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

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.