The Study And Application Of Gpr Data Processing Based On CEEMD And PCA

As a kind of high frequency electromagnetic method, ground penetrating radar(GPR) has been widely used in shallow engineering detection, detection of groundwater, earthquake disaster relief, and other fields. The working principle of GPR is transmitting high-frequency electromagnetic waves to the underground and receiving the reflected waves so as to realize the detection of the target. Its primary working object is underground complex lossy medium. In the process of transmission, the GPR signal is always disturbed by the surrounding environment and leads to scatter and dispersion phenomenon, which affect the target resolution and accuracy of GPR detection. Moreover, the measured GPR data is a kind of nonlinear and non-stationary signal. However, traditional signal analysis is mainly used to process steady data. Therefore, we often cannot get a satisfactory GPR result by traditional signal processing method. Complete ensemble empirical mode decomposition(CEEMD) is a kind of adaptive signal analysis method to analyze nonlinear and non-stationary signal, which can decompose the signal into a number of different intrinsic mode functions(IMFs), and each IMF component contains different frequency components of the original signal. Through Hilbert transform to each IMF, we can get the instantaneous parameters with physical meanings. These instantaneous parameters can be used to make frequency division processing for the complex GPR signal, by selecting the fixed frequency band of the IMF to remove the influence of high frequency noise and low frequency evanescent wave. The target signal and background medium of GPR data has some differences in signal energy and phase, through the application of principal component analysis(PCA), we can effectively isolate the target signal from the GPR signal and improve the resolution of the detection results. This article elaborates the basic principle of CEEMD and PCA method in detail, by combining the Hilbert- Huang transform, these two methods are applied to the GPR data processing.First of all, this paper introduces the basic principle and working process of the CEEMD method, compares and analyzes the differences on basic principle and implementation results with the conventional EMD and EEMD methods. Although EEMD overcomes the problem of mode mixing to a certain extent, this is an incomplete decomposition method and can’t reconstruct the original signal perfectly. CEEMD keeps the completeness of EMD and is able to achieve the separation of frequency spectrum better. Here, we use EMD, EEMD and CEEMD methods to make model test respectively to verify the superiority of CEEMD. It can be proved that the use of CEEMD is under the minimum influence of mode mixing, and reconstructs the original signal perfectly.Hilbert-Huang transform(HHT) is a time-frequency analysis method to deal with nonlinear and non-stationary signal, it can analyze signal from two aspects of time and frequency at the same time, so that the signal processing becomes more flexible. In this paper, the processing effect of HHT is shown through a simple synthetic signal, and verifies that the HHT is effective and intuitionistic to analyze signal and can provide the spectral images with higher time-frequency resolution and higher accuracy. Through synthetic signal and measured GPR data, we can demonstrate that the instantaneous spectrum based on CEEMD has higher time-frequency resolution and can reflect the different frequency components of the original signal in greater detail. In the application of the measured data, the CEEMD method is able to remove the noise and interference in the signal effectively and retain the effective signal.On the basis of the above methods, we use PCA method to extract target information. According to the correlation of signal, we get the eigenvalues and corresponding eigenvectors by decomposing the covariance matrix of GPR data, and make linear transformation for the GPR data to get the principal components(PCs). The lower-order PCs stand for the strong correlated target signals of the raw data, and the higher-order ones present the uncorrelated noise. Thus, we can extract the target signal and filter uncorrelated noise effectively by PCA. Here, this article applies PCA method on a through-wall radar data to extract the human body target signal, and we get the accurate vital sign information of human target, the result verifies the reliability of this method for GPR data processing and object identification.