Intelligent grading method for scrap steel

Through deep learning algorithms, the automatic identification and classification of scrap steel types is achieved, which solves the problems of low efficiency and safety risks of traditional manual classification, and realizes all-weather automated operation of intelligent scrap steel classification, improving the accuracy and efficiency of classification.

CN118967567APending Publication Date: 2024-11-15JIANGSU SHAGANG HIGH-TECH INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410936100.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Traditional artificial scrap grades have harsh environments, unstable results, extensive inspection processes and rely on experience, resulting in low efficiency and high safety risks, making it difficult to achieve all-weather automated operation.

Method used

The FCOS object detection model based on deep learning algorithm and the SOLOv2 instance segmentation algorithm are adopted to realize automatic identification and grading of scrap steel types. Through image acquisition and real-time monitoring, an intelligent grading system is built to achieve all-weather and full-process automated operation.

Benefits of technology

It improves the accuracy and efficiency of scrap steel grade, reduces manual intervention, reduces safety risks, realizes automated identification and digital statistics, unifies standards, and reduces disputes with suppliers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967567A_ABST
    Figure CN118967567A_ABST
Patent Text Reader

Abstract

The invention provides an intelligent grading method for waste steel, which comprises the following steps of: firstly, judging whether a task has a starting condition or not, distributing server resources for intelligent grading, identifying a car hopper and detecting a grab bucket in real time after the task is distributed, learning a waste steel grading standard by using an improved SOLOv2 deep learning algorithm for material type image instance segmentation, and finally, carrying out grading on the waste steel. A scrap steel intelligent grading instance segmentation model is constructed to automatically identify the material type, the impurity ratio, the dangerous explosion closed container and the like of the scrap steel of the whole vehicle, multi-angle comprehensive operation is carried out, it is ensured that grading is accurate, and the standard requirement of scrap steel grading is met. According to the invention, unmanned scrap steel grading is taken as a final target, and all-weather and full-process automatic operation can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of scrap steel grading in the metallurgical industry, and specifically but not limited to, relates to an intelligent scrap steel grading method. Background Art

[0002] Traditional steel smelting enterprises consume a large amount of scrap steel every year. The quality of scrap steel is directly related to the type, quality and smelting cycle of steel smelting, which in turn affects the efficiency of steel mills. The sources of scrap steel are complex, with various types and different materials.

[0003] The following problems exist in the manual scrap steel grading method: (1) The working environment of manual scrap steel grading is poor: Since scrap steel grading requires scrap steel inspectors to visually determine the scrap steel category, in order to observe the scrap steel in the truck bed closely, the scrap steel supply truck is on average 2-3 meters high, and inspectors need to get on and off the truck frequently, which is labor-intensive and has high safety risks. (2) The results of manual scrap steel grading are unstable: The sources of scrap steel are complex and varied, and the quality of manual scrap steel grading results depends entirely on the personal experience of inspectors. (3) The inspection process of manual scrap steel grading is rough: For each truckload of scrap steel delivered by the scrap steel supplier, experienced scrap steel inspectors can accurately give the specific category of each piece of scrap steel if there is enough time, thereby determining the type of scrap steel it belongs to. However, considering the daily planned purchase volume of scrap steel and the requirements of the production rhythm, the time is tight and the task is heavy. Scrap steel inspectors cannot accurately determine the specific category of each piece of scrap steel and calculate the proportion of the material type within a limited time. (4) Qualified graders need long-term experience accumulation: Due to the variety of scrap steel material types, qualified scrap steel graders need long-term experience accumulation and constantly enrich the types of material type identification in order to achieve accurate grading. This also makes the recruitment standards for scrap steel graders relatively high, and the recruited personnel need to have experience in scrap steel acceptance and classification. (5) Scrap steel is a low-carbon, environmentally friendly, recyclable renewable resource. Therefore, strengthening the recycling of scrap steel is an important measure for the steel industry to improve resource utilization efficiency, reduce resource consumption and pollution emissions.

[0004] In view of the business problems and challenges existing in manual scrap steel grading, a new scrap steel grading method needs to be provided to solve at least some of the above problems. Summary of the invention

[0005] In response to one or more problems in the prior art, the present invention proposes an intelligent scrap steel grading method, which realizes the functions of automatic identification of scrap steel material type, impurity ratio, hazardous and explosive closed containers, etc. The grading is accurate, meets the standard requirements for scrap steel grading, and can achieve all-weather, full-process automated operation.

