基于深度学习YOLOX算法的混凝土构件裂缝智能化检测方法
2024-05-21刘珂铖谢群李雁军
刘珂铖 谢群 李雁军



文章编号:1671-3559(2024)03-0341-09DOI:10.13349/j.cnki.jdxbn.20240024.003
摘要: 针对现有混凝土构件裂缝人工检测操作不仅费时、 费力,而且易出现错检、 误检、 漏检,以及部分位置难以开展检测的问题,提出一种基于深度学习YOLOX(You Only Look Once)算法的混凝土构件裂缝智能化检测方法;首先采集、 整理包含各类混凝土构件的典型裂缝图像,并通过图像数据增强建立Pascal VOC数据集,然后基于Facebook公司开发的深度学习框架Pytorch,利用数据集训练YOLOX算法,并进行裂缝识别和验证;将训练完成后YOLOX算法移植至搭载安卓系统的手机端,进行现场实时检测操作。结果表明:在迭代次数为700时,混凝土构件裂缝识别精度可达88.84%,能有效筛分混凝土构件表面裂缝,并排除其他干扰项,证明了所提出的方法对裂缝具有较高的识别精度和广泛的适用性;经试验测试,移植至手机端的YOLOX算法能在提升便携性的同时保证高效、准确的检测效果,具有良好的应用前景。
关键词: 深度学习; YOLOX(You Only Look Once)算法; 混凝土构件; 裂缝识别
中图分类号: TV544; TP18; TP391.41
文献标志码: A
开放科学识别码(OSID码):
Intelligent Detection Method for Cracks of Concrete Members Based on
Deep Learning YOLOX Algorithm
LIU Kecheng, XIE Qun, LI Yanjun
(School of Civil Engineering and Architecture, University of Jinan, Jinan 250022, Shandong, China)
Abstract: In view of the problem that the existing artificial detection technology for cracks of concrete members was not only time-consuming and laborious but also prone to misdiagnosis, false detection, miss detection, and difficulties in detecting some locations, an intelligent detection method for cracks of concrete elements based on deep learning YOLOX (You Only Look Once) algorithm was proposed. Typical crack images of various concrete members were firstly collected and sorted out, and a Pascal VOC dataset was established through image data enhancement. On the basis of deep learning framework Pytorch developed by Facebook, YOLOX algorithm was then trained and verified for crack identification by using the dataset. YOLOX algorithm after training was transplanted to the mobile phone terminal with Android system for on-site real-time detection. The results show that the crack identification accuracy concrete members can reach 88.84% when the iteration number is 700. Cracks on the surface of concrete members can be effectively screened out and other interference items can be excluded, which indicates that the proposed method has high identification accuracy and wide applicability for cracks. Through the test, YOLOX algorithm transplanted to mobile phone terminals can not only improve portability but also ensure the efficient and accurate detection effect, which has a good application prospect.
Keywords: deep learning; YOLOX (You Only Look Once) algorithm; concrete member; crack identification
收稿日期: 2022-12-21 網络首发时间:2024-01-24T18:14:34
基金项目: 国家自然科学基金项目(52108214); 山东省住房城乡建设科技计划项目(2020-K5-18); 建筑结构加固改造与地下空间
工程教育部重点实验室开放课题项目(MEKL202006)
第一作者简介: 刘珂铖(1999—),男,福建福州人。硕士研究生,研究方向为图像识别与智慧建造。E-mail:llwllkc0214@foxmail.com。
通信作者简介: 谢群(1979—),男,山东聊城人。教授,博士,博士生导师,研究方向为新材料结构、 新型装配式结构、 工程结构抗灾。
E-mail:cea_xieq@ujn.edu.cn。
网络首发地址: https://link.cnki.net/urlid/37.1378.N.20240124.1122.006
混凝土构件破坏开裂产生的裂缝是构件受力和变形最直观和主要的损伤形式[1]。……