Characterization Method for Feature Parameters of Corrosion Images Based on Machine Learning

Corrosion images of X80 steel under combined AC interference and cathodic protection were obtained through immersion tests. A machine learning-based method incorporating the graycomatrix () computational function was proposed for corrosion image feature extraction and analysis. The accuracy of this method was further validated by electrochemical measurements. The results show that under the combined action of AC interference (50 A/m², 60 Hz) and cathodic protection (-1.0 V, vs. SCE), the corrosion images of X80 steel could be divided into three stages: the first stage (0-20 h) was the stage of crater initiation and incubation, in which the number of pits increased rapidly; the second stage (20-40 h) was the accelerated corrosion stage, in which the number of pits decreased rapidly; and the third stage (40-60 h) was a stable corrosion stage, in which the number of pits slowly decreased and corrosion reached a stable state. As corrosion duration increased, the second moment of image exhibited similar variation trends to pit density, while the angular second moment of image showed an opposite trend to corrosion area ratio. Electrochmical testing results confirmed that both second moment of and angular second moment of image effectively characterize corrosion evolution, demonstrating the validity of the proposed methodology.