Road surface crack detection using deep learning in smart cities

Abstract

This study aims to develop an automatic and high-accuracy system for detecting cracks on asphalt road surfaces as part of smart city applications. The proposed method employs deep learning techniques, specifically a MobileNetV2-based convolutional neural network, trained on a large, balanced dataset of 40,000 images. Preprocessing steps such as resizing, normalization, and data augmentation were applied to improve model generalization. The system achieved 99.78% accuracy on the training set and 99.73% on the test set, demonstrating its capability for real-time operation on low-cost devices. The results indicate the proposed approach is a feasible and practical solution for urban infrastructure maintenance planning, with potential integration into mobile and embedded systems for real-world deployment.