Waste steel consumption recording method and system

Abstract

The invention relates to the technical field of image recognition, and particularly provides a waste steel consumption recording method and system, and the method comprises the steps: collecting a monitoring video, and recognizing an identity code marked on the surface of a hopper and a waste steel image in the hopper from the monitoring video; calling a pressure sensor signal corresponding to the identity code based on a preset corresponding relation between the identity code and a pressure sensor equipment name and the received pressure sensor signal; and converting the pressure sensor signal into a weight value, and recording the identity code and the weight value. According to the invention, waste steel consumption data can be recorded more accurately.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and specifically to a method and system for recording scrap steel consumption. Background Art

[0002] At present, the recording of scrap steel consumption mainly relies on manual recording by personnel in material preparation positions. However, this recording method has a series of problems and hidden dangers, which may affect the accuracy and reliability of the data.

[0003] People in material preparation positions often have multiple duties and may not be responsible for recording and confirming scrap steel consumption data. In a busy work environment, they may not always pay attention to the consumption of scrap steel, resulting in omissions or delays in data recording. In addition, due to the large number of types and furnace numbers of scrap steel, people in material preparation positions sometimes need to communicate with other positions to confirm the specific furnace number and material type. This communication method is not only inefficient, but also prone to errors, further increasing the risk of data distortion. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a scrap steel consumption recording method and system to solve the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides a method for recording scrap steel consumption, comprising: Collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video; Based on the preset correspondence between the identity code and the pressure sensor device name, and the received pressure sensor signal, retrieve the pressure sensor signal corresponding to the identity code; The pressure sensor signal is converted into a weight value, and the identity code and the weight value are recorded.

[0006] In an optional embodiment, collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video, includes: Using text extraction technology to identify the identity code marked on the surface of the hopper from the monitoring video; The type of scrap steel in the hopper is identified from the surveillance video using a pre-trained instance segmentation algorithm.

[0007] In an optional embodiment, using a pre-trained instance segmentation algorithm to identify the type of scrap steel in the hopper from the monitoring video includes: Use deep neural networks to extract edge features, texture features, and shape features from scrap steel images in the hopper; Constructing a feature pyramid network to fuse feature maps of different scales, wherein the feature pyramid network transmits high-level semantic information to low-level layers through a top-down path and lateral connections while retaining detail information of the low-level layers; A preset number of anchor points are set on each feature layer of the feature pyramid network. Each anchor point has multiple different aspect ratios and receptive field sizes. A target candidate box is generated on each feature layer by means of a sliding window. Extract features from each target candidate frame to obtain its corresponding feature vector, and use a classifier to classify the feature vector to obtain attribute information, including shape, size, structure, and material; The feature vector of the target candidate box is weighted and summed with the attribute information to obtain a fusion feature; Using a pre-trained level classifier to classify the fused features to obtain level information, the pre-trained level classifier adopts a strategy combining a cross entropy loss and a focal loss function during training; A regression operation is performed on each target candidate box. The CIOU loss function is used for training during the regression process, and the overlapping detection boxes are removed through the non-maximum suppression algorithm to retain the best target detection box.

[0008] In an optional embodiment, collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video, includes: Collect monitoring videos using multiple image acquisition devices deployed in the material preparation area and converter area; Converting the surveillance video into an image frame sequence; A visual algorithm is used to identify the action type of the hopper based on the image frame sequence, and the action type includes material preparation, material replenishment, passing and consumption.

[0009] In an optional embodiment, converting the pressure sensor signal into a weight value, and recording the identity code and the weight value, comprises: Track the action flow of the same hopper based on the identity code, as well as the scrap grade information and weight value corresponding to each action in the action flow; Confirm that the action flow of the hopper includes material preparation and consumption, then confirm that the hopper completes an action flow, and write the completed action flow and the corresponding scrap steel grade information and weight value into the database list.

