CN117493776B - Geophysical exploration data denoising method and device and electronic equipment - Google Patents
Geophysical exploration data denoising method and device and electronic equipment Download PDFInfo
- Publication number
- CN117493776B CN117493776B CN202311848272.2A CN202311848272A CN117493776B CN 117493776 B CN117493776 B CN 117493776B CN 202311848272 A CN202311848272 A CN 202311848272A CN 117493776 B CN117493776 B CN 117493776B
- Authority
- CN
- China
- Prior art keywords
- geophysical
- data
- dimensional
- data sequence
- sequence
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000012549 training Methods 0.000 claims abstract description 73
- 230000008569 process Effects 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims description 43
- 238000013135 deep learning Methods 0.000 claims description 39
- 238000011176 pooling Methods 0.000 claims description 23
- 238000013507 mapping Methods 0.000 claims description 16
- 238000013501 data transformation Methods 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 230000001131 transforming effect Effects 0.000 claims description 4
- 238000000926 separation method Methods 0.000 abstract description 5
- 239000000523 sample Substances 0.000 description 13
- 230000000694 effects Effects 0.000 description 10
- 238000001228 spectrum Methods 0.000 description 9
- 230000009286 beneficial effect Effects 0.000 description 6
- 238000010606 normalization Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008030 elimination Effects 0.000 description 2
- 238000003379 elimination reaction Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000035939 shock Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/36—Effecting static or dynamic corrections on records, e.g. correcting spread; Correlating seismic signals; Eliminating effects of unwanted energy
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/36—Effecting static or dynamic corrections on records, e.g. correcting spread; Correlating seismic signals; Eliminating effects of unwanted energy
- G01V1/364—Seismic filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Remote Sensing (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geophysics (AREA)
- Evolutionary Biology (AREA)
- Geology (AREA)
- Environmental & Geological Engineering (AREA)
- Acoustics & Sound (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention provides a geophysical exploration data denoising method, a geophysical exploration data denoising device and electronic equipment. According to the invention, the noise sharing frequency band with the signal can be effectively eliminated, the optimal separation threshold value of the signal and the noise is not required to be searched, and the training process on the cloud platform is efficient, low in cost and convenient for cooperation.
Description
Technical Field
The invention relates to the field of geophysical prospecting data processing, in particular to a geophysical prospecting data denoising method, a geophysical prospecting data denoising device and electronic equipment.
Background
Geophysical survey data includes useful geophysical survey signals, but also incorporates natural and artificial noise. Noise can reduce the signal-to-noise ratio of the signal, affecting subsequent analysis and interpretation of the geophysical prospecting signal. Denoising can improve signal quality and signal-to-noise ratio, so that the characteristics of geophysical prospecting signals are more obvious, and the analysis and understanding of geophysical prospecting data are facilitated.
Denoising the geophysical prospecting signals can improve the effects of other analysis algorithms. The input of many subsequent analysis algorithms requires clean data. The effect of these algorithms can be improved by performing denoising preprocessing.
Denoising the geophysical prospecting signals can improve accuracy of geophysical prospecting positioning and parameter estimation. In locating geophysical prospecting events and estimating geophysical prospecting wave parameters (e.g., time of day, waveform, etc.), it is necessary to identify and pick up the exact phase, and noise can interfere with this process. Denoising can make phase onset clearer and discernable, and helps to improve accuracy of geophysical prospecting positioning and parameter estimation.
Denoising the geophysical prospecting signals is also beneficial to identifying small signals. Small signals such as weak vibration and micro vibration are easily submerged by noise. Denoising makes these weak signals appear, which is beneficial to identify weak shocks, micro shocks and other low amplitude signals.
Denoising the geophysical prospecting signals can enhance the geophysical prospecting monitoring effect. Noise can mask geophysical prospecting events, degrading monitoring effectiveness. Denoising can improve monitoring quality, so that more geophysical prospecting events are detected. The method has great significance for geophysical exploration and early warning systems and the like.
Denoising the geophysical prospecting signals can also extract the earth background noise information. The background noise left after the removal of the useful signal also contains useful information of the earth's structure.
Filtering is a widely used method of suppressing noise in geophysical prospecting signal processing. This approach is effective if noise is concentrated primarily in a range outside the signal spectrum. However, if noise occupies the same frequency band as the effective signal, the filtering performance may be greatly reduced. Moreover, it is not easy to select appropriate filtering parameters. Unsuitable filtering parameters also substantially alter the geophysical survey signal, further reducing the quality of subsequent geophysical survey data analysis.
The geophysical survey data is also converted to time-frequency for noise reduction, and the noise correlation coefficients are processed in the time-frequency domain to maintain or enhance the signal-dependent coefficients to suppress noise. Such schemes require finding the optimal threshold for separating the signal from the noise, and finding the optimal threshold in the actual process is difficult, and if an improper threshold is used, it may result in signal attenuation or ineffective noise suppression.
In addition, performing extensive geophysical survey data processing and the like locally requires the purchase of expensive equipment such as GPU servers and is also often limited in computational efficiency and is not conducive to collaborative collaboration.
