Résumé:
The integration of deep learning techniques into satellite-based flood detection systems aims to
improve early warning and prevention measures. Flooding poses significant risks to human lives and
infrastructure, necessitating effective monitoring and mitigation strategies.
This project leverages deep learning techniques, specifically transfer learning, to enhance
satellite-based flood detection systems. By utilizing pre-trained models such as ResNet-50, VGG-16, and
MobileNet-V2, we analyze satellite imagery to detect and map flood-prone areas with high precision.
CNNs have the ability to can learn complex patterns, which significantly improves detection accuracy
automatically.
Our approach demonstrated that custom CNN architectures could achieve superior accuracy, with our
custom model reaching 96.7% accuracy. Transfer learning also proved effective, with models like
VGG-16 and MobileNet-V2 achieving 95% accuracy. However, the hybrid model showed a high accuracy
of 98% with the pre-trained DenseNet201 model, demonstrating its efficacy for classification.
Applying quantum computing techniques in flood detection systems could offer substantial
advantages, including enhanced computational power and speed. These advancements could lead to
more accurate and efficient future flood prediction and response strategies.