[0006] The technical solution to achieve the purpose of the present invention is:

[0007] A method for intelligent grading of scrap steel comprises the following steps:

[0008] Step 1: Obtain the scrap steel grading task status, determine whether the task meets the start conditions, and allocate server resources to the task if the start conditions are met;

[0009] Step 2: Use the target detection model based on the deep learning algorithm FCOS to identify the bucket and define the range of the bucket, detect the position and movement state of the grab bucket in real time, and determine whether the grab bucket starts to unload. When the grab bucket starts to unload, go to step 3;

[0010] Step 3: Real-time detection of the grab bucket and identification of whether the grab bucket is close to or far from the truck bucket; during the period from when the grab bucket is close to the truck bucket to when it is far from the truck bucket, the area where the grab bucket grabs the scrap steel in the truck bucket is recorded, and when the grab bucket is identified as being away from the truck bucket, the image acquisition process for the area where the grab bucket grabs the scrap steel in the truck bucket is triggered;

[0011] Step 4: Use the improved SOLOv2 deep learning algorithm for material type image instance segmentation to learn the scrap steel grading standard, build a scrap steel intelligent grading instance segmentation model, and identify the scrap steel material type based on the scrap steel intelligent grading instance segmentation model. Display the identified scrap steel and similar material types of warehoused scrap steel through pictures, and provide the location information of unqualified scrap steel. At this point, the grading cycle of the collected images is completed;

[0012] Step 5: Continuously monitor the task status. If the task is not completed, jump to step 3. If the task is completed, jump to step 6. The task completion means that all scrap steel in the truck bucket has been unloaded.

[0013] Step 6: Push the final grading results of all scrap steel to the system background, and the system background reduces the task volume of the server instance by 1.

[0014] Furthermore, in the intelligent scrap steel grading method of the present invention, the scrap steel grading task situation in step 1 includes: a server address, port, task id, task location, vehicle license plate number, task status, etc. corresponding to the task.

[0015] Furthermore, in the intelligent scrap steel grading method of the present invention, the image acquisition process in step 3 includes: controlling the camera to rotate and zoom to the area where the grab grabs the scrap steel in the bucket, and acquiring one or several images of the area, and then controlling the camera to return to a non-zoom state for photographing the entire vehicle.

[0016] Furthermore, in the scrap steel intelligent grading method of the present invention, the step 4 of constructing the scrap steel intelligent grading instance segmentation model comprises:

[0017] Step 4-1: Collect images of scrap steel materials in daily projects and perform image preprocessing;

[0018] Step 4-2: Use annotation tools to perform instance segmentation and annotation on the preprocessed image, and finely annotate the material type true value information of different areas;

[0019] Step 4-3: Save the material type true value file corresponding to the scene image and the data set file with instance segmentation labels in step 4-2, and output the original data set with annotations;

[0020] Step 4-4: Divide the data into training set and test set: randomly select 80% of the scrap steel material images as the training set and the remaining 20% ​​as the test set;

[0021] Step 4-5: Use the SOLOv2 instance segmentation algorithm to perform transfer learning on the annotated original data set based on the pre-trained model to obtain a high-precision scrap steel intelligent grading instance segmentation model.

[0022] Furthermore, in the intelligent steel scrap grading method of the present invention, the image of the steel scrap material type in step 4-1 is saved in a local folder image_org in png format, and the resolution of the image is 1920*1080.

[0023] Furthermore, in the intelligent scrap steel grading method of the present invention, the material type true value file and the data set file in step 4-3 are saved in the label_org folder.

[0024] Furthermore, in the intelligent scrap steel grading method of the present invention, the identification of scrap steel material types in the collected images in step 4 includes: identifying characteristic information such as data, thickness, area, etc. of different types of scrap steel in the mixture, and calculating the proportion of different types of the mixture.

[0025] Furthermore, in the scrap steel intelligent grading method of the present invention, the starting conditions in step 1 include: a vehicle loaded with scrap steel enters the unloading point, and the scrap steel intelligent grading task is started, and the task obtains the current idle interface.

