350 rub
Journal Neurocomputers №11 for 2010 г.
Article in number:
Modular neural network classifier for air target recognition
Keywords:
modular neural networks
classification
multilayered perceptron
identification of air targets
trajectory measurements
Authors:
A. V. Bobin, O. A. Mishulina, A. V. Slatin
Abstract:
In this paper, we solve the problem of air target type recognition based on trajectory measurements by radar. We propose a modular neural network classifier which produces a step-by-step solution on the type of tracked target: from a local decision based on current measurement to a final decision based on previous target movement. Neural network classifier provides improvement of the decision on target type during target tracking, simplicity of adjusting its parameters in training mode and interpretability of outputs for separate neural network modules.
Special attention is given to accuracy estimation of the decision on target type not by averaging of available sample data, but by analyzing current conditions of target tracking. We propose a special neural network to estimate accuracy of target type recognition. For this purpose we have elaborated the training procedure based on statistical properties of the classifier in the local neighborhood of the observed target trajectory.
Neural network technique of air target identification is illustrated on the simulated data that are adequate to real radar measurements. Information basis for modeling of neural network classifier is a set of trajectories of three recognizable types of missiles. These trajectories were obtained in different conditions of use and completely correspond to the dynamic properties of these missiles.
The simulation results have shown the principal possibility of air target identification on trajectory measurements without the use of special features of the reflected radar signal. Mathematical simulation has demonstrated high accuracy and stability of decision-making by the neural network classifier. High accuracy of air target identification is obtained not only on simulated data without noise in source measurements, but also on real radar measurements.
Pages: 54-62
References
- Fukunaga, K., Introduction to Statistical Pattern Recognition. SanDiego: AcademicPress. (Русский перевод: Фукунага К. Введение в статистическую теорию распознавания образов. М.: Наука. 1979. С. 368).
- Ту Дж., Гонсалес Р. Принципы распознавания образов. М.: Мир. 1978. С. 412.
- Айзерман М. А., Браверман Э. М., Розоноэр Л. И. Метод потенциальных функций в теории обучения машин. М.: Наука. 1970.
- Потапов А. С. Распознавание образов и машинное восприятие. Изд. Политехника. 2007. С. 552.
- Гитис Л. Х. Кластерный анализ в задачах классификации, оптимизации и прогнозирования. М.: МГГУ. 2001.
- Bishop, Chris. M., Neural Networks for Pattern Recognition. Oxford: University Press. 2005.
- Хайкин С. Нейронные сети: полный курс: Пер. с англ. М.: Вильямс. 2006. С. 1104.
- Осовский С. Нейронные сети для обработки информации. М.: Финансы и статистика. 2004. С. 344.