Subspace Analysis for Multi-temporal Disaster Mapping Using Satellite Imagery

TitleSubspace Analysis for Multi-temporal Disaster Mapping Using Satellite Imagery
Publication TypeConference Paper
Year of Publication2022
AuthorsOkorie, AM, Makrogiannis, S
EditorBebis, G, Li, B, Yao, A, Liu, Y, Duan, Y, Lau, M, Khadka, R, Crisan, A, Chang, R
Conference NameAdvances in Visual Computing
PublisherSpringer International Publishing
Conference LocationCham
ISBN Number978-3-031-20713-6

We propose an automated disaster mapping technique using pre- and post-disaster satellite imagery. We first find the geometric transformation for automatic image registration by matching regions represented by shape and intensity descriptors. We produce piece-wise constant approximations of the two images using the delineated regions. We perform linear subspace learning in the joint regional space and project the samples onto the orthogonal to tangent subspace to produce a change map and identify the outliers using statistical tests. We tested our method on multiple disaster datasets that is, four wildfire events and two flooding events. We validated our results by measuring the overlap score (DSC), and classification accuracy of our disaster map and ground-truth data. We performed comparisons to representative change detection techniques, namely Gabor Two-Level Clustering (G-TLC), and spectral index-based detection methods. Performance metrics indicated that the proposed Subspace Learning-based Disaster Mapping (SLDM) method produced more accurate change maps than the compared methods for multiple types of disaster events. Visual interpretation of the proposed SLDM method confirms its capacity for creating change maps for disaster mapping.