obtain video frames;

[0008] The video frames are inferred by the target detection model to obtain the recognition result, wherein the recognition result includes the category of the tank number, the recognition end marker, the coordinates of the upper left corner of the tank number recognition box and the coordinates of the lower right corner of the tank number recognition box.

[0009] When the width of the recognition box is greater than the width threshold and the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, an initial can number is generated by the can number sequence splicing algorithm and the initial can number is stored in the first temporary list. The width of the recognition box is the difference between the horizontal coordinate of the lower right corner of the can number recognition box and the horizontal coordinate of the upper left corner of the can number recognition box. The height of the recognition box is the difference between the vertical coordinate of the lower right corner of the can number recognition box and the vertical coordinate of the upper left corner of the can number recognition box. The coordinates of the effective recognition area are [first horizontal coordinate threshold, second horizontal coordinate threshold], where the first horizontal coordinate threshold is less than the second horizontal coordinate threshold.

[0010] When the x-coordinate of the upper left corner of the can number identification box is less than the first x-coordinate threshold or the x-coordinate of the upper left corner of the can number identification box is greater than the second x-coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold, the single can identification result is obtained through the multi-frame voting decision algorithm, stored in the second temporary list, and the first temporary list is cleared.

[0011] When the cumulative number of frames detected by the identification end marker reaches a preset period end frame threshold and the second temporary list is not empty, the identification period end algorithm performs deduplication on the second temporary list to obtain the deduplicated data and clears the second temporary list.

[0012] The deduplicated data is combined with the current timestamp to encapsulate the data, resulting in an identification record.

[0013] In one embodiment, the step of acquiring the video stream from the molten iron ladle to obtain video frames includes:

[0014] Send dynamic acquisition commands to an industrial-grade high-definition camera, wherein the dynamic acquisition commands include exposure parameters, white balance parameters and frame rate parameters that are adjusted in real time based on the ambient light intensity. The industrial-grade high-definition camera supports real-time streaming protocols and has image stabilization capabilities.

[0015] Receive the video stream of molten iron ladle returned by the industrial-grade high-definition camera, wherein the video stream of molten iron ladle has undergone image stabilization processing by the industrial-grade high-definition camera;

[0016] The integrity of the video stream from the molten iron ladle is verified to obtain the verification result. The integrity verification includes frame continuity, frame clarity, and frame motion blur-free status.

[0017] When the verification result meets the requirements, the inter-frame difference method is used to filter out the valid video frames, wherein the inter-frame difference method determines the valid video frames by comparing the pixel change rate of adjacent frames.

[0018] When the verification result does not meet the requirements, a parameter adjustment command is sent to the industrial-grade high-definition camera to reacquire the video stream of the molten iron ladle and verify it again until a valid video frame is extracted.

[0019] The valid video frames are then format-normalized to obtain video frames.

[0020] In one embodiment, the step of reasoning through the video frame using a target detection model to obtain the recognition result includes:

[0021] A target detection model with an attention mechanism is loaded. The target detection model is trained and generated based on a labeled dataset in the iron and steel metallurgy scenario. The labeled dataset includes image samples of molten iron ladle numbers under different light intensities, different levels of dirt coverage, and different shaking states, as well as image samples of molten iron ladles completely removed from the field of view.

[0022] The video frame is first enhanced by an adaptive histogram equalization algorithm to improve the contrast between the tank number area and the background, and then noise is removed by a median filtering algorithm to obtain the processed video frame.

[0023] The processed video frames are input into the feature pyramid network of the target detection model for feature extraction to obtain multi-scale tank number features.

[0024] The multi-scale can number features are classified by the classification branch of the target detection model to obtain a classification result. The classification result includes the can number numeric category or the recognition end marker. The can number numeric category is an integer from 0 to 9, and the recognition end marker is a preset non-0-9 integer category.

[0025] The multi-scale can number features are predicted by the regression branch of the target detection model, and the coordinates of the upper left corner and the lower right corner of the can number recognition box are output.

[0026] The consistency of the category determination result, the coordinates of the upper left corner of the tank number identification box, and the coordinates of the lower right corner of the tank number identification box are checked to obtain the identification result.

[0027] In one embodiment, the step of generating an initial can number and storing the initial can number in a first temporary list when the width of the recognition frame is greater than a width threshold, the height of the recognition frame is greater than a height threshold, and the can number recognition frame is within the effective recognition area includes:

[0028] Extract the coordinates of the top left corner and the bottom right corner of the can number recognition box from the recognition result, and calculate the width and height of the recognition box.

[0029] When the width of the recognition box is greater than the width threshold, the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, the can number number sequence splicing algorithm is called to first extract the horizontal coordinate of the upper left corner of the recognition box corresponding to each can number number, and sort the numbers in ascending order of the horizontal coordinates to obtain the number arrangement order.

[0030] The horizontal coordinate spacing and vertical coordinate deviation of adjacent tank number identification frames are calculated using the tank number sequence splicing algorithm. Abnormal numbers with horizontal coordinate spacing less than a preset spacing threshold or vertical coordinate deviation greater than a preset deviation threshold are removed to obtain valid tank number numbers.

[0031] The valid tank numbers are concatenated sequentially according to the numerical arrangement order to generate an initial tank number;

[0032] After confirming that the initial tank number is a pure numeric sequence, the initial tank number is written into the first temporary list, and the video frame acquisition timestamp corresponding to the initial tank number is recorded.

[0033] In one embodiment, the step of obtaining a single-can identification result through a multi-frame voting decision algorithm, storing it in a second temporary list, and clearing the first temporary list includes:

[0034] When the x-coordinate of the top left corner of the can number recognition box is less than the first x-coordinate threshold or the x-coordinate of the top left corner of the can number recognition box is greater than the second x-coordinate threshold, and the number of initial can numbers is greater than the list length threshold, the multi-frame voting decision algorithm is called to filter the format of all initial can numbers in the first temporary list to obtain valid initial can numbers, wherein the number of initial can numbers is the initial can number count stored in the first temporary list.

[0035] Frequency statistics are performed on the valid initial tank numbers, and the number of times each valid initial tank number appears and the order of its first appearance are recorded;

[0036] The single-can identification result is obtained based on the number of occurrences and the order of first occurrence.

[0037] The single-tank identification result is associated with the corresponding tank number identification time period information and stored in the second temporary list, wherein the tank number identification time period information is the video frame time range corresponding to the single-tank identification result;

[0038] Execute the list clear command to delete all initial tank number data in the first temporary list.

[0039] In one embodiment, the step of obtaining the single-can identification result based on the number of occurrences and the order of first occurrence includes:

[0040] If there is only one valid initial tank number that appears most frequently, the corresponding valid initial tank number will be determined as the single tank identification result;

[0041] If there are multiple valid initial tank numbers with the same frequency and all of them being the highest, the valid initial tank number that appears first in the first occurrence sequence is determined as the single tank identification result.

[0042] In one embodiment, the step of performing deduplication on the second temporary list, obtaining deduplicated data, and clearing the second temporary list when the cumulative number of frames of the identification end marker reaches a preset period termination frame threshold and the second temporary list is not empty, by using the identification period termination algorithm, includes:

[0043] Initialize the frame counter to 0 and obtain the preset period termination frame threshold;

[0044] The recognition cycle termination algorithm is invoked to monitor in real time whether the recognition result contains a recognition end marker. If a new can number category is detected, the frame counter is reset to 0. If the recognition end marker is continuously detected, the frame counter is controlled to accumulate frame by frame to obtain the cumulative frame count.

