М.H. Aksyaitov – Head of Research and Production Center,  Concern Granit-Electron JSC

Е.V. Egorova – Ph.D. (Eng.), Associate Professor, 

Department of Telecommunication Systems, Russian University of Technology MIREA

Е-mail: calipso575@gmail.com

А.N. Rybakov – Leading Expert,  FSUE VNIIA named N.L. Spirit

At present, the improvement and development of new models of aircraft and rocket technology requires an increase in sensitivity, noise immunity, resolution and accuracy in determining the coordinates and speeds of movement of targets for PLC. The recognition capabilities of radar images are largely due to the development of digital signal processing methods. Detection of ground targets is a complex radar task and is associated with the selection of the signal reflected from the target against the background of interference caused by reflections from the underlying surface and local objects, the effective dispersion surface (EPR) of which can be commensurate and even significantly exceed the EPR of the target. Improving methods for detecting and extracting signals from interference, developing and testing new methods and means of recognition, taking into account modern radio masking of objects, taking into account the issues of forecasting the radar visibility of objects of complex spatial form, the use of new radio wave bands and wideband pulse signals, allows solving a wide range of radar tasks. Based on the analysis of the recognition results of a specific image fragment: a test, in grayscale, radar images of the earth's surface using various recognition systems, the following features can be identified: from the point of view of speed, the Hamming neural network and its modification are most preferred. It should be noted that the neural network of back-propagation of error due to its more complex structure requires much more time for training and proper recognition of the object. The ability of the algorithm for each output to learn not only from the reference image, like the Hamming neural network, but also close to the reference, including noisy images, makes it possible in some cases to more reliably recognize objects, especially if there is preliminary information about the nature of noise and distortions. The proposed two-stage recognition procedure: «coarse» recognition based on the method invariant to the angle and verification of recognition reliability (refinement of the results based on any of the neural networks listed, trained by images from different angles), allows the object of the radar image to be recognized more reliably and with high quality. The prospects for the development of radar using complex noise-like broadband probing signals, including for detecting low-contrast ground objects, directly depend on further improvement of the recognition algorithm, which is invariant to the angle, which will allow reliable and high-quality recognition of radar image objects in a one-step process.

1. Obnaruzhenie i raspoznavanie ob"ektov radiolokatsii. Kollektivnaya monografiya. Pod red. A.V. Sokolova. M.: Radiotekhnika. 2006.  [in Russian]
2. Surovtsev I.S. Klyukin V.I., Pivovarova R.P. Neyronnye seti. Voronezh: Izd-vo VGU. 1994. [in Russian]
3. Korotkiy S. Neyronnye seti: algoritm obratnogo rasprostraneniya. URL: http://www.twirpx.com. [in Russian]
4. For A. Vospriyatie i raspoznavanie obrazov: Per. s frants. A.V. Seredinskogo. Pod red. G.P. Katysa. M.: Mashinostroenie. 1989. [in Russian]