Rotating target detection-based split disc liquid level fluctuation molten aluminum leakage detection method and system and medium

Abstract

The application discloses a kind of based on rotary target detection's shunt disc liquid level fluctuation aluminum liquid leakage detection method, system and medium.The method includes obtaining the video dataset of the surface of aluminum liquid production shunt disc and carrying out multi-classification abnormal label labeling, and is divided into training set and verification set;Based on rotary target detection model, build aluminum leakage video detection model, and complete the training of aluminum leakage video detection model using training set and verification set;Using the trained aluminum leakage video detection model, the video of real-time acquisition when aluminum liquid production is detected, when detecting aluminum leakage phenomenon, aluminum leakage alarm signal is sent out.The present application is based on the progressive relationship of point, line and surface, establishes a multi-level aluminum leakage visual feature visualization label system, and applies the rotary target detection model to the aluminum leakage visual detection of shunt disc, and the method is suitable for any installation position, any installation angle of shunt disc monitoring camera collected data.

Description

Technical Field

[0001] This invention belongs to the technical field of aluminum liquid leakage detection, specifically relating to a method, system, and medium for detecting aluminum liquid leakage by means of level fluctuation in a distribution plate based on rotating target detection. Background Technology

[0002] Aluminum alloys possess advantages such as low density, high specific strength, and high fracture toughness, making them widely used in aerospace, marine engineering, and transportation. Direct Chill Casting (DC) semi-continuous casting is currently the mainstream aluminum alloy casting process. However, DC casting is highly susceptible to accidents such as crystallizer cracking leading to aluminum leakage and molten metal overflow. Statistics show that nearly 900 aluminum processing (deep well casting) companies nationwide have experienced an average of one major safety accident involving more than three people annually over the past five years, causing severe casualties and economic losses. Aluminum leakage is a high-risk factor in aluminum processing (deep well casting). When high-temperature molten aluminum leaks from the manifold and comes into large-area contact with cooling water, it can lead to an explosion. Current methods for judging aluminum leakage in manifolds rely heavily on human experience, while related intelligent identification methods, such as thermal imaging aluminum leakage detection technology, suffer from large blind spots and low recognition rates. When aluminum leaks, obvious visual characteristics such as non-uniform settling and multiple bubbles will appear on the surface of the manifold. Therefore, this invention establishes a computer vision-based method for identifying abnormal aluminum liquid leakage in a distribution plate, based on the visual characteristics of aluminum liquid leakage, thus solving the problems of low accuracy and poor applicability of current identification methods. Summary of the Invention

[0003] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide a method, system, and medium for detecting aluminum liquid leakage by rotating target detection based on the level fluctuation of a manifold. From the perspective of process control, this invention studies the control methods and management approaches for aluminum liquid leakage before a deep well explosion. It proposes a new method for detecting aluminum liquid leakage by manifold based on computer vision, comprehensively analyzes the visual characteristics of the manifold surface during aluminum liquid leakage, and performs timely and accurate detection of aluminum liquid leakage. From the perspective of pre-control, this method avoids the occurrence of steam explosion reactions and provides new ideas for related research on controlling explosion accidents in aluminum processing (deep well casting).

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] One aspect of the present invention provides a method for detecting aluminum liquid leakage due to level fluctuations in a distribution plate based on rotating target detection, comprising the following steps:

[0006] A video dataset of the surface of the diversion plate during aluminum liquid production was obtained and multi-class anomaly labeling was performed, and the dataset was divided into training set and validation set.

[0007] A leaky aluminum video detection model was constructed based on the rotating target detection model, and the training set and validation set were used to complete the training of the leaky aluminum video detection model.

[0008] The trained aluminum leakage video detection model is used to detect real-time aluminum molten metal production videos. When aluminum leakage is detected, an aluminum leakage alarm signal is issued.

[0009] As a preferred technical solution, the multi-class anomaly labeling specifically involves:

[0010] Leakage vortexes and multiple floating bubbles appeared on the surface of the diversion plate, indicating an initial aluminum liquid leak that came into contact with water, i.e., a point-level leak.

[0011] A single row of non-uniform sedimentation appeared on the surface of the diversion plate, which was marked as an expansion of aluminum liquid leakage, i.e., a line-level leakage.

[0012] Molten aluminum overflowed directly from the surface of the manifold, which was marked as a lateral leak, i.e., a surface-level leak.

