Abstract
The topic of this work, a joint scientific program merging the CEA, the IMAG, the CNES (France) and the Naval Research Laboratories (USA), is the evaluation of connectionist techniques for on-board signal and image processing applications in radiative environments (e.g.: space). The objective is to define methods which improve the robustness with respect to radiations of electronic neural systems applied to artificial perception. Theoretical and simulation results are compared to two kinds of experiments: ① ground tests, performed in France and in the United States on electronic components and boards, ② an experiment on a neural artificial vision application embedded in a satellite (launch expected in July 1997). In this paper, we describe the first results of our work: after having verified on a natural texture classification application that training a neural network in noisy conditions leads to a significant improvement of robustness, we propose an interpretation of this phenomenon and suggest the use of simple activation functions, compatible with robustness.
This is a preview of subscription content, log in via an institution to check access.
Preview
Unable to display preview. Download preview PDF.
References
Pinçon, J. L.', Miniére, X., Lefeuvre, F., Classification of structured geophysical signals on-board satellite through a neural network approach. ICANN'95, Remote Sensing session, Paris, 1995.
Langenbacher, H., Zee, F., Daud, T., Thakoor, A., Radiation behavior of analog neural network chip. ICNN'96, Washington, 1996.
Assoum, A., Torres Moreno, J. M., Radi, N. E., Ecoffet, R., Velazco, R., Gordon, M. B., Robustness against SEU of digital implementations of Artificial Neural Networks. ICANN'95, VLSI & Dedicated Hardware session, Paris, 1995.
Velazco, R., Assoum, A., Cheynet, P., Ecoffet, R., SEU experiments on an artificial neural network implemented by means of a Transputer, IEEE Transactions on Nuclear Science, Vol. 6, N°43, Dec. 1996.
Muller, J.-D., Velazco, R., Gordon, M., Jubelin, B., Méroth, Y., Sensitivity and hardening of on-board neural systems (in French). NSI'96, Marly-le-Roi, France, 1996.
Kerlirzin, P., Multilayer Networks Robustness (in French). PhD Thesis, Univ. Paris XI, France, 1994.
Le Cun, Y., Generalization and Network Design Strategies. Connectionism in Perspective, Pfeifer et al. Eds, Elsevier Science Publishers B.V. (North-Holland), 1989.
Velazco, R., Cheynet, P., Muller, J.-D., Ecoffet, R., Artificial neural network robustness for on-board satellite image processing: results of SEU simulations and ground tests. IEEE Nuclear & Space Radiation Effects Conference NSREC'97, Snowmass Village, Colorado, July 1997.
Le Cun, Y., Denker, J. S., Solla, S. A., Optimal Brain Damage. Advances in Neural Information Processing Systems, Vol. 2, R. P. Lippmann, J. E. Moody & D. F. Touretzky Eds, Morgan Kaufmann, 1990.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1997 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Muller, J.D., Cheynet, P., Velazco, R. (1997). Analysis and improvement of neural network robustness for on-board satellite image processing. In: Gerstner, W., Germond, A., Hasler, M., Nicoud, JD. (eds) Artificial Neural Networks — ICANN'97. ICANN 1997. Lecture Notes in Computer Science, vol 1327. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0020316
Download citation
DOI: https://doi.org/10.1007/BFb0020316
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-63631-1
Online ISBN: 978-3-540-69620-9
eBook Packages: Springer Book Archive