Research On The Person And Vehicle Targets Identification Method Of Seismic Signal

With the development of the military technology, the technique of information reconnaissance in our country is developing in directions of three-dimensional, accurate and comprehensive. In the reconnaissance of the person and vehicle targets, vibration sensor acts as kind of commonly used ground sensor. It detects targets by recording seismic waves,and it has been shown that the moving person and the vehicle which are within 20m and 200m can usually be detected. The seismic signal by the person and vehicle targets is a typical non-stationary random signal, which is very important to analyze and process the signal. In recent years, empirical mode decomposition (EMD) has become the focus of the method for processing non-stationary random signal processing, and has been widely applied to seismic signal.Firstly, this paper introduces the seismic signal of person and the vehicle targets about denoising methods and identification methods, both in China and abroad. And the occurrence and propagation mechanism of seismic waves are studied. Describing the signal acquisition device of the vibration sensor and collecting the seismic signal of the person, the wheeled vehicle and the tracked vehicle, the signal is obtained, and the sample library is established.According to the characteristics of seismic signal, several classical time-frequency analysis methods are studied in detail, and the comparison is made by experiments. The analysis shows that the short time fourier transform (STFT) and the wigner-ville distribution (WVD)have the disadvantages on seismic signal.Then, because of the disadvantages of STFT and WVD, this paper uses ensemble empirical mode decomposition (EEMD) to solve that. Due to the influence of background noise and other external conditions, seismic signal by which person and vehicle produced is often subject to interference and difficult to identification. So the decomposition of intrinsic mode functions (IMF) high frequency components is used wavelet threshold algorithm for denoising. Through MATLAB simulation experiments, EEMD-wavelet threshold denoising method proposed in this paper is compared with other denoising methods. It verifies the method can effectively improve the signal to noise ratio, and simultaneously solve the problem of the loss of the effective information.Finally, for the complex and changeable environment, it is difficult to establish a large database, so the EEMD-support vector machine unbalance decision tree model is established:selecting the valid components of IMF and calculating normalized energy matrix as the input of SVM unbalance decision tree classifier. Person,wheeled vehicle and tracked vehicle are classified layer by layer through the SVM. At the same time, through the experiments prove that the EEMD-SVM unbalanced decision tree model is accurate and efficient for identification of person, wheeled vehicle and tracked vehicle. In addition, the MATLAB GUI interface is established,to simulate a system based on EEMD-SVM unbalance unbalanced decision tree model. A semi automatic recognition way of person and vehicle targets is realized.