[0010] In an optional embodiment, the method further comprises: Matching the action flow of the hopper with the operation data in the operation management platform to obtain the furnace number corresponding to the action flow of the hopper; The furnace number is written into the corresponding position of the database list.

[0011] In a second aspect, the present invention provides a scrap steel consumption recording system, comprising: An image recognition module, used to collect monitoring video and identify the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video; A signal retrieving module, configured to retrieve the pressure sensor signal corresponding to the identity code based on the correspondence between the preset identity code and the pressure sensor device name and the received pressure sensor signal; The data recording module is used to convert the pressure sensor signal into a weight value and record the identity code and the weight value.

[0012] In an optional embodiment, the image recognition module includes: A text recognition unit, used to recognize the identity code marked on the surface of the hopper from the monitoring video using text extraction technology; The image recognition unit is used to identify the type of scrap steel in the hopper from the monitoring video using a pre-trained instance segmentation algorithm.

[0013] In an optional implementation, the image recognition unit includes: A feature extraction subunit, for extracting edge features, texture features, and shape features from the scrap steel image in the hopper using a deep neural network; A first processing subunit is used to construct a feature pyramid network to fuse feature maps of different scales, wherein the feature pyramid network transmits high-level semantic information to low-level layers through a top-down path and lateral connections while retaining detail information of the low-level layers; A second processing subunit is used to set a preset number of anchor points on each feature layer of the feature pyramid network, each anchor point has a plurality of different aspect ratios and receptive field sizes, and generate a target candidate box on each feature layer by means of a sliding window; The first classification subunit is used to extract features from each target candidate frame to obtain its corresponding feature vector, and classify the feature vector using a classifier to obtain attribute information, wherein the attribute information includes shape, size, structure, and material; A feature fusion subunit, used for weighted summing of a feature vector of a target candidate box and the attribute information to obtain a fused feature; A second classification subunit is used to classify the fused features using a pre-trained level classifier to obtain level information, wherein the pre-trained level classifier adopts a strategy combining a cross entropy loss and a focal loss function during training; The regression processing subunit is used to perform regression operations on each target candidate box. The CIOU loss function is used for training during the regression process, and the non-maximum suppression algorithm is used to remove overlapping detection boxes and retain the best target detection box.

[0014] In an optional embodiment, the image recognition module includes: A video acquisition unit, used to acquire monitoring videos using multiple image acquisition devices deployed in the material preparation area and the converter area; A video conversion unit, used for converting the surveillance video into an image frame sequence; The action recognition unit is used to use a visual algorithm to identify the action type of the hopper based on the image frame sequence, and the action type includes material preparation, material replenishment, passing and consumption.

[0015] The beneficial effect of the present invention is that the scrap steel consumption recording method and system provided by the present invention realizes the automation of hopper tracking and material type identification through an automated monitoring system and intelligent algorithm, reduces manual intervention, and reduces the error rate. In this way, scrap steel consumption data can be recorded more accurately, and the accuracy of data analysis and decision-making can be improved.

[0016] In addition, the invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention.

[0019] Figure 2 It is a schematic diagram of an application scenario of a method according to an embodiment of the present invention.

[0020] Figure 3 is a schematic block diagram of a system according to an embodiment of the present invention.

[0021] Among them, 1. Camera; 2. Pressure sensor; 3. Operation management platform; 4. Hopper. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0024] The scrap steel consumption recording method provided in the embodiment of the present invention is executed by a computer device, and accordingly, the scrap steel consumption recording system runs in the computer device.

[0025] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention. Figure 1 The execution subject may be a scrap steel consumption recording system. According to different requirements, the order of the steps in the flow chart may be changed, and some may be omitted.

[0026] like Figure 1 As shown, the method includes: S1. Collect monitoring video, and identify the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video.

[0027] Real-time surveillance videos are collected through surveillance cameras deployed in key areas. These videos are not only used for security monitoring, but also in the scrap steel management system. The system uses advanced image recognition technology to automatically identify the identity codes on the surface of the hopper from the surveillance video. These identity codes are unique and are used to distinguish different hoppers. At the same time, the system can also capture images of scrap steel in the hopper, which provide important basis for subsequent analysis and processing.

