基于K-means++聚类算法和SSIM指标的金属板材腐蚀区域识别  

Corrosion Area Identification of Sheet Metal Based on K-means++Clustering Algorithm and SSIM Index

作　　者：龙梦翔 付桂翠[1] 万博 张钟庆 LONG Mengxiang;FU Guicui;WAN Bo;ZHANG Zhongqing(School of Reliability and Systems Engineering,Beihang University,Beijing 100191,China)

机构地区：[1]北京航空航天大学可靠性与系统工程学院,北京100191

出　　处：《中国腐蚀与防护学报》2024年第2期437-444,共8页Journal of Chinese Society For Corrosion and Protection

摘　　要：材料腐蚀后会在表面产生锈斑、裂纹和鼓泡等多种腐蚀特征现象,通过观察腐蚀特征现象可判断材料腐蚀程度。目前主要通过人工目测的方式对材料的腐蚀情况进行判断,但其存在结果无法量化、效率低下等不足。本文采用K-means++聚类算法对金属板材图像像素RGB值进行聚类,分离腐蚀区域和未腐蚀区域;采用图像结构相似性指标(SSIM)判断聚类各区域是否发生腐蚀。结果表明:将K-means++聚类中心数量k设定为5,可有效根据图像颜色分布划分出各聚类区域;相比峰值信噪比PSNR和均方误差MSE,结构相似性指标SSIM与图像是否发生腐蚀具有较强相关性,将SSIM指标阈值设定为0.95,可根据SSIM指标有效判断各聚类区域是否发生腐蚀;本文所用方法相比人为根据像素颜色划分腐蚀区域,具有更高的识别效率,且准确率不低于90%。本研究可用于金属板材环境试验后防腐性能自动化评价。Various corrosion characteristics such as rust spots and cracks will appear on the surface of metal plates after corrosion.Their corrosion degree can be determined by corrosion characteristics.At present,the corrosion degree of metal plates is mainly judged by manual visual inspection.But it has many non-ignorable shortcomings such as low consistency and low efficiency etc.In this paper,the RGB values of image pixels of corroded metal sheets were collected and then clustered by means of K-means++clustering algorithm,afterwards the relevant corroded-and uncorroded-regions were separated.Whether corrosion occurred or not was judged in each cluster area by means of image structural similarity index SSIM.The results show that setting the number of clustering centers‘k’to 5 can effectively delineate each clustering area based on the image color distribution.Compared to peak signal-to-noise ratio and mean square error,the structural similarity index SSIM is strongly correlated with the occurrence of erosion.Setting the SSIM index threshold at 0.95 can effectively judge whether erosion occurred in each cluster area.Compared to manually dividing corrosion areas based on pixel color,our method had a higher identification efficiency and an accuracy of not less than 90%.This research can be applied to automate the evaluation of the corrosion degree of metal sheets after environmental testing.

关 键 词：腐蚀评估 图像处理 K-means++算法 聚类 图像相似度 智能诊断 腐蚀面积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]