Disclosure of Invention
In view of the above, the invention aims to provide a geophysical exploration data denoising scheme based on deep learning on a cloud platform, which can effectively eliminate noise sharing frequency bands with signals, does not need to find an optimal separation threshold value of the signals and the noise, has high efficiency and low cost in a training process, and is convenient for cooperation.
According to an aspect of the present invention, there is provided a method of denoising geophysical prospecting data, the method comprising:
step 1, generating a time domain training data set on a cloud platform, wherein the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence;
step 2, a data conversion unit is established on a cloud platform, and the data conversion unit is called to respectively convert the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence based on the following steps of obtaining a time-frequency-domain two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence:
，
，
wherein,representing samples in a two-dimensional noiseless geophysical survey data sequence,representing samples in a two-dimensional noisy geophysical survey data sequence, N being the number of samples in a sliding rectangular window used in the transform to select samples of the time domain geophysical survey data sequence,representing an nth sample of the time domain noiseless geophysical survey data sequence in the current rectangular window,representing an nth sample point of the time domain noisy geophysical prospecting data sequence in a current rectangular window, traversing the whole time domain by sliding the rectangular window;
step 3, segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, wherein each data segment comprises a plurality of sample points, and establishing the following objective function:
，
wherein,for the number of data segments to be divided,for the mth two-dimensional noiseless geophysical survey data segment,for the mth two-dimensional noisy geophysical survey data segment,for defining the length of the estimated time window,representing a mapping function;
step 4, on the cloud platform, two-dimensional noisy geophysical exploration data segmentsAs input, two-dimensional noiseless geophysical prospecting data segmentAs a desired output to train a deep learning network learning mapping function：
And 5, uploading the actual geophysical exploration record to a cloud platform, converting the actual geophysical exploration record into a two-dimensional sequence by adopting the data conversion unit, segmenting the two-dimensional sequence, and then sending the obtained data segment into a trained deep learning network for denoising.
In some embodiments, in step 1, specifically:
generating a time domain noiseless geophysical prospecting data sequence in a time domain training data set by using a wavelet base;
generating a noise sequence by adopting a fast fractional difference algorithm based on Fourier transform;
the generated time-domain noise-free geophysical survey data sequence and the generated noise sequence are added to obtain a time-domain noisy geophysical survey data sequence.
In some embodiments, in step 2, the rectangular window has a size ranging from 32 samples, 64 samples, 128 samples, and 256 samples, and the sliding overlap ratio has a size ranging from 70% to 95%.
In some embodiments, in step 3, a method for defining an estimated time windowThe range of the value of (2) is 3-10.
In some embodiments, in step 4, the deep learning network is a Deep Supervision Network (DSN).
In some embodiments, the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 maximum pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 maximum pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 fully connected layers and 1 classification layer, wherein the first auxiliary supervision layer and the second auxiliary supervision layer comprise 1 fully connected layer and one output layer, the training process of the whole deep supervision network adopts a back propagation algorithm, and a random gradient descent optimizer is adopted for weight updating.
According to another aspect of the present invention, there is also provided a geophysical prospecting data denoising apparatus, the apparatus comprising:
the system comprises a training data generation unit, a data processing unit and a data processing unit, wherein the training data generation unit is used for generating a time domain training data set on a cloud platform, and the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence;
the data transformation unit is used for respectively transforming the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence on the cloud platform based on the following steps of:
，
，
wherein,representing samples in a two-dimensional noiseless geophysical survey data sequence,representing samples in a two-dimensional noisy geophysical survey data sequence, N beingThe number of samples in the sliding rectangular window used to select samples of the time domain geophysical survey data sequence in the transform,representing an nth sample of the time domain noiseless geophysical survey data sequence in the current rectangular window,representing an nth sample point of the time domain noisy geophysical prospecting data sequence in a current rectangular window, traversing the whole time domain by sliding the rectangular window;
the system comprises a segmentation and objective function construction unit, a segmentation and objective function construction unit and a control unit, wherein the segmentation and objective function construction unit is used for segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, each data segment comprises a plurality of sample points, and the following objective function is established:
，
wherein,for the number of data segments to be divided,for the mth two-dimensional noiseless geophysical survey data segment,for the mth two-dimensional noisy geophysical survey data segment,for defining the length of the estimated time window,representing a mapping function;
deep learning network training unit for training two-dimensional noisy geophysical exploration data segment on cloud platformAs input, two-dimensional noiseless geophysical prospecting data segmentAs a desired output to train a deep learning network learning mapping function：
The comprehensive calling denoising unit is used for calling the data transformation unit to transform the actual geophysical exploration record uploaded to the cloud platform into a two-dimensional sequence, calling the segmentation and objective function construction unit to segment the two-dimensional sequence, and then sending the obtained data segment into the trained deep learning network for denoising.
In some embodiments, in the training data generation unit, specifically:
generating a time domain noiseless geophysical prospecting data sequence in a time domain training data set by using a wavelet base;
generating a noise sequence by adopting a fast fractional difference algorithm based on Fourier transform;
the generated time-domain noise-free geophysical survey data sequence and the generated noise sequence are added to obtain a time-domain noisy geophysical survey data sequence.
