CN115828069A - End-to-end magnetic anomaly signal noise reduction method based on deep learning - Google Patents
End-to-end magnetic anomaly signal noise reduction method based on deep learning Download PDFInfo
- Publication number
- CN115828069A CN115828069A CN202211707589.XA CN202211707589A CN115828069A CN 115828069 A CN115828069 A CN 115828069A CN 202211707589 A CN202211707589 A CN 202211707589A CN 115828069 A CN115828069 A CN 115828069A
- Authority
- CN
- China
- Prior art keywords
- magnetic
- signal
- signals
- decomposition
- noise
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Abstract
The invention discloses an end-to-end magnetic anomaly signal noise reduction method based on deep learning. Firstly, performing wavelet decomposition and reconstruction on magnetic anomaly signals to realize band-pass filtering, and extracting signals of frequency bands where magnetic anomalies exist; then, decomposing the signals respectively by utilizing Orthogonal Basis Functions (OBF), and solving the decomposition coefficients on each orthogonal basis; then, taking the decomposition coefficient of each orthogonal basis as the input of a multi-channel deep learning network, taking the corresponding ideal magnetic signal as the output, and training to obtain an end-to-end noise reduction network model; and finally, processing the magnetic signal by using the trained noise reduction network model to obtain the noise-reduced magnetic abnormal signal. The method comprehensively utilizes the band-pass filtering, OBF decomposition and end-to-end deep learning network based on wavelet analysis to extract the characteristics of the acquired magnetic signals and reduce noise, and can effectively improve the signal-to-noise ratio of the magnetic abnormal signals.
Description
Technical Field
The invention relates to the technologies of signal processing, deep learning and the like, and belongs to the field of detection of magnetic anomaly signals.
Background
Magnetic Anomaly Detection (MAD) is an important method for detecting ferromagnetic targets and is widely applied to geological prospecting, sunken ship detection, aviation magnetic detection and the like. The main sources of noise in the magnetic signal are geomagnetic noise, sensor inherent noise, test platform interference noise, geological noise and the like. When the detection distance is long, the signal-to-noise ratio is extremely low, and magnetic anomaly signals are usually buried in strong geomagnetic noise, so that the performance of the traditional magnetic anomaly detection method is reduced.
With the development of the deep learning neural network, the neural network can learn and extract the deep features hidden behind the data by itself, and replaces the feature extraction work needing manual participation in the past. The end-to-end learning mode is that artificial sub-problem division is not carried out in the whole learning process, but the mapping from original input to expected output is completely given to a deep learning model for autonomous learning, a task of feature extraction is given to the model for doing, original data or slightly preprocessed data are directly input, and the model carries out feature extraction. By the method, the neural network can well learn the description of the features, so that the model can learn the feature operator needing manual design.
The invention discloses an end-to-end magnetic anomaly signal noise reduction method based on deep learning, aiming at the problem that a target signal is interfered and submerged by noise in magnetic anomaly detection. The method designs a deep learning model by utilizing a coding-decoding structure, adopts a decomposition coefficient obtained by subjecting a noise-containing magnetic anomaly signal to wavelet band-pass filtering and OBF decomposition as the input of the model, takes a corresponding ideal magnetic anomaly signal as the output, trains through a large amount of data, and automatically extracts the characteristics of magnetic anomaly data. And finally, inputting the magnetic signal to be denoised into a network model, wherein the output of the model is the denoised magnetic abnormal signal. Compared with the traditional method, the method has better self-adaptive capacity and robustness, and can greatly improve the signal-to-noise ratio of the magnetic anomaly signal.
