摘要
在液晶玻璃基板生产过程中容易产生夹杂、气泡、锡点、节瘤、裂痕等多种缺陷,这些缺陷严重降低了液晶玻璃的性能,本文提出了一种基于深度学习的液晶玻璃基板缺陷检测,在SSD目标检测网络的基础之上,引入了ResNet中的残差模块用于主干网络的特征提取,同时对提取到的特征进行了跨通道多尺度的融合,经实验表明该方法可以有效地改善SSD网络对缺陷检测的精度,特别是提高了对小目标的检测的检测精度。
In the production process of liquid crystal glass,it is easy to produce many defects such as inclusions,bubbles,tin spots,nodules,cracks and so on.These defects seriously reduce the performance of liquid crystal glass.This paper proposes a liquid crystal glass detection based on deep learning.On the basis of SSD target detection network,the residual module in ResNet is introduced for feature extraction of backbone network,and the extracted features are fused across channels and scales,Experiments show that this method can effectively improve the detection accuracy of SSD network for defects,especially for small targets.
作者
陈城
陈炜峰
张世杰
CHEN Cheng;CHEN Wei-feng;ZHANG Shi-jie(Nanjing University of Information Science and Technology,Nanjing 210000,China)
出处
《价值工程》
2023年第8期119-121,共3页
Value Engineering
关键词
液晶玻璃
缺陷检测
改进SSD
特征融合
liquid crystal glass
defect detection
improve SSD
feature fusion