Enhancing natural disaster image classification: an ensemble learning approach with inception and CNN models

Abstract

The core problem of this research is the rapid and accurate classification of natural disasters, which is essential for effective response and mitigation strategies. Existing detection methods are often time-consuming and costly. The purpose of this research is to introduce an innovative approach to the multi-class classification of natural disasters using image data from a Kaggle dataset encompassing Cyclone, Wildfire, Flood, and Earthquake incidents. The method used is an ensemble learning model that combines the strengths of the InceptionV3 model and a custom Convolutional Neural Network (CNN). The result of this study is an ensemble model that achieves a commendable accuracy of 92.79%, surpassing individual models and demonstrating the efficacy of combining diverse features extracted by InceptionV3 and CNN architectures. Additionally, a standalone CNN model is implemented, achieving an accuracy of 88.76%. The research concludes that the ensemble model’s superior performance makes it a valuable tool for the multi-class classification of natural disaster images.

Keywords

• Ensemble learning model
• InceptionV3
• convolutional neural network (CNN)
• cyclone
• wildfire
• earthquake