Disclosure of Invention
The invention discloses an end-to-end magnetic anomaly signal noise reduction method based on deep learning, which can reduce noise of magnetic anomaly signals. The method comprises the following implementation steps:
step 1: according to the magnetic dipole model, ideal magnetic anomaly signals under the conditions of different CPA (close Point of Approach) distances, different movement speeds and the like are constructed, and the calculation formula of the ideal magnetic anomaly signals is as follows:
B＝B x e x +B y e y +B z e z
l＝cos I cos D
m＝cos I sin D
n＝sin I
wherein B is the magnetic induction intensity of the target magnetic dipole, B x 、B y 、B z Is the component of B on the three axes of x, y and z, mu is the vacuum magnetic conductivity, p is the magnetic moment of the magnetic dipole, r is the linear distance from the magnetic dipole to the sensor, r is the magnetic dipole x 、r y 、r z The component of r on the three axes of x, y and z is shown, I is the geomagnetic inclination angle of the detection point, and D is the geomagnetic declination angle of the detection point;
step 2: constructing a magnetic abnormal signal containing noise, acquiring a pure geomagnetic background signal at a detection point, obtaining different pure geomagnetic background signals by moving a time window, and superposing the ideal magnetic abnormal signal obtained in the step (1) and the different pure geomagnetic background signals to obtain different magnetic abnormal signals containing noise, so as to construct a magnetic abnormal signal data set containing the geomagnetic background noise;
and step 3: performing wavelet band-pass filtering on the noise-containing magnetic anomaly signal obtained in the step (2), performing wavelet decomposition firstly, then setting the detail coefficient belonging to high frequency and the approximate coefficient belonging to extremely low frequency after the wavelet decomposition to zero, and then reconstructing to obtain a magnetic anomaly signal after the band-pass filtering;
and 4, step 4: carrying out Orthogonal Basis Function (OBF) decomposition on the filtered signals to obtain decomposition coefficients corresponding to three groups of orthogonal bases of each group of signals;
and 5: constructing an end-to-end deep learning network, constructing a training sample data set by using the decomposition coefficient obtained in the step 4, using the decomposition coefficient as the input of a network model, using a corresponding ideal magnetic anomaly signal as the output of the network model, and training to obtain an end-to-end noise reduction network model;
step 6: and (3) subjecting the magnetic signal to be denoised to wavelet band-pass filtering and OBF decomposition processing in the steps 3 and 4, then taking the decomposition coefficient as the input of the trained denoising network model, and taking the output of the denoising network model as the denoised magnetic anomaly signal.
Drawings
FIG. 1 is a flow chart of an end-to-end magnetic anomaly signal noise reduction method based on deep learning according to the present invention.
Fig. 2 shows a noise-containing magnetic anomaly signal according to the present invention.
Fig. 3 is a set of input signals for training an end-to-end deep learning network according to the present invention.
Fig. 4 is an output signal for training an end-to-end deep learning network according to the present invention.
Fig. 5 is a schematic diagram of a network structure of deep learning training according to the present invention.
Detailed Description
The following describes a technical scheme of an end-to-end magnetic anomaly signal noise reduction method based on deep learning in detail with reference to the accompanying drawings and specific embodiments. The method comprises the following implementation steps:
step 1: according to the magnetic dipole model, ideal magnetic anomaly signals under the conditions of different CPA (close Point of Approach) distances, different movement speeds and the like are constructed, and the calculation formula of the ideal magnetic anomaly signals is as follows:
B＝B x e x +B y e y +B z e z
l＝cos I cos D m＝cos I sin D n＝sin I
wherein B is the magnetic induction intensity of the target magnetic dipole, B x 、B y 、B z Is the component of B on the three axes of x, y and z, mu is the vacuum magnetic conductivity, p is the magnetic moment of the magnetic dipole, r is the linear distance from the magnetic dipole to the sensor, r is the magnetic dipole x 、r y 、r z The component of r on the three axes of x, y and z is shown, I is the geomagnetic inclination angle of the detection point, and D is the geomagnetic declination angle of the detection point.
Step 2: and (2) constructing a magnetic abnormal signal containing noise, firstly collecting a pure geomagnetic background signal at a detection point, obtaining different pure geomagnetic background signals by moving a time window, and superposing the ideal magnetic abnormal signal obtained in the step (1) and the different pure geomagnetic background signals to obtain different magnetic abnormal signals containing noise, as shown in fig. 2, so as to construct a magnetic abnormal signal data set containing the geomagnetic background noise.
And step 3: and (3) performing wavelet band-pass filtering on the noise-containing magnetic anomaly signal obtained in the step (2), performing wavelet decomposition on the signal to decompose the signal into 10 layers, then zeroing detail coefficients D1-D3 after the wavelet decomposition, simultaneously zeroing an approximation coefficient A10, equivalently only keeping middle detail coefficients D7-D10, and then reconstructing to obtain the magnetic anomaly signal after the band-pass filtering.
And 4, step 4: performing Orthogonal Basis Function (OBF) decomposition on the filtered signals to obtain decomposition coefficients corresponding to three groups of orthogonal bases of each group of signals, wherein the three orthogonal basis functions are as follows:
wherein w = D/R 0 D is the distance from the center point of the track when the target moves along the track, R 0 The distance between the target motion track and the sensor is the closest distance, namely the CPA distance.