[0028] S2. Based on the preset correspondence between the identity code and the pressure sensor device name, and the received pressure sensor signal, retrieve the pressure sensor signal corresponding to the identity code.

[0029] According to the correspondence between the preset identity code and the pressure sensor device name, the system searches for the pressure sensor that matches the currently recognized identity code. Once the corresponding pressure sensor is found, the system will immediately receive the signal sent by the sensor. These signals reflect the weight change of the scrap steel in the hopper and are the key data for calculating the weight of the scrap steel.

[0030] S3. Convert the pressure sensor signal into a weight value, and record the identity code and the weight value.

[0031] The received pressure sensor signal is converted into a specific weight value. This step is crucial because it ensures the accuracy and readability of the data. After the conversion is completed, the system will record the identity code and the corresponding weight value. This data is not only used to track the consumption of scrap steel in real time, but also provides a reliable basis for subsequent statistical analysis and decision-making. In this way, companies can more accurately grasp the use of scrap steel, optimize resource allocation, and improve production efficiency.

[0032] Please refer to Figure 2 , a QR code label is painted on the tail of each hopper 4 to uniquely identify each hopper 4. This ensures that the system can accurately track and identify each hopper 4. High-definition cameras 1 are installed in the preparation area, above the preparation area, and in the consumption area to obtain real-time images of the hopper operation and images of scrap steel added to the hopper. Pressure sensors 2 are installed on the force-bearing part of each hopper 4.

[0033] The scrap steel consumption recording system receives monitoring video and pressure values, identifies the identity code, scrap steel image, grade information and hopper action from the monitoring video, matches the identity code with the pressure sensor signal, and then analyzes the scrap steel consumption of the hopper. This information is recorded to achieve accurate recording of scrap steel consumption.

[0034] In addition, the scrap steel consumption recording system is connected to the operation management platform 3, and real-time data such as hopper weighing and furnace number are obtained from the connection operation management platform 3. This can ensure the accuracy and consistency of material type identification and weighing data.

[0035] In an embodiment of the present invention, based on step S1, a possible embodiment is given below to illustrate its specific implementation scheme in a non-limiting manner.

[0036] S101. Monitoring video preprocessing.

[0037] In the automated process of scrap steel management, in order to effectively monitor the status of the hopper and the scrap steel inside, multiple high-performance image acquisition devices are deployed in the material preparation area and the converter area. These devices work around the clock to capture dynamic images of key scenes in real time. These monitoring videos are then sent to the central processing system as the basis for subsequent analysis and processing.

[0038] First, the continuously played surveillance video is divided into a series of static image frames to form an image frame sequence. This processing method not only reduces the complexity of data processing, but also enables the system to analyze the surveillance images at each moment more carefully and capture fleeting details.

[0039] When the surveillance camera deployed at a key location captures the image of the hopper, the system immediately starts the image preprocessing process to ensure the accuracy of subsequent text recognition. This step is crucial because the original surveillance video may be affected by a variety of factors such as lighting changes, camera angles, dust or fog, resulting in reduced image quality and blurred text.

[0040] In order to ensure the accuracy of subsequent recognition, the brightness of the image is adjusted first. At the same time, the denoising algorithm is applied to effectively eliminate random noise and interference in the image, such as snowflakes and scratches, so that the edges of the text are clearer and the contours are more distinct.

[0041] In addition, image enhancement techniques such as sharpening and edge detection are used to further enhance the features of the text and improve its recognition in complex backgrounds. These pre-processing steps work together to enable the system to accurately capture the identity code on the hopper surface even in low light and harsh environments, laying a solid foundation for subsequent text recognition and information association.

[0042] S102. Using text extraction technology, identify the identity code marked on the surface of the hopper from the monitoring video.

