Recognition of corrosion image using machine learning

Igarashi, Takahiro  ; Otani, Kyohei   ; Aoyama, Takahito   

Evaluation of the integrity of large structures such as bridges and plants is one of the most important issues. In order to understand the corrosion mechanisms of steels using these large structures and to apply them to the integrity evaluation, many studies have been conducted, including exposure tests and repeated wet and dry tests in the laboratory. In addition, another research approach has been conducted to predict the corrosion state from surface images, which is considered to be one of the most useful methods to predict the amount of corrosion of actual materials that cannot be destructively inspected. In this study, we show that the FAST feature point extraction method can be used to predict the location of corrosion, although it is difficult to predict the absolute value of corrosion depth, by applying the FAST feature point extraction method to surface images of carbon steel specimens that have been subjected to repeated wet and dry tests before rust removal.