Corrosion grade classification: a machine learning approach

Corrosion tests allow to have information to indicate the state of materials in certain applications when environmental specifications are not met. It allows developing new material coatings to improve resistance to degradation. The area of materials inspection has relevance in construction, manufacture and medicine. In this work, we present an image corrosion classification method based on visual features and SVM (support vector machine). The feature extraction procedure includes SIFT (scale invariant transform features) and BOW (bag of words) approaches. The performance of classifiers is compared over the kernel function and the involved parameters. The experimental methodology known as the Tafel extrapolation method performed on each corrosion sample to find the corrosion rate and the corrosion current and voltage. A comparison between the Tafel test and the developed vision-based approach allows to see the high potential of the developed process to differentiate between pitting and general corrosion.