[0026] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0027] 1. The intelligent scrap steel grading method of the present invention realizes the automatic collection of scrap steel images and remote identification, improves work efficiency through automatic identification of scrap steel, and can realize digital statistics of various material types and provide the specific material type proportion distribution.

[0028] 2. The intelligent scrap steel grading method of the present invention adopts computer vision technology based on deep learning to model the types of scrap steel, and trains scrap steel recognition and material type determination models. At the same time, based on the collected images, the model is used to infer the scrap steel material type grade, which can realize the archiving of the determination results and facilitate the tracing back of historical scrap steel grading information.

[0029] 3. The intelligent scrap steel grading method of the present invention, by strictly controlling the quality of model training data, realizes the solidification of manual scrap steel grading experience, unifies the scrap steel grading standards and the consistency of scrap steel grading results, and is also conducive to reducing disputes with suppliers. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the present invention and are used together with the description to explain the embodiments of the present invention, but do not constitute a limitation of the present invention. In the accompanying drawings:

[0031] Figure 1 A schematic diagram of the deployment scheme of the intelligent scrap steel grading method of the present invention is shown.

[0032] Figure 2 A schematic diagram showing the use process of the intelligent scrap steel grading method of the present invention is shown. DETAILED DESCRIPTION

[0033] In order to further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, rather than limiting the claims of the present invention.

[0034] The description in this section is only for typical embodiments, and the present invention is not limited to the scope of the embodiments. The combination of different embodiments, the mutual replacement of some technical features in different embodiments, and the mutual replacement of the same or similar prior art means with some technical features in the embodiments are also within the scope of the description and protection of the present invention.

[0035] The main problem solved by the present invention is to realize intelligent grading of scrap steel through algorithms. The method first determines whether the task meets the starting conditions through algorithms and allocates server resources for intelligent grading. After the task allocation is completed, image recognition is adopted, and AI technology is used to learn the scrap steel grading standards. The scrap steel type, impurity ratio, and explosive-proof sealed containers of the whole vehicle are automatically identified, and comprehensive calculations are performed from multiple angles to ensure accurate grading and compliance with the standard requirements for scrap steel grading. The system takes unmanned scrap steel grading as its ultimate goal, and can achieve all-weather, full-process automated operation. The following is combined with Figure 1 , 2 The following describes in detail how the intelligent scrap steel grading method of the present invention can achieve intelligent scrap steel grading.

[0036] A method for intelligent grading of scrap steel comprises the following steps:

[0037] Step 1. After the scrap steel supplier's ship enters the dock and docks, the vehicle drives into the unloading point and triggers the scrap steel intelligent grading system on the mobile phone APP.

[0038] Step 2. The task distribution service randomly obtains the currently idle interface instance.

[0039] Step 3. The distribution service detects the health status of the acquired interface instance.

[0040] Step 4. If the current instance meets the startup requirements, a startup command is issued. If the command is successfully started, jump to step 5. If the command fails to start, jump to step 10. If the startup requirements are not met, the corresponding failure reason is returned, and the driver will jump to step 2 after retrying.

[0041] Step 5. Record the task status of the instance, including server address, port, task ID, task location, vehicle license plate number, task status, etc. At the same time, the corresponding server instance task amount is increased by 1 and jump to step 6. If the loading task needs to be paused, jump to step 11. After the loading task is completed, jump to step 13.

[0042] Step 6. The algorithm first completes the identification of the bucket and defines the scope of the bucket. It then identifies the position and motion state of the grab (magnetic crane) to determine whether the grab (magnetic crane) has started to unload, and then jumps to step 7 to trigger the process of collecting pictures.

[0043] Step 7. When the grab (magnetic crane) approaches or moves away from the truck bucket, it is necessary to detect in real time whether the grab (magnetic crane) appears in the monitoring screen and identify whether it is approaching or moving away from the truck bucket. In the period from approaching to moving away, the position of the grab (magnetic crane) is detected in real time, the area where the grab (magnetic crane) grabs scrap steel is recorded, and when the grab (magnetic crane) is identified to leave the state, the image acquisition command is triggered. The camera rotates and zooms, and controls the camera to zoom to the area according to the recorded scrap steel grabbing area, and captures one or several clear pictures. Then the camera returns to the non-zoom state of shooting the entire vehicle before, and proceeds to step 8.

