Classification of Disasters Using Deep Learning Techniques

Authors: S. Sreekanth, G. Teja Harshitha, K. Vagvevi , M. Vaishnavi

DOI Link: https://doi.org/10.22214/ijraset.2026.78995

Abstract

Theprojectiscenteredonclassificationandanalysis of zones affected by natural disasters using deep learning concepts. The project dataset was created by incorporating five different classes of natural disasters: cyclone, flooded, landslide, wildfire, and volcano. The dataset was preprocessed to ensure that only quality images of correct orientation are used for training. To ensure that data augmentation occurs for these classes as well, data augmentation was carried out for those classes by rotating the images, applying color jittering, and perspectives. The dataset was then trained using various deep learning models like DenseNet, EfficientNet, ResNet, along with accuracy, F1 score, confusion matrix, and ROC curve. The final model was created by ensembling different models for better classificationresults. Theapproachhashelpedinbetterdetection ofmisclassifieddisasterswhilealsocreatingarobustsystem for understanding images related to disasters using Grad-CAM visualization techniques, which can be used to quickly identify disasters.

Introduction

This paper presents a deep learning-based disaster zone classification and analysis system that automatically identifies natural disasters such as cyclones, floods, landslides, wildfires, and volcanic eruptions from satellite and aerial images. Traditional manual image analysis is slow and error-prone, making it unsuitable for emergency response. To address this, the proposed system employs advanced Convolutional Neural Network (CNN) architectures, including DenseNet, EfficientNet, and ResNet, to accurately classify disaster images. Targeted data augmentation techniques, such as rotation, color variation, and perspective transformation, improve the recognition of visually similar disaster types and enhance model generalization. Grad-CAM visualization is integrated to highlight affected regions, increasing model transparency and supporting disaster assessment. The workflow includes dataset collection, preprocessing, exploratory data analysis, model training, evaluation, and visualization. The system is designed for cloud deployment using frameworks like Flask or FastAPI, enabling real-time disaster monitoring through web or mobile applications. Experimental results show that EfficientNet-B0 achieved the highest accuracy (82.15%), outperforming DenseNet-121 (78.64%) and ResNet-34 (80.72%), while maintaining efficient processing times of 1.5–3 seconds per image. Overall, the proposed framework provides an accurate, scalable, and interpretable solution for automated disaster detection, supporting faster emergency response and disaster management.

Conclusion

The project successfully demonstrates the application of deep learning techniques in the classification of disaster zones using image data. It efficiently identifies various types of naturaldisasterssuchasfloods,cyclones,wildfires,landslides, and volcanoes using an EfficientNet-based CNN model. It demonstrates the potential of computer vision in supporting variousemergencyresponsesystemsanddisastermanagement operations. Comprehensive data preprocessing and augmentation were performed, focusing on enriching the diversity of the data set and thereby enhancing the robustness of the model, especially forclassesthatarelessrepresented.Anotherstepinthe methodology was exploratory data analysis, which aided in comprehending the data set and thereby forming a better strategy for data augmentation and balancing. Despite the challengesofdataimbalanceandsimilarityinimagesforsome disaster types, the model was able to train stably and attain a reasonableaccuracywithinaconstraineddatasetenvironment.

References

[1] H. Song, D. H. Lee, H. G. Baek, B. Bae & S. H. Park, “AutomaticClassification of Disaster Images Based on Deep Learning,” Journal ofKorean Institute of Communications and Information Sciences, vol. 48,no. 12, pp. 1633-1636, Dec. 2023. [2] L.Wen,Z.Xiao,X.Xu&B.Liu,“DisasterRecognitionandClassifica-tionBasedonImprovedResNet-50NeuralNetwork,”AppliedSciences,vol. 15, no. 9, article 5143, May 2025. [3] F. Alam, T. Alam, M. A. Hasan et al., “MEDIC: a multi-task learningdatasetfordisasterimageclassification,”NeuralComputing&Applica-tions, vol. 35, pp. 2609-2632, Jan. 2023. [4] S. Gardoll & O. Boucher, “Classification of tropical cyclone containingimagesusingaconvolutionalneuralnetwork:performanceandsensitiv-ity to the learning dataset,” Geoscientific Model Development (GMD),vol. 15, no. 18, pp. 7051-7064, 2022. [5] Z. Zou, H. Gan, Q. Huang, T. Cai & K. Cao, “Disaster Image Classifi-cation by Fusing Multimodal Social Media Data,” ISPRS InternationalJournal of Geo-Information, vol. 10, no. 10, article 636, 2021. [6] S. Sonang, Y. Yuhandri & M. Tajuddin, “Hybrid CNN Approach forPost-Disaster Building Damage Classification Using Satellite Imagery,”Journal of Applied Data Sciences, 2024. [7] A. Calantropio, F. Chiabrando, M. Codastefano & E. Bourke, “Deeplearning for automatic building damage assessment: application in post-disasterscenariosusingUAVdata,”ISPRSAnnalsofthePhotogramme-try,RemoteSensingandSpatialInformationSciences,vol.V-1-2021, pp.113-120,2021. [8] HarishK*,HarishAbinav&D.Kumar,“DeepLearning-DrivenDisasterandAerialImageSegmentationusingU-NetandMulti-U-NetArchitec-tures,” Atlantis Press, 2024. [9] “Disaster management using deep learning on social media,” Bhavana& P. Ramasubramanian, International Journal of Applied Science andEngineering, vol. 18, no. 2, June 2021. [10] “Leveraging CNNs and Ensemble Learning for Automated DisasterImageClassification,”A.Rathod,V.Pariawala,M.Surana&K.Saxena,2023

Copyright

Copyright © 2026 S. Sreekanth, G. Teja Harshitha, K. Vagvevi , M. Vaishnavi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download Paper

Paper Id : IJRASET78995

Publish Date : 2026-03-29

ISSN : 2321-9653

Publisher Name : IJRASET

DOI Link : Click Here