InSAR-DLPU: A benchmark dataset for deep learning-based synthetic aperture radar interferometry phase unwrapping [Software and Data Sets]
Lifan Zhou, Hanwen Yu
Abstract
This article presents a large and diverse benchmark dataset, InSAR-DLPU, consisting of 31,100 pairs of wrapped and absolute phase patches to support deep learning (DL) studies of phase unwrapping (PU) techniques in synthetic aperture radar interferometry (InSAR) signal processing. Each pair of patches in this dataset is generated by all the terrain data in China, provided by the 30-m-resolution NASA Shuttle Radar Topography Mission (SRTM) digital elevation map (DEM) database. To the best of our knowledge, InSAR-DLPU is the first public dataset offering a benchmark for DL-based InSAR PU tasks. Based on this dataset, we can extensively validate and compare existing deep convolutional neural networks (DCNNs). Moreover, owing to the high diversity of topographic features and noise levels, this dataset has better information richness, which can improve the generalization ability of DL-based PU methods. We have made InSAR-DLPU publicly available at <uri>https://github.com/zhoulifan/InSAR-DLPU</uri>, offering an important resource to support studies on DL-based PU in InSAR applications (e.g., topographic mapping and deformation monitoring).