Technology trends and capabilities for maximizing useful ...
Transcript of Technology trends and capabilities for maximizing useful ...
![Page 1: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/1.jpg)
Andreas Thorvaldsen*, Jānis Gailis, Michael Soukup,Arnoud Jochemsen, Christina Aas, S[&]T Norway
OBDP2019, ESTEC, February 26th, 2019
Technology trends and capabilities for maximizing
useful throughput per downlinked data unit by
enabling onboard image processing
https://stcorp.no
![Page 2: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/2.jpg)
2
Outline
Why process onboard?
HyperScout experience
The FONDA project
Training/test data generation
![Page 3: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/3.jpg)
3
The main reason
Image sensors == big data.
Image sensor data rates >> cubesat downlink data rates.
(NASA)
Limit the number of acquisitions
Timeliness of the data is poor
![Page 4: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/4.jpg)
4
The HyperScout mission“Miniaturized hyperspectral imager with its own brain”
European Space Agency (ESA) GSTP project.
2014: Development started 2016: Engineering model ready 2017: Flight model ready 2018: Launch onboard GOMX-4B
![Page 5: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/5.jpg)
5
Parameter Value
Orbit Sun synchronous, 500 km
Field of View 23º (ACT) x 16º (ALT)
Ground Sampling Distance
70m
Swath 200 x 150 km (ACT x ALT)
Spectral range 400 – 1000 nm
Spectral res. 15 nm
Dynamic range 12 bit
Number of bands ~45
SNR 50-100
Mass 1.3 kg
Power 9 W (peak)
The HyperScout instrument on GOMX-4B
![Page 6: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/6.jpg)
6
GOMX-4A and GOMX-4B
(Photo by ESA)
![Page 7: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/7.jpg)
7
First light
Commissioning image of HyperScout®. Flase colour single image of the southern Cuban costline. Image acquired on the 20th March 2018
(cosine)
![Page 8: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/8.jpg)
8
HyperScout on-board data processing
Image courtesy VITO
Data level Processing step
L1a Geometric calculations
L1b Radiometric corrections
L1c Data reshaping
L2a Projection and resampling
>=L2b Cloud masking, atmospheric
correction, segmentation, classification, change detection
![Page 9: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/9.jpg)
9
(slide with confidential content removed)
![Page 10: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/10.jpg)
10
10-band-image L2a at 150m
![Page 11: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/11.jpg)
11
(slide with confidential content removed)
![Page 12: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/12.jpg)
FONDA project
12
Flexible ONboard Data Analysis:
FPGA↑↑ / Watt
↑ / time
↓↓↓ / manhour (VHDL)
GPU↑ / Watt
↑↑ / time
↑ / manhour (C)
CPU↓ / Watt
↓ / time
↑↑↑ / manhour (Python, C++)
spectral calibration
atmospheric correction
haze cloud detection (CNN)
cloud shadow detection (CNN)
projection
land cover classification (CNN)
orchestration
landmark/object recognition
change detection
identify CNN training holes
prioritize downlink
radiometric processing
cloud masking (CNN)
ship detection (CNN)
iceberg detection (CNN)
![Page 13: Technology trends and capabilities for maximizing useful ...](https://reader034.fdocuments.in/reader034/viewer/2022042405/625dcd7674e13e791967a96c/html5/thumbnails/13.jpg)
Test/training data generation
13
Spectrally and spatially downsample data from existing satellite instrument into data for a future satellite instrument:
To make good training data, existing data should have at least the same Spatial resolution Spectral resolution Spectral range
Sentinel-2 possible alternative for RGB General purpose hyperspectral sources preferable:
Hyperion HICO CHRIS DESIS PRISMA
HyperScout processor used PROBA-V, Balloon and AVIRIS for test data