基于改进SAE网络的织物疵点检测算法

发布时间：2018-06-01 00:14

本文选题：深度学习 + 卷积自编码器　；参考：《电子测量与仪器学报》2017年08期

【摘要】：针对传统织物缺陷检测手工提取特征困难,疵点样本有限的问题,结合卷积自编码器(CAE),提出一种基于Fisher准则的栈式去噪自编码器算法(FSDAE)。首先从原始图像中截取若干小块图像,采用稀疏自编码器(SAE)训练,得到小块图像的稀疏性特征;其次利用该特征,初始化CAE网络参数,提取原始图像的低维特征;最后将该特征数据送入FSDAE网络进行疵点检测分类。分别对3类织物进行测试,实验结果表明,算法能够有效地提取织物图像的分类特征,且通过加入Fisher准则,提高了织物疵点的检测率。
[Abstract]:Aiming at the problem of traditional fabric defect detection which is difficult to extract features manually and limited defect samples, a stack de-noising self-encoder algorithm based on Fisher criterion is proposed. Firstly, some small images are intercepted from the original image, and the sparse self-encoder is used to obtain the sparse feature of the small block image, and then the CAE network parameters are initialized to extract the low-dimensional feature of the original image. Finally, the feature data is sent into FSDAE network for defect detection and classification. The experimental results show that the algorithm can effectively extract the classification features of fabric images, and the detection rate of fabric defects is improved by adding Fisher criterion.
【作者单位】：西安工程大学电子信息学院;
【基金】：国家自然科学基金(61301276) 陕西省工业科技攻关项目(2015GY034)资助
【分类号】：TP391.41;TS101.97

【相似文献】