第一届挑战赛B5-道路缺陷自动识别

摘摘摘摘要要要要：：：：道路缺陷的类别包括路面结构性破损和功能性破损，而路面结构性破损是最常见的，可以细分为龟裂、块裂、网裂、纵裂和横裂。研究和设计一套道路缺陷检测技术，开发高效、快速、准确的道路缺陷自动识别系统具有重要的理论和现实意义。虽然基于数字图像处理的路面破损检测和识别技术在国外已经有所发展，但是由于路面图像的复杂特性，数字图像处理算法仍在研究。 本文针对附件所提供的道路缺陷图像，运用MATLAB图形图像处理工具箱进行图形处理，包括图像的预处理，图像分割，图像特征提取，图像边缘检测等，并得到了预期效果。在分析比较传统方法对道路缺陷图像增强效果及不足之处的基础上，通过对实验分析，选用了梯度倒数加权平均滤波算法，它能有效地平滑图像背景、消除大部分噪声。在图像分割中，根据不同缺陷图像，分别进行了边缘检测和灰度阈值分割，提出了随机游动分割方法进行坑槽图像的分割，得到了较为满意的结果。 对缺陷图像进行特征提取，并分析计算了评判道路缺陷的特征参数，包括平均缝宽、裂缝宽度、块度、面积。采用BP神经网络的方法对缺陷图像进行模式识别，对识别的结果及学习率进行了深入讨论。 关键词关键词关键词关键词：：：：道路缺陷；破损检测；梯度倒数加权平均算法；随机游动分割算法；BP神经网络 The Automatic Pavement Identification Based OnData-Ming Technology Abstract:Road defect categories include pavement structural damage and functional damage,however,the former is the most common, and can be subdivided into the cracks, the block splitting networkcracking, longitudinal and transverse cracking. Research and design a road defect detection technology, it isof importance to develop the detect of efficient and accurate of road defects automatic identification system.Although based on digital image processing, pavement distress detecting and recognition technology hadbeen developed, but due to the complex nature of the road surface image, digital image processingalgorithms are still studying abroad. Rely on the defect image given by the Appendix, with MATLAB graphics image processingtoolbox,handling the defect image,including preprocessing,segmentation,feature extraction,edgedetection, and get the desired effect. Selected experimental analysis on the basis of the analysis of moretraditional methods of road defects image enhancement and deficiencies at the the gradient the Countdownweighted average filtering algorithm, can effectively smooth image background, eliminating most of thenoise. Were image segmentation based on different defect image edge detection and gray thresholdsegmentation, random walks segmentation pits image segmentation, and we get satisfactory results finally. Defect image feature extraction, analysis and calculation of the characteristic parameters of theevaluation of road defects, including average slit width, crack width, block, area. Using BP neural networkpattern recognition, image defect recognition results and learning rate were discussed in depth. Key words:road defects;distress detecting;GIWA algorithm;RW segmentation algorithm;BP NeuralNetworks 目录 1.挖掘目标挖掘目标挖掘目标挖掘目标······························································································· 4 2.分析方法与过程分析方法与过程分析方法与过程分析方法与过程···················································································· 4 2.1.总体流程····················································································································· 42.2.具体步骤····················································································································· 52.2.1缺陷图像预处理·································································································· 52.2.2缺陷图像分割······································································································ 82.2.3缺陷图像特征提取···························································································· 122.2.4BP神经网络模式识别······················································································ 152.3.结果分析··················································································································· 19 3.结论结论结论结论·····································································································19 4.参考文献参考文献参考文献参考文献······························································································20 1.挖掘目标挖掘目标挖掘目标挖掘目标 本次建模目标是利用附件所提供的道路缺陷图像，主要类型有：裂缝，龟裂，网裂，坑槽等缺陷。首先结合图形图像处理技术，对各类型图像进行特征提取和参数计算，并采用数据挖掘技术进行模式识别，以实现不同道路缺陷类别的自动识别。随着数字图像处理和模式识别技术的快速发展，不少研究人员开始试图利用图像识别技术来实现道路缺陷数据的调查研究。基于数字图像处理的道路路面病害自动识别技术可以及时掌握路面的破损情况，为路面的及时养护赢得时间的同时，也为养护资料的积累提供了方便，也充实了自动识别系统的数据库数据，从而为进一步制定道路养护规划决策提供科学依据。 2.分析方法与过程分析方法与过程分析方法与过程分析方法与过程 2.1.总体流程总体流程总体流程总体流程 本次建模主要包括3个模块，分别是数据库查询模块、图像处理模块、参数计算模块。其中数据库查询模块主要实现从前端数据库读取缺陷图像，以及图像处理完成并且参数提取以后的参数统计等功能；图像处理模块主要实现图像灰度变换、图像去噪、图像分割等功能；参数计算模块主要实现对缺陷图像类型的判断，以及对线性裂缝长度的计算和网状裂缝的面积计算等功能，具体实现过程如图1所示。 步骤一：利用MATLAB图形图像处理工具箱进行图像读取； 步骤二：编程进行图像处理模块的实现； 步骤三：对缺陷图像判断标准的参数进行计算并输出。 挖掘模型的具体流程如图2所示。 2.2.具体步骤具体步骤具体步骤具体步骤 2.2.1缺陷图像预处理缺陷图像预处理缺陷图像预处理缺陷图像预处理() 1灰度变换灰度变换灰度变换灰度变换 图像在形成、传输和记录的过程中，会由于成像系统、传输媒介和记录设备的不完善，使图像质量下降，形成退化图像，造成比较差的视觉效果和计算机处理上的困难。影响图像质量下降的因素包括成像系统和环境以及成像特点等，因此很难用一个显式的数学表达式来表征。()()() 我们假设道路图像,I x y由非均匀灰度(光照)的背景,bIx y，裂缝病害,nIx y以及石子、沥青等构成的噪声(),cIx y三部分相加组成，即 因此，如果可以找出背景信号，用原始图像减去背景就可以纠正灰度不均这一问题。从以上分析可以看出，关键在于背景图像的提取。在实际应用中，我们很难直接得到没有裂缝的背景，有时即使能得到，在时间轴上面的位置相距比较大，光照分布也往往不均匀。 经分析，我们考虑对背景子集双线性插值的方法来抽取拟合背景，算法步骤为：1Step 由原图像求取背景子集。对原图像分块，每块取一个背景点。光照不均在整幅图像表现明显，但在局部可以认为是近似均匀的。 2Step由子集做插值得到背景图像。由于图像的灰度变化是一个渐变的过程，考虑利用双线性插值来对四个相邻的像素进行插值，因为这样产生的表面是连续的。 ()2图像去噪图像去噪图像去噪图像去噪 中值滤波是一种非线性处理技术，能抑制图像中的噪声，由于它在实际运算过程中并不需要图像的统计特性，所以使用比较方便。它是基于图像的这样一种特性：噪声往往以孤立的点的形式出现，这些点对应的像素数很少，而图像则是由像素数较多、面积较大的小块构成。 中值滤波法虽然可以很好的