Datenbestand vom 23. März 2024

Warenkorb Datenschutzhinweis Dissertationsdruck Dissertationsverlag Institutsreihen     Preisrechner

aktualisiert am 23. März 2024

ISBN 9783843950084

84,00 € inkl. MwSt, zzgl. Versand


978-3-8439-5008-4, Reihe Informatik

Benjamin Bischke
Machine Learning-Aided Disaster Response - Using Remotely Sensed Imagery and Social Multimedia

214 Seiten, Dissertation Technische Universität Kaiserslautern (2021), Softcover, B5

Zusammenfassung / Abstract

The accelerating amount of natural disasters in recent years has caused immense socio-economic damage. Advances in disaster relief to aid a more effective emergency response are thus a central challenge of our time. This thesis explores how the shortage of disaster information can be solved with novel Machine Learning approaches. Thereby, a holistic view of disaster response is considered by studying information extraction methods for Social Media and Earth Observation (EO) content. While both sources provide vital in-situ information of disasters, their analysis from a methodological perspective is often challenging, e.g., current methods require manual labeling efforts, focus on particular modalities, and extract only a subset of the available information. The thesis emphasizes that Machine Learning methods have a clear potential to improve disaster response in practice by providing more accurate extraction results, faster response times, and novel perspectives of disaster events than existing approaches. In particular, we show the potential of using novel methods from Deep Learning to extract disaster-related information. Focusing on the analysis of EO data, this work achieves new state-of-the-art results on image segmentation datasets. Considering Social Multimedia, this thesis demonstrates its potential to contextually enrich and complement the information extracted from EO data - a vital aspect that most methods do not consider and allows us to go beyond state-of-the-art disaster map generation.