SpectraNet: Learned Recognition of Artificial Satellites from High Contrast Spectroscopic Imagery

Gazak, J. Zachary; McQuaid, Ian; Swindle, Ryan; Phelps, Matthew; Fletcher, Justin

doi:10.1109/wacv51458.2022.00246

Cited by 8 publications

(17 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…14 In the meantime, the effectiveness of computer vision against these problems (in which the visual context is highly ordered but beyond the visual capabilities of humans and the theoretical backing of physics) has been demonstrated. 1 This, paired with a proliferation of spectroscopy ready SDA telescopes, makes learned recognition for SDA enrichment a pressing need. 15,16…”

Section: Quantitative Vs Learned Spectroscopymentioning

confidence: 99%

“…Recognition models based on raw spectroscopic imagery generalize on small amounts of data−on the order of hundreds of observations per class. 1 Intuitively, this is because a spectrograph is neatly and specifically ordering information along the focal plane array−the information context is simple. Still, the dynamics of generalization need to be well understood for a given task before models are deployed.…”

Section: Generalizationmentioning

confidence: 99%

“…We adopt the definition of a concept as P (X, Y ), 19 P (X, Y ) = P (X)P (Y |X) = P (Y )P (X|Y ) (1) which represents the joint probability distribution between a dataset (X) and a model's outputs (Y). It follows from Eqn.…”

Section: Concept Driftmentioning

confidence: 99%

“…This period appears to end at roughly 250 examples / class, in good agreement with previous on sky experiments. 1 After about 10 weeks the experiment enters a "seasonal dilution" phase, in which added examples are seasonally shifted from the validation dataset. This is most visible in the Sparse experiment, where the dataset stops growing in size and the effects of dilution are visible by eye as generalization suffers by a few percent.…”

Section: Generalization Vs Training Baselinementioning

confidence: 99%

“…The resulting images, simple in form but complex in formation, are easily fit by CNNs. 1 Still, the spectra of objects change across observing season with the drifting position angle between an observer, satellite, and sun. Those seasonal changes affect what two dimensional projection of the satellite is illuminated by the sun, what projection is visible to an observer, and the angles between incident and reflected light.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Model validity dynamics in learned spectroscopic recognition of satellites

Gazak

Phelps

Swindle

et al. 2023

Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXIX

View full text Add to dashboard Cite

Recent work demonstrates that convolutional neural networks can be trained to recognize artificial satellites from spatially unresolved ground-based observations (SpectraNet). SpectraNet enables space domain awareness (SDA) catalogs to be enriched with object identity, a critical source of information for space domain stakeholders. As learned spectral SDA matures, conditions for training and deploying performant and calibrated neural network recognition algorithms must be measured. In this work we present a simulated three year baseline of observations using a longslit spectrograph on a single telescope. We use this dataset to develop a framework for measuring baseline data requirements for performant SpectraNet models, and for testing the performance of those models after deployment. On this limited (single telescope, longslit spectrograph) setup, the presented framework returns a performant model after three weeks of collections. Further, we find that a model can be deployed for a full annual cycle after twenty six weeks of data collection, and the model reaches maximum sustained inference performance after a year. Thus a SpectraNet powered longslit spectrograph can provide tactical inferences after a few weeks and be retrained to infer through seasonal variability during deployment. We find that the simulated system and dataset regularly exceed 82% classification accuracy, and discuss performance improvements with enhanced instrumentation and/or multi-telescope networks.

show abstract

Section: Quantitative Vs Learned Spectroscopymentioning

confidence: 99%

Section: Generalizationmentioning

confidence: 99%