摘要
为克服TFT-LCD表面缺陷检测中边缘模糊、对比度低、图像中存在重复纹理背景等噪声的干扰,提出了种基于全卷积神经网络的端到端的快速检测算法。该算法能够通过感受域获取原图信息,并生成低对比度特征图,然后将低对比度特征图映射到高对比度特征图上,最后通过高对比度特征图上的感受域重构出高对比度缺陷图像,并将缺陷筛选出来。
In order to overcome the interference of edge blur, low contrast, and repeat texture background in TFT-LCD surface defect detection, an end-to-end fast detection algorithm based on full convolution neural network was proposed. The algorithm can obtain the original image information through the sensing domain and generate the low contrast feature maps, and then map the low contrast feature maps to the high contrast feature maps. Finally, the high contrast defect images reconstructed based on the receptive field of the high contrast feature maps. then the defects were screened out. Compared with the current commonly used algorithms, this proposed method is the most prominent in the accuracy and speed of defect detection.
作者
欧先锋
晏鹏程
向灿群
张国云
吴健辉
涂兵
郭龙源
OU Xianfeng YAN Pengcheng XIANG Canqun ZHANG Guoyun WU Jianhui TU Bing GUO Longyuan(College of Information ＆ Communication Engineering Key Laboratory of Optimization ＆ Control for Complex Systems, Hunan Institute of Science and Technology, Yueyang 414006, China)
出处
《成都工业学院学报》
2017年第3期6-10,共5页
Journal of Chengdu Technological University
基金
国家自然科学基金项目(51704115)
湖南省自然科学基金项目(2017JJ3099
2016JJ2064)
湖南省科技计划项目(2016TP1021)
湖南省研究生创新项目(CX2016B670)
湖南省教育厅科学研究项目(16C0723)
关键词
表面缺陷检测
全卷积神经网络
深度学习
端到端
感受域
surface defect detection
Full Convolution Neural Networks
deep learning
end-to-end
receptive field
