Chair of Data Science in Earth Observation at the Technical University of Munich
Extreme weather events and climate changes increase the need for automated flood detection systems. In recent years, the deep learning models trained on earth observation data became promising approach for this problem producing state-of-the-art predictive results. Nevertheless, their usage in such critical real-world applications is hindered by the complex and non-linear internal inference mechanism that results in an intransparent decision process. In this research proposal, we plan to develop a novel explainable and unsupervised machine learning method which performs concept discovery for flood detection in the scope of a research visit of Ivica Obadic (PhD student at the Technical University of Munich) to the Lancaster Intelligent Robotic and Autonomous System Centre. The planned method aims to improve and optimize the existing pipelines for unsupervised concept discovery by utilizing the prototypes in the SwAV method which organize the latent space for efficient and discriminative feature representations. As an unsupervised method, it also aims to reduce the need for labelled data for accurate flood detection. The results of the proposed method will be benchmarked against the existing explainable machine learning methods for flood detection in terms of predictive performance and model interpretability.
Keywords: deep learning, unsupervised concept discovery, self-supervised learning, flood detection, earth observation
Scientific area: Explainable Machine Learning