Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A Grain Boundary Defects Detection Algorithm with Improved Localization Accuracy Based on EfficientDet

A Grain Boundary Defects Detection Algorithm with Improved Localization Accuracy Based on... Defects detection is one of the most important tasks in the materials industry. The existence of grain boundary defects causes the crystal structure to be susceptible to corrosion, which leads to a significant reduction in metal plasticity, hardness, and tensile strength. At present, some deep learning methods have been proposed to detect such problems based on HRTEM (high-resolution transmission electron microscope) images of crystal defects. However, they face the problem of low detection rate and low localization accuracy. In this paper, an improved detection algorithm has been proposed. Firstly, to balance the performance and complexity, the EfficientDet based network is adopted in the algorithm. Secondly, a weighted fusion module is introduced to the EfficientDet network to integrate the output of features from the backbone and BiFPN (bidirectional feature pyramid network) to achieve good detection accuracy. Finally, the location loss function of the network is substituted by \documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$CIoU$$\end{document} (complete intersection over union) loss, which can improve the defects localization accuracy. The experimental results show that compared to the initial algorithms, the AP (average precision) value of grain boundary defect detection can be improved by about 5%. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Automatic Control and Computer Sciences Springer Journals

A Grain Boundary Defects Detection Algorithm with Improved Localization Accuracy Based on EfficientDet

Download PDF
12 pages

Loading next page...
 
/lp/springer-journals/a-grain-boundary-defects-detection-algorithm-with-improved-mXZs3YZeEd

References (20)

Publisher
Springer Journals
Copyright
Copyright © Allerton Press, Inc. 2023. ISSN 0146-4116, Automatic Control and Computer Sciences, 2023, Vol. 57, No. 1, pp. 81–92. © Allerton Press, Inc., 2023.
ISSN
0146-4116
eISSN
1558-108X
DOI
10.3103/s0146411623010078
Publisher site
See Article on Publisher Site

Abstract

Defects detection is one of the most important tasks in the materials industry. The existence of grain boundary defects causes the crystal structure to be susceptible to corrosion, which leads to a significant reduction in metal plasticity, hardness, and tensile strength. At present, some deep learning methods have been proposed to detect such problems based on HRTEM (high-resolution transmission electron microscope) images of crystal defects. However, they face the problem of low detection rate and low localization accuracy. In this paper, an improved detection algorithm has been proposed. Firstly, to balance the performance and complexity, the EfficientDet based network is adopted in the algorithm. Secondly, a weighted fusion module is introduced to the EfficientDet network to integrate the output of features from the backbone and BiFPN (bidirectional feature pyramid network) to achieve good detection accuracy. Finally, the location loss function of the network is substituted by \documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$CIoU$$\end{document} (complete intersection over union) loss, which can improve the defects localization accuracy. The experimental results show that compared to the initial algorithms, the AP (average precision) value of grain boundary defect detection can be improved by about 5%.

Journal

Automatic Control and Computer Sciences – Springer Journals

Published: Feb 1, 2023

Keywords: HRTEM images; grain boundary defect detection; EfficientDet network; weighted fusion, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CIoU$$\end{document} loss

There are no references for this article.