[0045] When the cumulative number of frames reaches the preset period termination frame threshold and the second temporary list is not empty, a preset deduplication algorithm is invoked to extract the tank number identification time period information corresponding to each single tank identification result in the second temporary list.

[0046] Based on the tank number category and the tank number identification time period information, the priority of duplicate tank numbers is determined, and all single tank identification results are obtained after filtering.

[0047] All the filtered single-tank identification results are organized into an ordered dataset to obtain the deduplicated data.

[0048] Generate a list clearing command, which sequentially deletes all single-tank identification results and associated identification time period information from the second temporary list, thus completing the list reset.

[0049] Furthermore, to achieve the above objectives, this application also proposes a ladle number identification device based on multi-algorithm collaboration, the ladle number identification device based on multi-algorithm collaboration comprising:

[0050] The acquisition module is used to capture the video stream from the molten iron ladle and obtain video frames;

[0051] The recognition module is used to reason about the video frame through the target detection model to obtain the recognition result, wherein the recognition result includes the category of the tank number, the recognition end marker, the coordinates of the upper left corner of the tank number recognition box and the coordinates of the lower right corner of the tank number recognition box;

[0052] The can number generation module is used to generate an initial can number and store it in a first temporary list when the width of the recognition box is greater than a width threshold, the height of the recognition box is greater than a height threshold, and the can number recognition box is within the effective recognition area. The recognition box width is the difference between the horizontal coordinate of the lower right corner of the can number recognition box and the horizontal coordinate of the upper left corner of the can number recognition box. The recognition box height is the difference between the vertical coordinate of the lower right corner of the can number recognition box and the vertical coordinate of the upper left corner of the can number recognition box. The coordinates of the effective recognition area are [a first horizontal coordinate threshold, a second horizontal coordinate threshold], where the first horizontal coordinate threshold is less than the second horizontal coordinate threshold.

[0053] The can number result module is used to obtain a single can identification result through a multi-frame voting decision algorithm when the horizontal coordinate of the upper left corner of the can number identification box is less than a first horizontal coordinate threshold or the horizontal coordinate of the upper left corner of the can number identification box is greater than a second horizontal coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold. The result is then stored in the second temporary list and the first temporary list is cleared.

[0054] The optimization module is used to perform deduplication processing on the second temporary list when the cumulative number of frames of the identification end marker reaches a preset period end frame threshold and the second temporary list is not empty, by using the identification period end algorithm to obtain the deduplicated data and clear the second temporary list.

[0055] The results module is used to encapsulate the deduplicated data with the current timestamp to obtain the identification record.

[0056] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the molten iron ladle number identification method based on multi-algorithm collaboration as described above.

[0057] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the molten iron ladle number identification method based on multi-algorithm collaboration as described above.

[0058] This application acquires video streams from molten iron ladles and uses a target detection model to identify the ladle number category, end marker, and bounding box coordinates. Based on width and height thresholds and the determination of the effective recognition area, an initial ladle number is generated and stored in a temporary list. A multi-frame voting decision algorithm is used to optimize the recognition results, and a periodic termination algorithm is used for deduplication. Finally, the recognition record is encapsulated with a timestamp. This method does not rely on RFID hardware, is adaptable to complex industrial environments, improves the accuracy and reliability of molten iron ladle number identification, reduces maintenance costs, and increases production efficiency. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the first embodiment of the molten iron ladle number identification method based on multi-algorithm collaboration in this application.

[0061] Figure 2 This is a flowchart illustrating the second embodiment of the molten iron ladle number identification method based on multi-algorithm collaboration in this application.

[0062] Figure 3 This is a schematic diagram of the module structure of the molten iron ladle number identification device based on multi-algorithm collaboration, which is the first embodiment of the multi-algorithm collaborative molten iron ladle number identification method of this application.

[0063] Figure 4 This is a schematic diagram of the hardware operating environment involved in the method for identifying molten iron ladle numbers based on multi-algorithm collaboration in the embodiments of this application.

[0064] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0065] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0066] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0067] Currently, semi-supervised semantic segmentation mainly alleviates the problem of insufficient labeled data through data augmentation, loss function design, and training strategy optimization. For example, data augmentation techniques generate more diverse training samples through operations such as rotation, scaling, and cropping; in terms of loss function design, consistency regularization loss is introduced to require the model to output consistent prediction results for different augmented versions of unlabeled data; training strategy optimization includes methods such as adaptive threshold adjustment and pseudo-label generation to improve the utilization efficiency of unlabeled data. In addition, some studies have also attempted to improve the model's ability to identify minority classes by improving the model structure, such as introducing attention mechanisms or decoupling network structures.

[0068] While existing methods have alleviated the problem of insufficient labeled data to some extent, shortcomings remain. In practical applications, unlabeled data far outnumbers labeled data, making the assumption that this is untenable difficult to uphold. The class distribution of remote sensing imagery is inherently imbalanced; some minority classes may be extremely scarce in labeled data, making it difficult for models to learn effective features of these classes. Furthermore, although unlabeled data may contain abundant minority class samples, the lack of explicit supervision signals prevents these samples from generating reliable pseudo-labels, thus rendering them unusable. This distributional inconsistency not only weakens the effectiveness of semi-supervised learning but also further amplifies the inherent class imbalance problem in remote sensing semantic segmentation. Therefore, how to effectively mitigate pseudo-label bias caused by class imbalance, fully mine minority class information in unlabeled data, and enhance the model's ability to identify minority classes under limited labeling conditions has become an urgent problem to be solved.

[0069] Based on the above, this application provides a method for identifying ladle numbers based on multi-algorithm collaboration, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the molten iron ladle number identification method based on multi-algorithm collaboration of this application. In this embodiment, the molten iron ladle number identification method based on multi-algorithm collaboration includes steps S10~S60:

[0070] Step S10: Acquire the video stream from the molten iron ladle to obtain video frames.

[0071] It should be noted that the molten iron ladle video stream refers to a time-series collection of images generated by a camera device continuously capturing the movement of molten iron ladles at the production site. Through high-speed, continuous framing, it records the appearance of the molten iron ladle along the transport route, providing the most basic analog or digital signals for subsequent extraction of key visual features and generation of recognition commands. A video frame refers to a single still image obtained by breaking down the acquired continuous video stream into chronological order; it is the smallest independent information unit constituting the video data stream. Each image contains static pixel distribution information of the molten iron ladle at a specific moment.

[0072] Specifically, step S10 includes: sending a dynamic acquisition command to an industrial-grade high-definition camera; receiving a video stream of molten iron from the industrial-grade high-definition camera, wherein the video stream of molten iron has undergone image stabilization processing by the industrial-grade high-definition camera; performing integrity verification on the video stream of molten iron to obtain a verification result; when the verification result meets the requirements, using the inter-frame difference method to filter out valid video frames; when the verification result does not meet the requirements, sending a parameter adjustment command to the industrial-grade high-definition camera, re-acquiring the video stream of molten iron and verifying it again, until valid video frames are extracted; and performing format standardization processing on the valid video frames to obtain video frames.

