The week's pick
Random Articles
Reseach Article
Deep Learning and Explainable AI for Aerial Image based Flood Damage Detection
| International Journal of Computer Applications |
| Foundation of Computer Science (FCS), NY, USA |
| Volume 187 - Number 107 |
| Year of Publication: 2026 |
| Authors: Ramachandiran R., A. Sanjay, Gajulapalli Venkata Koushik |
10.5120/ijca20aa728f16aa
|
Ramachandiran R., A. Sanjay, Gajulapalli Venkata Koushik . Deep Learning and Explainable AI for Aerial Image based Flood Damage Detection. International Journal of Computer Applications. 187, 107 ( May 2026), 42-47. DOI=10.5120/ijca20aa728f16aa
Abstract
Disasters like floods are normally characterized by colossal losses in terms of human lives and property. Aerial photography is one of the extremely important tools in the evaluation of the damages and the identification of the impacted areas in the occurrence of floods. This issue is on the classification of aerial images in the postflood environment, and the images will be partitioned into varying areas with a great deal of accuracy using the aids of some of the deep learning networks, namely MobileNet and DenseNet. The objective of all the models is to classify the pictures into the six possible classes—buildings, flooded areas, forests, mountains, seas, and streets. It has Explainable Artificial Intelligence (XAI) technologies along with every model to ensure that there are such interpretability and transparency. The method employed is GradCAMgenerated Class Activation Maps. The result indicates that the deep learning models can distinguish the flooded areas among the rest of the critical elements in the postflood assessment and recovery planning. Besides, GradCAM interpretability enhances the model prediction’s reliability, which is a key element in the disaster mitigation and response interventions in real time. The performance evaluation of the given model is conducted on the basis of multiple measurements, and it is determined that this model is an addition to the already existing literature, which is geared towards the application of AIbased solutions that can support the process of postdisaster recovery.
References
- M. Rahnemoonfar, T. Chowdhury, A. Sarkar, D. Varshney, M. Yari and R. Murphy, “FloodNet: A High Resolution Aerial Imagery Dataset for Post Flood Scene Understanding,” arXiv preprint, Dec. 2020.
- B. Ghosh, S. Garg and M. Motagh, “Automatic Flood Detection from Sentinel1 Data Using Deep Learning Architectures,” ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. V32022, pp. 201202, May 2022.
- Takato (2020) investigates the classification of disasterrelated features in aerial images, focusing on regions affected by typhoons. The study utilizes GradCAM to provide visual explanations, helping to interpret how deep learning models identify and differentiate damaged areas.
- Chen et al. (2022) develop a lightweight convolutional neural network model designed to predict flood heights using radar echo data. The approach incorporates GradCAM to highlight important grid regions, improving both model interpretability and decisionmaking accuracy.
- Gebrehiwot et al. (2019) develop a deep convolutional neural network approach for mapping flood extents using UAVacquired imagery. Their method demonstrates the effectiveness of deep learning in accurately detecting floodaffected areas from aerial data.
- S. A. Raj, A. Chakravarty, K. Ghosh and S. Das, “Performance Comparison and Visualization of AIGeneratedImage Detection Techniques,” IEEE Access, vol. 10, pp. 1024610260, 2022.
- R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh and D. Batra, “GradCAM: Visual Explanations from Deep Networks via Gradientbased Localization,” arXiv preprint, Oct. 2016.
- Nam et al. (2025) explore the application of explainable artificial intelligence techniques for assessing flood susceptibility in Seoul. Their method integrates evolutionary algorithms with Bayesian AutoML optimization to enhance predictive performance while maintaining model transparency.
- This study (2025) presents a deep learningbased framework for classifying aerial images in postflood conditions. It combines robust neural network models with explainable AI techniques to improve classification reliability and provide insights into model predictions.
- Y. Liao, Y. Gao and W. Zhang, “Feature Activation Map: Visual Explanation of Deep Learning Models for Image Classification,” arXiv preprint, Jul. 2023
Index Terms
Keywords