High-resolution industrial radiography using convolutional neural networks
Yunu Ha, Seungwoo Ha, Jinwoo Kim, Hanbean Youn, Ho Kyung Kim · Journal of Instrumentation · 2020
For quality control of electronic products with high-density electronic packaging and corresponding multi-layer printed circuit boards, x-ray inspection is not optional anymore. For high-resolution industrial radiography, we develop a de-blur filter using serial cascades of convolutional neural networks. We increase the number of channels along the network depth to accept complex features even at deeper depths. Various ways to increase the channel numbers are described. The performance of the networks is quantified in comparisons with the state-of-the-art networks. The result shows a large correlation between the performance and the total number of weights built in a network. Including the detailed results, this paper discusses the limitation of the present study, and future study.