A Domain-based Late-Fusion for Disaster Image Retrieval from Social Media

We introduce a domain-specific and late-fusion algorithm to cope with the challenge raised in The MediaEval 2017 Multimedia Satellite Task. Several known techniques are integrated based on domainspecific criteria such as late fusion, tuning, ensemble learning, object detection using deep learning, and temporal-spatial-based event confirmation. Experimental results show that the proposed algorithm can overcome the main challenges of the proper discrimination of the water levels in different areas as well as the consideration of different types of flooding events.