Algorithmic and hardware provision for onboard radar data processing devices

Аuthors

Knyazeva V. V.^1*, Neretin E. S.^2**

1. Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia
2. Integration center branch of the Irkut Corporation, 5, Aviazionny pereulok, Moscow, 125167, Russia

*e-mail: knyazeva_valya@mail.ru
**e-mail: evgeny.neretin@ic.irkut.com

Abstract

Purpose: promptitude improvement of the existing onboard data processing devices for object recognition in radar images based on the optoelectronic apparatus and artificial intelligence methods. Research methods applied in the work: methods of data processing in complex systems, artificial intelligence techniques, methods of analysis of parallel computing algorithms, experimental methods, and test and inspection methods for samples of data measuring and control systems. The structure of the onboard data processing device was developed. The structure includes the optoelectronic matrix multiplier performing parallel optical matrix calculations that improves its promptitude when recognition of objects in the radar image on a flat stage. The algorithmic component of the device comprises Software for realizing operation of the modified neural networks along with database of object images that are to be detected. The process of object recognition on a flat stage is carried out by the onboard data processing device iteratively. One-dimensional and two-dimensional bipolar models of the Hopfield neural network were modified. Both models are intended for use in the onboard data processing devices based on the optoelectronic matrix multiplier for pattern recognition. The algorithms that realize the modified bipolar model of one-dimensional and two-dimensional Hopfield neural network were developed for recognition of objects in the radar images on a flat stage in the onboard data processing devices. The algorithms are implemented in the form of Software «Hopfield neural network simulator for on-board equipment». The Software testing and using it for simulation of simple examples fully confirmed its operability and adequacy of modified neural network models in the onboard data processing tasks. Simulation using radar images obtained with synthesized aperture radar confirmed the efficiency of the modified Hopfield neural ! ВЕСТНИК МАИ. Т.20. № 5
network models for solving tasks of object recognition in radar images on a flat stage in noise conditions. Two level architecture of firmware complex was developed for testing algorithms and hardware of optoelectronic components and perspective onboard modules of data processing devices including usage of the Hopfield neural network. The firmware complex designed allows carrying out testing of existing and promising onboard data processing devices. Originality: The onboard data processing device structure based on optoelectronic matrix multiplier and Hopfield neural network was developed; One-dimensional and two-dimensional bipolar Hopfield neural network models were modified for implementation in the onboard data processing devices based on the optoelectronic matrix multiplier for recognition of objects in radar images on a flat stage; Algorithms realizing the modified one- dimensional and two-dimensional bipolar Hopfield neural network models were developed. Algorithms applicable to the implementation in the onboard data processing devices based on optoelectronic matrix multipliers for object recognition in radar images on a flat stage; Structure of firmware complex for algorithms and hardware testing was developed. It provides testing of optoelectronic components and modules of perspective onboard data processing devices based on neural networks.

Keywords:

onboard radar-data processing units, neural networks, Hopfield model, vector-matrix multiplication, matrix-matrix multiplication, optical electronic devices testing

References

1. Ageev V.M., Pavlova N.V. Pribornye kompleksy letatelnykh apparatov iikh proektirovanie (Aircraft Instrumentation Systems and Their Design), Moscow, Mashinostroenie, 1990, 432p.
   
2. Pavlova N.V. Metody iskusstvennogo intellekta inovye informatsionnye tekhnologii vproektirovanii pribornykh kompleksov (Methods ofArtificial Intelligence and New Information Technologies inthe Design ofInstrumentation Systems), Moscow, MAI, 2000, 104p.
3. Neretin E.S. Vestnik Moskovskogo aviatsionnogo institute, 2011, vol.18, no.3, pp. 177-184.
   
4. Evtikhiev N.N., Onykii B.N., Perepelitsa V.V., Shcherbakov I.B. Neirokompyuter, 1994, no. 3-4, pp. 51- 59.
   
5. Tatuzov A.L. Neironnye seti vzadachakh radiolokatsii (Neural Networks inthe Problems ofRadiolocation), Moscow, Radiotekhnika, 2009, 432p.
6. Hopfield J.J. Neural Networks and Physical Systems with Emergent Collective Computational Abilities, USA, Proceedings ofthe National Academy ofSciences, 1982, vol.79, pp. 2554-2558.
   
7. RamachandranR., Gunasekaran N. Modifications of Two Dimensional Hopfield Model for Optical Neural Networks, International Conference onVision and Parallel Processing, ROVPIA96, University Sains Malaysia, Minden, Penang, Malaysia, 1996, pp. 408-413.
   
8. Farhat N.H., MiyaharaS., Lee K.S. Optical Analog of Two-Dimensional Neural Networks and their Application in Recognition of Radar Targets, AIP Conference Proceedings, 1986, no.151, pp. 146-152.
   
9. Kirpichev K.Yu., Neretin E.S. Elektronnyi zhurnal Trudy MAI, 2010, no 38, availableat: http:// www.mai.ru/science/trudy/eng/published.php?ID= 14158 (accessed 23.06.2010)