[0073] It should be noted that industrial-grade high-definition cameras refer to sensing devices capable of stable operation and outputting high-resolution visual signals in environments with high temperatures, dust, and strong electromagnetic interference. These devices possess high dynamic range light sensitivity adjustment capabilities, meaning they can finely adjust the image output quality according to rapid fluctuations in external lighting conditions. Industrial-grade high-definition cameras support real-time streaming protocols and have image stabilization functions. Real-time streaming protocols refer to application-layer protocols used to establish and control media streaming transmission over a network; they are standardized communication specifications that enable real-time video data interaction between the control terminal and the camera device. Image stabilization refers to the shake compensation algorithm or mechanical balancing mechanism built into the camera device, which improves image clarity by counteracting the physical vibrations generated by the operation of large equipment in industrial settings. Dynamic acquisition commands refer to a set of control commands containing real-time adjustment parameters, including exposure parameters, white balance parameters, and frame rate parameters adjusted in real-time based on the ambient light intensity. Integrity verification includes frame continuity, frame sharpness, and frame free motion. Inter-frame difference (IPD) is an algorithm that extracts moving targets by comparing pixel differences between adjacent images in a video sequence. Specifically, it calculates the rate of change of corresponding pixels between two consecutive frames and determines the valid video frames by comparing the pixel change rates of adjacent frames. Format standardization refers to the standardized conversion operation performed on the extracted original images, that is, standardizing the size, color space, and encoding format of the video frames according to the standards required by subsequent recognition algorithms.

[0074] Specifically, firstly, a dynamic acquisition command is sent to the industrial-grade high-definition camera. This command includes exposure parameters, white balance parameters, and frame rate parameters that are adjusted in real time according to the ambient light intensity. This ensures the camera can acquire high-quality video streams under different lighting conditions, while adjusting the frame rate guarantees the smoothness and real-time performance of the video stream. The industrial-grade high-definition camera supports the Real-Time Streaming Protocol (RTSP) and has image stabilization capabilities, effectively reducing image jitter caused by ladle movement or environmental vibrations. Secondly, the video stream from the ladle is received from the industrial-grade high-definition camera. Because the camera has image stabilization, the returned video stream has already undergone preliminary processing, reducing image blur or jitter caused by external factors, thus improving the quality of the video stream. Then, the integrity of the ladle video stream is verified, including frame continuity, frame sharpness, and the absence of motion blur. This step ensures that each frame in the video stream meets the requirements for subsequent processing, avoiding issues such as frame loss, blurring, or motion blur that could affect recognition accuracy. When the verification results meet the requirements, the inter-frame difference method is used to filter out valid video frames. The inter-frame difference method determines which frames contain important dynamic information by comparing the pixel change rate of adjacent frames, thus filtering out valid frames containing the movement of the molten iron ladle or other key changes, providing more valuable data for subsequent target detection. When the verification result does not meet the requirements, a parameter adjustment command is sent to the industrial-grade high-definition camera to reacquire the molten iron ladle video stream and verify it again. This process is repeated until a valid video frame that meets the requirements is extracted. This dynamic adjustment and verification mechanism ensures that high-quality video frames are always acquired in complex environments, thereby improving the robustness of the entire system. Finally, the selected valid video frames are format-normalized to ensure that the resolution, color format, and other parameters of the video frames meet the input requirements of the target detection model. Format normalization is necessary because different cameras and acquisition environments lead to inconsistent video frame formats, and normalization ensures the consistency of input data, thereby improving the model's recognition accuracy.

[0075] Step S20: The video frames are inferred using the object detection model to obtain the recognition result.

[0076] It should be noted that the object detection model is a lightweight model that incorporates an attention mechanism, improving the accuracy of ladle number recognition by assigning higher weights to key regions. The recognition results include the ladle number category, the end-of-recognition marker, and the coordinates of the top-left and bottom-right corners of the ladle number recognition box.

[0077] Further, step S20 includes: loading a target detection model with a fusion attention mechanism; first, enhancing the contrast between the can number region and the background by using an adaptive histogram equalization algorithm on the video frame, and then removing noise by using a median filtering algorithm to obtain a processed video frame; inputting the processed video frame into the feature pyramid network of the target detection model for feature extraction to obtain multi-scale can number features; performing category determination on the multi-scale can number features through the classification branch of the target detection model to obtain the category determination result; performing coordinate prediction on the multi-scale can number features through the regression branch of the target detection model to output the coordinates of the upper left corner and the lower right corner of the can number recognition box; performing consistency verification on the category determination result, the coordinates of the upper left corner and the lower right corner of the can number recognition box to obtain the recognition result.

[0078] It should be noted that the object detection model includes a feature pyramid network, a classification branch, and a regression branch. The feature pyramid network is a neural network topology used to extract multi-level spatial information from images, fusing deep semantics with shallow geometric details through a top-down path. The classification branch refers to the parallel computation path in the object detection model responsible for determining the logical attributes of the target, that is, determining the specific numerical label or state identifier corresponding to the current visual features through logical deduction. This branch maps the extracted image features to a preset numerical classification, achieving digital semantic transformation of numbers 0 to 9 and task termination signals. The regression branch refers to the computation module in the model used to predict the specific spatial location of the target, that is, accurately delineating the rectangular area where the ladle number is located by fitting coordinate offsets. This branch outputs specific geometric coordinate values, which are used to determine the precise physical boundaries of the recognition box in the pixel coordinate system of the video image, thereby guiding the target localization. The object detection model is trained and generated based on a labeled dataset in a steel metallurgy scenario. The labeled dataset includes image samples of molten iron ladle numbers under different lighting intensities, different levels of dirt coverage, different shaking states, and image samples of molten iron ladles completely removed from the field of view. Adaptive histogram equalization is a digital processing technique that improves visual clarity by locally enhancing image contrast. It dynamically adjusts the image based on the grayscale distribution of different regions. This algorithm effectively highlights the ladle number outline in dark or reflective areas on the ladle's outer wall, solving the problem of numbers being undetectable by conventional recognition algorithms due to uneven lighting or metal reflection. Median filtering is a non-linear smoothing method used to remove impulse noise from images. It uses the median of the grayscale values ​​of a pixel's neighborhood to replace noisy pixels for noise reduction. The category determination result includes the ladle number category or an end-of-recognition marker. The ladle number category is an integer from 0 to 9, while the end-of-recognition marker is a preset category of non-0-9 integers. The ladle number category refers to the specific numerical label corresponding to each digit in the identified ladle number, i.e., a specific character classification from the ten Arabic numerals from 0 to 9. Each category corresponds to a unique physical identifier sprayed on the ladle's outer wall, serving as the core data basis for generating complete ladle numbers and conducting industrial measurement. The recognition end marker is a specific visual feature or logical signal used to determine whether the current molten iron ladle recognition process is complete. In essence, it's a symbol in the video sequence representing that the target has completely passed or the recognition task has ended. This marker triggers a periodic termination algorithm, ensuring accurate delineation of recognition boundaries between different molten iron ladles.