[0013] As the preferred technical solution, the specific annotation is as follows:

[0014] A rectangle is used to annotate the data and generate a corresponding XML file. The XML file records the center coordinates of the rectangle, as well as the length, width, and rotation angle of the rectangle.

[0015] As a preferred technical solution, each row on the surface of the distribution plate is marked with labels at the point, line, and surface levels, and rotated at an appropriate angle so that the label box is tangent to the boundary of each circular flow hole in each row on the surface of the distribution plate, and does not include parts of other rows.

[0016] As a preferred technical solution, the rotating target detection model includes SCRDet, R3Det, ​​YOLOv3_OBB, YOLOv4_OBB, YOLOv5_OBB, YOLOv7_OBB, YOLOv8_OBB, and YOLOX_OBB.

[0017] As a preferred technical solution, the rotating target detection model is improved based on the Biformer module to construct an aluminum leakage video detection model, specifically as follows:

[0018] The YOLOv5_OBB neural network is used as the basic model. The YOLOv5_OBB neural network includes an input end, a backbone network, a neck network, and a head network.

[0019] The input terminal is used for Mosaic data augmentation, adaptive anchor box calculation, and adaptive image scaling.

[0020] The backbone network introduces a Focus structure and a CSP structure. The Focus structure is used to downsample the feature map, and the CSP structure is used to reduce the computational complexity and number of parameters of the model.

[0021] The neck network adopts an FPN+PAN structure to further improve the performance of target detection;

[0022] The head network uses a grid-based anchor and includes three detectors;

[0023] The Biformer module employs a two-layer routing attention mechanism, specifically: dividing the feature map into S*S regions to obtain the feature map for each region; pruning the correlation map by retaining the first k connections of each region to obtain an index matrix; calculating the Q vector in each index matrix and the K vector of a non-contiguous region in the feature map; and tensor-aggregating the K vector and V vector to obtain aggregated tensor key-value pairs; finally, performing attention calculations and local context enhancement on the aggregated key-value pairs.

[0024] By embedding the Biformer module into the neck network of the YOLOv5_OBB neural network before the second upsampling layer, global feature extraction is performed to obtain the aluminum leakage video detection model.

[0025] As a preferred technical solution, the training steps for the aluminum leakage video detection model are as follows:

[0026] The established label dataset is divided into a training set and a validation set in a 7:3 ratio;

[0027] Initialize the weights of the aluminum leakage video detection model using pre-trained COCO dataset hyperparameters;

[0028] The training and validation sets are input into the aluminum leakage video detection model, and the weights and biases of the aluminum leakage video detection model are updated by backpropagation with the goal of minimizing the loss.

[0029] The mean accuracy mAP@0.5 is set with an IoU threshold of 0.5, and the mean accuracy mAP@[0.5:0.95] is set at different IoU thresholds, specifically from 0.5 to 0.95, with a step size of 0.05; the ratio of mAP@0.5 to mAP@[0.5:0.95] is set to 1:9, the classification accuracy under each training round of the model is retained, and the weight file of the optimized aluminum leakage video detection model is recorded;

[0030] An aluminum leakage video detection model is established using the optimal weight file. Based on the real-time transmitted monitoring video of the distribution plate surface, the real-time detection effect is tested under the inference analysis model file, and the aluminum leakage video detection model is continuously improved.

[0031] As a preferred technical solution, the step of using a trained aluminum leakage video detection model to detect real-time acquired videos of aluminum molten metal production specifically involves:

[0032] Install distribution panel monitoring cameras on the production site;

[0033] The trained aluminum leakage video detection model is set up on a cloud server;

[0034] The video data collected by the monitoring camera of the distribution panel is uploaded to the cloud server in real time;

[0035] The trained aluminum leakage video detection model reads and analyzes the detection video in real time on the cloud server. When aluminum leakage is detected, an aluminum leakage alarm signal is issued. The aluminum leakage alarm signal is transmitted from the cloud server back to the alarm device on the production site to realize the alarm process.

[0036] Another aspect of the present invention provides a liquid level fluctuation aluminum liquid leakage detection system for a distribution plate based on rotating target detection, which is applied to the above-mentioned liquid level fluctuation aluminum liquid leakage detection method for a distribution plate based on rotating target detection, and includes a data acquisition module, a data annotation module, a model building module, a model training module and a model detection module;

[0037] The data acquisition module is used to acquire video datasets of the surface of the diversion plate during aluminum liquid production;

[0038] The data labeling module is used to perform multi-class anomaly labeling on the video dataset of the diversion plate surface during aluminum liquid production, dividing it into a training set and a validation set.