[0043] Use text extraction technology to perform text recognition on pre-processed images. This technology is based on deep learning algorithms and can automatically recognize and extract text information from images. During the recognition process, the system will perform detailed analysis and comparison on each character to ensure the accuracy of the identity code.

[0044] S103. Using a pre-trained instance segmentation algorithm, identify the type of scrap steel in the hopper from the monitoring video.

[0045] S103-1. Use a deep neural network to extract edge features, texture features and shape features from the scrap steel image in the hopper.

[0046] Deep neural networks (such as convolutional neural networks (CNNs)) are used to analyze the scrap steel images in depth, and edge features, texture features, and shape features are extracted from them. These features together constitute the basic representation of the scrap steel images, providing a rich source of information for subsequent processing. Edge features help identify the contours of objects, texture features reflect the details of the object's surface, and shape features reveal the overall structure of the object.

[0047] S103-2. Construct a feature pyramid network to fuse feature maps of different scales. The feature pyramid network transmits high-level semantic information to low-level layers through top-down paths and lateral connections while retaining detail information of the low-level layers.

[0048] The network structure of FPN usually includes a bottom-up convolutional neural network (CNN), a top-down path, and lateral connections.

[0049] Bottom-up convolutional neural network: This part is usually an existing convolutional neural network structure, such as ResNet, DenseNet, etc., which is used to extract image features. During the feature extraction process, the size of the feature map will gradually decrease, and the semantic information will gradually increase.

[0050] Top-down path: This part gradually reduces the high-level feature map to the same size as the low-level feature map through upsampling operations. Upsampling operations usually use interpolation methods, that is, based on the original image pixels, new elements are inserted between pixels using a suitable interpolation algorithm to expand the size of the original image.

[0051] Horizontal connection: Horizontal connection fuses feature maps of different scales at the same level. Specifically, the upsampled high-level feature map is fused with the low-level feature map generated from the bottom up. Usually, the channels are adjusted through a 1×1 convolution operation to match them, and then added or concatenated. The fused feature map contains both high-level semantic information and low-level detail information.

[0052] S103-3. Set a preset number of anchor points on each feature layer of the feature pyramid network, each anchor point has multiple different aspect ratios and receptive field sizes, and generate a target candidate box on each feature layer by means of a sliding window.

[0053] On each feature layer of FPN, the system sets a preset number of anchors. These anchors are rectangular boxes of fixed size and aspect ratio. They serve as the initial form of candidate targets and are used to locate possible targets in the image. Each anchor has multiple different aspect ratios and receptive field sizes, which can cover more possible target shapes and sizes.

[0054] By sliding the window, all anchor points are traversed on each feature layer to generate a series of potential target candidate boxes. The sliding window technique is a method of moving a small window (i.e., anchor point) on the image according to certain rules and processing the image content in the window at each position. Here, the sliding window technique is used to traverse the anchor points on the feature layer and generate candidate boxes.

[0055] Specifically, for each anchor point, the system generates a rectangular box of the corresponding size on the feature layer according to its aspect ratio and receptive field size. Then, through the sliding window method, all anchor points are traversed on the feature layer and corresponding candidate boxes are generated. These candidate boxes will cover the target areas that may exist in the image, providing a basis for subsequent classification and regression operations.

[0056] Specifically, the feature pyramid network generates target candidate boxes through the following steps: Data preprocessing: Perform necessary preprocessing operations on the images in the dataset, such as scaling and normalization.

[0057] Build a model: Use FPN to build a feature pyramid network model.

[0058] Generate candidate boxes: Set anchor points on each feature layer of the model and generate candidate boxes by sliding windows.

[0059] Training model: Use training data to train the model and adjust model parameters to optimize object detection performance.

[0060] Test model: Use test data to test the trained model and evaluate the performance of the model.

[0061] S103-4. Perform feature extraction on each target candidate frame to obtain its corresponding feature vector, use a classifier (support vector machine) to classify the feature vector to obtain attribute information, and the attribute information includes shape, size, structure, and material.

