Application of neural network technologies for recognizing distributed objects in radar images

Аuthors

Tonkikh A. N.

NaukaSoft Scientific and Production Association, 129085, Moscow, Godovikova street, 9, p. 4

e-mail: alex_tonkih@mail.ru

Abstract

The paper considers approaches to the construction of algorithms for solving the problem of detection and recognition of distributed artificial objects in radar images. A brief review of existing architectures of convolutional neural networks is carried out, their main advantages and applied solutions in the interests of improving performance are shown. For information support of training of neural network algorithms the necessity of application of mathematical modeling of radar images of distributed artificial objects is justified. In the course of the work the architecture of convolutional neural network for joint detection and recognition of objects in radar images is proposed. The analysis of the obtained results showed that the application of convolutional neural networks allows solving the problem of joint detection and recognition of objects in radar images. Such a solution is realized in real time. In this case, the main hardware and time costs are required at the stage of training of convolutional neural network. Comparability of the results of correct detection and recognition of the neural network trained on model radar images with the results of Faster R-CNN and VGG-16 allows us to speak about the suitability of mathematical modeling to inform the training process. The work can be continued by developing a convolutional neural network based on the Mask R-CNN model, where the problem of segmenting the possible region of the object location is solved. This approach will allow to use the shape of the object as an additional classification feature, thereby increasing the accuracy of identification. In addition, it is advisable to consider the inclusion in the architecture of the convolutional neural network of an additional branch that estimates the shadow of the object, which is in some cases more informative than the image of the object itself.

Keywords:

radar, radar image, mathematical modeling, convolutional neural network, detection, recognition