In some embodiments, the deep learning network is a Deep Supervision Network (DSN), in particular:
the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 maximum pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 maximum pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 full connection layers and 1 classification layer, wherein the first auxiliary supervision layer and the second auxiliary supervision layer comprise 1 full connection layer and one output layer, a reverse propagation algorithm is adopted in the training process of the whole deep supervision network, and a random gradient descent optimizer is adopted for weight updating.
According to another aspect of the present invention, there is also provided an electronic device including:
a memory storing executable instructions:
a processor executing the executable instructions in the memory to implement the geophysical survey data denoising method described above.
According to another aspect of the invention, there is also provided a computer readable storage medium storing a computer program which, when executed by a processor, implements the geophysical survey data denoising method described above.
The beneficial effects of the invention at least comprise:
(1) The one-dimensional time domain signal is transformed into a two-dimensional time-frequency domain, so that the time domain information and the frequency domain information of the signal can be simultaneously represented;
(2) When the signal and the noise share the frequency band, the signal and the noise are difficult to distinguish in the time domain, but after the signal and the noise are converted into the time-frequency domain, the signal and the noise can be more easily distinguished due to different time-frequency distribution;
(3) The time-frequency representation improves the sparsity of the signals, and is beneficial to the sparse representation of the signals learned by the denoising method;
(4) The constructed objective function denoises the current data segment through past, future and current data segments, so that the denoising effect is improved;
(5) Noise elimination is realized through training the deep learning network, and an optimal separation threshold value of signals and noise is not required to be found;
(6) Training the deep learning network to use the data segment instead of the whole frequency spectrum to infer the noise-free signal of interest can significantly reduce the amount of training data and improve the stability of the network;
(7) Matching with a deep learning network through specific time-frequency transformation, so that the prediction performance and training convergence are effectively improved;
(8) The geophysical exploration data denoising is realized on the cloud platform, expensive equipment is not required to be purchased locally, the training efficiency is high, and the cooperation is convenient.
The method and apparatus of the present invention have other features and advantages which will be apparent from or are set forth in detail in the accompanying drawings and the following detailed description, which are incorporated herein, and which together serve to explain certain principles of the present invention.
Drawings
The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular descriptions of exemplary embodiments of the invention as illustrated in the accompanying drawings wherein like reference numbers generally represent like parts throughout the exemplary embodiments of the invention.
FIG. 1 illustrates a flow chart of a method of denoising geophysical survey data according to one embodiment of the present invention.
FIG. 2 illustrates a signal-to-noise ratio comparison of a geophysical survey signal using one embodiment of the present invention and using the prior art.
Detailed Description
Preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Example 1
FIG. 1 shows a flow chart of a method of denoising geophysical survey data according to one embodiment of the present invention, which includes steps 1 through 5.
Step 1, generating a time domain training data set on a cloud platform, wherein the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence.
In some implementations, a time-domain noiseless geophysical survey data sequence in a time-domain training data set may be generated using a wavelet basis. For example, different time-domain noiseless geophysical survey data sequences may be generated using the following wavelet basis: biorthogonal (bio), daubechies (db), symlets (sym), coiflets (coif), meyer (meyr), anti-biorthogonal (rbior), fejer-Korovkin (fk), and the like.
In some embodiments, a fast fractional difference algorithm based on fourier transforms is employed to generate the noise sequence. Different noise sequences may be generated by adjusting parameters of the fast fractional difference algorithm.
The time-domain noiseless geophysical prospecting data sequence and the noise sequence may be added to obtain a time-domain noiseless geophysical prospecting data sequence.
The same time domain noiseless geophysical exploration data sequence is added with different noise sequences, so that different time domain noiseless geophysical exploration data sequences can be obtained; different time domain noiseless geophysical prospecting data sequences are added with the same noise sequence, and different time domain noiseless geophysical prospecting data sequences can be obtained.
The training dataset may include a plurality of time-domain noisy geophysical survey data sequences and a time-domain noisy geophysical survey data sequence corresponding to each time-domain noisy geophysical survey data sequence.
Step 2, a data conversion unit is established on a cloud platform, and the data conversion unit is called to respectively convert the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence based on the following steps of obtaining a time-frequency-domain two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence:
，
，
wherein,representing samples in a two-dimensional noiseless geophysical survey data sequence,representing two-dimensional noisy geophysical surveysSamples in the probe data sequence, N is the number of samples in a sliding rectangular window used to select samples of the time domain geophysical survey data sequence in the transform,representing an nth sample of the time domain noiseless geophysical survey data sequence in the current rectangular window,representing an nth sample of the time domain noisy geophysical survey data sequence in a current rectangular window, traversing the entire time domain by sliding the rectangular window.
Through the formula, geophysical exploration data can be transformed into a two-dimensional time-frequency domain, and the transformed sampling points are still real numbers, so that a corresponding deep learning network is conveniently designed, and the prediction performance of the deep learning network is improved.
In some embodiments, the sliding rectangular window used in the data transformation has a size ranging from 32, 64, 128, and 256 samples, and a sliding overlap ratio ranging from 70% to 95%. For example, in one example, the rectangular window has a size of n=128 and an overlap ratio of 85%.