[0062] S103-5. Perform weighted summation of the feature vector of the target candidate box and the attribute information to obtain a fusion feature.

[0063] In order to make full use of the feature vector and attribute information of the target candidate box, the system performs a weighted sum of the two to obtain a fusion feature. This step aims to integrate information of different dimensions to improve the accuracy and robustness of target detection.

[0064] S103-6. Use a pre-trained level classifier to classify the fused features to obtain level information, and the pre-trained level classifier adopts a strategy combining cross entropy loss and focal loss function during training.

[0065] Collecting datasets: First, you need to collect a dataset containing scrap steel target images and their corresponding grade labels. This dataset should be as comprehensive as possible, containing scrap steel targets of various shapes, sizes, structures, and materials to ensure the generalization ability of the model.

[0066] Data preprocessing: Preprocess the collected data, including image scaling, normalization, data augmentation, etc. These operations help improve the robustness and accuracy of the model.

[0067] Divide the dataset into training and test sets: Divide the dataset into training and test sets for model training and validation.

[0068] Choose a pre-trained model: Choose a pre-trained model suitable for image classification tasks, such as ResNet, VGG, Inception, etc. These models have been trained on large datasets and have shown good performance.

[0069] Build a grade classifier: Based on the pre-trained model, add a fully connected layer (or classification layer) to classify the fused features. The number of output nodes of this fully connected layer should match the number of grades of the scrap steel target.

[0070] Loss function design: Since the scrap steel target grade classification task may have data imbalance problems, a strategy combining cross entropy loss and focal loss function is selected as the loss function. Cross entropy loss is used to measure the difference between the model prediction result and the actual label, while focal loss is used to focus on difficult-to-classify samples and reduce the contribution of easy-to-classify samples to the loss, thereby solving the data imbalance problem.

[0071] Set training parameters: including learning rate, batch size, number of iterations, etc. The choice of these parameters has a great impact on the training effect of the model.

[0072] Training model: Input the training set into the model, and update the model's weight parameters through the back-propagation algorithm to minimize the loss function. During the training process, you can monitor the change in the loss value and the performance of the model on the validation set so that you can adjust the training parameters and model structure in a timely manner.

[0073] Model evaluation: Evaluate the trained model on the test set and calculate indicators such as accuracy, recall, and F1 score to measure the performance of the model.

[0074] Adjust hyperparameters: Based on the performance of the model on the test set, adjust hyperparameters such as learning rate and batch size to further improve the performance of the model.

[0075] Data augmentation: Try different data augmentation methods, such as rotation, flipping, cropping, etc., to increase the generalization ability of the model.

[0076] Ensemble learning: Integrate the prediction results of multiple models to improve the overall performance. This can be achieved through voting, weighted averaging, etc.

[0077] Deploy model: Deploy the trained model to the actual application scenario.

[0078] Real-time detection and classification: Use the deployed model to detect and classify scrap steel images collected in real time to obtain the grade information of scrap steel targets.

[0079] Result feedback and optimization: Based on the feedback results in actual applications, the model is further optimized and adjusted to improve its accuracy and reliability.

[0080] S103-7. Perform regression operation on each target candidate frame, use CIOU loss function for training in the regression process, and use non-maximum suppression algorithm to remove overlapping detection frames and retain the best target detection frame.

[0081] S104. Identify hopper action.

[0082] Combining the similarity of scrap steel image features of adjacent image frames in the image frame sequence and the synchronized pressure sensor signal fluctuation value to identify the hopper action, specifically: Material preparation identification: There is no scrap steel image in the previous image frame, but there is a scrap steel image in the next image frame, and the pressure value of the pressure sensor increases; Restocking identification: The similarity of the feature vectors of adjacent scrap steel image frames does not reach the set similarity threshold, and the pressure value of the corresponding period increases; Passing identification: The similarity of the feature vectors of adjacent scrap steel image frames reaches the set similarity threshold, and the fluctuation value of the pressure value in the corresponding period is within the set fluctuation range; Consumption identification: The similarity of the feature vectors of adjacent scrap steel image frames does not reach the set similarity threshold, and the pressure value of the corresponding period decreases.

