Deep learning-based road damage detection and severity assessment using explainable AI
Abstract
Road damages cost $100 billion annually, with the road-death tolls of several countries reaching up to 92% of the total deaths. The research is to evaluate and compare the MobileNetV2, ResNet50, EfficientNetB0 and InceptionV3 models to detect road damage with help of the RDD2022 dataset. On analysis, the modified InceptionV3 model shows better performance to classify road damage and predicting severity. Explainable Artificial Intelligence (XAI) like Grad-CAM and LIME are used for the model explainability. The study results show that the modified model shows an accuracy of 85% when compared to the existing models.
Fields of Study & Subjects
Infrastructure Maintenance and Monitoring
Advanced Neural Network Applications
Industrial Vision Systems and Defect Detection
Publication details
- Journal / source
- CRC Press eBooks
- Pages
- 550–556
- Publication date
- 30 January 2026
- Language
- EN
- Document type
- Book chapter