Opportunistic Bistatic SAR Image Classification Using Sub-aperture Decomposition

Abstract

In this article the classification of C -band opportunistic bistatic SAR images is assessed. The acquisition setup assumes Sentinel 1 spaceborne transmitter and the ground-based stationary receiver, COBIS located in an urban area. The synchronization/ echoed data spans many monostatic apertures; thus, an increased angular diversity may be exploited. The proposed framework is based on the use of sub-aperture decomposition of the backscatter in order to create the feature vector. Specifically, the goal of such an approach is to identify different scatterer signatures and associate them to specific semantic labels. Because the data dimension increased as a consequence of feature extraction approach, a technique for dimensionality reduction is suitable. The Doppler decomposition features are further embedded in the three-dimensional space in a semi-supervised manner. This operation is performed by UMAP algorithm. Finally, an unsupervised classification is achieved by DBSCAN. The presented results are obtained using manually labeled pixels from the following classes: forest, dam and water.