基于颜色特征的半监督聚类算法在铜片腐蚀等级识别中的应用  

Application of Semi-Supervised Clustering Algorithm Based on Color Feature to Corrosion Level Identification for Copper

作　　者：钱企豪 郑战光[1] 梁钊 伍鹏革 杜彭玉 QIAN Qihao;ZHENG Zhanguang;LIANG Zhao;WU Pengge;DU Pengyu(School of Mechanical Engineering,Guangxi University,Nanning 530004,China;State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body,College of Mechanical and Vehicle Engineering,Hunan University,Changsha 410082,China)

机构地区：[1]广西大学机械工程学院,南宁530004 [2]湖南大学机械与运载工程学院汽车车身先进设计制造国家重点实验室,长沙410082

出　　处：《腐蚀与防护》2023年第5期34-40,共7页Corrosion & Protection

基　　金：国家自然科学基金项目(51675110,51465002)。

摘　　要：提出一种基于半监督聚类算法的铜片腐蚀等级快速识别方法。该方法首先对于大量铜片腐蚀图像进行图像分割,使其尺寸归一化;然后通过滤波处理减弱异常值影响,利用颜色量化方法获取图像的颜色特征向量,并通过核主成分分析(KPCA)对颜色直方图信息进行降维处理;最后,将标准比色卡提取的颜色特征向量作为半监督k-means的初始聚类中心,结合预处理后腐蚀图像的颜色特征向量训练模型,得到每张图片对应的腐蚀等级。结果表明,通过该算法得到的铜片腐蚀等级分类结果与目测结果一致,说明该方法具有较高的准确性。A rapid identification method of copper corrosion level based on semi-supervised clustering algorithm was proposed.In this method,a larger number of images of corroded copper were segmented firstly in order to normalize their sizes.Then the influence of outliers was weakened by filter processing,and the color feature vector of the images was obtained by color quantization.The dimension of color histogram was reduced by the kernel principal component analysis(KPCA)method.Finally,the corresponding corrosion level of each image was obtained by taking the color feature vector extracted from the standard colorimetric card as the initial clustering center of semi-supervised k-means in combination with the color feature vector training model of pre-processed corrosion images.The results showed that classification results of corrosion level of copper by calculation through the algorithm corresponded well to the visual inspection results,indicating high accuracy of the method.

关 键 词：铜片腐蚀 颜色特征 图像预处理 半监督聚类 核主成分分析 

分 类 号：TG174[金属学及工艺—金属表面处理]