CN112713907B

CN112713907B - Marine CSEM noise reduction method and system based on dictionary learning

Info

Publication number: CN112713907B
Application number: CN202011537201.7A
Authority: CN
Inventors: 张鹏飞; 潘新朋; 邓明; 郭振威; 赵志者
Original assignee: Central South University; China University of Geosciences Beijing
Current assignee: Central South University; China University of Geosciences Beijing
Priority date: 2020-12-23
Filing date: 2020-12-23
Publication date: 2022-03-15
Anticipated expiration: 2040-12-23
Also published as: CN112713907A

Abstract

The present disclosure discloses a method and system for denoising marine CSEM based on dictionary learning, including: acquiring marine CSEM signals; determining the minimum transceiver distance position of the marine CSEM signal; selecting a set distance from the minimum transceiver distance position from the marine CSEM signals According to the obtained learning dictionary, the marine CSEM signal is sparsely reconstructed by the orthogonal matching pursuit algorithm, and the denoised marine CSEM signal is obtained. The learning dictionary is obtained by selecting the signal within the set distance from the minimum sending and receiving distance position, and the correlation between the dictionary and the signal is improved. noise, improving the noise reduction effect of the marine CSEM signal.

Description

Marine CSEM noise reduction method and system based on dictionary learning

Technical Field

The invention relates to the technical field of signal processing of marine controllable source electromagnetic methods, in particular to a marine CSEM noise reduction method and system based on dictionary learning.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

A Controlled Source Electromagnetic Method (CSEM) is a frequency domain active source electromagnetic method for distinguishing a target layer from a surrounding rock according to the difference of resistivity, and is widely applied to the exploration of seabed oil gas and natural gas hydrate. The working principle of the marine CSEM is shown in figure 1, receivers are distributed on the seabed according to a certain rule during operation, and transmitters advance along a certain track within the range of 25-100m away from the seabed under the towing of an operation ship and transmit electromagnetic signals.

The ocean controlled source electromagnetic method is very sensitive to a high-resistance layer in a high-conductivity background, and a target layer and surrounding rocks can be effectively distinguished by means of resistivity difference. The signal amplitude decays approximately exponentially as the transmit-receive distance increases. The signal is weak at the middle and long receiving and transmitting distances, and is easily influenced by various types of noise. Existing research has shown that noise can be classified into the following categories depending on the source of the noise:

(1) dipole oscillation noise. Under the action of ocean currents and other factors on the seabed, the receiving electrode arm generates irregular oscillation, and the noise intensity generated by the tiny oscillation is equivalent to the signal intensity of a target layer.

(2) Sea water movement noise. The waves on the sea surface and the ocean currents in the sea water cut the geomagnetic field magnetic induction lines to form an induction electromagnetic field in the motion process. The frequency is closely related to the wave or ocean current and is generally low.

(3) Natural electromagnetic field noise. Variations in the natural electromagnetic field can have an effect on signals in the receiver, such as sub-sea magnetotelluric signals.

(4) Random noise. Such noise includes both internal instrument noise and noise caused by external random interference.

(5) Air waves. Noise generated by an air-seawater interface in the process of electromagnetic wave propagation causes great interference to effective signals in shallow sea areas. Although there is a debate as to whether the airwave is noisy, it is still partially used as noise cancellation by some scholars.

At present, data noise reduction is focused on the aspects of traditional filtering, time domain filtering, transform domain noise reduction and the like, and noise of related research is set as random noise or Gaussian white noise. When the ocean controllable source electromagnetic method works, the method is influenced by various factors, and the current noise reduction method only limits the noise to be white Gaussian noise or has certain limitation only aiming at single type of noise.

Disclosure of Invention

The method and the system for reducing the noise of the marine CSEM signal can effectively reduce the noise in the marine CSEM signal and improve the noise reduction effect of the marine CSEM signal when the learning dictionary is used for reducing the noise of the collected marine CSEM signal.

In order to achieve the purpose, the following technical scheme is adopted in the disclosure:

in a first aspect, a marine CSEM noise reduction method based on dictionary learning is provided, including:

acquiring marine CSEM signals;

determining the minimum transceiving distance position of the marine CSEM signal;

selecting signals within a set distance from a minimum transceiving distance position from marine CSEM signals, and learning as samples to obtain a learning dictionary;

and carrying out sparse reconstruction on the marine CSEM signal through an orthogonal matching pursuit algorithm according to the obtained learning dictionary to obtain the noise-reduced marine CSEM signal.

