Abstract
Change detection using satellite data is frequently used in cases of natural disasters to identify the location and extent of the damage. However, current practical analysis for disaster monitoring largely depends on manual analysis by experts, creating a bottleneck of data that hinders effective and prompt response. This paper proposes an automatic change detection system that can execute entire processes of change detection without the intervention of human experts. The proposed system employs a knowledge-driven system based on ontology. The system stores knowledge modules, each of which can extract information or make an inference. At the time of change detection, the system selects the appropriate knowledge modules and constructs a Bayesian network for the inference of target change by interpreting the semantics of the target change and knowledge modules using ontology. The system extracts information and makes an inference, yielding probability images that indicate the confidence degree of the inference of the target change. This paper presents a demonstration of the proposed detection system by applying it without any modifications in the system to two cases of disaster monitoring: a mudslide and a flood.
