CLASSIFICATION OF CORROSION DETERIORATION ON WEATHERING STEEL BASED ON CNN

Abstract

This paper aims to develop an image processing for diagnosis of deterioration level of corroded weathering steel surface based on Convolutional Neural Network (CNN) analysis. The weathering steel corrosion test results, which had been obtained in differenent chloride environment, were used for this image processing. At first, the relationship between surface and gained weight or rust thickness were clarified to determine the deterioration level. Then, the CNN analysis was conducted for the dataset from each chloride environment and all chloride environments. As the result, there was no strong relationship between surface and gained weight or rust thickness. The accuracy of the CNN classifier created for each corrosion surface image obtained in each chloride environment is depending on the environment. Also, the accuracy of the CNN classifier using images obtained in all environments was low. But some deterioration level could be classified with high accuracy in the classification results for each deterioration level, therefore improvement may be expected.