Solutions to small datasets in defects detection based on Markov chain Monte Carlo simulations

Xin Yan Wang; Hua Fan

doi:10.1117/12.2679161

14 June 2023 Solutions to small datasets in defects detection based on Markov chain Monte Carlo simulations

Xin Yan Wang, Hua Fan

Proceedings Volume 12725, International Conference on Pure, Applied, and Computational Mathematics (PACM 2023); 127250S (2023) https://doi.org/10.1117/12.2679161
Event: International Conference on Pure, Applied, and Computational Mathematics (PACM 2023), 2023, Suzhou, China

Abstract

In the processes of industrial defects detections, deep neural networks have received much favor for its high efficiency and low labor cost. As is known, a big amount of dataset is the precondition to train any network. However, the datasets of industrial defects tended to be small and hard to obtain. To solve the problem above, we raise a thought of simulation based on MCMC aiming to sample from the posterior distribution of the defect’s images, which is viewed as the limit distribution of certain Markov Chain. Once we capture the posterior distribution, the law of the defects is revealed. 1Therefore, generating pseudo samples by the law is quite reasonable. It has been proved that the algorithm is practicable by advanced devices and economical in the long term. The significance of this work is to raise a practicable thought towards the problem of mini-batch industrial datasets. Code is available at https://github.com/MrSmallWang/solutions-tosmall- datasets.

Citation Download Citation

Xin Yan Wang and Hua Fan "Solutions to small datasets in defects detection based on Markov chain Monte Carlo simulations", Proc. SPIE 12725, International Conference on Pure, Applied, and Computational Mathematics (PACM 2023), 127250S (14 June 2023); https://doi.org/10.1117/12.2679161

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available