[0044] Step 8. Construct an instance segmentation model for intelligent scrap steel grading, and identify the specific material type of scrap steel for the captured images. In the scrap steel grading task, due to the characteristics of scrap steel's variable shapes and large size differences, model reasoning and feature extraction are performed on the images, and the result fusion and post-processing are completed to identify the data, thickness, area and other feature information of different types of scrap steel in the mixed materials, calculate the proportion of different types of mixed materials, and finally give the proportion of different material types. The results can be displayed through pictures, including the identified scrap steel and its similar warehoused scrap steel, and focus on the location of unqualified scrap steel and other information. At this point, a grading cycle is completed, and jump to step 9.

[0045] Step 9. The system continuously monitors the task status. If the task is not completed, jump to step 7. The whole vehicle needs to go through 18 to 40 grading cycles so that all scrap steel in the vehicle can be identified. If the task is completed, jump to step 13.

[0046] Step 10. If the command fails to start, the failure reason is returned to prompt the driver to try again. If the driver restarts the task, jump to step 2.

[0047] Step 11. The driver clicks to pause the task, and the task distribution service sends a task pause instruction to the corresponding server instance according to the task situation, and updates the corresponding task status in the system to pause. When the task can be continued, the driver clicks to continue the task and jumps to step 12.

[0048] Step 12. The task distribution service sends a continue instruction to the corresponding server instance according to the task situation, and updates the corresponding task status in the system to the start status, and jumps to step 7. When the loading task is completed, jump to step 13.

[0049] Step 13. The task distribution service sends an end instruction to the corresponding server instance according to the task situation. The server terminates the task to release server resources and updates the corresponding task status in the system to the end status. The intelligent grading algorithm pushes the final grading result to the system background, and the system background reduces the task amount of the server instance by 1.

[0050] Step 14. The scrap steel grader reviews and confirms the grade results through the system.

[0051] Step 15. The client's business personnel inquire about the grading results through the system.

[0052] At this point, the current vehicle loading process is completed.

[0053] Among them, the construction of the scrap steel intelligent grading instance segmentation model includes:

[0054] Step 1). Obtain pictures of various types of scrap steel collected in daily projects and save them in the local folder image_org in png format, where the resolution of the picture is 1920*1080;

[0055] Step 2). Use annotation tools to annotate the preprocessed image set, perform instance segmentation and annotation on the processed stockpile images, and finely annotate the material type true value information in different areas;

[0056] Step 3). Save the material area truth file corresponding to the scene image and the instance segmentation label file of the material participating target to the label_org folder;

[0057] Step 4). Divide the data into training set and test set; randomly select 80% of the images in the image_org folder as the training set and put them into the training folder, and the remaining 20% ​​as the test set and put them into the testing folder, then package the data set and put it into the corresponding directory of the deep learning model training environment in Huawei GPU training cloud server;

[0058] Step 5). By using the SOLOv2 instance segmentation algorithm, the pre-trained model is used for transfer learning under the condition of labeled data set to obtain a high-precision scrap steel intelligent grading instance segmentation model.

[0059] The intelligent scrap steel grading method of the present invention adopts a more intelligent end-side camera, which can adapt to the complex environment of the scrap steel grading site. During the loading process, the intelligent imaging model is used to identify the status of the steel grabber and the magnetic hanger, determine the scrap steel material change area and the best time to capture the picture, and reduce repeated sampling while ensuring that the captured image covers the material type of the entire vehicle. The precise positioning of the shooting area and the regional focus shooting can retain more scrap steel material type details, achieve a higher degree of restoration of the scrap steel material type, and provide high-quality data guarantee for subsequent scrap steel identification and grading. The side server is more stable and meets the real-time and high reliability requirements of scrap steel grading. Side deployment can reduce inference delay, achieve real-time result feedback, and reduce the impact of network delay on the system. The standard, unified, and open interface on the application side supports the docking and expansion of multiple scrap steel grading applications.

