Flood Detect Net: A Deep Learning Framework for Flood Detection Using Attention-Enhanced Squeeze Net and Autonomous Drones

Floods are one of the most severe natural disasters caused by climate change, resulting in widespread damage to infrastructure and human life. Traditional methods of flood detection are time consuming and have limited coverage, especially in remote regions. In response, the proposed work introduces FloodDetectNet, a light-weight deep learning model that can identify floods from imagery collected by autonomous drones. The system functions by employing entry-level drones to take pictures of areas prone to flooding, which are then sent to a local computing device for classification via communication protocols. FloodDetectNet leverages SqueezeNet as the backbone architecture and uses the Efficient Multi-Scale Feature Extractor block (EMSFE) and Convolutional Block Attention Module (CBAM) to enhance classification performance. The EMSFE block extracts features from different scales and the CBAM layer concentrates on important regions, enhancing detection accuracy. When tested on the self-curated dataset, FloodDetectNet achieved an overall accuracy of 96.11 %, with 100% accuracy for the flood class. This renders it an extremely reliable real-time flood detection tool. Additionally, automated alerts facilitate quick response by informing concerned authorities and boosts disaster management, even in resource-scarce areas.