MATHEMATICAL MODEL FOR THE PROBLEM OF CLASSIFICATION OF PAVEMENT DEFECTS

Abstract

The road surface defect detection and classification system based on machine learning algorithms is already very advanced and is increasingly proving its outstanding advantages. In this paper, some image segmentation algorithms used in practice are presented, compared and evaluated. In this study, we present the convolution neural network—VGG16 structure to classify pavement defects, with a graph-based method to optimize the image segmentation on the pavement defect image. This proposed method is intended to overcome limitations caused by objective factors, such as high sensitivity to data of certain types of light and noise dependence, such as defect data of the road surface. Three different datasets were collected from the Center for Telecommunication and Multimedia, INESC TEC - Portugal (1200 images), Irkutsk city - Russian Federation (800 images) and Thai Nguyen city - Vietnam (550 images). The classification results based on the VGG-16 methods of the datasets in turn are good because the curves have a state closer to 1 than 0.5.