A GPU-Parallel Image Coregistration Algorithm for InSar Processing at the Edge.

Area of Science:

• Geospatial Science
• Computer Science
• Electrical Engineering

Background:

• Interferometric Synthetic Aperture Radar (InSAR) processing, particularly image coregistration, is computationally intensive and typically performed in batch mode.
• The advent of low-power Graphics Processing Unit (GPU) accelerators presents an opportunity for edge computing applications.

Purpose of the Study:

• To develop and implement an efficient GPU-parallel algorithm for InSAR image coregistration.
• To adapt the algorithm for edge computing scenarios, focusing on performance and power efficiency.

Main Methods:

• Decomposition of the cross-correlation problem in InSAR image coregistration from a multilevel perspective.
• Design and implementation of a GPU-parallel algorithm targeting computational kernels.
• Analysis of algorithm accuracy, performance, and power efficiency across various settings, including edge computing.

Main Results:

• Significant speedup achieved in InSAR processing by leveraging GPU computing.
• Demonstrated applicability of the GPU-parallel algorithm in different scenarios, including edge computing.
• No loss of accuracy observed compared to traditional methods.
• Enabled onboard processing capabilities using System-on-Chip (SoC) hardware.

Conclusions:

• GPU-accelerated InSAR processing offers a viable and efficient alternative to traditional batch processing.
• Edge computing with GPUs can significantly enhance the speed and feasibility of InSAR analysis, including real-time applications.
• The developed algorithm is accurate, performant, and power-efficient, suitable for diverse deployment environments.