[0079] Specifically, firstly, an object detection model incorporating an attention mechanism is loaded. This model is trained on a dataset of steel metallurgical scenes, including extreme conditions such as strong light interference, dirt occlusion, and physical vibration, as well as samples of molten iron ladles completely removed from the field of view. This leverages prior features learned from specific environments to enhance robustness in complex and dynamic production environments. Secondly, adaptive histogram equalization and median filtering algorithms are sequentially applied to the acquired video frames. This involves first dynamically adjusting local contrast to enhance the outlines of previously dim or reflective ladle numbers, and then using nonlinear smoothing techniques to remove random salt-and-pepper noise generated by the industrial environment. This preserves the details of the digital edges while providing high-quality image input for subsequent feature extraction. Then, the processed video frames are fed into a feature pyramid network for deep convolution processing. Multi-level feature fusion technology generates multi-scale ladle number features covering different resolutions. This ensures stable recognition of ladle numbers at different operating distances and scaling ratios by integrating deep semantic information with shallow geometric details. Next, the classification branch performs attribute deduction on multi-scale features, outputting specific 0-9 digit categories or non-digit recognition end markers representing the removal of the molten iron ladle from the viewpoint. Simultaneously, the regression branch predicts coordinates on the feature map and calculates the pixel values ​​of the top-left and bottom-right corners of the ladle number recognition box. This parallel computing logic achieves simultaneous locking of the meaning of the ladle number and its spatial physical location. Finally, the output category determination result, status marker, and coordinates of both vertices are checked for consistency. This logical-level secondary matching verification eliminates abnormal prediction values ​​or false detection noise that do not conform to physical laws, thus encapsulating and outputting recognition results that truly reflect the identity information of the molten iron ladle.

[0080] Step S30: When the width of the recognition box is greater than the width threshold and the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, an initial can number is generated by the can number sequence splicing algorithm and the initial can number is stored in the first temporary list.

[0081] It should be noted that the width of the recognition frame is the difference between the x-coordinate of the bottom right corner and the x-coordinate of the top left corner of the recognition frame, and the height of the recognition frame is the difference between the y-coordinate of the bottom right corner and the y-coordinate of the top left corner of the recognition frame. The coordinates of the effective recognition area are [first x-coordinate threshold, second x-coordinate threshold]. The first x-coordinate threshold is the left boundary pixel value of the effective recognition area, used to define the left threshold point where the can number enters the effective recognition range. The second x-coordinate threshold is the right boundary pixel value of the effective recognition area, used to define the right threshold point where the can number leaves the effective recognition range, and the first x-coordinate threshold is less than the second x-coordinate threshold. The width threshold is a preset numerical standard for determining whether the horizontal dimension of the recognition frame is valid. Only when the width of the recognition frame exceeds this value can the can number be considered completely recognizable in the horizontal direction. The height threshold is a preset numerical standard for determining whether the vertical dimension of the recognition frame is valid. The height of the recognition frame must be greater than this value to confirm that the can number has no missing critical parts in the vertical direction.

[0082] Further, step S30 includes: extracting the coordinates of the upper left corner and the lower right corner of the can number recognition box from the recognition result, and calculating the width and height of the recognition box; when the width of the recognition box is greater than the width threshold, the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, the can number number sequence splicing algorithm is called to first extract the horizontal coordinate of the upper left corner of the recognition box corresponding to each can number, and sort the numbers in ascending order of the horizontal coordinates to obtain the number arrangement order; the horizontal coordinate spacing and vertical coordinate deviation of adjacent can number number recognition boxes are calculated by the can number number sequence splicing algorithm, and abnormal numbers with horizontal coordinate spacing less than a preset spacing threshold or vertical coordinate deviation greater than a preset deviation threshold are removed to obtain the effective can number numbers; the effective can number numbers are spliced ​​sequentially in the number arrangement order to generate the initial can number; after confirming that the initial can number is a pure number sequence, the initial can number is written into the first temporary list, and the video frame acquisition timestamp corresponding to the initial can number is recorded.

[0083] Specifically, firstly, the top-left and bottom-right pixel coordinates of the can number recognition box are extracted from the data output by the target detection model. Based on this, the physical width and height of the recognition box are calculated. The purpose is to filter out non-target candidate boxes caused by background interference, dust, or long-distance clutter through geometric dimensions, ensuring that all subsequent processing only targets are valid. Secondly, after confirming that the width and height of the recognition box both exceed preset thresholds and are within the valid recognition area, a digit sequence concatenation algorithm is called to extract the horizontal coordinate of the top-left corner of each digit detection box and arrange them in ascending order, thereby logically determining the physical reading order of the can number digits from left to right. Then, the algorithm further calculates the horizontal spacing and vertical height deviation between adjacent digit recognition boxes in the sequence, eliminating abnormal numbers with excessively small spacing due to detection overlap or excessive vertical offset due to false detection, ensuring that the final retained numbers have a high degree of linear consistency in spatial distribution. Finally, the selected valid numbers are concatenated in the determined order to generate the initial can number string, and it is confirmed that the generated initial can number string is a pure number sequence. If so, the initial tank number is written to the first temporary list, and the corresponding video frame capture timestamp is recorded. This step is to ensure that the tank numbers written to the list are valid and to record their source via timestamps for subsequent tracing and processing.

[0084] Step S40: When the x-coordinate of the upper left corner of the can number recognition box is less than the first x-coordinate threshold or the x-coordinate of the upper left corner of the can number recognition box is greater than the second x-coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold, the single can recognition result is obtained through the multi-frame voting decision algorithm, stored in the second temporary list, and the first temporary list is cleared.

[0085] It should be noted that the list length threshold is a preset numerical standard for determining whether the number of initial ladle numbers in the first temporary list is sufficient for voting decisions. This value can be adjusted to suit actual working conditions to ensure sufficient data to support the voting results. The multi-frame voting decision algorithm filters the optimal result by statistically analyzing the frequency of occurrence of multiple initial ladle numbers in the first temporary list. This algorithm can offset the random errors of single-frame recognition and improve the accuracy of ladle number recognition. The single-ladle recognition result is the final ladle number of a single molten iron ladle determined after filtering by the multi-frame voting decision algorithm. The second temporary list is a data storage container used to store the single-ladle recognition results of all molten iron ladles during a single transfer. The first temporary list is a data storage container used to temporarily store the initial ladle numbers generated by continuous frame recognition of a single molten iron ladle.

[0086] Specifically, the system first determines whether the x-coordinate of the top-left corner of the can number recognition frame is less than a first x-coordinate threshold or greater than a second x-coordinate threshold. These two thresholds define the boundaries of the effective recognition area. When the recognition frame exceeds this area, it indicates that the can number has been moved out of or is about to be moved out of the effective recognition range. Simultaneously, it checks whether the number of initial can numbers stored in the first temporary list is greater than a preset list length threshold. The list length threshold ensures sufficient frame data for accurate voting decisions, avoiding misjudgments due to insufficient data.

[0087] When the above conditions are met simultaneously, the multi-frame voting decision algorithm is invoked. This algorithm performs statistical analysis on all initial can numbers in the first temporary list, determining the final single-can identification result by calculating the frequency of occurrence of each initial can number. The purpose of this step is to leverage the advantages of multi-frame data to eliminate errors caused by single-frame noise or misidentification, thereby improving the accuracy and reliability of the identification.

