【作者】 曾燕华；

【导师】 李扬；

【作者基本信息】 广东工业大学 ， 测试计量技术及仪器， 2005， 硕士

【摘要】 交通运输是一个国家的经济命脉,道路和桥梁则是使交通运输能够畅通无阻的载体,尤其是桥梁,作为交通枢纽更是起着至关重要的作用。随着其服役时间的增加,桥梁结构难免会出现各种各样的疲劳和损伤从而引发坍塌等恶性事故,造成了巨大的经济损失和人身伤害。而大多数桥梁的结构缺陷主要产生在桥梁底部,人工检测和自动检测特别困难,因此如何对桥梁结构进行经常性的检查和健康诊断具有重要的理论意义和实用价值。 对于大型的混凝土桥梁,桥梁的裂缝是桥梁检测的重要项目之一。而采用单一的人工目测的检测方法是效率低下的。因此将计算机视觉应用于桥梁表面缺陷检测中,以提高工作效率,并对桥梁表面缺陷的客观评估具有重要意义。 本文主要研究工作是根据现代桥梁健康评估和检测方法,结合公路混凝土桥梁的现状与现行《规范》,应用数字图像处理和模式识别技术对桥梁表面缺陷图像进行检测和评估,并且对以下几个关键问题进行了研究:(1) 在采用高速CCD和图像采集卡进行缺陷图像采集过程中,分析缺陷图像采集过程中光照不均的问题,选用了合适的光源与光照方法。(2) 实现对缺陷图像的进行量化分割,即采用中值滤波与加权平均算法对量化后的缺陷图像进行噪声去除,同时选用结合迭代剪枝算法和形念学方法,以保存良好的缺陷特征信息。(3) 运用基于“块”图像的神经网络对缺陷图像进行识别分类:即将预处理完的图像进行“块”处理化,采用两种基于“块”图像的BP神经网络方法:基于图像的神经网络方法(INN)和基于直方图的神经网络方法(HNN)对缺陷图像进行分类识别。这两种方法都是在基于像素的神经网络方法(PixNN)的基础上发展起来的。通过对两种神经网络在不同隐含节点、学习效率和循环次数下分别进行性能测试。实验结果表明,采用基于直方图的神经网络方法能够较好的满足桥梁表面缺陷的识别分类,并确定了最优神经网络结构。(4) 对缺陷进行量化估算,并建立缺陷图像数据库,对缺陷图像进行分类、存储和管理。更多还原

【Abstract】 The transportation is a national economic lifeline, and paths and bridges enable the transportation to be the unimpeded carrier. Especially, bridges are playing a very important role. Along with their service time increase, the bridge structure can appear unavoidably various weary and damage, which lead to malignant accident, huge economic loss and person injures. But the majority bridge fault of construction mainly produces in the bridge base, where both manual examination and automatic detection are especially difficult. Therefore, how to carry the regular inspection and the health check-up on the bridge structure has the important theory significance and the practical value.Regarding the large-scale concrete bridge, the bridge crack is one of the main contents in bridge detection and measurement. But the simplex manual range estimate examination method is low efficiency. Therefore applying the digital image technology and the pattern recognition methods in the bridge surface defect examination, in order to enhance the working efficiency, and it is extremely important for the external evaluation of bridge surface flaw.In the thesis, according to the health evaluation and inspecting method of modern bridges, the digital image technology and the pattern recognition methods combined with existing criterions are provided to detect the surface-flaw in bridges. And discuss some keys thereinafter: (1) High speed CCD and the image-capturing card are utilized to gather the flaw images: By analyzing the uneven illuminations in image capturing, suitable lighting mode is chosen. (2) The median filter and the weighted average algorithm are used to getting rid of noises. And the iteration clipper algorithm and morphology are carried to preserve more available characters of image. (3) After the "block" processing carried on the image, the thesis proposes two block-based BP neural network methods: image-based nerve network method (INN) and histogram-based neural network method (HNN). These two methods are all developed based on the更多还原