EUVIMEDEUVIMED entdecken
Crossref

A Generalized Software Framework for Deep Neural Network Inference in Space Applications

Rafael Polanco-Segovia, Scott A. Zemerick, Piyush M. Mehta

IEEE Aerospace and Electronic Systems Magazine · 2025-12 · 40 (12) · 56-68

DOI: 10.1109/maes.2025.3574111 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Aerospace and Electronic Systems Magazine, 2025-12. Verfasst von Rafael Polanco-Segovia und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 22 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recent advances in hardware and software technology have made it possible to implement more resource-demanding deep learning algorithms in constrained environments. This creates opportunities to use deep learning for aerospace applications on increasingly smaller spacecraft. This work shows our Neural Network Execution Framework, which aims to provide a cross-platform and reusable framework to deploy trained neural networks for aerospace applications. This framework will execute the networks’ inference process regardless of the deep learning creation platform, for several flight software frameworks, operating systems, and hardware configurations. Starting with a specific implementation of our neural-based compression algorithm used to process Solar Dynamics Observatory images, and using NASA's Core Flight System, we recognized the potential to generalize this implementation to integrate more flight frameworks and execute different deep learning aerospace applications. Users and organizations can use our framework to create reusable deep learning deployment and execution solutions, instead of implementing one-off solutions for each specific developed aerospace application. This approach allows developers to focus on their deep learning models, rather than the deployment process on the flight platform. This work demonstrates the design, implementation, and testing of our framework for executing PyTorch and TensorFlow neural networks, using Core Flight System and a low-cost single-board computer to emulate a space-like hardware–software configuration. This software demonstration aims to highlight the benefits of our framework for implementing deep learning in aerospace missions. In addition, for comparative purposes, the performance of the networks in this hardware–software configuration is contrasted with the performance obtained in a conventional computer.

Zitieren

APA

Polanco-Segovia, R., Zemerick, S. A., & Mehta, P. M. (2025). A Generalized Software Framework for Deep Neural Network Inference in Space Applications. IEEE Aerospace and Electronic Systems Magazine, 40(12), 56-68. https://doi.org/10.1109/maes.2025.3574111

Vancouver

Polanco-Segovia R, Zemerick SA, Mehta PM. A Generalized Software Framework for Deep Neural Network Inference in Space Applications. IEEE Aerospace and Electronic Systems Magazine. 2025;40(12):56-68. doi:10.1109/maes.2025.3574111

BibTeX
@article{PolancoSegovia2025,
  author = {Polanco-Segovia, Rafael and Zemerick, Scott A. and Mehta, Piyush M.},
  title = {A Generalized Software Framework for Deep Neural Network Inference in Space Applications},
  journal = {IEEE Aerospace and Electronic Systems Magazine},
  year = {2025},
  volume = {40},
  number = {12},
  pages = {56--68},
  doi = {10.1109/maes.2025.3574111}
}
Kostenlos starten →