[0088] Finally, the single-tank identification result is stored in the second temporary list, and the first temporary list is cleared. Clearing the first temporary list frees up storage space for the identification process of the next tank number, ensuring the efficient operation of the system. Simultaneously, the single-tank identification result is stored in the second temporary list to prepare for further processing and the generation of the final result.

[0089] Step S50: When the cumulative number of frames that indicate the end of identification reaches the preset period termination frame threshold and the second temporary list is not empty, the second temporary list is deduplicated by the identification period termination algorithm to obtain the deduplicated data and the second temporary list is cleared.

[0090] It should be noted that the recognition cycle termination algorithm is used to determine whether the recognition work for a single molten iron ladle transfer is complete. Its core mechanism involves monitoring the cumulative number of frames from the recognition end marker and the status of the second temporary list to achieve precise closed-loop control of the recognition cycle. The recognition end marker is a preset integer category other than 0 to 9 (such as 10, 11, 99, etc.), specifically used to mark the state where the molten iron ladle has completely moved out of the camera's field of view, and is the core signal triggering the recognition cycle termination judgment. The cumulative frame count is the number of video frames continuously detected by the recognition end marker, serving as a quantitative basis for determining whether the molten iron ladle has completely left the recognition range. The preset cycle termination frame threshold is a preset standard for determining whether the recognition cycle has ended. This value is calibrated by combining the molten iron ladle's moving speed and the effective recognition area width to ensure that the cycle only terminates after the molten iron ladle has completely left.

[0091] Deduplication is an operation that filters the single-tank identification results stored in the second temporary list. The core is to remove duplicate tank number records and retain only the unique tank number information.

[0092] Further, step S50 includes: initializing the frame counter to 0 and obtaining a preset period termination frame threshold; calling the recognition period termination algorithm to monitor in real time whether the recognition result contains a recognition end marker; if a new can number digit category is detected, the frame counter is reset to 0; if a recognition end marker is continuously detected, the frame counter is controlled to accumulate frame by frame to obtain the cumulative frame count; when the cumulative frame count reaches the preset period termination frame threshold and the second temporary list is not empty, a preset deduplication algorithm is called to extract the can number recognition time period information corresponding to each single can recognition result in the second temporary list; the priority of duplicate can numbers is determined based on the can number digit category and the can number recognition time period information to obtain all single can recognition results after filtering; all single can recognition results after filtering are organized into an ordered dataset to obtain the deduplicated data; a list clearing instruction is generated to delete all single can recognition results and associated recognition time period information in the second temporary list in turn to complete the list reset.

[0093] It should be noted that the frame counter is a counting tool used to count the number of consecutive frames that indicate the end of recognition. Its initial value is fixed at 0, and its value changes directly reflect the continuous detection duration of the end of recognition. The preset periodic end frame threshold is a frame count standard calibrated in advance based on the ladle's moving speed and the effective recognition area width. Only when the cumulative frame count reaches this value is the recognition process for a single ladle transfer considered complete. The new ladle number category is an integer category within the range of 0 to 9 identified by the target detection model, representing that there are still identifiable ladle numbers in the video frames. This is a key condition for triggering the frame counter reset. The cumulative frame count is the value accumulated frame by frame by the frame counter as it continuously detects the end of recognition, i.e., the total number of video frames in which the end of recognition appears consecutively. This value is used to compare with the preset periodic end frame threshold. The preset deduplication algorithm is a pre-configured algorithm for processing the second temporary list data. Unlike simply deleting duplicates, it prioritizes duplicate ladle numbers based on the ladle number recognition time period information. The ladle number identification time period information is the video frame time range corresponding to the single ladle identification result. It represents the time interval from when a single ladle number enters the valid identification area to when it leaves, and is composed of the video frame acquisition timestamps corresponding to the first and last initial ladle numbers in the first temporary list. The priority for duplicate ladle numbers is a filtering rule set for single ladle identification results of the same number category in the second temporary list, prioritizing the retention of single ladle identification results with more complete identification time periods. The ordered dataset is a structured data set formed by organizing the filtered single ladle identification results according to the ladle transfer order, with the data arrangement conforming to the usage logic of production scheduling.

[0094] Specifically, the frame counter is first initialized to 0 and a preset period termination frame threshold is loaded. A period termination recognition algorithm is used to monitor the status bits in the video stream in real time. Once a new ladle number category is detected, the system immediately resets the counter to 0 to maintain monitoring activity. When a recognition end marker is continuously detected, the counter is incremented frame by frame. This is done to determine whether the molten iron ladle has truly and completely moved out of view through a continuous negative feedback delay confirmation mechanism, effectively filtering false task termination signals caused by light and shadow fluctuations or brief obstructions. Secondly, when the accumulated frame count reaches the preset threshold and the second temporary list is not empty, a deduplication process is initiated, extracting the recognition time period information corresponding to each recognition result in the list. The program prioritizes duplicate records based on the completeness of the ladle number, recognition frequency, and time period distribution, eliminating redundant recognition items caused by ladle dwelling, reversal, or multiple triggers, and retaining the set of results with the highest confidence. This is done to ensure that each physical ladle corresponds to only one unique and accurate digital file in a complex production scheduling environment, avoiding the risk of duplicate measurement. Finally, the filtered single-tank identification results are organized into an ordered dataset according to time or process logic, and a list clearing instruction is generated simultaneously to completely erase the historical data and associated metadata information in the second temporary list. This not only achieves standardized output of identification results, but also completes the normalization reset of system logic and cache space for the start of the next identification cycle.

[0095] Step S60: Encapsulate the deduplicated data with the current timestamp to obtain the identification record.

[0096] It should be noted that the current timestamp records the time information when the deduplication process of the identification process is completed, accurate to the millisecond level. It is directly linked to the system's real-time clock and is used to mark the specific time node when the ladle number identification result is generated, providing a time basis for production traceability. The identification record is a complete data record formed after the packaging operation is completed. It contains all the unique ladle number information for a single trip of molten iron ladle transfer and the corresponding generation timestamp. It is the final output of the entire ladle number identification process and can be directly used for production scheduling, molten iron traceability, and other tasks.

[0097] Specifically, the process begins by acquiring deduplicated data, which consists of the final single-tank identification results after filtering and processing, ensuring the uniqueness and accuracy of each tank number. Then, this data is encapsulated with the current timestamp. The timestamp records the specific time of each identification result, providing a time reference for subsequent traceability and analysis. Finally, the deduplicated data and timestamps are combined into a complete identification record. This process aims to generate a structured and standardized identification result, facilitating storage, retrieval, and subsequent production scheduling.

[0098] This embodiment acquires video streams from molten iron ladles and uses a target detection model to identify the ladle number category, end marker, and identification box coordinates. Based on width and height thresholds and the determination of the effective identification area, an initial ladle number is generated and stored in a temporary list. A multi-frame voting decision algorithm is used to optimize the identification results, and a periodic termination algorithm is used for deduplication. Finally, the identification record is encapsulated with a timestamp. This method does not rely on RFID hardware, is adaptable to complex industrial environments, improves the accuracy and reliability of molten iron ladle number identification, reduces maintenance costs, and increases production efficiency.

