EVALUATION OF RUST APPEARANCE ON WEATHERING STEEL USING CONVOLUTIONAL NEURAL NETWORK

Abstract

The evaluation of the corrosion state of weathering steel is mainly based on its appearance and the criteria are not very quantitative, often leading to a difficult/incorrect judgement. In the present study, we have tried to construct a simple, practical method for evaluating the corrosion state of weathering steel by employing the image classification AI. This class of AI, which utilizes a convolutional neural network(CNN), is known as a powerful tool for image classification problems. To be specific for the construction of the present AI, the structure of the CNN model was carefully studied from a viewpoint of classification tendency and the majority voting ensemble was conducted. It was revealed then that the pooling method is more influential on the classification of the corrosion state than the convolutional-layer structure. Moreover, the majority voting ensemble was found useful in reducing misclassification.