一种基于频域变换的无监督车身漆膜缺陷检测算法  

An algorithm for unsupervised defect detection in car body coating based on frequency domain transform

作　　者：杜超 刘桂华[1] DU Chao;LIU Guihua(Southwest University of Science and Technology,Information Engineering Institute,Mianyang 621000,China)

机构地区：[1]西南科技大学信息工程学院,四川绵阳621000

出　　处：《电镀与涂饰》2020年第6期344-351,共8页Electroplating & Finishing

基　　金：国家自然科学基金(11602292)。

摘　　要：针对现有的2D汽车漆膜缺陷检测算法不能很好地检测出颜色暗、对比度低和较小的漆膜缺陷的问题,提出了一种基于频域变换的漆膜缺陷检测算法。首先通过对其主频率进行零掩码来消除图像背景中频谱显示较高梯度值分量的图像背景噪声。然后使用小波收缩法对傅里叶逆变换后的图像进行去噪。通过小波系数进行阈值修改,提取缺陷的高频信息,最后对重建后的图像使用阈值分割法分割出缺陷区域。对不同尺度和类型缺陷样本进行实验的结果表明,该算法检测准确率为95.01%,且具备良好的鲁棒性。Aiming at the problem that the existent 2D automobile coating defect detection algorithms can not detect the dark color,low contrast,and small defects well,an algorithm based on frequency domain transform was proposed.Firstly,the image background noise with high gradient component in the image background is eliminated by zero mask of its main frequency,and then the inverse Fourier transform image is denoised by the wavelet shrinkage method.Secondly,the high frequency information of defect is extracted by threshold modification of wavelet coefficients.Finally,the defect region is segmented from the reconstructed image by threshold segmentation method.The experiments on defect samples of different scales and types showed that this algorithm had a detection accuracy of 95.01%and good robustness.

关 键 词：漆膜 缺陷检测 对比度 离散傅里叶变换 小波收缩 阈值分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TQ639.8[自动化与计算机技术—计算机科学与技术]