[0099] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The method for identifying ladle numbers based on multi-algorithm collaboration, step S40, further includes steps S201 to S205:

[0100] Step S201: When the x-coordinate of the upper left corner of the can number recognition box is less than the first x-coordinate threshold or the x-coordinate of the upper left corner of the can number recognition box is greater than the second x-coordinate threshold, and the number of initial can numbers is greater than the list length threshold, the multi-frame voting decision algorithm is called to perform format filtering on all initial can numbers in the first temporary list to obtain valid initial can numbers.

[0101] It should be noted that the initial tank number count is a statistical count of the initial tank numbers stored in the first temporary list. It is a value obtained after counting all the initial tank numbers stored in the first temporary list, reflecting the total number of initial tank numbers generated by single-tank continuous frame recognition.

[0102] Specifically, the algorithm first determines whether the x-coordinate of the top-left corner of the can number recognition box is less than a first x-coordinate threshold or greater than a second x-coordinate threshold. These two thresholds define the effective area for can number recognition. When the recognition box exceeds this area, it indicates that the can number may have moved out of or is about to move out of the effective recognition range. Simultaneously, it checks whether the number of initial can numbers stored in the first temporary list is greater than a preset list length threshold. This threshold ensures sufficient frame data for multi-frame voting decisions, avoiding misjudgments due to insufficient data. When both conditions are met, the multi-frame voting decision algorithm is invoked. The algorithm first performs format filtering on all initial can numbers in the first temporary list, removing can numbers that do not meet the format requirements (such as non-pure numeric sequences), thus obtaining valid initial can numbers. This step ensures the data quality of subsequent processing, preventing incorrectly formatted can numbers from affecting the accuracy of the final result.

[0103] Step S202: Count the frequency of valid initial tank numbers and record the number of times each valid initial tank number appears and the order in which it first appears.

[0104] It should be noted that the occurrence count refers to the number of times a single valid initial can number appears repeatedly in the first temporary list. A higher value indicates that the can number has been confirmed more times in multi-frame recognition, which is a key quantitative indicator for determining the reliability of the can number. The first occurrence order refers to the order in which each valid initial can number is first stored in the first temporary list. It is used as a supplementary basis for priority determination when multiple valid initial can numbers have the same occurrence count.

[0105] Specifically, the first step is to perform frequency statistics on the valid initial can numbers after format filtering. This step involves iterating through all valid initial can numbers in the first temporary list and counting the number of times each can number appears. The purpose of this is to determine which can numbers are repeatedly identified across multiple frames, thereby improving the reliability of the identification results. Then, the order in which each valid initial can number first appears is recorded. Simultaneously with frequency statistics, the time sequence or frame number of each can number's first appearance is recorded. This step ensures that, in cases of identical frequencies, the earliest appearing can number is selected as the final result based on its time sequence. This approach effectively handles the problem of repeated identification caused by noise or transient interference, ensuring the stability and consistency of the identification results.

[0106] Step S203: Obtain the single-can identification result based on the number of occurrences and the order of first occurrences.

[0107] It should be noted that if there is only one valid initial tank number with the highest frequency, the corresponding valid initial tank number will be determined as the single tank identification result; if there are multiple valid initial tank numbers with the same frequency and all of them are the highest, the valid initial tank number that appears first in the order of first appearance will be determined as the single tank identification result.

[0108] Specifically, the frequency statistics are first checked to determine if there is a single valid initial can number that appears most frequently. If only one such can number exists, it is directly identified as the single-can recognition result. This is because a high frequency indicates that the can number has been repeatedly identified across multiple frames, possessing high reliability, and therefore can be directly used as the final result. Then, if multiple valid initial can numbers have the same frequency and are all the highest, further judgment is needed. In this case, the earliest appearing valid initial can number in the order of first appearance is identified as the single-can recognition result. The logic behind this step is that the earliest appearing can number is more likely to be the correct recognition result, as subsequent repeated recognitions may be misjudged due to brief noise or interference. This method effectively resolves conflicts in cases of identical frequencies, ensuring the highest reliability of the final single-can recognition result.

[0109] Step S204: Associate the single tank identification result with the corresponding tank number identification time period information and store it in the second temporary list.

[0110] Specifically, the first step is to associate the individual can identification result with the corresponding can number identification time period information. This step involves matching the timestamp or frame number of the individual can identification result with the identification time period information (e.g., the time range between the first and last appearances of the can number), ensuring that each identification result has a clear temporal context. This is done to provide more complete identification information, facilitating subsequent tracing and analysis. Then, the associated individual can identification result and its corresponding can number identification time period information are stored in a second temporary list. This step is to temporarily save the reliable identification results selected by the multi-frame voting decision algorithm, preparing for the subsequent identification cycle termination process. Storing the identification results in the second temporary list ensures that these results are not lost in subsequent processing and can be further processed and analyzed together with other identification results.

[0111] Step S205: Execute the list clearing command to delete all initial tank number data in the first temporary list.

[0112] Specifically, the first step is to issue a list clearing command, which clears all data in the first temporary list. After executing this command, all initial ladle number data stored in the first temporary list will be deleted one by one. This step aims to free up storage space for the next ladle number identification cycle, ensuring efficient system operation and avoiding processing delays or errors caused by data accumulation. By clearing the first temporary list, a new identification cycle can begin, thus enabling continuous and stable identification of molten iron ladle numbers.

[0113] In this embodiment, when the can number recognition frame exceeds the effective area and the initial number of can numbers is sufficient, the initial can numbers in the first temporary list are format-filtered using a multi-frame voting decision algorithm. The frequency and first appearance order of the effective can numbers are counted to determine the single can recognition result. The recognition time period information is associated with the result and stored in the second temporary list. Finally, the first temporary list is cleared. This effectively improves the recognition accuracy and reliability, ensures efficient system operation, and reduces misjudgments and data redundancy.

[0114] Based on the first embodiment of this application, this application also provides a ladle number identification device based on multi-algorithm collaboration, please refer to... Figure 3 The device includes:

[0115] Acquisition module 10 is used to acquire the video stream of the molten iron ladle and obtain video frames;

[0116] The recognition module 20 is used to infer the video frame through the target detection model to obtain the recognition result, which includes the category of the tank number, the recognition end mark, the coordinates of the upper left corner of the tank number recognition box and the coordinates of the lower right corner of the tank number recognition box;

[0117] The can number generation module 30 is used to generate an initial can number and store it in a first temporary list when the width of the recognition box is greater than the width threshold, the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area. The recognition box width is the difference between the horizontal coordinate of the lower right corner of the can number recognition box and the horizontal coordinate of the upper left corner of the can number recognition box, the recognition box height is the difference between the vertical coordinate of the lower right corner of the can number recognition box and the vertical coordinate of the upper left corner of the can number recognition box, and the coordinates of the effective recognition area are [first horizontal coordinate threshold, second horizontal coordinate threshold], where the first horizontal coordinate threshold is less than the second horizontal coordinate threshold.

[0118] The can number result module 40 is used to obtain a single can identification result through a multi-frame voting decision algorithm when the horizontal coordinate of the upper left corner of the can number identification box is less than the first horizontal coordinate threshold or the horizontal coordinate of the upper left corner of the can number identification box is greater than the second horizontal coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold. The result is stored in the second temporary list and the first temporary list is cleared.

