Session
Technical Poster Session 1
Location
Utah State University, Logan, UT
Abstract
This work demonstrates the possibility of enabling onboard processing of SAR data in real-time through the adoption of an innovative focusing technology coupled with object detection, using limited computational resources. Our approach aims to provide a coarse focused product onboard to unlock real-time monitoring capabilities, complementing the ground-based detailed focusing algorithms.
The focusing algorithm transforms the Level-0 raw signal into Level-1 Single-Look-Complex (SLC) data. It consists of a two-layers hybrid architecture: a traditional Fast-Fourier Transform (FFT) algorithm for range processing and a Deep Neural Network (DNN), trained to solved the azimuth processing task, which provides scalability and modularity benefits. After focusing, an object detection network is trained to detect the presence of ships in the SLC data.
The whole processing chain has been optimized and deployed on different embedded devices, including NVIDIA Jetson Nano, and NVIDIA Jetson Xavier, to demonstrate the feasibility of running the overall pipeline onboard the future generations of SAR missions.
Unlocking Onboard SAR Processing: Focusing and Ship Detection on Sentinel-1 IW Data
Utah State University, Logan, UT
This work demonstrates the possibility of enabling onboard processing of SAR data in real-time through the adoption of an innovative focusing technology coupled with object detection, using limited computational resources. Our approach aims to provide a coarse focused product onboard to unlock real-time monitoring capabilities, complementing the ground-based detailed focusing algorithms.
The focusing algorithm transforms the Level-0 raw signal into Level-1 Single-Look-Complex (SLC) data. It consists of a two-layers hybrid architecture: a traditional Fast-Fourier Transform (FFT) algorithm for range processing and a Deep Neural Network (DNN), trained to solved the azimuth processing task, which provides scalability and modularity benefits. After focusing, an object detection network is trained to detect the presence of ships in the SLC data.
The whole processing chain has been optimized and deployed on different embedded devices, including NVIDIA Jetson Nano, and NVIDIA Jetson Xavier, to demonstrate the feasibility of running the overall pipeline onboard the future generations of SAR missions.