[0039] The model building module is used to build a video detection model for aluminum leakage based on the rotating target detection model;

[0040] The model training module is used to train the aluminum leakage video detection model based on the training set and the validation set;

[0041] The model detection module is used to detect real-time aluminum molten metal production videos based on a trained aluminum leakage video detection model, and to issue an aluminum leakage alarm signal when aluminum leakage is detected.

[0042] In another aspect, the present invention also provides a storage medium storing a program, characterized in that: when the program is executed by a processor, it implements the above-described method for detecting aluminum liquid leakage by fluctuation of the diversion plate based on rotating target detection.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] (1) Based on the progressive relationship of points, lines and surfaces, this invention establishes a multi-level visual labeling system for aluminum leakage visual features;

[0045] (2) This invention applies the rotating target detection (YOLOv5_OBB) system method to the visual detection of aluminum liquid leakage in the manifold. This method is applicable to the data collected by the monitoring camera of the manifold at any installation position and any installation angle.

[0046] (3) This invention uses the Biformer attention mechanism based on the Transformer architecture to extract global features of molten aluminum leakage using a GPU-friendly dense matrix multiplication. This module facilitates server-side inference acceleration to better meet the actual needs of model deployment. Attached Figure Description

[0047] Figure 1 This is a flowchart of the aluminum liquid leakage detection method based on rotating target detection for level fluctuation in a diversion plate according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the visual calibration tag library for aluminum liquid leakage in the diversion plate according to an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the structure of the rotating target detection model according to an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of the YOLOv5_OBB neural network based on the Biformer module according to an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of the structure of the aluminum liquid leakage detection system based on rotating target detection for level fluctuation in a diversion plate according to an embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram of the structure of the storage medium according to an embodiment of the present invention. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0054] Example

[0055] like Figure 1 As shown, this embodiment provides a method for detecting aluminum liquid leakage due to level fluctuations in a distribution plate based on rotating target detection, including the following steps:

[0056] S1, such as Figure 2 As shown, a video dataset of the surface of the flow divider during aluminum liquid production was obtained and multi-classified anomaly labeling was performed, dividing it into a training set and a validation set;

[0057] In the actual casting process, due to the unequal heat exchange between the high-temperature molten aluminum liquid and the cooling water in the crystallizer of the distribution plate, the molten aluminum liquid leaks from the guide hole of the crystallizer, and the leaked molten aluminum liquid flows into the casting well quickly.

[0058] The process of molten aluminum leakage is a gradual one:

[0059] Initially, the flow rate of leaking molten aluminum is small, flowing only in a small portion around the leak point, manifesting as a leakage vortex on the surface of the distribution plate. After the leaking molten aluminum comes into contact with water, the liquid water turns into gaseous vapor, which surges upward from the leak point, manifesting as multiple floating bubbles on the surface of the distribution plate. This can be described as a leak at a certain guide hole of the distribution plate crystallizer, i.e., a point-level leak.

[0060] As the leak expands further and the area of ​​the leak opening increases, a large amount of molten aluminum rapidly seeps down from the leak opening, manifesting as a single row of non-uniform settling on the surface of the distribution plate, or as rows of aluminum leaking downwards, i.e., a line-level leak.

[0061] In addition, when the flow rate of molten aluminum is too large, some of the molten aluminum overflows directly from the surface of the distribution plate, which is manifested as a side leakage of the distribution plate. The leaking molten aluminum can be seen directly on the side of the distribution plate. At this time, it is a surface leakage, that is, a surface-level leakage.

[0062] Therefore, based on the above, a three-level data labeling system of points, lines, and surfaces is established. The collected video data is decomposed frame by frame into images, and data visualization annotations are performed based on the state characteristics of the distribution panel surface.

[0063] Furthermore, in this embodiment, the rolabelimg software is used for data annotation. This tool is specifically designed for annotating datasets used in rotating target detection tasks. Each annotated bounding box generates a corresponding XML file containing the center coordinates of the bounding box, as well as its length, width, and rotation angle. Based on a three-level label database of points, lines, and surfaces, each row on the flow distribution plate surface is labeled, rotating at an appropriate angle to ensure that the labeled bounding box is as tangent as possible to the boundary of each circular flow outlet in each row on the flow distribution plate surface, reducing redundant information and avoiding including parts of other rows. Additionally, labelImg and labelme software can also be used for annotation. These two software programs annotate horizontal bounding boxes. rolabelimg adds angle information to labelimg, making it suitable for rotating target detection.

