基于深度学习的建筑表面裂缝缺陷识别  

Recognition of cracks and defects on building surface based on deep learning

作　　者：陈墨[1] 杨沛[1] 陈丽君 CHEN Mo;YANG Pei;CHEN Li-jun(Anhui Technical College of Industry and Economy,Hefei 230051,China)

机构地区：[1]安徽工业经济职业技术学院,合肥230051

出　　处：《齐齐哈尔大学学报（自然科学版）》2021年第5期57-61,66,共6页Journal of Qiqihar University(Natural Science Edition)

摘　　要：针对传统建筑表面裂缝缺陷中存在识别效率低下、且识别精度较低的问题,为此,提出了基于深度学习的建筑表面裂缝缺陷识别方法。首先采集建筑表面裂缝缺陷的图像数据,然后将采集图像采样与量化,以实现缺陷图像的数字化处理。将上述数字化后的图像数据进行阈值分割、滤波以及以及增强等,完成建筑表面图像的预处理;构建R-CNN深度学习模型,模型结构分为四部分,包括输入图像模块、生成模块、提取卷积特征模块以及分类和边框回归模块。将建筑表面裂缝图像输入构建的深度学习模型中,完成建筑表面裂缝缺陷的识别。实验结果表明,采用所提方法识别建筑表面裂缝缺陷的效率较高,且识别的精度较好。Aiming at the problems of low identification efficiency and low identification accuracy in the traditional building surface crack defects.Therefore,a deep learning based crack defect identification method for building surface is proposed.Firstly,image data of building surface cracks and defects are collected,and then the collected images are sampled and quantified to realize digital processing of defect images.Threshold segmentation,filtering and enhancement of the digitized image data are carried out to complete the preprocessing of the building surface image.R-CNN deep learning model is built.The model structure is divided into four parts,including image input module,generation module,convolutional feature extraction module,and classification and border regression module.The image of building surface cracks was input into the deep learning model to complete the identification of building surface cracks and defects.Experimental results show that the proposed method is efficient and accurate in identifying cracks on building surface.

关 键 词：深度学习 建筑表面裂缝 缺陷识别 阈值分割 卷积特征模块 

分 类 号：TU755.7[建筑科学—建筑技术科学] TP399[自动化与计算机技术—计算机应用技术]