GPU4EO challenge 2015: boosting SAR interferometry

Recent advances in GPU computing and programming hold promise for time efficient implementation of imaging algorithms, thus enhancing the development of advanced Emergency Management Services based on Earth Observation technologies. In the framework of the RIESCO project, Planetek Italia Srl has promoted the GPU4EO Challenge 2015, whose goal was to stimulate the intensive use of GPUs in the field of image processing and more specifically in the field of remote sensing applications, in order to enhance the synergy between research and business. In particular, Synthetic Aperture Radar (SAR) Differential Interferometry is a powerful and consolidated technique for ground displacements detection and geohazard monitoring, but impose strong requirements in terms of hardware resources. To this end, the Delft Object-oriented Radar Interferometric Software (DORIS) was selected as target application. This open source tool, developed by the DEOS Institute of the Delft University of Technology, is specific for the interferometric processing of satellite SAR images and is composed of a number of computing processes to be optimized and ported to GPU platforms, in order to reduce the overall execution time. GPU4EO has been a sprint initiative, from April to mid-June 2015, which involved research teams coming from different Italian universities. A reference hardware platform was provided to all the participants by the ReCaS data center, based on NVIDIA K40 board, for testing the developments and assessing the performance achieved. In this presentation, the most effective results achieved by the winner team, from University of Salento, will be presented. The main issues have been the data dependency within loops, the optimal use of device memory hierarchy, the isolation of the I/O operations and the use of existent optimized numerical library for GPU like cuFFT. The results achieved within the challenge timeframe have been very encouraging, reaching more than 5x of performance improvement.

–