Step 3, segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, wherein each data segment comprises a plurality of sample points, and establishing the following objective function:
，
wherein,for the number of data segments to be divided,for the mth two-dimensional noiseless geophysical survey data segment,for the mth two-dimensional noisy geophysical prospectionThe data field is used for the data field,for defining the length of the estimated time window,representing the mapping function.
The inventors found in the study that when deep learning network training is performed using two-dimensional time-frequency data, the training set was too large. Because there are infinite numbers of spectrograms of geophysical prospecting signal amplitude and positional variations in the time-frequency domain, the training set needs to be rich enough to cover all the possibilities, which is further exacerbated by random noise. To solve this problem, the inventors creatively propose to use the data segment instead of the entire time-frequency domain for training, so that the training set can achieve sufficient training with less data.
Since geophysical prospecting data can be written:
,
wherein,representing a noiseless geophysical prospecting signal,the noise is represented by a characteristic of the noise,representing a received/recorded sequence of geophysical survey data.
Based on the above equation, the following objective function can be established:
wherein,which represents the desired effective signal(s),the signal representing the recording is represented by a signal,the mapping function is represented as a function of the mapping,the square of the L2 norm is shown.
In order to enhance the reliability of the prediction, the inventors modified the above-mentioned commonly employed objective function to establish the following objective function:
，
wherein,for the number of data segments to be divided,for the mth two-dimensional noiseless geophysical survey data segment,for the mth two-dimensional noisy geophysical survey data segment,for defining the length of the estimated time window,representing the mapping function.
According to the invention, when denoising the current data, the current data is denoisedFrequency spectrum within time window、、……、And futureFrequency spectrum within time window、、……、Current spectrumAre taken into consideration to give the current spectrumDenoising, and its predictive performance is significantly better than the prior art.
In some embodiments, a method for defining an estimated time windowThe range of the value of (2) is 3-10. In one example of this, in one implementation,7 may be taken.
Step 4, on the cloud platform, two-dimensional noisy geophysical exploration data segmentsAs input, two-dimensional noiseless geophysical prospecting data segmentAs a desired output to train a deep learning network learning mapping function。
In the prior art, correlation coefficients may typically be calculated and compared to a threshold to separate the signal from noise, a disadvantage of which has been described above. The inventor of the invention adopts a deep learning network to learn the mapping function f, does not need to search the optimal separation threshold value, and can obtain better prediction performance.
In some embodiments, the deep learning network is a Deep Supervision Network (DSN). Through intensive research, the inventor considers that the deep supervision network is adopted in the invention to match the data transformation algorithm in the step 2, and compared with a Convolutional Neural Network (CNN) which is more commonly adopted, the deep supervision network can obtain better denoising effect.
In some embodiments, the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 maximum pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 maximum pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 fully connected layers and 1 classification layer, wherein the first auxiliary supervision layer and the second auxiliary supervision layer comprise 1 fully connected layer and one output layer, the training process of the whole deep supervision network adopts a back propagation algorithm, and a random gradient descent optimizer is adopted for weight updating.
This is a neural network structure specifically designed by the inventors based on the other parts of the present invention. With n=128 in step 2, in step 3An example is described. The input is 15 transformed data segments with dimensions 128x15x1. Feature extraction is performed by two convolution layers, each layer may use a 3×3 filter, may include 64 channels, and each convolution layer is followed by Batch Normalization (BN) and leaky linear rectification unit (LeakyReLU) to activate and stabilize the training process. Followed by a maximum pooling layer to reduce the spatial dimension. In the middle part of the network a first auxiliary supervisory layer is introduced, which may comprise a fully connected layer with 128 units and an output layer with 128 units, the presence of this layer being aimed at providing additional supervisory signals to guide the learning process of the network. Next, the two convolutional layers, each using a 3 x 3 filter, were continued with the channel number maintained at 128 and Batch Normalization (BN) and LeakyReLU activation. Followed by a maximumThe layers are pooled to further reduce the spatial dimension. In the latter half of the network, a second auxiliary supervisory layer is introduced, similar in structure to the first auxiliary supervisory layer, also comprising a fully connected layer having 128 units and an output layer having 128 units. Followed by a global averaging pooling layer for aggregating spatial information. Following two fully connected layers, containing 256 and 128 units, respectively, and Batch Normalization (BN) and LeakyReLU activation were performed. The final classification layer contains 128 units for outputting the final predictions of the network.
The training process of the whole network can use a back propagation algorithm and adopts optimizers such as random gradient descent and the like to update the weight. The final output is a tensor with a dimension of 1x1x128, representing the predicted characteristics of the network to the input, i.e., the denoised data segment.
The number of samples selected per iteration may be 512 in size. After each iteration, the gradient penalty may be calculated. When the gradient loss is not reduced after successive rounds of iteration, then termination of training may be considered. For example, the training may be terminated after 6 consecutive iterations without any reduction in gradient loss.
After the training is terminated, the trained deep learning network may be validated. For example, a time-domain noisy geophysical prospecting data sequence which does not participate in the training may be selected from the time-domain training data set, subjected to data transformation to obtain a time-frequency-domain two-dimensional noisy geophysical prospecting data sequence, segmented to serve as an input of the deep learning network, and the output of the deep learning network is compared with the expected output to verify the training effect of the deep learning network.
