Human-visual-perception-like intensity recognition for color rust images based on artificial neural network

Abstract

Abstract Digital image processing has been applied to the assessment of steel bridge coating quality since the late 1990s. Most previously developed methods cluster rust images into two or three groups before calculating the rust percentage. Since two- or three-group clustering might not properly reflect the rust intensity or rusting severity on a rust image, the artificial-neural-network-based rust intensity recognition approach (ANNRI) is proposed in this paper. ANNRI integrates the root-mean-square standard deviation (RMSSTD) and artificial neural network (ANN) to cluster a rust image based on its rust intensity or rusting severity. RMSSTD measures the similarity of rust colors on a rust image, and an ANN trained with the results of a human visual rust inspection experiment would generate the optimal number of clusters for rust intensity recognition. Together with a pre-defined rust color spectrum, ANNRI is able to perform human-visual-perception-like rust intensity recognition and screen out background noises. According to the experiments conducted in this study, the proposed ANNRI can discriminate rust intensity much better than the existing methods with a fixed number of clusters.