CNN-Based Building Type Classification into EF Scale Categories to Support Tornado Damage Assessments

Accurate assessment of tornado impacts requires detailed information on building characteristics that influence vulnerability. This study investigates a convolutional neural network (CNN)-based approach to classify nonresidential buildings into enhanced Fujita (EF) damage categories using image data. Three deep learning architectures—CNN, ConvNeXt, and ResNet50—were evaluated on a data set representing 11 EF building categories. ResNet50 achieved the highest overall accuracy of 83%, with F1 scores above 0.75 in 8 of 11 categories. The results demonstrate the feasibility of image-based classification as an initial step toward enhancing structural data sets used in hazard modeling and disaster planning (e.g., FEMA Hazus-MH, engineering fragility curve, etc). This work highlights the potential of deep learning methods to supplement structural data. Future research will expand model scalability and incorporate spatial metadata to improve applicability for operational risk assessment.