Study on Classification of Moving Vehicles Based on Micro-Doppler Effect
|School||Xi'an University of Electronic Science and Technology|
|Course||Signal and Information Processing|
|Keywords||Micro-Doppler Radar automatic target recognition (RATR) Wheeled andtracked vehicles Narrow band radar Short dwell time Adaptive clutter suppression Feature extraction Hierarchical classification Robust classification|
Micro-Doppler effect was firstly introduced in coherent laser radar system. It containsinformation about the characteristics of interested moving target, and can be seen as a uniquesignature of the target. The research on micro-Doppler provides a new prospect for radarautomatic target recognition (ATR) and it is also a complementary to existing methods.Wheeled and tracked vehicles often undertake different tasks in battle, which determines theyhave different threat degrees, thus the discrimination of the two types of vehicles is importantto a successful tactic. Since vehicles have typical micro motion structures which will inducemicro-doppler modulation on radar signal, micro-Doppler information contained in theirreturned signals can be utilized to discriminant different moving vehicles. Furthermore, fornarrow band radar system, the research of micro-Doppler effect is beneficial to improve theclassification function. In addition, the research in short dwell time condition could help tosolve the contradiction of time resource distribution between various modern radar functions.This dissertation is supported by Advanced Defense Research Programs of China.Considering the engineering background of the narrow band radar target classification withinshort dwell time, the dissertation studies micro-Doppler signal pre-processing, micro-Dopplersignal analysis and micro-Doppler feature extraction for classification of moving vehicles.The main work can be summarized as follows:1. The principle of micro-Doppler effect and micro motion of rotation are introduced fora single scattering point and a target. Accordingly, the signal models of wheeled and trackedvehicles are established. Based on the signal model, measured data are analyzed todemonstrate distinctions between vehicles. The analysis results show the possibility of vehicleclassification using micro-Doppler signatures.2. Some effective pre-preprocessing techniques for micro-Doppler signals of movingvehicles are studied. Firstly, clutter suppression is analyzed to show the purpose of clutterpre-preprocessing, which is simultaneously clutter suppression and signal preservation forclassification task. A band-limited CLEAN algorithm is introduced to realize the purpose, thismethod has low computation burden and requires only one frame of signal. Another techniqueis based on the assumption that clutter around target is stable. In this case, generalizedmatched filter (GMF) can be employed to suppress clutter. It is adaptive to the clutter due tothe utilization of priori information. Then the normalization of main bulk velocity is discussed.Since the change of bulk velocity usually affects the location of corresponding component andthe width of target spectrum in Doppler domain, which is undesired for classification, a velocity normalization method is proposed to eliminate the influence of the change of bulkvelocity. Finally, a nonlinear transform is introduced to enhance the function ofmicro-Doppler component in classification.3. After analyzing the distinction between wheeled and tracked vehicles, a Dopplerspectrum based classification method is proposed. In this method,5features which isdescribed the figure of Doppler spectrum is extracted, and further classification process isimplemented. The classification results verify the effectiveness of the proposed method.4. Information about energy distribution is used to discriminate wheeled and trackedvehicles. By the analysis of micro-Doppler signals, the returned signal of vehicles can beconsidered as the summation of harmonic components. Accordingly, the harmonic analysisbased method and eigenvalue spectrum based method are proposed. Although use differentapproaches to process micro-Doppler signals, these two methods essentially utilize the similarenergy distribution information to distinguish wheeled and tracked vehicles. The experimentresults show that energy distribution features well depict the distinction between wheeled andtracked vehicles.5. Information associated with target structure is used to discriminate different types ofvehicles. In the returned signals of moving vehicles, the relation between differentmicro-Doppler frequency components reflects the structure information of micro motion parts.A hierarchical classification method based on empirical mode decomposition (EMD) ofmicro-Doppler signatures is proposed for moving vehicles, in which EMD is utilized fordecomposing the more detailed motion components of moving vehicles. Since the velocity ofthe upper track is always twice as large as the bulk velocity, which is determined by thespecial structure of track, this unique feature of tracked vehicle is utilized in the first stage ofour hierarchical structure to elementarily identify the tracked vehicle data. Then, if thefrequency induced by the upper track does not exist from some observation aspect-sectors, themicro-Doppler components are further characterized as the discriminative feature for the twokinds of vehicles in the second classification stage. In addition, clutter suppression can beachieved without extra pre-processing due to the adaptive decomposition characteristic ofEMD. Experiment results verify the effectiveness of the method and show the introduction ofstructure information is beneficial to improve classification performance.6. The robust classification of moving vehicles is discussed. For moving vehicleclassification within short dwell time, how to handle ground clutter and receiver white noiseto obtain robust classification performance, and how to extract discriminative informationfrom micro-Doppler signatures as much as possible should be considered. According to theprinciple of compressive sensing (CS), classification method based on sparse representation is proposed. The method can not only adaptively remove ground clutter while preserve originalsignal as much as possible but also improve the classification performance especially underlow SNR conditions.