In a second aspect, a marine CSEM noise reduction system based on dictionary learning is provided, including:

the signal acquisition module is used for acquiring marine CSEM signals;

the minimum transceiving distance position determining module is used for determining the minimum transceiving distance position of the marine CSEM signal;

the learning dictionary acquisition module is used for selecting a signal within a set distance from the position with the minimum transceiving distance from the marine CSEM signal, and learning the signal as a sample to obtain a learning dictionary;

and the noise-reduced signal acquisition module is used for carrying out sparse reconstruction on the marine CSEM signal through an orthogonal matching pursuit algorithm according to the acquired learning dictionary to obtain the noise-reduced marine CSEM signal.

In a third aspect, an electronic device is provided, which includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, where the computer instructions, when executed by the processor, perform the steps of a marine CSEM noise reduction method based on dictionary learning.

In a fourth aspect, a computer-readable storage medium is provided for storing computer instructions which, when executed by a processor, perform the steps of a method for marine CSEM noise reduction based on dictionary learning.

Compared with the prior art, the beneficial effect of this disclosure is:

1. according to the method, the learning dictionary is obtained by learning through selecting the signals within the set distance from the minimum transceiving distance position, the correlation between the marine CSEM signals and the learning dictionary is improved, so that when the marine CSEM signals are subjected to sparse coding and noise reduction reconstruction through the learning dictionary, a better sparse expression effect can be obtained, the noise in the marine CSEM signals can be effectively reduced, the noise reduction effect of the marine CSEM signals is improved, and the influence of noise types is avoided.

Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.

FIG. 1 is a diagram of the working principle of marine CSEM;

fig. 2 is a flow chart of the learning of the marine CSEM noise reduction dictionary disclosed in embodiment 1 of the present disclosure;

FIG. 3 is a graph of the model and response constructed in example 1 of the present disclosure;

FIG. 4 is a graph of the noise and noisy marine CSEM signal referred to in example 1 of the present disclosure;

fig. 5 is a graph of reconstruction errors under different dictionaries obtained by experiments in embodiment 1 of the present disclosure;

fig. 6 is a graph of noise reduction results obtained experimentally in embodiment 1 of the present disclosure under different dictionaries;

fig. 7 is a diagram of a R2 site five-component time series referred to in example 1 of the present disclosure;

fig. 8 is a graph of the actual signal characteristics and convergence obtained in embodiment 1 of the present disclosure;

FIG. 9 is a diagram of 0.5Hz MVT from three dictionaries obtained in example 1 of the present disclosure;

FIG. 10 is a diagram of 1.5Hz MVT under three dictionaries obtained in example 1 of the present disclosure.

The specific implementation mode is as follows:

the present disclosure is further described with reference to the following drawings and examples.

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

In the present disclosure, terms such as "upper", "lower", "left", "right", "front", "rear", "vertical", "horizontal", "side", "bottom", and the like indicate orientations or positional relationships based on those shown in the drawings, and are only relational terms determined for convenience in describing structural relationships of the parts or elements of the present disclosure, and do not refer to any parts or elements of the present disclosure, and are not to be construed as limiting the present disclosure.

In the present disclosure, terms such as "fixedly connected", "connected", and the like are to be understood in a broad sense, and mean either a fixed connection or an integrally connected or detachable connection; may be directly connected or indirectly connected through an intermediate. The specific meanings of the above terms in the present disclosure can be determined on a case-by-case basis by persons skilled in the relevant art or technicians, and are not to be construed as limitations of the present disclosure.

Example 1

A marine Controlled Source Electromagnetic Method (CSEM) is a frequency domain electromagnetic exploration method for detecting the condition of a seabed stratum according to the difference of stratum resistivity, and is widely applied to seabed oil and gas evaluation and gas hydrate exploration. The amplitude of marine CSEM signals is approximately exponentially attenuated along with the increase of the receiving and transmitting distance, the signals are weak at the middle and far receiving and transmitting distances and are easily influenced by noise, and the noise reduction is an important ring in marine CSEM application. The embodiment provides a marine CSEM noise reduction method based on dictionary learning in a compressed sensing framework from signal characteristics, wherein in the compressed sensing framework, the noise reduction effect of a signal is directly related to a selected dictionary, the stronger the correlation between the selected dictionary and the signal is, the better the noise reduction effect of the signal is, and a learning dictionary obtained by learning a sample set through a dictionary learning algorithm generally has better correlation. Firstly, selecting signals or pure signal segments within a set distance from a minimum transceiving distance position to form a sample set. And then, learning the sample set by adopting a dictionary learning K-SVD algorithm to obtain a learning dictionary. And finally, carrying out sparse expression and noise reduction reconstruction on the marine CSEM signal by an Orthogonal Matching Pursuit (OMP) algorithm according to the obtained learning dictionary. And performing a numerical simulation experiment, and verifying the effectiveness of the marine CSEM noise reduction method based on dictionary learning by comparing with other traditional noise reduction methods. And finally, the noise reduction method is applied to field data of the Qiongnan basin, and the result shows that the method can effectively reduce the influence of noise in signals.

