A PROCEDURE FOR ISAR OBJECTS RECOGNITION WITH NEURAL NETWORKS ALGORITHM
With the development of inverse synthetic theory and the development of effective methods for obtaining high-quality image images in Inverse Synthetic Aperture Radars (ISARs), new possibilities are provided to obtain images of high-resolution flying objects. The current task is to solve the problems related to the automatic recognition of these images and the need to create algorithms to ensure the practical realization and information provision ISARs working on this principle.
In the development of means for more efficient detection of the images obtained by observing different flying objects, it is necessary to perform optimizations at the stage of modeling of the reflected complex trajectory signal. To improve the recognition process and information analysis in this area, simulation experiments were performed to restore high-quality images. The objects to be recognized in these images should use models with a detailed 3D structure that are as close as possible to real-world aircraft.
Based on the results known, obtained in the simulation observation experiments and analytical procedures on a 2D and 3D geometry object, this work proposes to introduce geometry adjustments of the 2D model in order to achieve a higher degree of similarity to the predicted Real reflected signal. The analysis of this approach, based on experimental simulation results and the Z buffer algorithm, gives grounds for defining different shading zones for the radar signal at a different angle of the object relative to the ISAR system.
When implementing a recognition algorithm, the availability of a ready database with a large number of detailed flying object models, their characteristics, their structure features and other indicators is essential for the rapid solution of the particular task. With these recognition system requirements and the availability of many models, it is essential that the image obtained from the ISAR can be classified by the maximum number of benchmarks.
In this article, a neural network is selected in which neuronal learning in the individual layers is performed by a backward-propagation algorithm. This type of neural network has proven metrics for recognizing symbols, printed text, and solid images.
This article offers a neural network architecture for automatic recognition of Inverse Synthetic Aperture Radar objects represented in images with high level of additive noise. A full explanation of the procedures of two-layer neural network architecture creating and training is described. The neural network is depicted in MATLAB and SIMULINK environment.
[a] F. Benedetto, F. Riganti Fulginei, A. Laudani, G. Albanese, Automatic Aircraft Target Recognition by ISAR Image Processing based on Neural Classifier, International Journal of Advanced Computer Science and Applications,Vol. 3, No.8, pp 98-103, 2012.
Boulay T., Lagoutte J., Mohammad-Djafari A., Gac N., A Fuzzy-Logic Based Non Cooperative Target Recognition, Signal Image Technology and Internet Based Systems (SITIS), 2012 Eighth International Conference, Nov. 2012.
Lazarov A.D., Minchev C.N., Correlation-autofocusing-spectral 2-D ISAR image reconstruction from linear frequency modulated signals, Digital Avionics Systems Conference, 2002. Proceedings, pp. 27-31 Oct. 2002.
Lazarov A. D., ISAR Signal Modeling and Image Reconstruction with Entropy Minimization Autofocussing, 25th Digital Avionics Systems Conference, IEEE/AIAA, 2006.
Li, J., Wu R., Victor,C.C. Robust autofocus algorithm for ISAR imaging of moving targets. IEEE Trans., AES, vol.37, No3, July 2001.
Minchev C. N., Slavyanov K.S., An opportunity for improved modeling and information analysis in ISAR systems, “Machines, Technologies, Materials”, ISSN 1313-0226. ISSUE 7/2013, pp. 24-28, 2013.
Tan Y., Yang J., Li L., Xiong J., Data fusion of radar and IFF for aircraft identification, Journal of Systems Engineering and Electronics ( Volume: 23, Issue: 5, Oct. 2012 ), pp. 715 – 722. 2012.