References

1. Shkol'nyi L.A., Tolstov E.F., Detkov A.N., Karpov O.A. Radiolokatsionnye sistemy vozdushnoi razvedki, deshifrirovanie radiolokatsionnykh izobrazhenii (Radar systems of air reconnaissance, interpretation of radar images). Moscow: VVIA im. prof. N.E. Zhukovskogo Publ., 2008. 531 p.
2. Kondratenkov G.S., Potekhin V.A., Reutov A.P., Feoktistov Yu.A. Radiolokatsionnye stantsii obzora Zemli (Earth observation radar stations). Moscow: Radio i svyaz' Publ., 1983. 272 p.
3. Shkolny L.A., Tonkikh A.N., Yasko S.I. Algorithm of automatic detection and recognition of noncompact group objects by radar data. Radiotekhnika. 2008. No. 6. P. 32. (in Russ.)
4. Vlasov S.O. et al. Solving the problem of object detection using neural network technologies. Preprinty IPM im. M.V. Keldysha. 2023. No. 16. 27 p. (In Russ.)
5. Dzyuba V.A. Investigation of the algorithms and methods of detection and recognition on the basis of the neural network systems. Sovremennye nauchnye issledovaniya i innovatsii. 2021. No. 12. (In Russ.). URL: https://web.snauka.ru/issues/2021/12/97280
6. Yartsev I.M., Prokhorov A.G., Dmitriev I.N. Tekhnologiya sozdaniya sistemy neirosetevogo avtomaticheskogo raspoznavaniya odinochnykh nazemnykh i nadvodnykh ob''ektov vooruzhenii, voennoi i spetsial'noi tekhniki na osnove sistemy iskhodnykh dannykh, poluchennykh metodom matematicheskogo modelirovaniya (Technology of creating a system of neural network automatic recognition of single ground and surface objects of armaments, military and special equipment on the basis of the system of initial data obtained by mathematical modeling). Moscow: Yuniti-Dana Publ., 2023. P. 299-315.
7. Savchenko B.S., Sobolev A.V. Technology of detection and classification of objects of artificial origin on highly detailed images acquired by space radars with synthetic antenna aperture. Materialy XXXIII Vserossiiskogo simpoziuma «Radiolokatsionnoe issledovanie prirodnykh sred» (Sankt-Peterburg, 19–20 aprelya 2023): sbornik trudov. Saint-Petersburg: Voenno-kosmicheskaya akademiya imeni A.F. Mozhaiskogo Publ., 2024. P. 146-151.
8. Nenashev V.A. Nenashev S.A. Classification and recognition of ground objects in the stream of radar frames on the basis of the neural network approach in the front viewing area of the airborne radars of the multi-position system. Trudy Mezhdunarodnoi konferentsii po komp'yuternoi grafike i zreniyu "Grafikon". 2023. No. 33. P. 572-580. (In Russ.). DOI: 10.20948/graphicon-2023-572-580
9. Moving and Stationary Target Acquisition Recognition (MSTAR). Program Review. Denver. CO. Nov., 1996.
10. Novak L.M., Owirka G.J., Brower W.S., Weaver A.I. The Automatic Target-Recognition System in SAIP. Lincoln Laboratory Journal. 1997. Vol. 10, No. 2. P. 187-201.
11. Novak L.M., Owirka G.J., Brower W.S. An Efficient Multi-Target SAR ATR Algorithm. Asilomar Conference. Pacific Grove. CA. November 1998.
12. Owirka G.J., Verbout S.M., Novak L.M. Template-based SAR ATR performance using different image echancement techniques. In Algorithms for Synthetic Aperture Radar Imagery VI. Proc. of SPIE. 1999. Vol. 3721, P. 302-319.
13. Gavrilov K.YU., Kamenskii K.V., Malyutina O.A. Trajectory signal modelling in the aperture synthesis radar based on optical images of the Earth surface. Trudy MAI. 2021. No. 118. (In Russ.). URL: https://trudymai.ru/eng/published.php?ID=158252. DOI: 10.34759/trd-2021-118-12
14. Kuznetsova V.A., Tonkikh A.N. Software complex of radar data modeling. Collection of scientific articles. VII Mezhdunarodnaya nauchno-prakticheckaya konferentsiya «Aktual'nye problemy i perspektivy razvitiya radiotekhnicheskikh i infokommunikatsionnykh sistem» «Radioinfokom-2023»: sbornik trudov. Moscow: MIREA Publ., 2023. P. 43-46.
15. Simonyan K., Zisserman A. Very Deep Convolutional Networks for Large-Scale Image Recognition. In: Bengio Y., LeCun Y. (eds.) // Proceedings of the 3rd International Conference on Learning Representations (ICLR-2015). San Diego, CA, USA; 2015. DOI: https://doi.org/10.48550/arxiv.1409.1556
16. Akshay Kapoor, Rishi Shah, Rushi Bhuva, Tejas Pandit. Understanding inception network architecture for image classification. University of Waterloo, February 2020. DOI: 10.13140/RG.2.2.16212.35204
17. He K., Zhang X., Ren S., Sun J. Deep Residual Learning for Image Recognition. Computer Vision and Pattern Recognition, Dec 2015. DOI: 10.48550/arXiv.1512.03385
18. Brock A., De S., Smith S., Simonyan K. High-Performance Large-Scale Image Recognition Without Normalization. Feb 2021. DOI: 10.48550/arXiv.2102.06171
19. Joseph Redmon, Ali Farhadi. YOLO9000: Better, Faster. University of Washington, Institute for AI/ 2016. DOI: 10.48550/arXiv.1612.08242
20. Girshick R. Fast R-CNN. Computer Vision and Pattern Recognition, Sep 2015. DOI: 10.48550/arXiv.1504.08083
21. Babokin M.I., Efimov A.V., Zaitsev S.E., Karpov O.A. et al. Kondor-E spacecraft with RSA and its capabilities. Issledovanie Zemli iz kosmosa. 2017. No. 3. P. 85-95. (In Russ.). DOI: 10.7868/S0205961417030010
22. Gusev S.N., Sakhno I.V., Khubbiev R.V. Evaluation technique for virtual objects on radar images formation quality. Trudy MAI. 2019. No. 104. (In Russ.). URL: https://trudymai.ru/eng/published.php?ID=102169
23. Kamenskii K.V. Trajectory instabilities and onboard navigation system characteristics influence on synthetic aperture radar image quality. Trudy MAI. 2022. No. 125. (In Russ.). URL: https://trudymai.ru/eng/published.php?ID=168186. DOI: 10.34759/trd-2022-125-14
24. Sentsov A.A., Nenashev V.A., Ivanov S.A., Turnetskaya E.L. Combining the generated radar images with a digital map of the area in on-board systems for operational monitoring of the earth's surface. Trudy MAI. 2021. No. 117. (In Russ.). URL: https://trudymai.ru/eng/published.php?ID=156227. DOI: 10.34759/trd-2021-117-08

Download