[0119] The optimization module 50 is used to perform deduplication on the second temporary list when the cumulative number of frames that detect the end of recognition reaches the preset period termination frame threshold and the second temporary list is not empty, by using the recognition period termination algorithm to obtain the deduplicated data and clear the second temporary list.

[0120] The result module 60 is used to encapsulate the deduplicated data with the current timestamp to obtain the identification record.

[0121] The molten iron ladle number identification device based on multi-algorithm collaboration provided in this application adopts the molten iron ladle number identification method based on multi-algorithm collaboration in the above embodiments, which can solve the technical problem of how to improve the accuracy and reliability of molten iron ladle number identification. Compared with the prior art, the beneficial effects of the molten iron ladle number identification device based on multi-algorithm collaboration provided in this application are the same as the beneficial effects of the molten iron ladle number identification method based on multi-algorithm collaboration provided in the above embodiments, and other technical features in the molten iron ladle number identification device based on multi-algorithm collaboration are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0122] This application provides a ladle number identification device based on multi-algorithm collaboration. The ladle number identification device based on multi-algorithm collaboration includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the ladle number identification method based on multi-algorithm collaboration in the above embodiment 1.

[0123] The following is for reference. Figure 4This document illustrates a structural schematic diagram of a molten iron ladle number identification device suitable for implementing embodiments of this application. The molten iron ladle number identification device in this application embodiment can include, but is not limited to, mobile terminals such as laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The illustrated molten iron ladle number identification device based on multi-algorithm collaboration is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0124] like Figure 4 As shown, the ladle number identification device based on multi-algorithm collaboration may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the ladle number identification device based on multi-algorithm collaboration. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the multi-algorithm collaborative ladle number identification device to communicate wirelessly or wiredly with other devices to exchange data. Although various multi-algorithm collaborative ladle number identification devices are shown in the figures, it should be understood that implementation or possession of all shown is not required. More or fewer of these devices may be implemented alternatively.

[0125] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0126] The molten iron ladle number identification device based on multi-algorithm collaboration provided in this application, employing the molten iron ladle number identification method based on multi-algorithm collaboration in the above embodiments, can solve the technical problem of how to improve the accuracy and reliability of molten iron ladle number identification. Compared with the prior art, the beneficial effects of the molten iron ladle number identification device based on multi-algorithm collaboration provided in this application are the same as those of the molten iron ladle number identification method based on multi-algorithm collaboration provided in the above embodiments, and other technical features in this molten iron ladle number identification device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0127] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0129] This application provides a computer-readable medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the molten iron ladle number identification method based on multi-algorithm collaboration in the above embodiments.

[0130] The computer-readable medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0131] The aforementioned computer-readable medium may be included in a multi-algorithm collaborative molten iron ladle number identification device; or it may exist independently and not be assembled into a multi-algorithm collaborative molten iron ladle number identification device.

[0132] The aforementioned computer-readable medium carries one or more programs that, when executed by a multi-algorithm collaborative molten iron ladle number identification device, enable the device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java and C++—and conventional procedural programming languages—such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0133] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, all blocks in the flowcharts or block diagrams may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that all blocks in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based implementations that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0134] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0135] The readable medium provided in this application is a computer-readable medium, which stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for identifying molten iron ladle numbers based on multi-algorithm collaboration. This method can solve the technical problem of how to improve the accuracy and reliability of molten iron ladle number identification. Compared with the prior art, the beneficial effects of the computer-readable medium provided in this application are the same as those of the molten iron ladle number identification method based on multi-algorithm collaboration provided in the above embodiments, and will not be repeated here.

[0136] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for identifying molten iron ladle numbers based on multi-algorithm collaboration.

[0137] The computer program product provided in this application can solve the technical problem of how to improve the accuracy and reliability of molten iron ladle number identification. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the molten iron ladle number identification method based on multi-algorithm collaboration provided in the above embodiments, and will not be repeated here.

[0138] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A molten iron ladle number recognition method based on multi-algorithm cooperation, characterized in that, The method includes: Acquire video streams from molten iron ladles to obtain video frames; The video frames are inferred by the target detection model to obtain the recognition result, wherein the recognition result includes the category of the tank number, the recognition end marker, the coordinates of the upper left corner of the tank number recognition box and the coordinates of the lower right corner of the tank number recognition box. When the width of the recognition box is greater than the width threshold and the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, an initial can number is generated by the can number sequence splicing algorithm and the initial can number is stored in the first temporary list. The width of the recognition box is the difference between the horizontal coordinate of the lower right corner of the can number recognition box and the horizontal coordinate of the upper left corner of the can number recognition box. The height of the recognition box is the difference between the vertical coordinate of the lower right corner of the can number recognition box and the vertical coordinate of the upper left corner of the can number recognition box. The coordinates of the effective recognition area are [first horizontal coordinate threshold, second horizontal coordinate threshold], where the first horizontal coordinate threshold is less than the second horizontal coordinate threshold. When the x-coordinate of the upper left corner of the can number identification box is less than the first x-coordinate threshold or the x-coordinate of the upper left corner of the can number identification box is greater than the second x-coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold, the single can identification result is obtained through the multi-frame voting decision algorithm, stored in the second temporary list, and the first temporary list is cleared. When the cumulative number of frames detected by the identification end marker reaches a preset period end frame threshold and the second temporary list is not empty, the identification period end algorithm performs deduplication on the second temporary list to obtain the deduplicated data and clears the second temporary list. The deduplicated data is combined with the current timestamp to encapsulate the data, resulting in an identification record.

2. The method of claim 1, wherein, The steps of acquiring the video stream from the molten iron ladle to obtain video frames include: Send dynamic acquisition commands to an industrial-grade high-definition camera, wherein the dynamic acquisition commands include exposure parameters, white balance parameters and frame rate parameters that are adjusted in real time based on the ambient light intensity. The industrial-grade high-definition camera supports real-time streaming protocols and has image stabilization capabilities. Receive the video stream of molten iron ladle returned by the industrial-grade high-definition camera, wherein the video stream of molten iron ladle has undergone image stabilization processing by the industrial-grade high-definition camera; The integrity of the video stream from the molten iron ladle is verified to obtain the verification result. The integrity verification includes frame continuity, frame clarity, and frame motion blur-free status. When the verification result meets the requirements, the inter-frame difference method is used to filter out the valid video frames, wherein the inter-frame difference method determines the valid video frames by comparing the pixel change rate of adjacent frames. When the verification result does not meet the requirements, a parameter adjustment command is sent to the industrial-grade high-definition camera to reacquire the video stream of the molten iron ladle and verify it again until a valid video frame is extracted. The valid video frames are then format-normalized to obtain video frames.

