Synthetic Aperture Radar (SAR) remote sensing has been widely used as one of the most effective tools for responding to earthquake disasters. In general, damaged-building detection with SAR data has been conducted based on change detection using temporal SAR data acquired in the same observation mode. However, it is not always possible to use SAR data obtained with the appropriate observation mode in unexpected events such as natural disasters. This study aims to detect earthquake-induced damaged buildings using temporal SAR data having different observation modes. We presented a contextual change analysis method to map damaged buildings based on novel textural features. This study was conducted using the bi-temporal Komapsat-5 data obtained in different polarization modes. Experimental results for the area severely damaged by the 2016 Kumamoto earthquake showed that the proposed textural analysis can improve detectability in building-damaged areas while maintaining low false alarm rates in agricultural areas. According to the grid-based accuracy analysis, the proposed method can successfully detect the damaged areas with a detection rate of about 72.5% and false alarms of about 6.8% even on challenging data sets.

Change detection using SAR data has been an active topic in various applications. Because conventional change detection identifies signal changes in single-pol radar observations, they cannot separately detect different kinds of change associated with different ground parameters. In this study, we investigated the comprehensive use of dual-pol parameters and proposed a novel dual-pol-based change detection framework utilizing different dual-pol scatter-type indicators. To optimize the exploitation of dual-pol change information, we presented a two-step processing strategy that divides the multiclass change detection process into a binary detection step that identifies the presence of changes and the classification step that distinguishes the types of change. In the detection stage, each dual-pol parameter was considered as an independent information source. Assuming potential conflict between dual-pol parameters, a disjunctive combination of detection results from different dual-pol parameters was applied to obtain the final detection result. In the classification step, an unsupervised change classification strategy was proposed based on the change direction and magnitude of the dual-pol parameters within the change class. Experimental results exhibited significantly improved detectability across a wide change spectrum compared with previous dual-pol-based change detection approaches. They also demonstrated the possibility of distinguishing different semantic changes without in situ ground data.