This paper by Airbus presents two artificial intelligence methods to make space optics instruments more precise. The current problem is that measurements have to be highly accurate, but they reach complexity limitations. Humans tend to make mistakes, and mistakes can prove to be very costly.
1) Random telegraphic signal processing
Test data set is used for machine learning, including the pixel time-series to be checked for RTS. If the pixel checked is RTS pixel, the pixel is flagged and the RTS behaviour characterised. If not, the program moves on.
2) Ghost stray light processing
Ghost straylight is difficult to fully correct by post processing and very time-consuming to detect. GSL also pushes up the volume of images needed for calibration. Machine learning can help automatically segment ghost stray light, reducing the volume of images needed and leading to better precision.
Conclusion
Machine learning allows for more accuracy with more speed. Testing the random telegraphic signal processing, the Airbus-propoused machine learning method generated the equivalent output to classical time-series histogram 10 hours in under 30 minutes.
The AI applied for the space spectrometer for ghost stray light detection is Airbus 2019 patent pending.
Sources:
Ferrato, M., Rivière, R., Candeias, H., Schmid, S., & Krauser, J. (2021). Space optics instrument optimization and characterization with artificial intelligence. International Conference on Space Optics — ICSO 2020, 11852, 2502–2512. https://doi.org/10.1117/12.2600054