基于YOLO V5算法的建筑外立面渗漏红外图像识别方法  

Infrared image identification method of building facade leakage based on YOLO V5 algorithm

作　　者：吴玉龙[1] 王嘉浩 丁勇[2] WU Yulong;WANG Jiahao;DING Yong(Kunshan Construction Engineering Quality Inspection Center,Kunshan 215337,China;School of Science,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]昆山市建设工程质量检测中心,昆山215337 [2]南京理工大学理学院,南京210094

出　　处：《无损检测》2023年第10期54-58,共5页Nondestructive Testing

基　　金：中央级公益性科研院所基本科研业务费专项资金(Y322008);国家重点研发计划(2022YFC3005502);国家自然科学基金(51979174);国家自然科学基金联合基金项目(U2040221)。

摘　　要：采用深度学习技术中的YOLO V5目标识别算法对红外成像仪中采集到的渗漏区域红外图像进行识别。对于红外渗漏目标而言,不同背景条件、不同时间的红外渗漏目标样本量较少且难以采集,给深度学习模型的训练造成了很大的困难。深度学习需要较多的检测目标数据量进行训练,为了减少对真实渗漏红外图像数量的需求,结合仿真渗漏红外图像与真实渗漏红外图像来制作数据集,作为深度学习的样本进行训练。试验结果表明,所提出的数据集制作与识别方法,对建筑外立面红外图像中渗漏区域的识别准确率达87.6%。In this paper,the YOLO V5 target recognition algorithm of the depth learning technology was used to identify the leakage region of the infrared imager from collected infrared images.For infrared leaky targets,the sample size of infrared leaky targets with different background conditions and different time was small and difficult to collect,which made the training of deep learning model very difficult.In order to reduce the need of real infrared leakage image,this paper combined simulated infrared leakage image with real infrared leakage image to make data set,train these data set as a sample for deep learning.The experimental results showed that the accuracy of the proposed method was 87.6% for identifying the leakage area in the infrared image of the building facade.

关 键 词：建筑外立面 红外图像 深度学习 渗漏识别 

分 类 号：TU761.11[建筑科学—建筑技术科学] TP39.141[自动化与计算机技术—计算机应用技术] TG115.28[自动化与计算机技术—计算机科学与技术]