[0083] In an embodiment of the present invention, based on step S2, a possible embodiment is given below to illustrate its specific implementation scheme in a non-limiting manner.

[0084] First, a detailed database needs to be established to store the identity code of each pressure sensor (such as a unique identifier or serial number) and its corresponding device name (such as "Hopper A-Pressure Sensor 1").

[0085] This database should contain all pressure sensors registered in the system, ensuring that each sensor has a unique identification code associated with it.

[0086] In order to ensure the accuracy and timeliness of the database, we need to synchronize and update the information in the database regularly or in real time.

[0087] This may include adding newly registered pressure sensors, removing retired sensors, or updating the device name or location information of existing sensors.

[0088] Each pressure sensor in the system sends its measured pressure signal to a central processing unit (such as a PLC, computer, or cloud server) either periodically or on demand.

[0089] These signals are usually transmitted via wired or wireless means and contain key information such as sensor identity, measurement time and pressure value.

[0090] After receiving the signal, the central processing unit will first parse it and extract key data such as identity code, measurement time and pressure value.

[0091] The analysis process may involve steps such as signal format conversion, data cleaning and verification to ensure the accuracy and completeness of the data.

[0092] Next, the system will search for the sensor device name that matches the identity code in the pre-established correspondence database based on the parsed identity code.

[0093] This process is usually implemented through efficient database query algorithms, such as hash table lookup or index lookup, to ensure that the target sensor is found quickly and accurately.

[0094] Once a matching sensor device name is found, the system will retrieve the pressure signal record corresponding to the sensor from the database.

[0095] This may include historical signal data, current signal data, and any additional information associated with the signal (such as sensor status, calibration data, etc.).

[0096] After retrieving the signal, the system may perform additional data verification steps to ensure the accuracy and consistency of the data.

[0097] After verification, the data will be stored in the specified location for subsequent analysis, processing or reporting.

[0098] In an embodiment of the present invention, based on step S3, a possible embodiment is given below to illustrate its specific implementation scheme in a non-limiting manner.

[0099] When the system confirms that the action flow of the hopper is complete and includes the two key actions of material preparation and consumption, it will be considered that the hopper has completed a complete action flow.

[0100] At this point, the system generates a record containing key data such as action flow, scrap grade information and weight value.

[0101] Finally, the system will write the generated records into the database to form a complete database list.

[0102] This list can be used for subsequent tasks such as data analysis, report generation, and decision support.

[0103] System advantages and application prospects: Improve production efficiency: By automatically identifying and tracking the movement process of the hopper, the system can significantly reduce manual intervention and errors and improve production efficiency.

[0104] Optimize resource allocation: Based on the grade information and weight value of scrap steel, the system can arrange production plans and resource allocation more accurately to reduce production costs.

[0105] Enhanced data visualization: Through data visualization technology, the system can present complex production data in an intuitive and easy-to-understand way, helping managers better understand the production status.

[0106] Support intelligent decision-making: Based on big data analysis and artificial intelligence technology, the system can provide enterprises with intelligent decision-making support and help them achieve digital transformation and intelligent upgrading.

[0107] In some embodiments, the scrap steel consumption recording system may include a plurality of functional modules composed of computer program segments. The computer programs of the various program segments in the scrap steel consumption recording system may be stored in a memory of a computer device and executed by at least one processor to perform (see Figure 1 Description) Function of scrap steel consumption record.

[0108] In this embodiment, the scrap steel consumption recording system can be divided into multiple functional modules according to the functions it performs, such as Figure 3 As shown. The functional modules of the system 300 may include: an image recognition module 310, a signal acquisition module 320 and a data recording module 330. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0109] An image recognition module, used to collect monitoring video and identify the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video; A signal retrieving module, configured to retrieve the pressure sensor signal corresponding to the identity code based on the correspondence between the preset identity code and the pressure sensor device name and the received pressure sensor signal; The data recording module is used to convert the pressure sensor signal into a weight value and record the identity code and the weight value.

