Numerous factors, such as a rapid urbanization, increased development and economic activity in flood plains, and climate change are responsible for floods. Utilizing satellite images to analyze the flood areas is an effective technique to accomplish flood area analysis. Houses with completely or partially surrounded by flooded water will help the rescue team to prioritize the rescue areas. Therefore, in this article two custom architectures based on convolutional neural networks (CNN) along with pretrained VGG16 and MobileNet architecture is evaluated for classifying the completely or partially surrounded houses. In this work, a new dataset is prepared from the xBD dataset that comprised of satellite images with flood affected houses. The results showed that MobileNet v4 architecture produces better results with an accuracy of 75%.
Deep neural network-based clustering algorithms have proven potential for clustering high dimensional and real-world data because of their tremendous representa
1 January 2026 in Fusion: Practice and Applications
Floods are considered the most significant disaster that occurs frequently worldwide. Floods result from heavy rainfall, snow melts, or storm surges caused due to climate change. Identifying custom UAV flood image datasets for classifying them into extreme or low floodwater level images would assist the rescue teams in prioritizing based on floodwater levels and deploying the resources to perform effective rescue operations. This work uses a Convolutional Block Attention Module (CBAM) to enhance water-level classification performance using pre-trained CNN models, Vision (ViT), and Shifted window (Swin) transformers. This article proposed a hybrid transformer that combines best-performing pre-trained CNN models with Vision transformers for improved classification performance. Experimental results demonstrate the model’s effectiveness in classifying the images into extreme floodwater or low floodwater levels by leveraging features from attention-based transformer models.
Floods are the most common natural disaster when excess water submerges the dry ground. Floods affect human lives, livestock, etc. Many people usually are stuck in flood-prone areas. To help the rescue team decide the prioritized rescue locations, this article classifies the flooded houses captured by satellite as completely or partially surrounded by flood water. This will help the rescue team decide on houses needing urgent attention. Therefore, this article uses various pre-trained Convolutional Neural Network (CNN) models, such as Visual Graphics Group (VGG16), Residual Network (ResNet101v2), Densely Connected Convolutional Neural Network (DenseNet201), InceptionNetv3, etc., with optimized hyperparameters for post-flood image classification. ResNet101v2 obtained an accuracy of 86.87 %. This article also investigates many hybrid models by combining two pre-trained models in parallel. Among them, the hybrid CNN model using MobileNetv2 and DenseNet201 achieved an accuracy of 85.83 %.