Classification of Damaged Road Images Using the Convolutional Neural Network Method

Abstract

Objective: Automatic identification is carried out with the help of a tool that can take an image of road conditions and automatically distinguish the types of road damage, the location of road damage in the image and calculate the level of road damage according to the type of road damage.Design/method/approach: Identification of damaged roads usually uses manual RCI system which requires high cost. In this study, a comparison framework is proposed to determine the performance of the image pre-processing model on the image classification algorithm.Results: Based on 733 image data classified using the CNN method from 4 models of pre-processing stages, it can be concluded that training from grayscale images produces the best level of accuracy with a training accuracy value of 88% and validation accuracy reaching 99%.Authenticity/state of the art: Testing of 4 pre-processing models against the classification algorithm used as a comparison resulted in the best algorithm/method for managing road images.