Thomas Gemmer, Aptima, Inc.; Charlotte Shabarekh, Aptima, Inc.
Keywords: Predictive SSA, Non Traditional Sources
Abstract:
Increasingly, eye witness observations about world and space events can be found in open source intelligence (OSINT) ranging from widely-used social media platforms such as Twitter to niche astronomy websites such as SeeSat. While much of Space Situational Awareness (SSA) continues to be performed using the United States Air Forces Space Surveillance Network (SSN), there is an increased interest in incorporating Non-Traditional Sources (NTS), such as OSINT and commercial sensors. However, working with OSINT provides challenges for exploitation because of low quality data products such as low-resolution imagery from amateur astronomers, unreliable tweets and inaccurate geo-tagging. Additionally, fusing and correlating OSINT with traditional sensor data products is challenging because of inconsistencies in meta-data and reporting standards. For instance, backyard astronomers widely range in how they report observations and sometimes do not correlate observations to catalog IDs. Furthermore, exploiting OSINT requires the consideration of the trustworthiness of sources to manage deception and misinformation.
In this paper, we fuse simulated observations from NTS with simulated observations from the SSN to improve our satellite maneuver prediction system which was presented at AMOS 2016 (Shabarekh et al, 2016). Our satellite maneuver prediction technology is built on an unsupervised machine learning algorithm called the Interval Similarity Model (ISM) which lends itself well for dealing with ambiguity and noisy data inherent in NTS. Its probabilistic approach can incorporate weights to award or penalize observations from specific sensors or data types and can in fact learn those biases to discover which sources are consistent with one another. We present experiments that compare the baseline system which only used SSN data with the new system that incorporates NTS and SSN observations. Our initial findings show that the updated algorithm improves better than the results in the 2016 paper and the probabilistic nature of this approach supports the bias or discounting of sources to manage issues of trustworthiness of NTS.
Date of Conference: September 11-14, 2018
Track: Poster