Towards a Semantic Understanding of very High- Resolution Satellite Images: the Case of Major Disasters

The lack of knowledge on damage extent and damage level of affected areas following a major disaster impedes the delivery of the necessary support in guiding rescue teams on the ground, delimiting the extent and level of damaged buildings, spotting the best location for refugee camps, and selecting effective access roads. The increased accessibility of VHR satellite imagery offers new perspectives for the remote sensing and disaster management communities. RS technologies allow fast, effective, and accurate observations of the affected areas. However, these observations need to be rapidly inspected and interpreted to deliver the necessary support. The International Charter "Space and Major Disasters" is activated for this purpose to provide the rescue teams with ready damage maps prepared by means of manual processing and interpretation of satellite images by photo interpreters. A complex, lengthy, and demanding task, which is also subject to errors and subjectivity. Automatic/semiautomatic tools are good alternatives. Automatic processing offers the required prompt treatment intended in such critical situations, nonetheless, it generally presents a semantic gap drawback. The objective of this work is the incorporation of semantics into RS and GIS applications to express and represent expert knowledge in an automatic way. A global ontology that allows geographic and disaster-related knowledge representation, expressivity, and discovery is developed with expert knowledge in remote sensing, disasters, and geographic domains. The approach is based on (i) the conceptualisation of domain knowledge and information surrounding the context, (ii) the development of a global ontology including eight sub-ontologies representing the characteristics of the different related interdomains, (iii) the development of an ontology-based VHR satellite image classification technique based on GEOBIA, and (iv) the application of the ontology and the previous classification results for change detection and damage assessment. A case study on Haiti 2010 earthquake is demonstrated, and the strengths and limitations of the approach are discussed. The results validate the impact of the ontologies in the geographic, remote sensing, and disaster management fields