And 5, uploading the actual geophysical exploration record to a cloud platform, converting the actual geophysical exploration record into a two-dimensional sequence by adopting the data conversion unit, segmenting the two-dimensional sequence, and then sending the obtained data segment into a trained deep learning network for denoising.
And (2) carrying out inverse transformation processing corresponding to the data transformation in the step (2) on the output of the deep neural network, and converting the two-dimensional data into one dimension to obtain a geophysical exploration data sequence after time domain denoising.
According to the embodiment, the noise in the same frequency as the effective geophysical prospecting signal can be eliminated by transforming the one-dimensional time domain signal to the two-dimensional time-frequency domain, and the noise is removed through the deep learning network, so that the best threshold for separating the signal and the noise is not needed to be sought like the prior art, and the embodiment is implemented on a cloud platform, so that the local equipment cost is remarkably saved, the training process is faster and more efficient, mass data is conveniently stored, and team collaborative development is convenient. In particular, the embodiment constructs a specific data transformation unit, which is favorable for improving the prediction performance and training convergence speed of the subsequent deep learning network, and is particularly suitable for being matched with a specific deep supervision network. The embodiment also can reduce the training data amount and improve the network stability by dividing the sequence into a plurality of data segments for training. The present embodiment also takes into account not only the current spectrum but also the spectrum of the past and future time periods when constructing the objective function, and its prediction performance is significantly better than that of the prior art.
Example 2
An embodiment of the invention discloses a geophysical prospecting data denoising device, which comprises a training data generating unit, a data transformation unit, a segmentation and objective function constructing unit, a deep learning network training unit and a comprehensive calling denoising unit.
The training data generation unit is used for generating a time domain training data set on the cloud platform, wherein the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence.
The data transformation unit is used for respectively transforming the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence on the cloud platform based on the following steps of:
，
，
wherein,representing samples in a two-dimensional noiseless geophysical survey data sequence,representing samples in a two-dimensional noisy geophysical survey data sequence, N being the number of samples in a sliding rectangular window used in the transform to select samples of the time domain geophysical survey data sequence,representing an nth sample of the time domain noiseless geophysical survey data sequence in the current rectangular window,representing an nth sample of the time domain noisy geophysical survey data sequence in a current rectangular window, traversing the entire time domain by sliding the rectangular window.
The system comprises a segmentation and objective function construction unit, a segmentation and objective function construction unit and a control unit, wherein the segmentation and objective function construction unit is used for segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, each data segment comprises a plurality of sample points, and the following objective function is established:
，
wherein,for the number of data segments to be divided,for the mth two-dimensional noiseless geophysical survey data segment,for the mth two-dimensionalA segment of noisy geophysical survey data,for defining the length of the estimated time window,representing the mapping function.
Deep learning network training unit for training two-dimensional noisy geophysical exploration data segment on cloud platformAs input, two-dimensional noiseless geophysical prospecting data segmentAs a desired output to train a deep learning network learning mapping function。
The comprehensive calling denoising unit is used for calling the data transformation unit to transform the actual geophysical exploration record uploaded by the cloud platform into a two-dimensional sequence, calling the segmentation and objective function construction unit to segment the two-dimensional sequence, and then denoising the obtained data segment through the trained deep learning network.
In some embodiments, in the training data generation unit, specifically:
generating a time domain noiseless geophysical prospecting data sequence in a time domain training data set by using a wavelet base;
generating a noise sequence by adopting a fast fractional difference algorithm based on Fourier transform;
the generated time-domain noise-free geophysical survey data sequence and the generated noise sequence are added to obtain a time-domain noisy geophysical survey data sequence.
In some embodiments, the deep learning network is a Deep Supervision Network (DSN).
In some embodiments, the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 maximum pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 maximum pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 fully connected layers and 1 classification layer, wherein the first auxiliary supervision layer and the second auxiliary supervision layer comprise 1 fully connected layer and one output layer, the training process of the whole deep supervision network adopts a back propagation algorithm, and a random gradient descent optimizer is adopted for weight updating.
Further details and advantages of the present embodiment may be referred to the corresponding descriptions in the foregoing embodiments, and are not repeated here.
Example 3
According to another aspect of the present invention, there is also provided an electronic apparatus. The electronic device includes:
a memory storing executable instructions:
a processor executing the executable instructions in the memory to implement a geophysical survey data denoising method according to the present invention.
In particular, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, random Access Memory (RAM) and/or cache memory (cache), and the like. The non-volatile memory may include, for example, read Only Memory (ROM), hard disk, flash memory, and the like.
The processor may be a Central Processing Unit (CPU) or other form of processing unit having data processing and/or instruction execution capabilities, and may control other components in the electronic device to perform the desired functions. In one embodiment of the invention, the processor is configured to execute the computer readable instructions stored in the memory.
The detailed description of the present embodiment may refer to the corresponding description in the foregoing embodiments, and will not be repeated herein.