[0064] S2. Construct a video detection model for aluminum leakage based on the rotating target detection model, such as... Figure 3 As shown, the model is trained using the training and validation sets;

[0065] Furthermore, in this embodiment, the rotating target detection models used include: (1) those formed by fusing deep learning networks such as the ResNet series with rotation angle loss, such as SCRDet, R3Det, ​​etc. (2) those formed by fusing rotation angle loss with YOLO series for rotating target detection, such as YOLOv3_OBB, YOLOv4_OBB, YOLOv5_OBB, YOLOv7_OBB, YOLOv8_OBB, and YOLOX_OBB, etc.

[0066] Furthermore, in this embodiment, the rotating target detection model is improved based on the Biformer module to construct an aluminum leakage video detection model, as detailed below:

[0067] YOLOv5s_OBB, version 6.0 of YOLO, was selected as the base model.

[0068] YOLOv5s_OBB is the network model with the smallest depth and feature map width in the YOLOv5_OBB series, exhibiting low computational complexity and parameter count, making it suitable for scenarios with limited computing resources. Its network structure mainly consists of an input layer, a backbone network, a neck network, and a head network. The input layer employs Mosaic data augmentation, adaptive anchor box calculation, and adaptive image scaling techniques. The backbone network introduces a Focus structure and a CSP structure. The Focus structure primarily downsamples the feature map to improve the network's non-linear expressiveness and reduce the number of parameters. The CSP structure effectively reduces the model's computational complexity and parameter count. The neck network uses an FPN+PAN structure to further improve object detection performance. The head network mainly utilizes grid-based anchors to perform object detection on feature maps at different scales, including three Detect detectors.

[0069] Because the local receptive field of convolutional computation results in insufficient extraction of contextual features, the global feature extraction is inadequate, leading to low detection accuracy for small targets. Therefore, in this embodiment, the Biformer module is embedded into the base model. After multiple ablation experiments, it was determined that this module performs the best global feature extraction before the second upsampling layer in the Neck of the YOLOv5s_OBB neural network, and this module was selected as the final training model.

[0070] The basic structure of the Biformer module is as follows: Figure 4 As shown, its core content is the Bi-level Rounding Attention (BRA) mechanism. The BRA module constructs a region affinity graph through an adjacency matrix, collects key-value pairs from the top k relevant windows, and implements filtering operations by applying fine-grained token-to-token attention to the regions.

[0071] The BRA module divides the feature map into S*S regions, obtaining a feature map for each region. The correlation map is pruned by retaining the first k connections of each region, resulting in an index matrix. For each Q vector in the index matrix, it is computed with a K vector from a non-contiguous region in the feature map. To ensure the continuity of GPU data loading, the K and V vectors are tensor-aggregated to obtain aggregated tensor key-value pairs. Attention calculations and local context enhancement are then performed on these aggregated key-value pairs to obtain the final attention mechanism.

[0072] The training steps for the aluminum leakage video detection model are as follows:

[0073] The established label dataset is divided into a training set and a validation set in a 7:3 ratio;

[0074] Initialize the weights of the aluminum leakage video detection model using pre-trained COCO dataset hyperparameters;

[0075] Input the training and validation sets into the model, and backpropagate to update the weights and biases of the aluminum leakage video detection model with the goal of minimizing the loss.

[0076] The IoU threshold (IoU threshold is a measurement standard used in object detection to calculate the area of ​​overlap between two regions divided by the area of ​​the two regions. It is an important indicator in NMS and mAP algorithms, used to filter redundant predicted boxes. The confidence threshold is used to filter out boxes in the predicted box that have a confidence score lower than 0.5. In this embodiment, it is set to 0.5, that is, the area union of the predicted box and the ground truth box is set to 0.5, and predicted boxes with a score less than 0.5 are directly filtered out) is set to 0.5 of mAP@0.5 (mAP@0.5 refers to the average accuracy of multiple categories) and the mean The average mAP@[0.5:0.95] (referring to the average mAP at different IoU thresholds, i.e., IoU thresholds from 0.5 to 0.95, with a step size of 0.05, including 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, and 0.95) is set to 1:9. The classification accuracy is retained for each epoch (model training round, i.e., the number of iterations), and the weight file of the optimized aluminum leakage video detection model is recorded.

