Research Of Airborne Transient Electromagnetic Denoising Methods Based On Subspace Analysis

Airborne transient electromagnetic method (ATEM) is a kind of airbornegeophysical exploration method which is based on the helicopter carrier and theelectromagnetic induction principle. Because it is of larger exploration depth andwider detection range and can overcome complex terrain, airborne transientelectromagnetic method has been widely applied in geological mapping, mineralexploration, environmental monitoring and other fields. However, due to the widebandwidth of ATEM signal and the weak effective energy of the later-channel data,the measured data is often contaminated by random noise, sferics noise andman-made noise, etc. If these noises can not be eliminated effectively, the quality ofthe data will be reduced which may impact the accuracy of the latter inversioninterpretation. At present, although there are a variety of ATEM denoising methods,there are no methods can suppress all kinds of noises simultaneously. Subspaceanalysis method, one of the statistical methods, which has good performance fordifferent noises, can enhance the signal to noise ratio of ATEM data. Therefore, twotypical subspace analysis methods are mainly studied in this paper: principalcomponent analysis (PCA) and independent component analysis (ICA), which areapplied to the ATEM denoising. Specifically, the main research contents were shownin the following:（1）The denoising method based on principal component analysis: Principalcomponent analysis works with the statistical properties at first or second order. First,make the mean of data be zero and get its covariance matrix; Then decompose thecovariance matrix and calculate the projection of the raw data in the subspace spannedby the feature vector to get all main components of the data; at last, select the maincomponents on behalf of the useful signal to reconstruct the de-noised data.Throughout the algorithm implementation process, the most effective technique is tochoose the effective signal components for reconstruction accurately. This paperadopts curve trend comparison method and signal-to-noise separation criterion whichbased on L-curve method to extract valid signal components. On the one hand, we accurately separate the signal into geological components and noise components. Onthe other hand, the two methods used together can play a mutual verification role.（2）The denoising method based on the independent component: PCA methodonly uses second-order statistics without taking into account the higher-orderstatistical properties. So it cannot describe the details of data. However, theindependent component analysis can analyze the higher-order statistical properties,which can obtain more detailed information of the data. This paper adopts the fastfixed-point algorithm (FastICA algorithm) which based on the maximizenon-Gaussian to estimate the independent component (IC). The ambiguity of ICAmodel itself will lead to a serious distortion of the effective signal amplitude, so werepair the amplitude after the ICA denoising in the paper. The restoration approachesfirstly calculates the inversion of the separation matrix to achieve an approximation ofthe original mixed matrix. Then the independent components which are estimated byFastICA are multiplied by the corresponding coefficients from the inverse matrix toget the denoised data. The amended ICA denoising method can not only suppress theprimary noises, but also maintain the amplitude effectively.（3）Through the forward modeling and experimental data, we analyze andevaluate the denoising performance of the principal component analysis andindependent component analysis. For ATEM data, the principal component analysishas better denoising effectiveness than independent component analysis through thecomparison of experiments.