[0110] Optionally, as an embodiment of the present invention, the image recognition module includes: A text recognition unit, used to recognize the identity code marked on the surface of the hopper from the monitoring video using text extraction technology; The image recognition unit is used to identify the type of scrap steel in the hopper from the monitoring video using a pre-trained instance segmentation algorithm.

[0111] Optionally, as an embodiment of the present invention, the image recognition unit includes: A feature extraction subunit, for extracting edge features, texture features, and shape features from the scrap steel image in the hopper using a deep neural network; A first processing subunit is used to construct a feature pyramid network to fuse feature maps of different scales, wherein the feature pyramid network transmits high-level semantic information to low-level layers through a top-down path and lateral connections while retaining detail information of the low-level layers; A second processing subunit is used to set a preset number of anchor points on each feature layer of the feature pyramid network, each anchor point has a plurality of different aspect ratios and receptive field sizes, and generate a target candidate box on each feature layer by means of a sliding window; The first classification subunit is used to extract features from each target candidate frame to obtain its corresponding feature vector, and classify the feature vector using a classifier to obtain attribute information, wherein the attribute information includes shape, size, structure, and material; A feature fusion subunit, used for weighted summing of a feature vector of a target candidate box and the attribute information to obtain a fused feature; A second classification subunit is used to classify the fused features using a pre-trained level classifier to obtain level information, wherein the pre-trained level classifier adopts a strategy combining a cross entropy loss and a focal loss function during training; The regression processing subunit is used to perform regression operations on each target candidate box. The CIOU loss function is used for training during the regression process, and the non-maximum suppression algorithm is used to remove overlapping detection boxes and retain the best target detection box.

[0112] Optionally, as an embodiment of the present invention, the image recognition module includes: A video acquisition unit, used to acquire monitoring videos using multiple image acquisition devices deployed in the material preparation area and the converter area; A video conversion unit, used for converting the surveillance video into an image frame sequence; The action recognition unit is used to use a visual algorithm to identify the action type of the hopper based on the image frame sequence, and the action type includes material preparation, material replenishment, passing and consumption.

[0113] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiment.

[0114] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or modules, which can be electrical, mechanical or other forms.

[0115] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0116] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0117] Although the present invention has been described in detail with reference to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention.

Claims

1. A method for recording scrap steel consumption, characterized in that: include: Collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video; Based on the preset correspondence between the identity code and the pressure sensor device name, and the received pressure sensor signal, retrieve the pressure sensor signal corresponding to the identity code; The pressure sensor signal is converted into a weight value, and the identity code and the weight value are recorded.

2. The method according to claim 1, characterized in that Collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video, including: Using text extraction technology to identify the identity code marked on the surface of the hopper from the monitoring video; The type of scrap steel in the hopper is identified from the surveillance video using a pre-trained instance segmentation algorithm.

3. The method according to claim 2, characterized in that The type of scrap steel in the hopper is identified from the surveillance video using a pre-trained instance segmentation algorithm, including: Use deep neural networks to extract edge features, texture features, and shape features from scrap steel images in the hopper; Constructing a feature pyramid network to fuse feature maps of different scales, wherein the feature pyramid network transmits high-level semantic information to low-level layers through a top-down path and lateral connections while retaining detail information of the low-level layers; A preset number of anchor points are set on each feature layer of the feature pyramid network. Each anchor point has multiple different aspect ratios and receptive field sizes. A target candidate box is generated on each feature layer by means of a sliding window. Extract features from each target candidate frame to obtain its corresponding feature vector, and use a classifier to classify the feature vector to obtain attribute information, including shape, size, structure, and material; The feature vector of the target candidate box is weighted and summed with the attribute information to obtain a fusion feature; Using a pre-trained level classifier to classify the fused features to obtain level information, the pre-trained level classifier adopts a strategy combining a cross entropy loss and a focal loss function during training; A regression operation is performed on each target candidate box. The CIOU loss function is used for training during the regression process, and the overlapping detection boxes are removed through the non-maximum suppression algorithm to retain the best target detection box.