Example 4
According to another aspect of the invention, there is also provided a computer readable storage medium storing a computer program which, when executed by a processor, implements a geophysical survey data denoising method according to the invention.
A computer-readable storage medium according to an embodiment of the present invention has stored thereon non-transitory computer-readable instructions. When executed by a processor, perform all or part of the steps of the methods of embodiments of the invention described above.
The computer-readable storage medium described above includes, but is not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or removable hard disk), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).
It should be understood by those skilled in the art that, in order to solve the technical problem of how to obtain a good user experience effect, the present embodiment may also include well-known structures such as a communication bus, an interface, and the like, and these well-known structures are also included in the protection scope of the present invention.
The detailed description of the present embodiment may refer to the corresponding description in the foregoing embodiments, and will not be repeated herein.
Example 5
A specific example of the present invention is given below to explain the advantageous effects of the present invention.
In this example, the size of the sliding rectangular window used for data transformation is taken as n=128, and the size is taken as an objective functionFor illustration, the input to the deep supervisory network is 15 data segments, with dimensions 128x15x1. Feature extraction is first performed by two convolutional layers, each of which may use a 3 x 3 filter, which may contain 64 channels. Batch Normalization (BN) and leakage linear rectification units (LeakyReLU) are used to activate and stabilize the training process after each convolution layer. Followed by a maximum pooling layer to reduce the spatial dimension. In the middle part of the network, a first auxiliary supervisory layer is introduced, which comprises a fully connected layer with 128 units and an output layer with 128 units to provide additional supervisory signals for guiding the learning process of the network. Next, the two convolutional layers, each using a 3 x 3 filter, were continued with the channel number maintained at 128 and Batch Normalization (BN) and LeakyReLU activation. Followed by a maximum pooling layer to further reduce the spatial dimension. In the latter half of the network, a second auxiliary supervisory layer is introduced, similar in structure to the first auxiliary supervisory layer, also comprising a fully connected layer having 128 units and an output layer having 128 units. Followed by a global averaging pooling layer for aggregating spatial information. Following two fully connected layers, containing 256 and 128 units, respectively, and Batch Normalization (BN) and LeakyReLU activation were performed. The final classification layer contains 128 units, outputting the final predictions of the network.
The training process of the whole network uses a back propagation algorithm, and adopts optimizers such as random gradient descent and the like to update the weight. The final output is a tensor with a dimension of 1x1x128, representing the predicted characteristics of the network to the input, i.e., the denoised data segment.
FIG. 2 illustrates a signal-to-noise ratio comparison of a geophysical survey signal using one embodiment of the present invention and using the prior art. The ordinate in fig. 2 represents the signal-to-noise ratio after denoising in dB (decibel), and the abscissa represents the record number. The prior art by contrast is a denoising scheme based on convolutional neural networks, which does not employ the data conversion method as the present invention, nor does it divide the sequence into data segments, which employ an objective function in a more common form than the present invention. As can be seen from fig. 2, the effect of implementing the present invention is significantly better than that of the prior art.
In summary, the beneficial effects of the embodiments of the present invention at least include:
(1) The one-dimensional time domain signal is transformed into a two-dimensional time-frequency domain, so that the time domain information and the frequency domain information of the signal can be simultaneously represented;
(2) When the signal and the noise share the frequency band, the signal and the noise are difficult to distinguish in the time domain, but after the signal and the noise are converted into the time-frequency domain, the signal and the noise can be more easily distinguished due to different time-frequency distribution;
(3) The time-frequency representation improves the sparsity of the signals, and is beneficial to the sparse representation of the signals learned by the denoising method;
(4) The constructed objective function denoises the current data segment through past, future and current data segments, so that the denoising effect is improved;
(5) Noise elimination is realized through training the deep learning network, and an optimal separation threshold value of signals and noise is not required to be found;
(6) Training the deep learning network to use the data segment instead of the whole frequency spectrum to infer the noise-free signal of interest can significantly reduce the amount of training data and improve the stability of the network;
(7) Matching with a deep learning network through specific time-frequency transformation, so that the prediction performance and training convergence are effectively improved;
(8) The geophysical exploration data denoising is realized on the cloud platform, expensive equipment is not required to be purchased locally, the training efficiency is high, and the cooperation is convenient.
The foregoing description of embodiments of the invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the technical improvement of the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
Claims (10)
1. A method of denoising geophysical survey data, the method comprising:
step 1, generating a time domain training data set on a cloud platform, wherein the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence;
step 2, a data transformation unit is established on a cloud platform, and the data transformation unit is called to respectively transform the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence based on the following steps of obtaining a time-frequency-domain two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence:
，
，
wherein,representing sample points in a two-dimensional noiseless geophysical prospecting data sequence, < >>Representing samples in a two-dimensional noisy geophysical survey data sequence, N being the number of samples in a sliding rectangular window used in the transform to select samples of the time domain geophysical survey data sequence,/>An nth sample representing a time domain noiseless geophysical survey data sequence in a current rectangular window,/->Representing an nth sample point of the time domain noisy geophysical prospecting data sequence in a current rectangular window, traversing the whole time domain by sliding the rectangular window;
step 3, segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, wherein each data segment comprises a plurality of sample points, and establishing the following objective function:
，
wherein,for the number of divided data segments, +.>For the mth two-dimensional noiseless geophysical prospecting data segment,>for the mth two-dimensional noisy geophysical prospecting data segment,>for defining the length of the estimated time window, +.>Representing a mapping function;
step 4, on the cloud platform, two-dimensional noisy geophysical exploration data segmentsAs input, two-dimensional noiseless geophysical prospecting data segment +.>As a desired output to train a deep learning network learning mapping function +.>：
And 5, uploading the actual geophysical exploration record to a cloud platform, converting the actual geophysical exploration record into a two-dimensional sequence by adopting the data conversion unit, segmenting the two-dimensional sequence, and then sending the obtained data segment into a trained deep learning network for denoising.
