A.A. Romanov - Post-graduat Student, Moscow Aviation Institute (National Research University); Engineer, JSC «Corporation «Vega» (Moscow). E-mail: alexromanoviv@ya.ru B.G. Tatarsky - Dr.Sc. (Eng.), Professor, Moscow Aviation Institute (National Research University); Director of Research and Education Center, JSC «Corporation «Vega» (Moscow)

The key step of earth\'s surface radar image stitching is relative orientation. Using the coordinates of the overlap region points we can calculate orientation within overlap region. To obtain desired transformation of other image regions it is necessary to use extrapolation in solving problems which succeeded artificial neural networks. In this paper, the research and simulation of existing architectures of predicting and approximating artificial neural networks were done. To evaluate the performance of obtained neural networks several pairs of overlapping radar images were formed with typical geometric distortions - offset, blur, zoom, rotate and projective transformation. The three-layer cascade neural networks trained by the method based on the algorithm of Levenberg-Marquardt are able to solve the task. The disandvantage of the neural network application is a large error in case of small radar image overlapping region caused small and incomplete training set.

 

1. Verba V.S., Neronskijj L.B., Osipov I.G., Turuk V.EH. Radiolokacionnye sistemy zemleobzora kosmicheskogo bazirovanija. M.: Radiotekhnika. 2010. 675 s.
2. SHkolnyjj L.A., Tolstov E.F., Detkov A.N., Karpov O.A., JAkovlev A.M., Titov M.P., Filatov A.A., Tonkikh A.N., Cvetkov O.E., Arkhangelskijj A.S. Radiolokacionnye sistemy vozdushnojj razvedki, deshifrirovanie radiolokacionnykh izobrazhenijj: uchebnik dlja kursantov VVIA imeni professora N.E. ZHukovskogo. M.: Izd. VVIA im. prof. N.E. ZHukovskogo. 2008. 531 s.
3. Osovskijj S. Nejjronnye seti dlja obrabotki informacii / Per. s polsk. I.D. Rudinskogo. M.: Finansy i statistika. 2002. 344 s.
4. Soldatova O.P., Semenov V.V. Primenenie nejjronnyjj setejj dlja reshenija zadach prognozirovanija // EHlektronnyjj mnogopredmetnyjj nauchnyjj zhurnal «Issledovano v Rossii». 2006. T. 9. MFTI. S. 1270-1276.
5. JUsupov A.N. Nejjrosetevaja approksimacija uslovnojj funkcii chastoty // Nejjrokompjutery: razrabotka, primenenie. 2015. № 8. S. 93-98.
6. KHantimirov R.I. Prognozirovanie nagruzki v oblachnojj vychislitelnojj srede s ispolzovaniem nejjrosetejj EHlmana, obuchaemykh sistemojj iskusstvennogo immuniteta // Nejjrokompjutery: razrabotka, primenenie. 2015. № 3. S. 59-64.
7. Galushkin A.I. Nejjronnye seti: osnovy teorii. M.: Gorjachaja linija - Telekom. 2010. 496 s.
8. Gorban A.N. Obobshhennaja approksimacionnaja teorema i vychislitelnye vozmozhnosti nejjronnykh setejj // Sibirskijj zhurnal vychislitelnojj matematiki. 1998. № 1. T. 1. S. 12-24.
9. Fercev A.A. Uskorenie obuchenija nejjronnojj seti dlja raspoznavanija izobrazhenijj s pomoshhju tekhnologii NVIDIA CUDA // Vestn. Samarskogo gos. tekhn. un-ta. Ser. Fiz.-mat. nauki. 2012. Vyp. 1(26). S. 183-191.
10. Kondratjuk A.V., Krisilov V.A. Metod povyshenija chuvstvitelnosti nejjronnykh setejj, obuchaemykh s uchitelem, v zadachakh prognozirovanija vremennykh rjadov // Izvestija JUzhnogo federalnogo universiteta. Tekhnicheskie nauki. 2006. T. 71. № 16.  S. 65-69.
11. Kruglov V.V., Borisov V.V. Iskusstvennye nejjronnye seti. Teorija i praktika. Izd. 2-e, stereotip. M.: Gorjachaja linija - Telekom. 2002. 382 s.
12. Beskorovajjnyjj V.V., Soboleva E.V. Identifikacija chastnojj poleznosti mnogofaktornykh alternativ s pomoshhju S-obraznykh funkcijj // Bionika intellekta. KHarkov: KHNURE. 2010. № 1(72). S. 50-54.