Spacecraft Identification by Multispectral Signature Analysis Using Neural Networks.
Abstract
This study examines the feasibility of: (1) identifying satellites by their spectral signatures and (2) developing an algorithm to automate the process. The efforts of this study focus on solving the problem of crosstagging deep space objects. Crosstagging is the misnaming of a satellite which occurs when the identity of a tracked satellite is unknown or when the identities of several satellites are commingled. This problem can occur due to variations in satellite orbits and/or delays between data collection. Sunlight reflecting off of a satellite creates a spectral signature. Satellite signatures may differ due to geometry and material properties. Spectral signatures of seven satellites were simulated using an image simulation software package and high fidelity satellite models. These spectra took into account atmospheric degradation and were simulated for a variety of orbital parameters and different imaging times. These simulated signatures trained a neural network to identify the satellite. The trained network was able to accurately identify satellites based on their spectral signatures. This technology has application to the space analyst needing to identify satellites beyond the range of resolved imaging and detect anomalies on these objects.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1997
- Accession Number
- ADA324846
Entities
People
- Conrad J. Poelman
- Stephanie R. Meltzer