GPU-accelerated SAR processing tools development
Programme
GSTP
Programme Reference
GT11-028EO
Prime Contractor
CGI EESTI AS
Start Date
End Date
Status
Closed
Country
Estonia
Objectives
The objective of this activity is to develop a set of SAR pre-processing tools that maximize the benefit of GPUs as tools for massively parallelizing the processing, allowing fast generation of analysis-ready data.
Description
The activity aims to develop a toolbox providing capabilities for accelerated interactive processing of SAR imagery utilizing the massively parallel processing capability of GPU-s.
The toolbox shall be integrated into existing cloud-based processing platforms that are compliant with the standards agreed within the ESA Network of Exploitation Platforms (e.g. the TEP-s). A pre-selection of processing algorithms to be implemented within the activity has been made. These shall be implemented on the GPUs on the first basis. The archive and data handling mechanisms shall be developed in a way that the resulting GPU SAR toolbox is expandable with additional processing algorithms for both Sentinel-1 and other SAR data.
To maximize the benefit from GPU parallelization, development of a custom archive solution is required. The archive must be able to store imagery in tiles small enough to be moved onto GPU-s from the hard drive without any extra tiling effort when processing is initiated. The exact solution for the archive along with resulting requirements to the hardware shall be dependent on the use cases for the GPU toolbox and shall be specified early within the activity.
To achieve this, the following activities are required:
-
- Use case specification;
- Trade-off analysis to define detailed software architecture;
- Development of a custom light-weight archive for storing tiled imagery in a shared database;
- Development of a tiling mechanism to allow distributed storage of imagery and quick access by GPUs;
- GPU implementation of pre-processing algorithms (e.g. co-registered stack generation);
- GPU implementation of SAR algorithms providing analysis-ready data (e.g. coherence, intensity and polarimetric stack generation) and inserting them in datacube-like infrastructures;
- Development of API-s to display processed data in an user-friendly manner;
- Development of a lightweight user interface that could be integrated into existing platform UI-s.
The development is planned to take an iterative approach, providing the baseline software architecture and first results early within the activity, to allow further trade-off analysis between performance and interoperability be held.
Application Domain
Earth Observation
Technology Domain
6 - RF Subsystems, Payloads and Technologies
7 - Electromagnetic Technologies and Techniques
Competence Domain
9-Digital Engineering
Keywords
3-Platforms / Others
Initial TRL
TRL 3
Target TRL
TRL 5
Achieved TRL
TRL N/A