Research On Noise Reduction Of Transient Electromagnetic Signal Based On Dictionary Learning And Noise Classification

The transient electromagnetic method is an active measurement method based on the principle of electromagnetic induction.It uses transient electromagnetic signals to analyze the difference in resistivity between geoelectric body,and then achieves the purpose of detecting them.However,the transient electromagnetic signals are easily disturbed by various types of noise.A transient electromagnetic signal can be divided into early,middle,and late fields.In the late field of a signal,noise even drowns out the effective transient electromagnetic signal,which results in limited signal utilization and affects the detection of deep targets.The noise reduction of the transient electromagnetic signal refers to the use of various methods to suppress noise as much as possible and extract the effective transient electromagnetic signal.The current noise reduction methods for the transient electromagnetic signal are mostly only effective for specific noises,which is not enough for situations where multiple noises are likely to exist simultaneously in a real environment.In recent years,dictionary learning and noise classification techniques have received high attention in the field of transient electromagnetic signal noise reduction.Dictionary learning can extract the characteristics of noise and the pure transient electromagnetic signal respectively,and use their characteristics to recover the effective transient electromagnetic signal from the noisy transient electromagnetic signal.The noise classification technology can identify the types of noise in the noisy transient electromagnetic signal,and assist the use of dictionary learning noise reduction methods in a real environment.The supervised and semi-supervised noise reduction methods of non-negative matrix factorization(NMF)are taken as the main line to introduce the main work of this paper.First,a noise classifier based on convolutional neural networks is proposed.The amplitude spectrum and phase spectrum obtained by performing short-time Fourier transform on the four kinds of noises to be classified are input as features to the built 15-layer classifier for classification.Experimental results show that the classification effect of this classifier is good.Secondly,a supervised algorithm based on NMF was applied to the transient electromagnetic signal to reduce the noise.In the training phase,NMF is performed on the noise and the pure transient electromagnetic signal respectively to obtain the atomic dictionary that characterizes their respective characteristics.Furthermore,the noise reduction model is used to reduce the noise of the noisy transient electromagnetic signal.In the real transient electromagnetic signal denoising process,the trained classifier is used to classify the late field of the real transient electromagnetic signal,determine the type of real noise,and then use the computer to simulate the corresponding noise so that the noise reduction method can be used in a real environment.Experimental results verify the effectiveness of the method.Finally,a semi-supervised noise reduction method with NMF is introduced.In the training phase,only the pure transient electromagnetic signal is subjected to NMF to obtain an atomic dictionary characterizing its characteristics.Similarly,the corresponding noise reduction model is used to reduce the noise of the noisy transient electromagnetic signal.Experimental results verify the effectiveness of the method.