[0077] An aluminum leakage video detection model is established using the optimal weight file. Based on the real-time transmitted RTSP video monitoring of the distribution panel surface (referring to the transmission of video data via the RTSP protocol), the real-time detection effect is tested and the aluminum leakage video detection model is continuously improved using detect.py (referring to the inference analysis model file, which uses the trained model to detect and analyze the real-time transmitted RTSP video stream).

[0078] S3. Use the trained aluminum leakage video detection model to detect the real-time video of aluminum liquid production. When aluminum leakage is detected, issue an aluminum leakage alarm signal.

[0079] Install distribution panel monitoring cameras on the production site;

[0080] The trained aluminum leakage video detection model is set up on a cloud server;

[0081] The video data collected by the monitoring camera of the distribution panel is uploaded to the cloud server in real time;

[0082] The trained aluminum leakage video detection model reads and analyzes the detection video in real time on the cloud server. When aluminum leakage is detected, an aluminum leakage alarm signal is issued. The aluminum leakage alarm signal is transmitted from the cloud server back to the alarm device on the production site to realize the alarm process.

[0083] like Figure 5 As shown, in another embodiment of this application, a liquid level fluctuation aluminum liquid leakage detection system based on rotating target detection is provided. The system includes a data acquisition module, a data annotation module, a model building module, a model training module, and a model detection module.

[0084] The data acquisition module is used to acquire video datasets of the surface of the diversion plate during aluminum liquid production;

[0085] The data labeling module is used to perform multi-class anomaly labeling on the video dataset of the diversion plate surface during aluminum liquid production, dividing it into a training set and a validation set.

[0086] The model building module is used to build a video detection model for aluminum leakage based on the rotating target detection model;

[0087] The model training module is used to train the aluminum leakage video detection model based on the training set and the validation set;

[0088] The model detection module is used to detect real-time aluminum molten metal production videos based on a trained aluminum leakage video detection model, and to issue an aluminum leakage alarm signal when aluminum leakage is detected.

[0089] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above. This system is applied to the aluminum liquid leakage detection method based on rotating target detection of the diversion plate liquid level fluctuation in the above embodiments.

[0090] like Figure 6 As shown, in another embodiment of this application, a storage medium is also provided, storing a program that, when executed by a processor, implements a method for detecting aluminum liquid leakage due to level fluctuations in a diversion plate based on rotating target detection, specifically:

[0091] A video dataset of the surface of the diversion plate during aluminum liquid production was obtained and multi-class anomaly labeling was performed, and the dataset was divided into training set and validation set.

[0092] A leaky aluminum video detection model was constructed based on the rotating target detection model, and the training set and validation set were used to complete the training of the leaky aluminum video detection model.

[0093] The trained aluminum leakage video detection model is used to detect real-time aluminum molten metal production videos. When aluminum leakage is detected, an aluminum leakage alarm signal is issued.

[0094] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0095] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for detecting aluminum liquid leakage due to level fluctuation in a diversion plate based on rotating target detection, characterized in that, Includes the following steps: We acquired a video dataset of the surface of the flow distribution plate during aluminum molten metal production and performed multi-class anomaly labeling, dividing it into training and validation sets. Specifically: Leakage vortexes and multiple floating bubbles appeared on the surface of the diversion plate, indicating an initial aluminum liquid leak that came into contact with water, i.e., a point-level leak. A single row of non-uniform sedimentation appeared on the surface of the diversion plate, which was marked as an expansion of aluminum liquid leakage, i.e., a line-level leakage. Molten aluminum overflowed directly from the surface of the distribution plate, which was marked as a lateral leak, i.e., a surface-level leak. A leaky aluminum video detection model was constructed based on the rotating target detection model, and the training set and validation set were used to complete the training of the leaky aluminum video detection model. An improved rotating target detection model based on the Biformer module is constructed to create a video detection model for aluminum leakage, specifically: The YOLOv5_OBB neural network is used as the basic model. The YOLOv5_OBB neural network includes an input end, a backbone network, a neck network, and a head network. The input terminal is used for Mosaic data augmentation, adaptive anchor box calculation, and adaptive image scaling. The backbone network introduces a Focus structure and a CSP structure. The Focus structure is used to downsample the feature map, and the CSP structure is used to reduce the computational complexity and number of parameters of the model. The neck network adopts an FPN+PAN structure to further improve the performance of target detection; The head network uses a grid-based anchor and includes three detectors; The Biformer module employs a two-layer routing attention mechanism, specifically: dividing the feature map into S*S regions to obtain the feature map for each region; pruning the correlation map by retaining the first k connections of each region to obtain an index matrix; calculating the Q vector in each index matrix and the K vector of a non-contiguous region in the feature map; and tensor-aggregating the K vector and V vector to obtain aggregated tensor key-value pairs; finally, performing attention calculations and local context enhancement on the aggregated key-value pairs. The Biformer module is embedded into the neck network of the YOLOv5_OBB neural network before the second upsampling layer to perform global feature extraction, resulting in a leaky aluminum video detection model. The trained aluminum leakage video detection model is used to detect real-time aluminum molten metal production videos. When aluminum leakage is detected, an aluminum leakage alarm signal is issued.

