2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS | 2021
A Study of Recovering Polsar Information from Single-Polarized Data Using DNN
Abstract
High-resolution (HR) SAR images contain rich details thus offer better capabilities on describing the target spatial properties. Besides, Polarimetric SAR (PolSAR) data can convey physical characteristics embedded in the polarimetric information. However, HR and polarimetric are two conflicting properties since acquiring PolSAR data will limit the pulse repetition frequency (PRF) for each polarization hence limit the resolution. Tackling this issue, recovering HR full-polarized (FP) data from HR single-polarized (SP) data using DNN is a meaningful approach. This paper studies the ability of DNN to recover FP information from SP data of different resolutions and different polarization channels in various areas. We use UAVSAR L-band data for experiments. Results demonstrate the feasibility of using DNN trained by lower-resolution FP data to recover HR FP data from HR SP data. It is showed that the performance is more notable over natural surfaces than in built-up areas, and HH polarization has a better reconstruction result compared with other channels. Besides, the recovery capability in different regions is also compared, which provides some inspirations for FP and HR SAR data applications.