Author:
Teng Bowen,Zhao Hongjian,Jia Pei,Yuan Junfeng,Tian Chenghua
Abstract
Abstract
The defect detection of ceramic sanitary ware is often done manually, which is inefficient and unhealthy. The development of deep learning technology makes a non-contact and high-efficient method of inspection possible. In this paper, based on VGG-16 network, we proposed an optimized method for ceramic sanitary ware defect detection. We carried out pre-processing to denoise and enhance the original image data, improved activation function of MReLU, and used transfer learning method to train the model. The results of the test on the ceramic sanitary ware defection data set showed that the accuracy can reach 97.48%, which is 6.46% higher than that of the model using the ReLU activation function. The detection speed can reach 7 fps, which could meet the requirements of industrial online real-time production.
Subject
General Physics and Astronomy
Reference12 articles.
1. Deep learning [J];Lecun;Nature,2015
2. Progress and Prospects of the Application of Deep Learning in Target Visual Detection [J];Zhang;Acta Automatica Sinica,2017
3. Gradient-based learning applied to document recognition [J];Lecun;Proceedings of the IEEE,1998
4. Handwritten Digit Recognition with a Back-Propagation Network [J];Cun;Advances in neural information processing systems,1997
5. A convolutional neural network cascade for face detection [C];Li,2015
Cited by
5 articles.
订阅此论文施引文献
订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献