A New Neuro-Fuzzy Approach for Post-earthquake Road Damage Assessment Using GA and SVM Classification from QuickBird Satellite Images

Tracking damaged roads and damage level assessment after earthquake is vital in finding optimal paths and conducting rescue missions. In this study, a new approach is proposed for the semi-automatic detection and assessment of damaged roads in urban areas using pre-event vector map and both pre and post-earthquake QuickBird images. In this research, damage is defined as debris of damaged buildings, presence of parked cars and collapsed limbs of trees on the road surface. Various texture and spectral features are considered and a genetic algorithm is used to find the optimal features. Subsequently, a support vector machine classification is applied to the optimal features to detect damages. The proposed method was tested on QuickBird pan-sharpened images from the Bam earthquake and the results indicate that an overall accuracy of 93% and a kappa coefficient of 0.91 were achieved for the damage detection step. Finally, an appropriate fuzzy inference system (FIS) and also an Adaptive Neuro-Fuzzy Inference System are proposed for the road damage level assessment. These results show that ANFIS has achieved overall accuracy of 94% in comparison with 88% of FIS. The obtained results indicate the efficiency and accuracy of the Neuro-Fuzzy systems for road damage assessment.