Flood Detection and Mapping: A Critical Review of Methods, Challenges and Future Prospects
摘要
Globally, flood is one of the most devastating natural disasters. The loss is registered not only in the form of human life or infrastructural damage but also of the entire local bio-diversity. A thorough technological intervention is necessary for rescue operators to arrest casualties and such damage. Over the years, remote sensing due to its large area coverage has proved as an efficient technique for flood detection and mapping but with its own set of limitations. In recent years, machine learning and deep learning has shown significant progress in remote sensing tasks. This paper critically reviews the key potentials of remote sensing and stateofthe-art machine learning and deep learning techniques applied to optical and synthetic aperture radar satellite imagery for flood detection and mapping. Initially, remote sensing methods are reviewed followed by experimental results on synthetic aperture radar satellite imagery to underscore the inadequacies of conventional methods. Next, the usage of machine learning techniques and their limitations in flood-related investigations are discussed highlighting the need to explore advanced methods for analyzing satellite imagery. Additionally, the paper focuses on deep learning methods, their usage in past studies and benchmark datasets followed by in depth discussion on potential challenges regarding the adoption of deep learning methods on remote sensing satellite imagery. In the end, future research prospects of flood detection and mapping are presented to the investigators.