Image Classification of Natural Disasters Using Different Deep Learning Models

Abstract

Natural disasters continue to affect the world through wildfires, cyclones, earthquakes, and floods. The advent of photography has provided us with valuable images of how disasters happen and their impact. Many deep-learning models have been developed to classify images. However, the classification of natural disasters still lags. Through transfer learning, eleven existing deep learning models and two optimizers were adapted, analyzed and tested on images based on natural disasters. We explore the impact of data augmentation on deep learning model performance. Based on experimental results, ResNet-50 coupled with SGDM optimizer achieved an accuracy of 98.6%. However, AlexNet converge faster in 4109 seconds, compared to all adopted deep learning models.