In the embodiment, a dictionary learning algorithm is adopted to further promote the noise reduction research of the marine controllable source electromagnetic data under the compressed sensing framework. Theoretically, a learning dictionary obtained by learning the signal features has better correlation compared with other given dictionaries, and a better noise reduction effect can be obtained.

A marine CSEM noise reduction method based on dictionary learning comprises the following steps:

acquiring marine CSEM signals;

Further, the marine CSEM signals form an MVT curve, and the position with the minimum signal transmitting-receiving distance is determined through the MVT curve.

Further, according to the moving speed of the mobile transmitting station and the set distance from the minimum transmitting-receiving distance position, the intercepting time of the signals is calculated, and the signals are intercepted from the marine CSEM signals according to the intercepting time of the signals, so that dictionary learning is carried out.

Furthermore, azimuth correction is carried out on the collected marine CSEM signals, and an MVT curve is formed through the signals after the azimuth correction.

And further, learning the sample by a K-SVD algorithm.

Further, in the dictionary learning, the minimum error is taken as a standard, and the minimum error decomposition term is selected as an updating term and a corresponding sparse coefficient of the dictionary atom through the singular value decomposition of the error matrix.

Furthermore, carrying out sparse representation on the marine CSEM signal, carrying out sparse coding on the marine CSEM signal through an orthogonal matching tracking algorithm according to the obtained learning dictionary, and reconstructing the signal after the sparse coding to obtain the noise-reduced marine CSEM signal.

A marine CSEM noise reduction method based on dictionary learning is disclosed, the process is shown in figure 2, the key point of applying a dictionary learning algorithm to the marine CSEM noise reduction process is selection and construction of a sample set, and signals with fixed length are captured as sample data in a numerical simulation experiment. In marine CSEM operation, a mode of fixing a receiving station and moving a transmitting station is mostly adopted, and the amplitude of marine CSEM signals is approximately exponentially attenuated along with the increase of the receiving and transmitting distance. In the stage of small transmitting-receiving distance, the amplitude of the marine CSEM effective signal is large and is less influenced by noise, so that the signal within the set distance from the position with the minimum transmitting-receiving distance is selected as a training sample to carry out dictionary learning.

And substituting the constructed training sample into a K-SVD dictionary learning process, and continuously updating dictionary atoms through sparse decomposition and iterative computation when the initial dictionary is given to the DST dictionary. And carrying out sparse decomposition and reconstruction on the target signal under a learning dictionary, setting reconstruction precision and maximum iteration times, and finishing the noise reduction process. Wherein the iteration precision or the maximum iteration number is determined by numerical simulation experiments. The target signal is a captured marine CSEM signal.

The learning process of the K-SVD dictionary is as follows:

sparse coding

Assuming finite length one-dimensional discrete signal

Can be regarded as R^NIn the space, the column vector with dimension Nx1 has elements x (N), and N is 1,2, … N. From mathematical knowledge, R^NAny signal in space can be represented by Nx 1-dimensional column vector

Such a set of uncorrelated vectors constitutes R^NA base under space. If the base vectors are orthonormal, then an orthogonal base is formed by taking the base vector as a column vector, and the matrix is represented by NxN, any signal in the space can be represented as

Wherein

Is the projection of the signal under the basis function matrix.

And

is the expression of the same signal under different basis functions, if at

Only K elements are needed to approximate the signal well, then the signal is represented

At the basis function

The following is K sparse. At this time

Where epsilon represents the error. At this point, the solution

Problem translation for sparse representation

In the formula | · | non-conducting phosphor₀To solve for the zero norm operation, the number of non-zero sparseness in the vector is represented. The formula (3) is an NP-hard problem, which is often converted into a solution for easy resolution

Wherein | · | purple₂To solve the 2-norm operation, the positive square root of the sum of the absolute values of the vector elements is represented. Equation (4) can be solved by a matching pursuit algorithm. In this embodiment, an orthogonal matching pursuit algorithm is used for solving, and the basic flow is as follows:

(1) initialization r₀＝y,Λ₀＝φ,t＝1；

(2) Finding an index λ_tSo that

(3) Let Λ_t＝Λ_t-1∪{λ_t}；

(4) Calculating [ phi ]_λ:λ∈Λ_tY-shaped space orthogonal projection P_t；

(5) Computing a new approximation x_tAnd residual r_t:x_t＝P_ty,r_t＝y-x_t

(6) t is t +1, if t < K, return to (2)

(7) Obtaining an estimate

At index Λ_KA non-zero element of a location, and the measurement vector at that location is approximated as:

the marine CSEM emission signal generally adopts square waves or combined bipolar square waves, the emission signal is sparse in a frequency domain, and a noise spectrum is disordered and generally considered to be incapable of being sparsely represented. The noise reduction problem of the marine CSEM data is converted into a sparse expression problem of noise-containing signals under a specific dictionary. The method is characterized in that a DST dictionary and a DST-Wavelet cascade dictionary are adopted to carry out noise reduction research on marine controllable source electromagnetic data, and a certain noise reduction effect is achieved. However, a given dictionary can only provide sparse representation of signals to a certain extent, and a good sparse representation method should be more flexible, simpler and adaptive, and can adaptively select a proper basis function according to the characteristics of the signals and complete signal decomposition to obtain a concise representation result. Therefore, the dictionary learning algorithm is introduced into the embodiment, and the self-adaptive sparse domain matrix is constructed for a certain signal characteristic to complete the sparse representation of the signal.

(II) dictionary learning

The purpose of dictionary learning is to obtain an adaptive dictionary aiming at signal characteristics in a machine learning mode, so that the dictionary is more consistent with the inherent characteristics of signals, and simultaneously, observation signals of the same kind can obtain optimal sparse expression under the dictionary.

The most commonly used sparse representation algorithm at present is a K-SVD algorithm, which is firstly proposed in 2006, is derived from K clustering analysis in a machine learning algorithm, takes the minimum error as a standard, and selects a minimum error decomposition item as an updating item of a dictionary atom and a corresponding sparse coefficient by Singular Value Decomposition (SVD) of an error matrix. The K-SVD algorithm avoids the inversion operation of the matrix, so the operation complexity is low, the convergence speed is high, and the K-SVD algorithm becomes the most common dictionary learning algorithm at present. The K-SVD objective function is:

wherein

In order to train the resulting dictionary,

in the form of a matrix of the original signals,

for sparse representation matrices, T₀To give the target sparsity, | ·| non-woven phosphor_FRepresenting the Frobenius norm. Hypothetical dictionary

And fixing, then converting the problem into a sparse expression problem for solving signals under a fixed dictionary, and then converting the punishment item into:

equation (5) can be transformed into N problems that can be solved independently by the OMP algorithm.

Assuming a matrix of coefficients

And dictionary

All are fixed, take the kth atom in the dictionary and take it as

The k column of the coefficient matrix associated therewith is labeled

Where T represents transpose, then the penalty term can be expressed as:

wherein

The terms represent the error matrix except for the k-th term. If the singular value decomposition algorithm is directly used for the error matrix, the k-th item coefficient is not sparse any more.

Definition of ω_kUsing elements as a set of directed samples

Index of (1), then

Definition of

Is Nx | omega_kMatrix of size | where the coordinate is (ω)_k(i) I) the value of the element is 1, and the values of the other elements are 0, then

Compressing the original coefficient matrix in the same way

Operation will error matrix

Compression is performed. The original problem is changed into that:

to pair

Direct application of SVD algorithm

Then obtain the result

Namely a dictionary

The entry is updated.

The K-SVD algorithm comprises the following steps:

(1) initialization: setting a normalized initial dictionary matrix D⁰∈R^n×K，J＝1；

(2) Sparse coding: for each by orthogonal matching pursuit algorithm

Under the dictionaryIs sparse representation of

I.e. to solve for

(3) And (3) dictionary updating: for the

Is updated step by step every K (K is 1,2, … K), specifically:

is used

Calibrating atoms to be updated;

② calculating error matrix

Selecting omega in error matrix_kCalibrated new error matrix of atomic composition

Fourthly, for new error matrix

Using singular value decomposition

Selecting dictionary update terms

As a matrix

And updates the corresponding coefficients

(4)J＝J+1。

At present, the dictionary learning algorithm is not applied to marine CSEM data noise suppression, so the dictionary learning algorithm is introduced into marine CSEM data noise reduction for the first time.

The marine CSEM noise reduction method based on dictionary learning provided by the embodiment is subjected to simulation test.

