基于集成学习算法的带钢表面缺陷分类算法研究  

Research on classification algorithm of strip surface defects based on ensemble learning algorithm

作　　者：宗德祥[1] 蒋渝[2] 何永辉[1] ZONG Dexiang;JIANG Yu;HE Yonghui(Research Institute,Baoshan Iron & Steel Co. ,Ltd. , Shanghai 201999, China;Cold Rolling Plant,Baoshan Iron & Steel Co. ,Ltd. , Shanghai 201900, China)

机构地区：[1]宝山钢铁股份有限公司中央研究院,上海201999 [2]宝山钢铁股份有限公司冷轧厂,上海201900

出　　处：《宝钢技术》2021年第3期16-21,共6页Baosteel Technology

摘　　要：介绍了集成学习算法的原理和应用,针对工业现场特别是带钢表面状态的特殊性,即具备正常带钢表面,又含有不影响使用的伪缺陷带钢表面以及含有真实缺陷的带钢表面这一复杂现象,提出了一种基于集成学习算法进行缺陷过滤并结合多尺度卷积、特征金字塔与视觉注意力机制和传统特征的深度学习网络算法模型。通过对比验证,集成学习算法具有较高的准确性和鲁棒性,能够满足工业现场需求。The principle and application of the ensemble learning algorithm are introduced.Aiming at the particularity of the surface condition of the strip steel in the industrial field,it has both the normal strip surface and the pseudo-defect strip surface that does not affect the use and the strip surface with real defects.A deep learning network algorithm model based on ensemble learning algorithm for defect filtering combined with multi-scale convolution,feature pyramid and visual attention model and traditional features is proposed.Through comparison and verification,the Ensemble learning algorithm has high accuracy and robustness,and can meet the needs of sites.

关 键 词：集成学习 目标分类 带钢表面检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]