4. The method according to claim 3, characterized in that Collecting monitoring video, and identifying the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video, including: Collect monitoring videos using multiple image acquisition devices deployed in the material preparation area and converter area; Converting the surveillance video into an image frame sequence; A visual algorithm is used to identify the action type of the hopper based on the image frame sequence, and the action type includes material preparation, material replenishment, passing and consumption.

5. The method according to claim 4, characterized in that Converting the pressure sensor signal into a weight value, and recording the identity code and the weight value, comprises: Track the action flow of the same hopper based on the identity code, as well as the scrap grade information and weight value corresponding to each action in the action flow; Confirm that the action flow of the hopper includes material preparation and consumption, then confirm that the hopper completes an action flow, and write the completed action flow and the corresponding scrap steel grade information and weight value into the database list.

6. The method according to claim 5, characterized in that The method further comprises: Matching the action flow of the hopper with the operation data in the operation management platform to obtain the furnace number corresponding to the action flow of the hopper; The furnace number is written into the corresponding position of the database list.

7. A scrap steel consumption recording system, characterized in that: include: An image recognition module, used to collect monitoring video and identify the identity code marked on the surface of the hopper and the image of the scrap steel in the hopper from the monitoring video; A signal retrieving module, configured to retrieve the pressure sensor signal corresponding to the identity code based on the correspondence between the preset identity code and the pressure sensor device name and the received pressure sensor signal; The data recording module is used to convert the pressure sensor signal into a weight value and record the identity code and the weight value.

8. The system according to claim 7, characterized in that The image recognition module comprises: A text recognition unit, used to recognize the identity code marked on the surface of the hopper from the monitoring video using text extraction technology; The image recognition unit is used to identify the type of scrap steel in the hopper from the monitoring video using a pre-trained instance segmentation algorithm.

9. The system according to claim 8, characterized in that The image recognition unit comprises: A feature extraction subunit, for extracting edge features, texture features, and shape features from the scrap steel image in the hopper using a deep neural network; A first processing subunit is used to construct a feature pyramid network to fuse feature maps of different scales, wherein the feature pyramid network transmits high-level semantic information to low-level layers through a top-down path and lateral connections while retaining detail information of the low-level layers; A second processing subunit is used to set a preset number of anchor points on each feature layer of the feature pyramid network, each anchor point has a plurality of different aspect ratios and receptive field sizes, and generate a target candidate box on each feature layer by means of a sliding window; The first classification subunit is used to extract features from each target candidate frame to obtain its corresponding feature vector, and classify the feature vector using a classifier to obtain attribute information, wherein the attribute information includes shape, size, structure, and material; A feature fusion subunit, used for weighted summing of a feature vector of a target candidate box and the attribute information to obtain a fused feature; A second classification subunit is used to classify the fused features using a pre-trained level classifier to obtain level information, wherein the pre-trained level classifier adopts a strategy combining a cross entropy loss and a focal loss function during training; The regression processing subunit is used to perform regression operations on each target candidate box. The CIOU loss function is used for training during the regression process, and the non-maximum suppression algorithm is used to remove overlapping detection boxes and retain the best target detection box.

10. The system according to claim 9, characterized in that The image recognition module comprises: A video acquisition unit, used to acquire monitoring videos using multiple image acquisition devices deployed in the material preparation area and the converter area; A video conversion unit, used for converting the surveillance video into an image frame sequence; The action recognition unit is used to use a visual algorithm to identify the action type of the hopper based on the image frame sequence, and the action type includes material preparation, material replenishment, passing and consumption.

Images