Implementation of Convolutional Neural Network for Road Damage Detection and Classification in Surabaya City

Abstract

Road damage is a significant infrastructural problem that impacts the safety of road users and economic efficiency. The current road damage detection system, which relies on manual inspection, has limitations in speed and accuracy. Therefore, this study proposes the use of a conventional Convolutional Neural Network (CNN) to enhance accuracy and efficiency in the detection and classification of road damage in Surabaya City. The methods applied include data preprocessing and basic data augmentation techniques such as rotation and flipping. The dataset used comes from CV. Wastu Kencana Teknik, consisting of four road damage classes: potholes, surface delamination, cracks, and edge cracks. The implementation of the CNN model with standard configurations shows potential for application in an AI-based road infrastructure monitoring system. The model evaluation was performed using a confusion matrix and ROC-AUC, indicating that the model has stable and accurate classification performance. With these results, the model has the potential to enhance the effectiveness of detection and decision-making in road maintenance.