3. The method of claim 1, wherein, The step of reasoning through the video frame using the object detection model to obtain the recognition result includes: A target detection model with an attention mechanism is loaded. The target detection model is trained and generated based on a labeled dataset in the iron and steel metallurgy scenario. The labeled dataset includes image samples of molten iron ladle numbers under different light intensities, different levels of dirt coverage, and different shaking states, as well as image samples of molten iron ladles completely removed from the field of view. The video frame is first enhanced by an adaptive histogram equalization algorithm to improve the contrast between the tank number area and the background, and then noise is removed by a median filtering algorithm to obtain the processed video frame. The processed video frames are input into the feature pyramid network of the target detection model for feature extraction to obtain multi-scale tank number features. The multi-scale can number features are classified by the classification branch of the target detection model to obtain a classification result. The classification result includes the can number numeric category or the recognition end marker. The can number numeric category is an integer from 0 to 9, and the recognition end marker is a preset non-0-9 integer category. The multi-scale can number features are predicted by the regression branch of the target detection model, and the coordinates of the upper left corner and the lower right corner of the can number recognition box are output. The consistency of the category determination result, the coordinates of the upper left corner of the tank number identification box, and the coordinates of the lower right corner of the tank number identification box are checked to obtain the identification result.

4. The method of claim 1, wherein, The step of generating an initial can number and storing the initial can number in the first temporary list when the width of the recognition frame is greater than the width threshold, the height of the recognition frame is greater than the height threshold, and the can number recognition frame is within the effective recognition area includes: Extract the coordinates of the top left corner and the bottom right corner of the can number recognition box from the recognition result, and calculate the width and height of the recognition box. When the width of the recognition box is greater than the width threshold, the height of the recognition box is greater than the height threshold, and the can number recognition box is within the effective recognition area, the can number number sequence splicing algorithm is called to first extract the horizontal coordinate of the upper left corner of the recognition box corresponding to each can number number, and sort the numbers in ascending order of the horizontal coordinates to obtain the number arrangement order. The horizontal coordinate spacing and vertical coordinate deviation of adjacent tank number identification frames are calculated using the tank number sequence splicing algorithm. Abnormal numbers with horizontal coordinate spacing less than a preset spacing threshold or vertical coordinate deviation greater than a preset deviation threshold are removed to obtain valid tank number numbers. The valid tank numbers are concatenated sequentially according to the numerical arrangement order to generate an initial tank number; After confirming that the initial tank number is a pure numeric sequence, the initial tank number is written into the first temporary list, and the video frame acquisition timestamp corresponding to the initial tank number is recorded.

5. The method of claim 1, wherein, The steps of obtaining the single-can identification result through a multi-frame voting decision algorithm, storing it in a second temporary list, and clearing the first temporary list include: When the x-coordinate of the top left corner of the can number recognition box is less than the first x-coordinate threshold or the x-coordinate of the top left corner of the can number recognition box is greater than the second x-coordinate threshold, and the number of initial can numbers is greater than the list length threshold, the multi-frame voting decision algorithm is called to filter the format of all initial can numbers in the first temporary list to obtain valid initial can numbers, wherein the number of initial can numbers is the initial can number count stored in the first temporary list. Frequency statistics are performed on the valid initial tank numbers, and the number of times each valid initial tank number appears and the order of its first appearance are recorded; The single-can identification result is obtained based on the number of occurrences and the order of first occurrence. The single-tank identification result is associated with the corresponding tank number identification time period information and stored in the second temporary list, wherein the tank number identification time period information is the video frame time range corresponding to the single-tank identification result; Execute the list clear command to delete all initial tank number data in the first temporary list.

6. The method of claim 5, wherein, The step of obtaining the single-tank identification result based on the number of occurrences and the order of first occurrences includes: If there is only one valid initial tank number that appears most frequently, the corresponding valid initial tank number will be determined as the single tank identification result; If there are multiple valid initial tank numbers with the same frequency and all of them being the highest, the valid initial tank number that appears first in the first occurrence sequence is determined as the single tank identification result.

7. The method of claim 1, wherein, The step of performing deduplication on the second temporary list, obtaining deduplicated data, and clearing the second temporary list when the cumulative number of frames detected by the identification end marker reaches a preset period end frame threshold and the second temporary list is not empty, by using the identification period end algorithm includes: Initialize the frame counter to 0 and obtain the preset period termination frame threshold; The recognition cycle termination algorithm is invoked to monitor in real time whether the recognition result contains a recognition end marker. If a new can number category is detected, the frame counter is reset to 0. If the recognition end marker is continuously detected, the frame counter is controlled to accumulate frame by frame to obtain the cumulative frame count. When the cumulative number of frames reaches the preset period termination frame threshold and the second temporary list is not empty, a preset deduplication algorithm is invoked to extract the tank number identification time period information corresponding to each single tank identification result in the second temporary list. Based on the tank number category and the tank number identification time period information, the priority of duplicate tank numbers is determined, and all single tank identification results are obtained after filtering. All the filtered single-tank identification results are organized into an ordered dataset to obtain the deduplicated data. Generate a list clearing command, which sequentially deletes all single-tank identification results and associated identification time period information from the second temporary list, thus completing the list reset.

8. A molten iron ladle number recognition device based on multi-algorithm cooperation, characterized in that, The device includes: The acquisition module is used to capture the video stream from the molten iron ladle and obtain video frames; The recognition module is used to reason about the video frame through the target detection model to obtain the recognition result, wherein the recognition result includes the category of the tank number, the recognition end marker, the coordinates of the upper left corner of the tank number recognition box and the coordinates of the lower right corner of the tank number recognition box; The can number generation module is used to generate an initial can number and store it in a first temporary list when the width of the recognition box is greater than a width threshold, the height of the recognition box is greater than a height threshold, and the can number recognition box is within the effective recognition area. The recognition box width is the difference between the horizontal coordinate of the lower right corner of the can number recognition box and the horizontal coordinate of the upper left corner of the can number recognition box. The recognition box height is the difference between the vertical coordinate of the lower right corner of the can number recognition box and the vertical coordinate of the upper left corner of the can number recognition box. The coordinates of the effective recognition area are [a first horizontal coordinate threshold, a second horizontal coordinate threshold], where the first horizontal coordinate threshold is less than the second horizontal coordinate threshold. The can number result module is used to obtain a single can identification result through a multi-frame voting decision algorithm when the horizontal coordinate of the upper left corner of the can number identification box is less than a first horizontal coordinate threshold or the horizontal coordinate of the upper left corner of the can number identification box is greater than a second horizontal coordinate threshold, and the number of initial can numbers stored in the first temporary list is greater than the list length threshold. The result is then stored in the second temporary list and the first temporary list is cleared. The optimization module is used to perform deduplication processing on the second temporary list when the cumulative number of frames of the identification end marker reaches a preset period end frame threshold and the second temporary list is not empty, by using the identification period end algorithm to obtain the deduplicated data and clear the second temporary list. The results module is used to encapsulate the deduplicated data with the current timestamp to obtain the identification record.

9. A molten iron ladle number recognition device based on multi-algorithm cooperation, characterized in that, The device includes: a memory, a processor, and a multi-algorithm collaborative molten iron ladle number identification program stored in the memory and running on the processor, the multi-algorithm collaborative molten iron ladle number identification program being configured to implement the steps of the multi-algorithm collaborative molten iron ladle number identification method as described in any one of claims 1-7.

10. A storage medium, characterized by The storage medium stores a ladle number identification program based on multi-algorithm collaboration. When the ladle number identification program based on multi-algorithm collaboration is executed by the processor, it implements the steps of the ladle number identification method based on multi-algorithm collaboration as described in any one of claims 1-7.

Images