(one) analog signal construction

The background model is shown in fig. 3 (a), and has a sea level as the origin of coordinates and a vertically downward direction as the positive direction of the z-axis. Air layer above the origin, resistivity 10¹³ohm-m, extending up to infinity. The sea water layer with the depth of 1000m below the origin is provided with the resistivity of 0.3ohm-m, and the surrounding rock layer with the resistivity of 1ohm-m is provided below the sea water layer. The classical model is as shown in (b) of fig. 3, and the main difference from the background model is that there is a high resistivity 100ohm-m barrier with a thickness of 100m at a depth of 2000m to simulate the reservoir.

Using the above model, the Magnitude-distance (MVO) curve of the response signal at the transmission frequency of 0.1Hz is shown in fig. 3 (c). In the figure, curve 1 represents the signal MVO curve under the background model, and curve 2 represents the signal MVO curve under the classical model. In this embodiment, the MVO curve is used as an evaluation criterion of the noise reduction effect.

Setting the ship's cruising speed to 1m/s, the distance-varying signal can be converted into a time series of amplitude variations with time. And if the sampling frequency of the signal is 1Hz, the sinusoidal signal with the frequency of 0.1Hz is subjected to amplitude modulation of the signal so as to simulate the actual signal received inside the receiver. Knowing that the amplitudes of the air wave and the magnetotelluric attenuate with the increase of the depth of the seawater, the average depth of the study object adopted in the embodiment exceeds 1000m, and therefore the influence of the two factors is ignored. Assuming that all the noises are additive noises, four kinds of noises are added to the forward analog signal, which are respectively:

(1) random noise, set amplitude 1X 10^-15The average value of integer multiples of V/(Am2) is 0, as shown in fig. 4 (a).

(2) The internal noise of the instrument, which is the noise of internal devices such as electrodes and electronic amplifiers, is set to a signal having an amplitude of 1% of the amplitude of the effective signal, as shown in fig. 4 (b).

(3) The third type is impulse noise with amplitude of 1 × 10^-13V/(Am²) And 30 impulse noises randomly distributed at each position to simulate the oscillation of the dipole, as shown in fig. 4 (c).

(4) The fourth type of noise is sinusoidal signals with frequencies of 0.01Hz, 0.02Hz, 0.03Hz and 0.04Hz, respectively, randomly added to the analog signals to simulate the noise caused by the movement of seawater, as shown in (d) of fig. 4.

In marine CSEM, in addition to a magnitude-distance (MVO) curve, a magnitude-time (MVT) curve is also often used as a criterion for evaluating the noise influence or the noise reduction effect. Fig. 4 (e) is a comparison graph of MVT curves of a noise-containing signal added with four types of noise and an original marine CSEM signal, and it is not difficult to find out that after 15,000s, the influence of the noise is gradually obvious and appears as a spike pulse, and after 22,000s, the signal is basically submerged in the noise.

To further investigate the influence of noise on the signal, four types of noise are added to the signal without noise, and then a normalized influence analysis chart is obtained, i.e. the amplitude of the signal with noise is divided by the amplitude of the original signal at the corresponding position, and the result is shown in (f) of fig. 4. Wherein type1 to type4 respectively represent the influence of four types of noise, and finally the all added curve represents the total influence after all the noise is added. Between 10,000s and 15,000s, significant noise effects have occurred. The sharp-pointed fluctuation is very obvious between 15,000s and 20,000 s. The 20,000s to 25,000s section, in addition to the more significant spike noise, has a section of signal significantly larger than the surrounding signal due to the increased amplitude of the noisy signal caused by the added marine motion noise. After 25,000s, the signal is substantially drowned out in noise.

The noise-containing signal is subjected to a short-time fourier transform with a window size of 60s, and the resulting spectrogram is shown in fig. 4 (f), and in the initial stage of the signal, the signal rapidly attenuates, conforming to the content shown by the MVT curve. There are distinct rectangular blocks around 10,000s, 15,000s and 22,000s, which represent the four frequencies of sea motion noise added by this embodiment. The whole spectrogram has bar-shaped vertical lines caused by impulse noise, and after 25,000s, the signal is not obvious, i.e. the signal is submerged in the noise.

The known signal is an amplitude modulation signal of a sine signal, so that a DST dictionary, a DST-Wavelet cascade dictionary and a learning dictionary are selected to respectively develop a reconstructed signal convergence experiment. Wherein, the Wavelet part in the DST-Wavelet cascade dictionary selects a Symlet dictionary which is commonly used for noise reduction of marine CSEM data.

Given that the amplitude of the received signal increases with time and approximates exponential decay, the near zone (500-.

