복소 독립 성분 분석 기법을 이용한 헬리콥터 특징 추출 연구

Abstract

because Various helicopters that are used by belligerent nations are major threats to the ground operation. It is very important to avoid the damage in advance and recognize the exact model of the helicopter by using all-weather devices such as the radar. Because unique blade length, the number of blades and rotation velocity are decided according to the purpose of aggression, weight of helicopter, etc., major parameters related to helicopter blades may become very important components to recognize helicopters. Micro-Doppler analysis is usually used to extract features of blades. Micro-Doppler means Doppler frequency by own motion of targets except translation motion. Features of blade length, rotation velocity and other parameters can be estimated by applying feature estimation technique based on bistatic radar to Doppler feature analysis. Because Micro-Doppler by blades is acquired in the form of flashes, the period cannot be accurately estimated when the exact number of blades is not known a priori. In addition, the initial position of blades should be known to perform the feature estimation based on the bistatic geometry. In this paper, we propose an efficient technique suitable for micro-Doppler signal based on real scattering mechanisms and feature estimation based on the bistatic geometry. In the proposed method, signals of baldes are separated by complex independent component analysis (ICA) and then matching between the time-frequency image of a blade and the 2-D window image is applied to analyze Micro-Doppler features. Finally, micro-Doppler estimation technique based on the bistatic geometry is applied. Applying a unified algorithm to targets with various blade numbers, the proposed method separates blade signals completely and estimates the number of blades. Using the separated blade signals, Doppler frequency was estimated successfully with 1.342 % error by using the window matching method. In addition, using the feature estimation method based on the bistatic geometry, the blade length and the rotation velocity were estimated with 0.9892 % and 0.015 %errors, respectively.