基于图像识别的水下结构表观缺陷定量评估  

Quantitative evaluation of apparent defects on underwaterstructures based on image recognition

作　　者：董营营 胡俊华 杨森 邹辰浩 张朝 DONG Yingying;HU Junhua;YANG Sen;ZOU Chenhao;ZHANG Chao(Shandong Expressway Group,Sichuan Leyi Highway Co.,Ltd.,Chengdu 610095,China;Shandong Expressway Engineering Testing Co.,Ltd.,Jinan 250013,China;China Institute of Water Resources and Hydropower Research,Beijing 100038,China)

机构地区：[1]山东高速集团四川乐宜公路有限公司,四川成都610095 [2]山东高速工程检测有限公司,山东济南250013 [3]中国水利水电科学研究院,北京100038

出　　处：《山东建筑大学学报》2025年第2期110-117,共8页Journal of Shandong Jianzhu University

基　　金：北京市自然科学基金青年项目(8244059);国家重点研发计划(2023YFF0906100);流域水循环模拟与调控国家重点实验室开放研究基金(IWHR-SKL-KF202410);流域水循环模拟与调控国家重点实验室开放研究基金(IWHR-SKL-KF202218);中国博士后科学基金(2022M713473)。

摘　　要：常规检测方法不适用水下混凝土结构表观缺陷的定量评估。文章基于丰满老坝试验数据,围绕水下图像增强和缺陷识别问题,研究融合暗通道(Dark Channel)算法和视网膜皮层(Retinex)理论的(Dark-Retinex, DR)图像增强算法,探索边缘检测方法。结果表明:DR算法可有效抑制水下图像的背景噪声,解决图像偏色的问题,增强图像的对比度、色彩亮度和照明度,从而突显裂缝、剥落、麻面、蜂窝等水下混凝土结构的表观缺陷;边缘检测方法实现了表观缺陷识别及其特征参数提取,可定量地评估裂缝的长度和宽度,以及剥落、麻面、蜂窝等缺陷的面积,精度为毫米级;DR算法结合边缘检测方法,解决了水下混凝土结构表观缺陷的自动检测和定量评估的难题。Conventional detection methods are not suitable for quantitative evaluation of apparent defects in underwater concrete structures.Based on the experimental data of Fengman old dam,this study focuses on underwater image enhancement and defect recognition.The Dark-Retinex(DR)algorithm which integrates the Dark Channel algorithm and the Retinex algorithm was studied,and the edge detection method is explored.The results show that the DR algorithm can effectively suppress the background noises and color deviation and enhance the contrast,color brightness and illumination of the underwater images.Therefore,the apparent defects of underwater concrete structures,including cracks,peelings,pits and honeycombs,are highlighted.The edge detection method realizes the recognition of apparent defects and the extraction of feature parameters,which can quantitatively evaluate the length and width of cracks,as well as the area of defects such as peeling,pitting,and honeycomb,with an accuracy of millimeter level.The Combination of the DR algorithm and the edge detection method overcome the difficulty in automatic detection and quantitative evaluation on the apparent defects of underwater concrete structures.

关 键 词：表观缺陷评估 水下图像处理 Dark-Retinex算法 图像边缘检测 

分 类 号：TV331[水利工程—水工结构工程]