Quantification and prediction of carbon steel surface corrosion based on machine vision

Corrosion detection plays a critical role in assessing material degradation and predicting performance changes to ensure equipment safety. Traditional manual inspection methods are highly subjective, while instrumental analysis tends to be destructive and costly. To address these limitations, this study proposes a machine vision-based method for detecting and predicting the composition of corrosion products on carbon steel subjected to corrosion after different durations. The corrosion products were analyzed using Raman spectroscopy. A generative adversarial network (GAN) was employed to expand the corrosion image dataset, and a convolutional neural network (CNN) was used for automatic corrosion classification. The K-means algorithm was then applied to segment different corrosion product regions. The total proportion of each corrosion product on the carbon steel surface was calculated by correlating the rusted area with the product distribution, and the corrosion evolution was analyzed using interpolation. Experimental results confirm that this method accurately quantifies corrosion product proportions, offering a novel strategy for intelligent corrosion detection.