Study on detection of dent defects of polariser based on deep convolutional generative adversarial network
by Pengfei Shi
International Journal of Materials and Product Technology (IJMPT), Vol. 68, No. 1/2, 2024

Abstract: The existing techniques of polariser detection only concern whether the polarisers have defects or not and do not classify them as specialised. In addition, lightweight CNN architectures proposed for defect classification of polarisers are based on limited samples. In order to attack the aforementioned issues, a novel grating imaging mechanism based on an adsorption transport platform is designed for a certain defect, dent. Multi-scale negative samples with dent defects and positive samples with other defects or not are expanded by a deep convolutional generative adversarial network (DCGAN). O sets, 64_10000 sets and 128_10000 sets (referred to as the original data, 64*64 generated data and 128*128 generated data) are trained on multiple convolutional neural networks (AlexNet, VGGNet, GoogLeNet, ResNet, SqueezeNet, MoblieNet, ShuffleNet) respectively, the obtained models are then validated on two new samples. Empirically show that ResNet obtained by 64G+128G perform better than others, classification accuracy rate of the new model is up to 94.94%.

Online publication date: Thu, 22-Feb-2024

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Materials and Product Technology (IJMPT):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com