Show simple item record

contributor authorLiu, Tianyuan
contributor authorBao, Jinsong
contributor authorWang, Junliang
contributor authorZhang, Yiming
date accessioned2022-02-04T14:23:52Z
date available2022-02-04T14:23:52Z
date copyright2020/01/03/
date issued2020
identifier issn1530-9827
identifier otherjcise_20_2_021005.pdf
identifier urihttp://yetl.yabesh.ir/yetl1/handle/yetl/4273578
description abstractMachine vision has a wide range of applications in the field of welding. The rise of convolutional neural network (CNN) provides a new way to extract visual features of welding. Due to the limitation of the small size of our molten pool dataset, the regularization of the CNN model is necessary to prevent overfitting. We propose a coarse-grained regularization method for convolution kernels (CGRCKs), which is designed to maximize the difference between convolution kernels in the same layer. The algorithm performance was tested on our self-made dataset and other public datasets. The results show that the CGRCK method can extract multi-faceted features. Compared with L1 or L2 regularization, the proposed method works great on CNNs and introduces little overhead cost to the training.
publisherThe American Society of Mechanical Engineers (ASME)
titleA Coarse-Grained Regularization Method of Convolutional Kernel for Molten Pool Defect Identification
typeJournal Paper
journal volume20
journal issue2
journal titleJournal of Computing and Information Science in Engineering
identifier doi10.1115/1.4045294
page21005
treeJournal of Computing and Information Science in Engineering:;2020:;volume( 020 ):;issue: 002
contenttypeFulltext


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record