基于异域数据联邦学习的金属板表面划痕检测  

Scratch Detection of Metal Plate Surface Based on Federated Learning of Exotic Data

作　　者：蔡剑锋 柏俊杰 向洪成 胡林 周涛琪 高帅 CAI Jianfeng;BAI Junjie;XIANG Hongcheng;HU Lin;ZHOU Taoqi;GAO Shuai(School of Electrical Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)

机构地区：[1]重庆科技学院电气工程学院,重庆401331

出　　处：《安阳工学院学报》2022年第6期40-44,共5页Journal of Anyang Institute of Technology

基　　金：中国高校产学研创新基金“异构智能计算项目”(2020HY06001)。

摘　　要：由于真实工业场景下的部分工业缺陷太少,导致深度学习应用在表面缺陷检测困难。针对金属板材表面缺陷检测的真实工业环境中所存在的小样本问题,本文提出了异域数据联邦检测模型,采用轻量级网络MobileNet-YOLOv3,有效利用不同领域的划痕数据,分场景训练检测模型。实验结果表明,通过异域数据结合训练的检测模型拥有一个很好的划痕检测效果,金属板训练的本地模型的mAP为94.95%,比传统数据增广方法训练模型mAP高5%,比YOLOv3的推理速度减少44.8%。Due to the lack of some industrial defects in real industrial scenarios,it is difficult for deep learning applications to detect surface defects.Aiming at the small sample problem in the real industrial environment of metal sheet surface defect detection,we propose a federated detection model for exotic data in this paper,and use the lightweight network MobileNet-YOLOv3 and scratch data from different fields to train the detection model by scene.The experimental results show that the detection model trained by combining exotic data has a good scratch detection effect.The mAP of the local model trained by the metal plate is 94.95%,which is 5%higher than that of the traditional data augmentation method training model,and the inference speed is 44.8%lower than that of YOLOv3.

关 键 词：小样本问题 异域数据 划痕检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]