By Chris Forrester
Brian Weimer, Partner, Telecom Team Leader Sheppard Mullin, said his panel could addressed conventional imagery from visible through to hyperspectral sensing. The expert panel addressed the challenges during a session (‘Tooling Commercial Observation Services for Government Customers’) at the Silicon Valley Space Week’s MilSat Symposium held at Mountain View on October 24.
Frank Avila, Deputy Director, Source Operations Grp, National Geospatial Intelligence Agency was a buyer of all aspects of imagery. He compared conventional imagery with hyperspectral to a highly detailed police fingerprint in the amount of detail possible. However, he stressed that common standards was vital especially given that the Agency was looking at hundreds of satellites, and as well as needing the image it wanted to access the core information especially for first responders in times of catastrophe. Calibration of the assets was crucial. He added that there was a shortage of human talent and the NGA College had been established to help fill the knowledge gap.
Tina Ghataore, CEO, Aerospacelab Inc. already has a series of satellites flying but wanted to satisfy both government and commercial requirements. She said they constantly looked at how the end result image would be interpreted and, for example, was the image a vehicle covered by camouflage netting? Her analytic tools had that skillset. She said was a daily visit enough, or more frequent visits essential, and what was the geometric accuracy?
Alan Campbell, Principal Space Products Solutions Architect, Amazon Web Services (AWS) said that they were handling hundreds of Terabytes of data. The traditional model was for its satellite to pass over a target and for the data to be handled in a 15 minute window, but end-users wanted speedier latency and they were now satisfying those demands. AWS was enthusiastically using AI and machine learning team and in particular to enhance accuracy.
Tara Gattis, Strategic GEOINT Advisor & SME, Orbital Sidekick, was using AWS for its Cloud storage and pipeline delivery and said Orbital Sidekick would be adding new technology to its next iteration of satellites. She said that the processing pipeline was good once the image was received and the data could then be processed within an hour, but the bottleneck was frequently the download from the satellite.
Askash Parekh, CCO, Pixxel, was also using AWS although used its own ‘sandbox’ platform could be able to make it super-easy for customers to access hyperspectral imagery. It was not a downstream analyst player, but provided its data and tooling set for examining the data and the imagery. It is fair to say that hyperspectral is complex, he said. “At Pixxell we normally manage the data’s interpretation in 72 hours, but we are working to handle different demands depending on the user, and are handling most request in 6 hours, but can manage down to 30 minutes when the demand exists. But imagine the amount of data coming in per image. It can be Terabytes so compression is crucial, and then it is a question of how much could be spent in analysing 25 or more bands of data and thus achieving a sweet spot.