2. The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection according to claim 1, characterized in that, The specific annotation is as follows: A rectangle is used to annotate the data and generate a corresponding XML file. The XML file records the center coordinates of the rectangle, as well as the length, width, and rotation angle of the rectangle.

3. The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection according to claim 1, characterized in that, Label each row on the surface of the distribution plate according to the point, line and surface level labels, and rotate it at an appropriate angle so that the label box is tangent to the boundary of each circular flow hole in each row on the surface of the distribution plate, and does not include part of the area of ​​other rows.

4. The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection according to claim 1, characterized in that, The rotating target detection models include SCRDet, R3Det, ​​YOLOv3_OBB, YOLOv4_OBB, YOLOv5_OBB, YOLOv7_OBB, YOLOv8_OBB, and YOLOX_OBB.

5. The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection according to claim 1, characterized in that, The training steps for the aluminum leakage video detection model are as follows: The established labeled dataset is divided into a training set and a validation set in a 7:3 ratio; Initialize the weights of the aluminum leakage video detection model using the pre-trained COCO dataset hyperparameters; The training and validation sets are input into the aluminum leakage video detection model, and the weights and biases of the aluminum leakage video detection model are updated by backpropagation with the goal of minimizing the loss. The mean accuracy mAP@0.5 is set with an IoU threshold of 0.5, and the mean accuracy mAP@[0.5:0.95] is set at different IoU thresholds, specifically from 0.5 to 0.95, with a step size of 0.05; the ratio of mAP@0.5 to mAP@[0.5:0.95] is set to 1:9, the classification accuracy under each training round of the model is retained, and the weight file of the optimized aluminum leakage video detection model is recorded; An aluminum leakage video detection model is established using the optimal weight file. Based on the real-time transmitted monitoring video of the distribution plate surface, the real-time detection effect is tested under the inference analysis model file, and the aluminum leakage video detection model is continuously improved.

6. The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection according to claim 1, characterized in that, The process of using a trained aluminum leakage video detection model to detect real-time aluminum molten metal production videos specifically involves: Install distribution panel monitoring cameras on the production site; The trained aluminum leakage video detection model is set up on a cloud server; The video data collected by the monitoring camera of the distribution panel is uploaded to the cloud server in real time; The trained aluminum leakage video detection model reads and analyzes the detection video in real time on the cloud server. When aluminum leakage is detected, an aluminum leakage alarm signal is issued. The aluminum leakage alarm signal is transmitted from the cloud server back to the alarm device on the production site to realize the alarm process.

7. A liquid level fluctuation and aluminum liquid leakage detection system based on rotating target detection, characterized in that, The method for detecting aluminum liquid leakage by level fluctuation in a diversion plate based on rotating target detection, as described in any one of claims 1-6, includes a data acquisition module, a data annotation module, a model building module, a model training module, and a model detection module; The data acquisition module is used to acquire video datasets of the surface of the diversion plate during aluminum liquid production; The data labeling module is used to perform multi-class anomaly labeling on the video dataset of the diversion plate surface during aluminum liquid production, dividing it into a training set and a validation set. The model building module is used to build a video detection model for aluminum leakage based on the rotating target detection model; The model training module is used to train the aluminum leakage video detection model based on the training set and the validation set; The model detection module is used to detect real-time aluminum molten metal production videos based on a trained aluminum leakage video detection model, and to issue an aluminum leakage alarm signal when aluminum leakage is detected.

8. A storage medium storing a program, characterized in that: When the program is executed by the processor, it implements the method for detecting aluminum liquid leakage based on rotating target detection in a diversion plate according to any one of claims 1-6.

Images