The OBF decomposition coefficient is calculated by the formula:
wherein Sig is a noisy magnetic anomaly signal. The three sets of decomposition coefficients of the signal shown in fig. 2 after OBF decomposition are shown in fig. 3.
And 5: and (4) constructing an end-to-end deep learning network, constructing a training sample data set by using the decomposition coefficient obtained in the step (4), taking the decomposition coefficient as the input of the network model, taking the corresponding ideal magnetic anomaly signal as the output of the network model, and training to obtain an end-to-end noise reduction network model. For example, the decomposition coefficient shown in fig. 2 is input as a model, and the ideal magnetic anomaly signal shown in fig. 4 is trained as an output of the model.
The end-to-end deep learning network is an encoding-decoding network, and the structure of the encoding-decoding network is shown in fig. 5, and the encoding-decoding network is composed of 4 convolutional layers and 4 deconvolution layers which are connected in series, wherein the four convolutional layers respectively adopt 128, 256, 512 and 512 4 × 1 convolution kernels, and the four deconvolution layers respectively adopt 512, 256 and 128 4 × 1 convolution kernels. The input signal passes through the convolutional layer and the deconvolution layer in sequence and finally reaches the output layer. The loss function in the model is a cross entropy function, and the calculation formula is as follows:
where x is the value of the true tag,
is the probability of prediction, i.e.
Representing the difference of the prediction probability and the sample label.
Step 6: and (3) subjecting the magnetic signal to be denoised to wavelet band-pass filtering and OBF decomposition processing in the steps 3 and 4, then taking the decomposition coefficient as the input of the trained denoising network model, and taking the output of the denoising network model as the denoised magnetic anomaly signal.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. This need not be, nor should it be exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (3)
1. The invention discloses an end-to-end magnetic anomaly signal noise reduction method based on deep learning, which is characterized by comprising the following implementation steps of:
step 1: according to the magnetic dipole model, ideal magnetic anomaly signals under the conditions of different CPA (close Proximity Approach) distances, different motion speeds and the like are constructed, and the calculation formula of the ideal magnetic anomaly signals is as follows:
B＝B x e x +B y e y +B z e z
l＝cosIcosD
m＝cosIsinD
n＝sinI
wherein B is the magnetic induction intensity of the target magnetic dipole, B x 、B y 、B z Is the component of B on the three axes of x, y and z, mu is the vacuum magnetic conductivity, p is the magnetic moment of the magnetic dipole, r is the linear distance from the magnetic dipole to the sensor, r is the magnetic dipole x 、r y 、r z The component of r on the three axes of x, y and z is shown, I is the geomagnetic inclination angle of the detection point, and D is the geomagnetic declination angle of the detection point;
and 2, step: constructing a magnetic abnormal signal containing noise, acquiring a pure geomagnetic background signal at a detection point, obtaining different pure geomagnetic background signals by moving a time window, and superposing the ideal magnetic abnormal signal obtained in the step (1) and the different pure geomagnetic background signals to obtain different magnetic abnormal signals containing noise, so as to construct a magnetic abnormal signal data set containing the geomagnetic background noise;
and 3, step 3: performing wavelet band-pass filtering on the noise-containing magnetic anomaly signal obtained in the step (2), performing wavelet decomposition firstly, then setting the detail coefficient belonging to high frequency and the approximate coefficient belonging to extremely low frequency after the wavelet decomposition to zero, and then reconstructing to obtain a magnetic anomaly signal after the band-pass filtering;
and 4, step 4: carrying out Orthogonal Basis Function (OBF) decomposition on the filtered signals to obtain decomposition coefficients corresponding to three groups of orthogonal bases of each group of signals;
and 5: constructing an end-to-end deep learning network, constructing a training sample data set by using the decomposition coefficient obtained in the step 4, taking the decomposition coefficient as the input of a network model, taking a corresponding ideal magnetic anomaly signal as the output of the network model, and training to obtain an end-to-end noise reduction network model;
step 6: and (3) subjecting the magnetic signal to be denoised to wavelet band-pass filtering and OBF decomposition processing in the steps 3 and 4, then taking the decomposition coefficient as the input of the trained denoising network model, and taking the output of the denoising network model as the denoised magnetic anomaly signal.
2. The method for reducing noise of the end-to-end magnetic anomaly signal based on deep learning as claimed in claim 1, wherein the number of the orthogonal basis functions in the step 4 is 3, and the formulas are respectively as follows:
wherein w = D/R 0 And D is the track running time and track in the targetDistance of the center points, R 0 The distance between the target motion track and the sensor is the closest distance, namely the CPA distance.