2. The method according to claim 1, characterized in that in step 1, in particular:
generating a time domain noiseless geophysical prospecting data sequence in a time domain training data set by using a wavelet base;
generating a noise sequence by adopting a fast fractional difference algorithm based on Fourier transform;
the generated time-domain noise-free geophysical survey data sequence and the generated noise sequence are added to obtain a time-domain noisy geophysical survey data sequence.
3. The method according to claim 1, wherein in step 2, the rectangular window has a size ranging from 32, 64, 128 and 256 samples, and the sliding overlap ratio has a size ranging from 70% to 95%.
4. The method according to claim 1, characterized in that in step 3, a time window for estimation is definedThe range of the value of (2) is 3-10.
5. The method of claim 1, wherein in step 4, the deep learning network is a Deep Supervision Network (DSN).
6. The method according to claim 5, characterized in that the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 max-pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 max-pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 fully-connected layers and 1 classification layer, the first auxiliary supervision layer and the second auxiliary supervision layer each comprise 1 fully-connected layer and one output layer, the training process of the whole deep supervision network adopts a back propagation algorithm, and a random gradient descent optimizer is adopted for weight updating.
7. A geophysical prospecting data denoising apparatus, comprising:
the system comprises a training data generation unit, a data processing unit and a data processing unit, wherein the training data generation unit is used for generating a time domain training data set on a cloud platform, and the time domain training data set comprises a time domain noisy geophysical exploration data sequence and a corresponding time domain noiseless geophysical exploration data sequence;
the data transformation unit is used for respectively transforming the time-domain noisy geophysical exploration data sequence and the time-domain noiseless geophysical exploration data sequence on the cloud platform based on the following steps of:
，
，
wherein,representing sample points in a two-dimensional noiseless geophysical prospecting data sequence, < >>Representing samples in a two-dimensional noisy geophysical survey data sequence, N being the number of samples in a sliding rectangular window used in the transform to select samples of the time domain geophysical survey data sequence,/>An nth sample representing a time domain noiseless geophysical survey data sequence in a current rectangular window,/->Representing an nth sample of a time-domain noisy geophysical survey data sequence in a current rectangular windowTraversing the entire time domain by sliding a rectangular window;
the system comprises a segmentation and objective function construction unit, a segmentation and objective function construction unit and a control unit, wherein the segmentation and objective function construction unit is used for segmenting a two-dimensional noisy geophysical exploration data sequence and a two-dimensional noiseless geophysical exploration data sequence on a cloud platform, each data segment comprises a plurality of sample points, and the following objective function is established:
，
wherein,for the number of divided data segments, +.>For the mth two-dimensional noiseless geophysical prospecting data segment,>for the mth two-dimensional noisy geophysical prospecting data segment,>for defining the length of the estimated time window, +.>Representing a mapping function;
deep learning network training unit for training two-dimensional noisy geophysical exploration data segment on cloud platformAs input, two-dimensional noiseless geophysical prospecting data segment +.>As a desired output to train a deep learning network learning mapping function +.>：
The comprehensive calling denoising unit is used for calling the data transformation unit to transform the actual geophysical exploration record uploaded to the cloud platform into a two-dimensional sequence, calling the segmentation and objective function construction unit to segment the two-dimensional sequence, and then sending the obtained data segment into the trained deep learning network for denoising.
8. The apparatus according to claim 7, characterized in that in the training data generation unit, in particular:
generating a time domain noiseless geophysical prospecting data sequence in a time domain training data set by using a wavelet base;
generating a noise sequence by adopting a fast fractional difference algorithm based on Fourier transform;
the generated time-domain noise-free geophysical survey data sequence and the generated noise sequence are added to obtain a time-domain noisy geophysical survey data sequence.
9. The apparatus according to claim 7, wherein the deep learning network is a Deep Supervision Network (DSN), in particular:
the deep supervision network is designed to sequentially comprise 2 convolution layers, 1 maximum pooling layer, a first auxiliary supervision layer, 2 convolution layers, 1 maximum pooling layer, a second auxiliary supervision layer, a global average pooling layer, 2 full connection layers and 1 classification layer, wherein the first auxiliary supervision layer and the second auxiliary supervision layer comprise 1 full connection layer and one output layer, a reverse propagation algorithm is adopted in the training process of the whole deep supervision network, and a random gradient descent optimizer is adopted for weight updating.
