摘要
为解决带钢生产过程中产品表面缺陷自动检测问题,通过分析不同类型视觉检测算法特点,选取Faster-RCNN、YOLOv4和CenterNet 3种算法,采用Python语言实现3种算法设计并应用于带钢表面缺陷检测中。通过对带钢表面6种典型缺陷1800张图像进行训练和测试,YOLOv4和Faster-RCNN算法的识别精度达70%以上,在带钢实际生产过程中具有较高应用价值。实验对比不同缺陷的识别精度,对于斑块、划痕、麻点、夹杂等边缘清晰对比度高的缺陷,适合采用机器识别算法进行检测。
In order to solve the automatic detection problem of surface defects in strip steel production,the characteristics of different types of visual detection algorithms are analyzed,three algorithms:Faster-RCNN,yolov4 and centernet are selected,and Python language is used to realize the three kinds of algorithms,and the application of the algorithms in strip steel surface defect detection.By training and testing 1800 images of six typical defects on the surface of strip steel,the recognition accuracy of yolov4 and Faster-RCNN algorithm can reach more than 70%,which has high application value in the actual production process of strip steel.The recognition accuracy of different defects is compared by experiments.Such clear edges of high contrast defects as patches,scratches,pitting and inclusion,etc.are suitable for detection with machine recognition algorithm.
作者
杜孟新
毕玉
杜鹏昊
DU Mengxin;BI Yu;DU Penghao(Instrumentation Technology and Economy Institute,Beijing 100055,China;Beijing Technology and Business University,Beijing 100048,China)
出处
《火力与指挥控制》
CSCD
北大核心
2022年第8期132-135,共4页
Fire Control & Command Control
基金
国家重点研发计划重大专项科技创新2030-“新一代人工智能”重大资助项目(2018AAA0101801)。
