Deep Architecture Based Spalling Severity Detection System Using Encoder-Decoder Networks
Deep Architecture Based Spalling Severity Detection System Using Encoder-Decoder Networks
复制标题
DOI：
10.1007/978-3-031-20716-7_26
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Tamanna Yasmin;C. Le;Hung M. La
中科院分区：
文献类型：
--
作者：
Tamanna Yasmin;C. Le;Hung M. La
混凝土结构的适当维护是避免民用基础设施出现任何危险情况的重要问题。剥落是桥梁和建筑物表面混凝土的重要病害。正确检测剥落的严重程度，可以及时发现和维护有害剥落，避免事故的发生。虽然以前的工作都是在表面缺陷，如裂缝和剥落，很少有解决剥落的严重程度检测。在本文中，我们提出了一种基于深度学习的方法，通过逐像素分类，根据严重程度检测剥落的确切位置。我们的网络将每个像素标记为无剥落、小剥落或大剥落。为了得到最优的深度架构，我们测试了几种编码器-解码器网络来比较和分析检测过程的性能。
Proper maintenance of concrete structures is a significant issue to avoid any hazardous situation in civil infrastructure. Spalling is a significant surface concrete distress in bridges and buildings. Correctly detecting the severity level of spalling can make it happen to detect and maintain the harmful spalling promptly to avoid any accidents . While previous works have been on surface defects, like cracks and spallings, few have addressed spalling severity detection. In this paper, we have proposed a deep learning-based approach to detect the exact location of spalling according to severity level by using pixel-by-pixel classification. Our network labels each pixel as no-spalling, small, or large spalling. To get the optimal proposed deep architecture, we tested several encoder-decoder networks to compare and analyze the performance of the detection processes.