Error probability minimization while object identification by onboard computing system of an unmanned aerial vehicle

Computing machinery, complexes and computer networks


Аuthors

Protasov V. I.1*, Sharonov A. V.2**, Sharnin M. M.3, Klimenko A. S.3

1. National University of Science and Technology "MISIS", 4, Leninsky Prospect, Moscow, 119991, Russia
2. Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia
3. Institute of Physicotechnical Informatics, 6, Zavodskoy prospect, Protvino, Moscow Region, 142281, Russia

*e-mail: protasov_vi@mail.ru
**e-mail: anatoly.sharonov@yandex.ru

Abstract

This work set and solved the problem of unmanned aerial vehicles (UAVs) computing power sharing to ensure reliable identification of a number of objects, employing neuron network identification.

The main requirement imposed on a group of neural networks’ joint operation is related to incorrect detection of objects in complicated cases, when this probability in a single network is sufficiently great.

The authors found the conditions, which fulfillment defines the possibility of forming a group of neural networks, solving the problem of detection of any degree of complexity and returning a wrong answer with the probability not exceeding the preset small value.

Theoretical justification of neural networks’ joint operation organization is based on evolutionary solutions adjustment method.

On its first stage populations of separate neuron networks solutions were formed, and possible answers were generated, where the room was left for discarding the answer. At the second and subsequent stages the exchange of variants was carried out, and neuron networks, which «discarded» the answer, select, in their judgement, the right answers from proposed answers, or refuse to answer again. This iteration process continues until the majority of neuron networks give coinciding answers.

It is quite clear that this answer ca be of three types: <<right>>, <<wrong>> or <<cannot decide>>. The probability values of such answers depend on the number neural networks in the group, problem complexity, initial probabilities of right and wrong answers generation by single neuron networks, and of probabilities of right or wrong selection of foreign answers at the stages of solutions adjustment.

To obtain analytical solution of the set problem of defining conditions which fulfillment a group of neuron networks goes wrong with the probability not exceeding the preset small value, the authors employed the result of Condorcet jury theorem and Rasch model. The authors proved the theorem on inconsistent solution of a problem of arbitrary complexity, obtained by a group of neuron networks, probability tending to zero, as well as the theorem on limit value of existence problems’ complexity, which cannot obtain the proper solutions with the pre-set probability.

The judgement was exercised, that the proved theorems bear universal character and can be implemented to the group of natural intellectual objects, such as performing the task on context scientific citation number building in complicated cases.

The paper describes “error-free” objects detection technology by a UAVs group. Preliminary processing of images under detection was carried out according to the algorithm presented in [11].

Computer simulation of images detection by a group of neural networks confirmed the workability of the approach under discussion and allowed draw inferences on probability value of correct objects detection significant increase in simple cases, and reduce practically to zero the probability value of incorrect detection.

Keywords:

correct decision making probability, detection, network decision making system, unmanned aerial vehicles, neural network, evolutionary solutions adjustment method, Rasch model, context detection

References

  1. Zhuravlev Yu.I. Kibernetika, 1977, no.4. pp. 5–17.

  2. Zhuravlev Yu.I. Kibernetika, 1977, no.6. pp. 21–27.

  3. Zhuravlev Yu.I. Kibernetika, 1978, no.2. pp. 35–43.

  4. Protasov V.I. Konstruirovanie metasistemnykh perekhodov (Construction of metasystem transitions), Moscow, Izd-vo instituta fiziko-tekhnicheskoi informatiki, 2009, 186 p.

  5. Сondorcet, marquis Marie-Jean-Antoine-Nicolas de Caritat. Essai sur l’application de l’analyse à la probabilité des décisions rendues à la pluralité des voix. Imprimerie Royale, Paris, 1785.

  6. Rasch G. Probabilistic Models for Some Intelligence and Attainment Tests /Expanded Edition, with Foreword and Afterword by B.D. Wright. Chicago: University of Chicago Press, 1980.

  7. Protasov V.I., Potapova Z.E., Osipchuk O.K. Trudy ХХ Baikal'skoi Vserossiiskoi konferentsii «Informatsionnye i matematicheskie tekhnologii v nauke i upravlenii», Irkutsk, ISEM SO RAN, 2015, pp. 164-175.

  8. Ezhela V.V, Klimenko S.V., Raikov A.N.,. Sharnin M.M. A Semantic Approach to the Evaluation of the Quality of Academic Publications. ISSN 0005-1055, Automatic Documentation and Mathematical Linguistics. 2015, vol. 49, No. 4. pp. 117-121.

  9. Malyshev O.V., Khmarov I.M,. Kondrashov N.G. Vestnik Moskovskogo aviatsionnogo instituta, 2011, no.1, vol.18, pp. 142-148.

  10. Maksimov N.A, Sharonov A.V. Nauchnyi vestnik Moskovskogo gosudarstvennogo tekhnicheskogo universiteta grazhdanskoi aviatsii, 2014, №207(09), pp. 67-76.

  11. Kalaev I.A., Protasov V.I., Shapowal V.G. An Efficient Method for the Recognition of Three- Dimensional Objects from a Contour Segment // Pattern Recognition and Image Analysis. 1998, vol.8. No 2, pp. 196-197.

  12. Gupta L., Sayeh M.R. Neural networks for planar shape classification. IEEE. 1988. p. 936.

  13. Kim V.V., Krylov I.G. Trudy MAI, 2012, no. 62: http://www.mai.ru/science/trudy/eng/published.php?ID=35507

  14. Sharonov A.V., Maksimov N.A., Sincha D.P. Trudy MAI, 2011, no.49: http://www.mai.ru/science/trudy/eng/published.php?ID=28119&PAGEN_2=2


Download

mai.ru — informational site MAI

Copyright © 2000-2019 by MAI

Вход