摘要
钢包管理是炼钢厂生产管理的重要环节,而钢包号识别是钢包管理的关键。针对钢包灰色表面直接喷涂白色颜料的现实,基于YOLOv5神经网络框架,开发了多钢包和多钢包的多目标检测算法。首先对钢包图像进行钢包真实框和钢包号真实框标注,然后生成多钢包图像训练集,利用预测框参数矩阵将锚框转变为预测框,利用非极大值抑制方法筛选预测框,获取钢包区域图像和钢包号区域图像;最后利用残差网络进行数字识别,给出钢包号和相应概率。结果表明,本算法的钢包区域识别准确率高达98%,钢包号数字区域识别准确率达到98%,钢包号识别准确率可达到99%,说明此算法能够有效地实现钢包号的精确识别,为炼钢厂钢包智能调度管理提供可靠的数据支持。
Ladle management is an important part of production management in steelmaking mills,and the recognition for ladle number plays an important role in ladle management.It focuses on the reality of white pigment spraying on the gray surface of ladles directly.Based on YOLOv5 neural network framework,a multi-objective detection algorithm is developed to identify multiple ladles and multiple ladle numbers.Firstly,the ladle images are marked by ground truth box of ladle and ladle number,and then a multi-ladle image training set is generated.The anchor box is transformed into a prediction box by the prediction box parameter matrix,and the non-maximum suppression method is applied to choose the prediction boxes,to obtain ladle area images and ladle number area images;Finally,the residual network is applied to identify the number,and to provide the ladle number and the related probability.The results show that,the precision of ladle area recognition in this algorithm is 98%,the precision of ladle number area recognition is 98%,and the precision of ladle number recognition can reach 99%.This indicates that this algorithm can effectively achieve accurate recognition of ladle numbers and provide reliable data support for intelligent scheduling and management of ladles in steelmaking mill.
作者
莫瑷玮
张爽
雷洪
余翔
郭金溥
马毓
MO Aiwei;ZHANG Shuang;LEI Hong;YU Xiang;GUO Jinpu;MA Yu(Key Laboratory of Electromagnetic Processing of Materials,Ministry of Education,Northeastern University,Shenyang110819,China;School of Metallurgy,Northeastern University,Shenyang 110819,China;College of Information Science and Engineering,Northeastern University,Shenyang 110819,China)
出处
《工业加热》
2026年第1期29-33,38,共6页
Industrial Heating
关键词
钢包号
锚框
预测框
真实框
多目标检测
ladle number
anchor box
prediction box
ground truth box
multi-object recognition
