<![CDATA[ Unmonitored road infrastructure conditions can lead to delayed maintenance actions and pose safety risks to road users. This study aims to develop an automated classification system for detecting road damage levels based on visual image data using deep learning methods. Three Convolutional Neural Network (CNN) architectures were evaluated in this research: VGG19, MobileNetV2, and EfficientNetB0. Each model was assessed based on training and validation accuracy, loss values, and confusion matrix performance. Experimental results indicate that the VGG19 and MobileNetV2 model achieved the best performance in classifying road images into four categories: good, moderate, minor damage, and severe damage, showing more stable accuracy and generalization compared to the other models. This model was then integrated into the GIS ASA mobile application, a real-time machine learning-based tool designed to detect road conditions. The classification results from the mobile app are subsequently visualized through the GIS ASA web platform, enabling spatial and interactive monitoring of road damage. This study demonstrates that the application of deep learning technologies offers an efficient solution for road condition mapping and monitoring. Future improvements may include dataset expansion, field validation, and additional GIS features to support more accurate decision-making in transportation infrastructure management.]]>
Copyrights © 2025
