Change detection is a fundamental approach in utilization of satellite remote sensing image, especially in multi-temporal
analysis that involves for example extracting damaged areas by a natural disaster. Recently, the amount of data obtained
by Earth observation satellites has increased significantly owing to the increasing number and types of observing sensors,
the enhancement of their spatial resolution, and improvements in their data processing systems. In applications for
disaster monitoring, in particular, fast and accurate analysis of broad geographical areas is required to facilitate efficient
rescue efforts. It is expected that robust automatic image interpretation is necessary. Several algorithms have been
proposed in the field of automatic change detection in past, however they are still lack of robustness for multi purposes,
an instrument independency, and accuracy better than a manual interpretation.
We are trying to develop a framework for automatic image interpretation using ontology-based knowledge representation.
This framework permits the description, accumulation, and use of knowledge drawn from image interpretation. Local
relationships among certain concepts defined in the ontology are described as knowledge modules and are collected in
the knowledge base. The knowledge representation uses a Bayesian network as a tool to describe various types of
knowledge in a uniform manner. Knowledge modules are synthesized and used for target-specified inference. The results
applied to two types of disasters by the framework without any modification and tuning are shown in this paper.