摘要
圆钢表面缺陷是影响圆钢质量的重要因素。随着钢铁生产轧制节奏提升和智能化升级,人工检测、传统表面缺陷检测等方法已经难以同时满足多种类缺陷、高速在线检测等方面的需求。因此,设计了适用于圆钢表面缺陷检测的成像系统,提出分类优先网络与目标检测网络融合的圆钢表面缺陷检测方法,并将一种非缺陷样本加入网络模型的训练以提升检测精度。试验结果及应用效果表明,该方法针对凹坑、裂纹、耳子、划伤、翘皮等表面缺陷的准确识别率达到95.61%,能有效减少缺陷误报、漏报的问题。
Surface defect of round steel is an important factor affecting the quality of round steel. With the improvement of rolling rhythm and intelligent upgrading of iron and steel production, manual detection, traditional surface defect detection and other methods have been difficult to meet the needs for multiple types of defects and high-speed online detection at the same time. Therefore, an imaging system suitable for round steel surface defect detection was designed. A round steel surface defect detection method based on fusion of classification priority network and object detection network was proposed, and non-defect sample was added to the training of network model to improve the detection accuracy. The experimental results and application effect show that the accurate recognition rate of this method for surface defects such as pits, cracks, ears, scratch and warping skin is 95.61%, which can effectively reduce the problem of false alarm and missing report of defects.
作者
邓能辉
侯睿
叶俊明
DENG Neng-hui;HOU Rui;YE Jun-ming(Design and Research Institute Co.,Ltd.,University of Science and Technology Beijing,Beijing 100083,China)
出处
《中国冶金》
CAS
北大核心
2022年第12期113-121,共9页
China Metallurgy
关键词
圆钢
表面缺陷检测
深度学习
分类优先网络
目标检测
round steel
surface defect detection
deep learning
classification priority network
object detection