[0060] The description and application of the present invention here are illustrative, and it is not intended to limit the scope of the present invention to the above-mentioned embodiments. The effects or advantages involved in the specification may not be reflected in the actual experimental examples due to the uncertainty of specific condition parameters or other factors, and the effects or advantages are not used to limit the scope of the invention. The deformation and change of the embodiments disclosed here are possible, and the replacement of the embodiments and the various equivalent parts are well known to those of ordinary skill in the art. It should be clear to those skilled in the art that the present invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials and parts without departing from the spirit or essential features of the present invention. Other deformations and changes can be made to the embodiments disclosed here without departing from the scope and spirit of the present invention.

Claims

1. A method for intelligent grading of scrap steel, characterized in that: The following steps are involved: Step 1: Obtain the scrap steel grading task status, determine whether the task meets the start conditions, and allocate server resources to the task if the start conditions are met; Step 2: Use the target detection model based on the deep learning algorithm FCOS to identify the bucket and define the range of the bucket, detect the position and movement state of the grab bucket in real time, and determine whether the grab bucket starts to unload. When the grab bucket starts to unload, go to step 3; Step 3: Real-time detection of the grab bucket and identification of whether the grab bucket is close to or far from the truck bucket; during the period from when the grab bucket is close to the truck bucket to when it is far from the truck bucket, the area where the grab bucket grabs the scrap steel in the truck bucket is recorded, and when the grab bucket is identified as being away from the truck bucket, the image acquisition process for the area where the grab bucket grabs the scrap steel in the truck bucket is triggered; Step 4: Use the improved SOLOv2 deep learning algorithm for material type image instance segmentation to learn the scrap steel grading standards and build an instance segmentation model for intelligent scrap steel grading; Based on the scrap steel intelligent grading instance segmentation model, the scrap steel type is identified for the image collected in step 3, and the identified scrap steel and similar scrap steel of the incoming warehouse are displayed through pictures, and the location information of unqualified scrap steel is given. At this point, the grading cycle of the image collected this time is completed; Step 5: Continuously monitor the task status. If the task is not completed, jump to step 3. If the task is completed, jump to step 6. The task completion means that all scrap steel in the truck bucket has been unloaded. Step 6: Push the grading results of all scrap steel to the system background, and the system background will reduce the task volume of the server instance by 1.

2. The intelligent grading method for scrap steel according to claim 1, characterized in that: The scrap steel grading task situation described in step 1 includes: the server address, port, task ID, task location, vehicle license plate number, and task status corresponding to the task.

3. The intelligent scrap steel grading method according to claim 1, characterized in that: The image acquisition process in step 3 includes: controlling the camera to rotate and zoom to the area where the grab grabs the scrap steel in the bucket, and acquiring one or several images of the area, and then controlling the camera to return to a non-zoom state for photographing the entire vehicle.

4. The intelligent scrap steel grading method according to claim 1, characterized in that: The construction of the intelligent scrap steel grading instance segmentation model in step 4 includes: Step 4-1: Collect images of scrap steel materials in daily projects and perform image preprocessing; Step 4-2: Use annotation tools to perform instance segmentation and annotation on the preprocessed image, and finely annotate the material type true value information of different areas; Step 4-3: Save the material type true value file corresponding to the scene image and the data set file with instance segmentation labels in step 4-2, and output the original data set with annotations; Step 4-4: Divide the data into training set and test set: randomly select 80% of the scrap steel material images as the training set and the remaining 20% ​​as the test set; Step 4-5: Use the SOLOv2 instance segmentation algorithm to perform transfer learning on the annotated original data set based on the pre-trained model to obtain a high-precision scrap steel intelligent grading instance segmentation model.

5. The intelligent scrap steel grading method according to claim 4, characterized in that: The image of the scrap steel material in step 4-1 is saved in the local folder image_org in png format, and the resolution of the image is 1920*1080.

6. The intelligent scrap steel grading method according to claim 4, characterized in that: In step 4-3, the material type true value file and data set file are saved in the label_org folder.

7. The intelligent scrap steel grading method according to claim 1, characterized in that: In step 4, the scrap steel type recognition of the collected images includes: recognizing the characteristic information of data, thickness, and area of ​​different types of scrap steel in the mixture, and calculating the proportion of different types of the mixture.

8. The intelligent scrap steel grading method according to claim 1, characterized in that: The starting conditions in step 1 include: a vehicle loaded with scrap steel enters the unloading point and starts the scrap steel intelligent grading task, and the task obtains the current idle interface.