Natural hazard damage detection based on object-level support vector data description of optical and SAR Earth observations

Abstract

ABSTRACTThe Earth’s land covers are exposed to several types of environmental changes, issued by either human activities or natural disasters. On 11 March 2011, an earthquake occurred at about 130 km of the east coast of Sendai City in Japan. This earthquake has been followed by a huge tsunami, which caused devastating damages over the wide areas in the eastern coastlines of Japan. In order to manage such crises quickly and efficiently, change maps of affected areas are really crucial for damage estimation and proposing the essential services. Therefore, an automatic extraction of inundated and damaged areas from satellite images, with less user interaction, is essential and helpful. So far, the existing change detection (CD) approaches have a low degree of automation and are not optimal and applicable to high resolution optical and radar remote-sensing data. In order to resolve these problems, an integrated object-level CD method based on an object-based classifier and support vector data description method is proposed. In addition, parameter determination of the proposed method is addressed automatically by using an inter-cluster distance based approach. In order to evaluate the efficiency of the proposed method and extract the damaged areas, various optical and radar remote-sensing images from before and after of Sendai 2011’s tsunami, acquired by IKONOS, Radarsat-2, and TerraSAR-X, were used. The accuracy analysis of results showed a great flexibility for CD of by finding nonlinear object-level solutions to the problem. Furthermore, the comparative analysis of experimental results from IKONOS, Radarsat-2, and TerraSAR-X (kappa coefficient (κ): 0.85, 0.82, 0.76) and Support Vector Machine-based CD techniques (κ = 0.83, 0.73, 0.84), showed that the accuracy of the change maps is relatively improved. Finally, we came to the conclusion that the proposed method is considerably automated and less influenced by the errors in the classification process.