Image Classification of Ground-Level Post-Disaster Data Using DenseNet201 for Disaster Damage Assessment in Quezon Province

The assessment of disaster-related structural damages has been important, and it is one of the most prevalent tasks undertaken to address the effects of disasters. However, the traditional and current method of damage assessment utilized by DRRM offices or councils is still manual. Therefore, the study sought to improve on this approach by developing an image classification model to classify the level of damage (partially or totally damaged) in a structure caused by a disaster. The study entails collecting ground-level post-disaster images as input, preprocessing the data before modeling, constructing, and training the model based on the supplied data, assessing model performance, and finally deploying it into a mobile application to demonstrate the feasibility of the study. The model developed was based on the DenseNet201 architecture and was tested using a different set of image datasets. The model improved significantly as training and validation accuracies increased while training and validation losses decreased, and it was able to provide an accuracy score of 90% when classifying images based on the given classes.