A Comparison of Machine Learning Approaches for Classifying Flood-Hit Areas in Aerial Images

Abstract

Floods caused due to climatic changes have become one among the most devastating natural hazards. Immediate relief operations play an important role in saving numerous lives during flood-hit time. Various technologies are used for quick response, one being the use of drones. As drones take the aerial images of the floodhit areas, we have proposed a method of classifying aerial images to identify flood-hit areas using various classifiers such as SVM, fine tree, KNN, and neural networks. Their performances are compared, and it is observed that SVM classifier outperforms the remaining classifiers with almost 93.1% due to its simplicity though neural networks minimize the amount of training to a larger extent. This classification of images is then used to identify the flood-affected areas to spot the extent of floods. Keywords SVM • Fine tree • KNN • Neural networks • Classification • Disaster response