Fig. 5 (d) shows reconstructed error curves under three dictionaries of the near-region signal, and the three dictionaries all show better convergence. The convergence of the learning dictionary is best, and the convergence is basically finished after several iterations. In fig. 5, (e) and (f) are reconstruction error curves of the middle and far zone signals under three dictionaries, respectively, and the shapes of the curves are similar to those of (a) in fig. 5. In the near, middle and far three-section signals, the three dictionaries all show better convergence performance, the convergence performance under the learning dictionary is optimal, and the learning dictionary can effectively extract signal characteristics and provide optimal sparse expression for target signals.

And (3) selecting the noise-containing signals constructed in the foregoing, and developing a noise reduction experiment of the compressed sensing algorithm under the traditional dictionary and the learning dictionary. In contrast, conventional short-time fourier transform denoising and wavelet transform denoising are introduced. The band-pass frequency is set to be 0.05-0.2Hz, the stop band cut-off frequency is set to be 0.01Hz and 0.2Hz, the pass band ripple is not more than 2dB, and the stop band attenuation is more than 40 dB. The wavelet transform selects Sym4 wavelet function, divides the target signal into 4 layers, and selects the threshold value of each layer according to the principle of maximum and minimum.

In addition, the iteration termination conditions of the orthogonal matching pursuit algorithm are two, wherein one is that the iteration reaches the maximum iteration times; the other is that the reconstruction error is smaller than the specified error. The amplitude of the marine CSEM signal decays approximately exponentially with time, while the amplitude of the noise is substantially constant, so that the signal-to-noise ratio decreases with increasing distance. The effective signal can be well restored by setting a smaller error threshold value at a position where the signal is larger, but the effective signal is easily influenced by noise at a position where the signal-to-noise ratio is lower. And setting the maximum iteration times of the three dictionaries to be 10 by combining the convergence performance test. The signals are intercepted by adopting a 60s time window, short-time Fourier transform denoising, wavelet change denoising and compressed sensing reconstruction denoising under three dictionaries are adopted for each segment of signals, and the obtained result is shown in figure 6.

Fig. 6 (a) is a graph of MVT after short-time fourier transform noise reduction, and it is easy to find a noise-containing curve, and a certain noise reduction effect is obtained, but the effect is not obvious. Fig. 6 (b) is a graph of MVT after wavelet transform denoising, and compared with the graph of MVT of a signal containing noise, the wavelet transform has limited denoising effect on spike pulses. Fig. 6 (c) is a comparison graph of the noise reduction signal MVT curve and the noise-free signal MVT using the classical DST dictionary. FIG. 6 (d) is a diagram comparing the MVT curve of the noise-reduced signal under the DST-Wavelet cascade dictionary with the curve without noise. Overall, the coincidence effect of the two curves is better, and the spike pulse signal is better suppressed. But there is a section between 20,000s and 25,000s where the reconstructed signal amplitude is significantly higher than the noise free signal. Here, the fourth part of the added sea motion noise is recovered as a valid signal by mistake in the DST-Wavelet dictionary. At this time, the signal amplitude is low, the noise amplitude is large, and part of the noise is mistaken for a valid signal and is reconstructed. Fig. 6 (e) is a comparison graph of a noise reduction signal MVT curve and a noise free signal MVT curve of the learning dictionary obtained by training using the K-SVD algorithm. Compared with the noise-containing signal MVT diagram given in the foregoing, the noise reduction effect is better on the whole, and the spike noise signal is suppressed to a certain extent, but the influence is not completely eliminated.

In conclusion, the noise reduction effect of the reconstructed signals under the three dictionaries is better than that of short-time Fourier transform and wavelet transform. The marine CSEM noise reduction method based on dictionary learning has the best effect, not only effectively reduces the noise level, but also inhibits the influence of spike pulse noise to a certain extent.

The advantage of using numerical simulation data for noise reduction analysis is that the signal is known to be free of any noise, and can be used as an evaluation criterion to convert the noise reduction effect of various algorithms into quantitative calculations. Common evaluation criteria are signal to noise ratio (SNR), Mean Square Error (MSE), Mean Absolute Error (MAE), and the like. Considering that the amplitude of the marine CSEM signal is approximately exponentially attenuated along with the receiving and transmitting distance, SNR and MAE are selected as the noise reduction effect quantitative evaluation standards, and the formulas are respectively as follows:

because the analog signal data volume is large and the medium-distance is obviously influenced by noise, the data of 13,000s-13,200s and 19,000s-19,200s are selected as target signals, and the quantitative calculation results of the noise reduction effect of various algorithms are listed in table 1.

TABLE 1 comparison table of noise reduction results of various algorithms

According to the definition of SNR and MAE, the larger the SNR value is, the smaller the MAE value is, the better the noise reduction effect is. The method has the advantages that the noise reduction effect is most obvious, the essential characteristics of signals can be mastered by a learning dictionary through a learning algorithm, and better expression is provided for signal sparse representation and noise reduction.

In conclusion, the noise reduction effect of the study dictionary under the compressed sensing framework is the best. The superiority of the learning dictionary algorithm as the ocean controllable source electromagnetic data noise reduction algorithm is proved in a numerical simulation experiment.

The examples are analyzed by a marine CSEM noise reduction method based on dictionary learning.

Actual data were from the south-east range of qiong. In operation the transmitter is dragged and dropped to a height of 50m from the seafloor, starting at a position 5,000m from one end of the receiver array and ending at a position 5,000m from the other end. During towing operation, the ship speed is towed at a constant speed of 2 knots (61.7m/min) along the operation line. The transmission frequency adopts a bipolar square wave signal with the combined frequency of 0.5Hz and 1.5Hz, the main energy of the signal is concentrated in 0.5Hz and 1.5Hz, and the signal has a wider frequency band range and controllable phase. The monitoring system timely adjusts the navigation parameters of the ship, the horizontal offset distance of the emission source relative to the observation profile is kept within 200m, and the distance between the emission unit and the seabed is kept within 100 m.

In the area of southeast and John, 4,500m exploration lines are arranged along the NW-SE direction, wherein 10 receiving stations (R1-R10) are included, the distance between the stations is 500m, and the data of the second station is selected for research. The 5 components recorded in the R2 site are Ex, Ey, Ez, Hx, and Hy, respectively, as shown in fig. 7.

As shown in fig. 7, the time periods from 13:00 to 13:30 are time periods with a small transceiving distance, the signal in the receiver has obvious response, and the effective signal in other time periods has low amplitude and is not obvious in the time sequence. The frequency components of 0.5Hz and 1.5Hz were extracted from the signal and expressed as MVT plots, respectively, and the results are shown in FIG. 8.

Fig. 8 (a) is an MVT chart at 0.5Hz, and the ordinate is a logarithmic coordinate, which can show more information than the time-series chart. Wherein it is readily apparent that the time period which is comparatively little affected by noise lies between 12:00 and 14: 00. Fig. 8 (b) is a 1.5Hz frequency MVT graph, which is less affected by noise during the 12:00 to 14:00 time period, similar to (a). Therefore, data in the time period are selected to form a training sample set, and a dictionary is trained through a K-SVD algorithm and recorded as DL. Two sections of signals are intercepted from the time sequence diagram for reconstruction experiments, and the intercepting positions are shown as circles in the diagram. Fig. 8 (c) and (d) are respectively time-series signals after being intercepted, where the signal transmission and reception distance shown in (c) is large and is affected by noise to some extent. (d) The signal receiving and transmitting distance is closer, and the influence degree of noise is smaller. Fig. 8 (e) shows the signal convergence experimental results of the DST dictionary, the DST-Wavelet cascade dictionary, and the DL dictionary, where the convergence performance of the DST dictionary and the DST-Wavelet cascade dictionary is similar and better than that of the learning dictionary DL. The signal at this stage is affected by noise more seriously, and the DST dictionary and the DST-Wavelet dictionary recover part of the noise, so the convergence error is smaller. Fig. 8 (f) shows the convergence result of the signal shown in (d) under three dictionaries, wherein the convergence performance of the DST dictionary is similar to that of the DST-Wavelet cascade dictionary, the convergence of the learning dictionary is faster in the initial stage, and then the speed is slower, but the convergence effect is better than that of the signal shown in (c).

The noise reduction operation is performed on the Ey component of the R2 site under the DST dictionary, the DST-Wavelet cascade dictionary and the learning dictionary, and the extraction of the 0.5Hz MVT curve and the 1.5Hz MVT curve from the obtained results are shown in fig. 9 and fig. 10, respectively.

In fig. 9, (a) and (b) are respectively 0.5Hz MVT curves under the DST dictionary and the DST-Wavelet cascade dictionary, and the noise reduction effect is not obvious, which is represented by mixing the original noise-containing signal and the noise-reduced signal amplitude point. Fig. 9 (c) is a graph of the noise reduction result under the learning dictionary, which has a relatively obvious noise reduction effect compared to the first two given dictionaries, and shows that the signal point after noise reduction is obviously located below the original signal point in the time period of 10:30 to 12: 00.

In fig. 10, (a), (b), and (c) are 1.5Hz MVT curves after noise reduction of the DST dictionary, the DST-Wavelet cascade dictionary, and the learning dictionary, respectively. Compared with a 0.5Hz MVT graph, the three dictionaries in the 1.5Hz MVT graph show better noise reduction effect, wherein the noise reduction effect of the DST dictionary is similar to that of a DST-Wavelet cascade dictionary. The learning dictionary has the best noise reduction effect, the noise level of the noise-reduced signal is obviously lower than that of the original signal in a time period of 14:00 to 15: 30.

In conclusion, the dictionary learning can obtain the self-adaptive dictionary aiming at the signal through the learning of the data set, so that the dictionary is more consistent with the inherent characteristics of the signal, and the signals of the same type can obtain a better noise reduction effect under the dictionary.

According to the marine CSEM signal denoising method, the learning dictionary is obtained by learning through selecting the signals within the set distance from the minimum transceiving distance position, the correlation between the marine CSEM signal and the learning dictionary is improved, so that when the marine CSEM signal is subjected to sparse coding and denoising reconstruction through the learning dictionary, a better sparse expression effect can be obtained, the noise in the marine CSEM signal can be effectively reduced, and the denoising effect of the marine CSEM signal is improved.

Example 2

In this embodiment, a marine CSEM noise reduction system based on dictionary learning is disclosed, including:

the signal acquisition module is used for acquiring marine CSEM signals;

Example 3

In this embodiment, an electronic device is disclosed, which comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the steps of the method for marine CSEM noise reduction based on dictionary learning disclosed in embodiment 1.

Example 4

In this embodiment, a computer readable storage medium is disclosed for storing computer instructions which, when executed by a processor, perform the steps of a method for marine CSEM noise reduction based on dictionary learning as disclosed in embodiment 1.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims

1. A marine CSEM noise reduction method based on dictionary learning is characterized by comprising the following steps:

acquiring marine CSEM signals; wherein the CSEM signals represent marine controllable source electromagnetic signals;

selecting a signal within a set distance from a minimum transceiving distance position from marine CSEM signals as a sample to learn to obtain a learning dictionary;

wherein, from marine CSEM signal, choose to study as the sample apart from the signal in minimum transmitting-receiving distance position settlement distance, obtain the study dictionary, include: calculating the interception time of the acquired signal according to the moving speed of the mobile transmitting station and the set distance from the minimum transceiving distance position, intercepting the signal from the marine CSEM signal according to the interception time of the signal, and performing dictionary learning;

and sparsely representing the target marine CSEM signal to be denoised, sparsely encoding the marine CSEM signal by an orthogonal matching pursuit algorithm according to the obtained learning dictionary, and reconstructing the sparsely encoded signal to obtain the denoised marine CSEM signal.

2. The marine CSEM noise reduction method based on dictionary learning as claimed in claim 1, wherein the marine CSEM signal is formed into an MVT curve, and the minimum position of the signal transmitting and receiving distance is determined through the MVT curve; the MVT curve represents a magnitude versus time curve.

3. The method of claim 1, wherein orientation correction is performed on the collected marine CSEM signals, and the MVT curve is formed by the orientation-corrected signals.

4. The method of claim 1, wherein the samples are learned by a K-SVD algorithm.

5. The marine CSEM noise reduction method based on dictionary learning as claimed in claim 1, wherein in the dictionary learning, the smallest error decomposition term is selected as the updating term of the dictionary atom and the corresponding sparse coefficient by taking the smallest error as a standard and performing singular value decomposition on an error matrix.

6. A marine CSEM noise reduction system based on dictionary learning, comprising:

the signal acquisition module is used for acquiring marine CSEM signals; wherein the CSEM signals represent marine controllable source electromagnetic signals;

and the noise-reduced signal acquisition module is used for carrying out sparse representation on the target marine CSEM signal to be subjected to noise reduction, carrying out sparse coding on the marine CSEM signal through an orthogonal matching tracking algorithm according to the obtained learning dictionary, and carrying out sparse reconstruction on the sparsely coded signal to obtain the noise-reduced marine CSEM signal.

7. An electronic device comprising a memory and a processor, and computer instructions stored on the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the steps of a method for marine CSEM noise reduction based on dictionary learning of any one of claims 1-5.

8. A computer readable storage medium storing computer instructions which, when executed by a processor, perform the steps of a method for marine CSEM noise reduction based on dictionary learning according to any one of claims 1 to 5.