Top-Hat变换序列分析及在制件缺陷自动检测中的应用  

Top-Hat Transform Sequence Analysis and Application in Automatic Detection of Products Defects

作　　者：毛锋[1] 张树有[1] 

机构地区：[1]浙江大学CAD&CG国家重点实验室,杭州310027

出　　处：《机械工程学报》2010年第13期155-164,共10页Journal of Mechanical Engineering

基　　金：教育部博士点基金(200803350031);浙江省自然科学基金重点(Z1080339)资助项目

摘　　要：制件中存在缺陷会带来很大的危害,通过对制件缺陷图像信号进行分析处理可以实现制件质量的自动监控。提出Top-Hat变换序列分析方法。在结构元素尺寸驱动下对信号进行迭代开运算和Top-Hat变换,构建信号的TH(Top-Hat变换)序列,同时获得信号的O-TH(Open-TopHat)分解树,并以树的叶子结点能量为幅值建立信号的对象分析谱。通过分析,得出信号中各种对象在谱中的行为,从而获得最佳结构元素尺寸和最佳分割阈值。然后用该最佳尺寸的结构元素对原信号进行Top-Hat变换,并以最佳阈值进行分割,准确地定位信号中目标的位置。在此基础上,通过比较TH序列中的元素与最佳变换结果,从而获得修补边界,并以此对初步分割结果进行目标尺寸扩展,得到完整的目标。最后,综合以上过程,给出基于Top-Hat变换序列分析的图像目标提取算法。方法在多种制件图像、各种缺陷类型中进行应用验证。结果表明,该方法通用性强,有很强的抗噪性和抗背景干扰能力,缺陷提取准确度高、鲁棒性好。Defects in parts bring great harms.To implement automatic monitoring of parts quality,images of parts are processed and analyzed.A method of Top-Hat transform sequence analysis is proposed.Firstly,iterative open operation and Top-Hat transformation of the signal are computed by using size driven structure elements and the O-TH（Open-TopHat） decomposition tree is obtained.Objects analysis spectrum is established whose amplitude is the energy of the O-TH tree leaf nodes.Then,the optimum size of the structure element and the optimum segmentation threshold are obtained by analyzing the behavior of each object of the signal in the spectrum.Thus,all the targets in the signal are located accurately.On this basis,the preliminary segment result is patched by comparing the elements in the TH sequence with the Top-Hat transform result which uses the optimum structure element,and the targets in the signal are detected completely.Finally,object extraction algorithm based on Top-Hat transform sequence analysis is designed.The method is verified by experiments including several kinds of parts and different defects.The results prove that the method has good generality and strong immunity and anti-background interference abililty,high accuracy in defect extraction and good robustness.

关 键 词：TOP-HAT变换 结构元素尺寸 分割阈值 缺陷检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]