10. An electronic device, the electronic device comprising:
a memory storing executable instructions:
a processor executing the executable instructions in the memory to implement the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311848272.2A CN117493776B (en) | 2023-12-29 | 2023-12-29 | Geophysical exploration data denoising method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311848272.2A CN117493776B (en) | 2023-12-29 | 2023-12-29 | Geophysical exploration data denoising method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117493776A CN117493776A (en) | 2024-02-02 |
| CN117493776B true CN117493776B (en) | 2024-03-01 |
Family
ID=89680430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311848272.2A Active CN117493776B (en) | 2023-12-29 | 2023-12-29 | Geophysical exploration data denoising method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117493776B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101354885A (en) * | 2007-01-16 | 2009-01-28 | 哈曼贝克自动***股份有限公司 | Active noise control system |
| CN107194895A (en) * | 2017-05-27 | 2017-09-22 | 上海海洋大学 | A kind of safely outsourced fusion denoising method for multiframe remote sensing images |
| CN110353656A (en) * | 2019-07-12 | 2019-10-22 | 东南大学 | A kind of wearable ECG monitor system and its monitoring method based on cloud framework |
| CN116009065A (en) * | 2021-10-21 | 2023-04-25 | 中国石油化工股份有限公司 | Seismic data noise removing method, electronic equipment and medium |
| CN116165713A (en) * | 2021-11-24 | 2023-05-26 | 中国石油天然气集团有限公司 | Method and device for denoising seismic data of DAS in well |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220015711A1 (en) * | 2020-07-20 | 2022-01-20 | Board Of Regents, The University Of Texas System | System and method for automated analysis and detection of cardiac arrhythmias from electrocardiograms |
| WO2022193327A1 (en) * | 2021-03-19 | 2022-09-22 | 深圳市韶音科技有限公司 | Signal processing system, method and apparatus, and storage medium |
-
2023
- 2023-12-29 CN CN202311848272.2A patent/CN117493776B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101354885A (en) * | 2007-01-16 | 2009-01-28 | 哈曼贝克自动***股份有限公司 | Active noise control system |
| CN107194895A (en) * | 2017-05-27 | 2017-09-22 | 上海海洋大学 | A kind of safely outsourced fusion denoising method for multiframe remote sensing images |
| CN110353656A (en) * | 2019-07-12 | 2019-10-22 | 东南大学 | A kind of wearable ECG monitor system and its monitoring method based on cloud framework |
| CN116009065A (en) * | 2021-10-21 | 2023-04-25 | 中国石油化工股份有限公司 | Seismic data noise removing method, electronic equipment and medium |
| CN116165713A (en) * | 2021-11-24 | 2023-05-26 | 中国石油天然气集团有限公司 | Method and device for denoising seismic data of DAS in well |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117493776A (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598166B (en) | Wavelet denoising method for adaptively determining wavelet layering progression | |
| CN112735460B (en) | Beam forming method and system based on time-frequency masking value estimation | |
| CN109031422A (en) | A kind of seismic signal noise suppressing method based on CEEMDAN and Savitzky-Golay filtering | |
| CN113221781A (en) | Carrier signal detection method based on multitask deep convolutional neural network | |
| CN113392732B (en) | Partial discharge ultrasonic signal anti-interference method and system | |
| CN111399057B (en) | Seismic data noise suppression method based on non-convex sparse constraint | |
| Zhang et al. | Birdsoundsdenoising: Deep visual audio denoising for bird sounds | |
| Hidayat et al. | A Modified MFCC for Improved Wavelet-Based Denoising on Robust Speech Recognition. | |
| CN109765608B (en) | Coal seam roadway anchor rod vibration noise suppression method based on joint dictionary | |
| CN109658944B (en) | Helicopter acoustic signal enhancement method and device | |
| CN117493776B (en) | Geophysical exploration data denoising method and device and electronic equipment | |
| CN117574062A (en) | Small loop transient electromagnetic signal denoising method based on VMD-DNN model | |
| Ou et al. | Frame-based time-scale filters for underwater acoustic noise reduction | |
| WO2016197629A1 (en) | System and method for frequency estimation | |
| CN113341463B (en) | Non-stationary blind deconvolution method for pre-stack seismic data and related components | |
| CN102509268B (en) | Immune-clonal-selection-based nonsubsampled contourlet domain image denoising method | |
| CN112363217A (en) | Random noise suppression method and system for seismic data | |
| CN114358040A (en) | Disturbing signal denoising method based on adaptive estimation threshold method | |
| CN112578439B (en) | Seismic inversion method based on space constraint | |
| Geetha et al. | Microseismic signal denoising based on variational mode decomposition with adaptive non-local means filtering | |
| CN113009564A (en) | Seismic data processing method and device | |
| Tanwar et al. | Hard component detection of transient noise and its removal using empirical mode decomposition and wavelet‐based predictive filter | |
| Anderson et al. | Joint deconvolution and classification with applications to passive acoustic underwater multipath | |
| CN116705049A (en) | Underwater acoustic signal enhancement method and device, electronic equipment and storage medium | |
| CN111354372A (en) | Audio scene classification method and system based on front-end and back-end joint training |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |