Corrosion Analysis Through an Adaptive Preprocessing Strategy Using The K-Means Algorithm

Corrosion is the process of the deterioration of metals through chemical and electrochemical reactions. The early identification of high-risk areas affected by corrosion allows time for making decisions to prevent disasters that could result from the phenomenon of corrosion. In this paper, the K-means clustering algorithm is used with various color-manipulation techniques to process and segment images of corroded metal surfaces. The goal of our research is to identify corrosion patterns in images. As an experiment, an image set of corroded surfaces demonstrating various levels of corrosion was processed. In the first phase, the images were processed with the K-means algorithm and then processed with color-manipulation techniques. The results show that this process provides relevant information regarding the surface being analyzed, successfully isolating in every test case the zone or area most affected by corrosion and highlighting the dimension of the problem. Based on our results, we recommend using the HSV technique for preprocessing images if the corroded area comprises less than 50% of the image, and an average value of k = 4 and k = 8 is recommended for obtaining significant patterns. This is important because identifying these types of patterns makes it possible for experts to make decisions without submitting these corroded metal surfaces to a second or third process or to manual analysis, and thus analytical complexity is reduced regarding time, costs, and resources.