Binary Classification of Corrosion in Images through the LibAUC Library

Abstract

Corrosion is a critical problem that damages metal surfaces in different environments, with a significant economic impact and safety risks. It mainly affects industrial installations and people's physical integrity, which is why it is very important to detect it. In the present work, corrosion classification was carried out using image analysis with a library named LibAUC. In search of state-of-the-art, this library has been used in other areas, but not for corrosion. The methodology consisted of the following: collection of images, image preprocessing, modification of the library code for compatibility with updated libraries, adaptation of the deep model learning of melanoma classification for corrosion classification, execution of the model with training and validation images. The metric used for the performance of the model was the AUC (Area Under ROC), which achieved a value of 0.9973. It is concluded that the LibAUC library has a high performance for binary corrosion classification.