AlleyFloodNet: A Ground-Level Image Dataset for Rapid Flood Detection in Economically and Flood-Vulnerable Areas

Urban flooding in economically and environmentally vulnerable areas—such as alleyways, lowlands, and semi-basement residences—poses serious threats to lives and property. Existing flood detection research has largely relied on aerial or satellite-based distant-view imagery. While some studies have explored ground-level images, datasets specifically focused on flood-vulnerable areas remain scarce. To address this gap, we introduce AlleyFloodNet, a ground-level image dataset designed to support rapid and accurate flood classification in high-risk urban environments. The dataset reflects a variety of real-world conditions, enabling deep learning models to better recognize floods in complex urban settings. We fine-tuned classification models using AlleyFloodNet and compared their performance to models fine-tuned on FloodNet, a widely used UAV-based dataset. Results show that models trained on AlleyFloodNet significantly outperform those trained on FloodNet when applied to ground-level flood images. This demonstrates the importance of viewpoint-specific data in improving detection accuracy for localized flooding. By constructing a dataset tailored to economically and flood-vulnerable areas, this study contributes to the development of practical flood detection systems that aim to reduce disaster impacts and enhance protection for at-risk communities.