The OBF decomposition coefficient is calculated by the formula:
wherein Sig is a noisy magnetic anomaly signal.
3. The method for reducing noise of end-to-end magnetic anomaly signals based on deep learning according to claim 1, wherein the end-to-end deep learning network constructed in the step 5 is an encoding-decoding network, the structure of which is shown in fig. 5, and the encoding-decoding network is composed of 4 convolutional layers and 4 deconvolution layers which are connected in series, wherein the number of convolutional kernels in the four convolutional layers is n, 2n, 4n and 4n, the number of convolutional kernels in the deconvolution layers is 4n, 2n and n, and the sizes of the convolutional kernels are all 4 × 1; the input signal sequentially passes through the convolution layer and the deconvolution layer and finally reaches the output layer; the loss function in the model is a cross entropy function, and the calculation formula is as follows:
where x is the true tag value and,
is a predicted probability, i.e.
Representing the difference of the prediction probability and the sample label.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211707589.XA CN115828069A (en) | 2022-12-29 | 2022-12-29 | End-to-end magnetic anomaly signal noise reduction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211707589.XA CN115828069A (en) | 2022-12-29 | 2022-12-29 | End-to-end magnetic anomaly signal noise reduction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115828069A true CN115828069A (en) | 2023-03-21 |
Family
ID=85519270
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211707589.XA Pending CN115828069A (en) | 2022-12-29 | 2022-12-29 | End-to-end magnetic anomaly signal noise reduction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115828069A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117076858A (en) * | 2023-08-18 | 2023-11-17 | 东华理工大学 | Deep learning-based low-frequency geomagnetic strong interference suppression method and system |
-
2022
- 2022-12-29 CN CN202211707589.XA patent/CN115828069A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117076858A (en) * | 2023-08-18 | 2023-11-17 | 东华理工大学 | Deep learning-based low-frequency geomagnetic strong interference suppression method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Seismic signal denoising and decomposition using deep neural networks | |
| CN113935467B (en) | DAS (data acquisition system) well exploration data noise suppression method based on iterative multi-scale attention network | |
| CN106441288A (en) | Adaptive wavelet denoising method for accelerometer | |
| CN108961181B (en) | Shearlet transform-based ground penetrating radar image denoising method | |
| CN115828069A (en) | End-to-end magnetic anomaly signal noise reduction method based on deep learning | |
| CN111190227A (en) | Low signal-to-noise ratio seismic data denoising method based on residual convolution generation countermeasure model | |
| CN112842348B (en) | Automatic classification method for electrocardiosignals based on feature extraction and deep learning | |
| CN112818876B (en) | Electromagnetic signal extraction and processing method based on deep convolutional neural network | |
| CN111580161A (en) | Earthquake random noise suppression method based on multi-scale convolution self-coding neural network | |
| CN115877461A (en) | Desert earthquake noise suppression method based on multi-scale attention interaction network | |
| CN110989005B (en) | Weak magnetic anomaly self-adaptive real-time detection method based on scalar magnetometer array | |
| CN111650653A (en) | Magnetic resonance signal denoising method based on noise correlation and wavelet threshold method | |
| CN112084845A (en) | Low-frequency 1/f noise elimination method based on multi-scale wavelet coefficient autocorrelation | |
| CN114371474A (en) | Intelligent radar signal sorting method and system based on convolution noise reduction self-encoder | |
| CN116184502A (en) | Underground DAS noise suppression method based on subspace projection attention network | |
| CN115951418A (en) | Magnetic anomaly detection method based on deep reinforcement learning | |
| Fang et al. | Mask-guided model for seismic data denoising | |
| CN115436993A (en) | Seismic signal P wave detection method based on PPNet network | |
| CN114706128A (en) | Microseism data noise suppression method and system, storage medium and seismic information processing equipment | |
| CN115097391A (en) | Synthetic aperture radar interference suppression method based on automatic encoder | |
| CN114358040A (en) | Disturbing signal denoising method based on adaptive estimation threshold method | |
| CN107609530A (en) | One kind is with brill orientation electromagnetic resistivity imaging features extracting method | |
| CN109991657B (en) | Seismic data high-resolution processing method based on inverse binary recursive singular value decomposition | |
| Liao et al. | Automatic first arrival picking combining transfer learning and the short-time Fourier transform | |
| Zhang et al. | Seismic Random Noise Suppression Based on MIRNet With Dense Feature Fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |