CN110221346B - Data noise suppression method based on residual block full convolution neural network - Google Patents
Data noise suppression method based on residual block full convolution neural network Download PDFInfo
- Publication number
- CN110221346B CN110221346B CN201910608174.9A CN201910608174A CN110221346B CN 110221346 B CN110221346 B CN 110221346B CN 201910608174 A CN201910608174 A CN 201910608174A CN 110221346 B CN110221346 B CN 110221346B
- Authority
- CN
- China
- Prior art keywords
- residual blocks
- noise
- seismic data
- network
- data
- 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
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000001629 suppression Effects 0.000 title claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims description 12
- 230000009977 dual effect Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 4
- 230000029087 digestion Effects 0.000 abstract description 2
- 239000000284 extract Substances 0.000 abstract 1
- 238000013527 convolutional neural network Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000009466 transformation Effects 0.000 description 3
- 230000005764 inhibitory process Effects 0.000 description 2
- 230000016776 visual perception Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Images
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
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/30—Noise handling
- G01V2210/32—Noise reduction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geology (AREA)
- Environmental & Geological Engineering (AREA)
- Acoustics & Sound (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Noise Elimination (AREA)
- Error Detection And Correction (AREA)
Abstract
The invention discloses a data noise suppression method based on a residual block full convolution neural network. The training set and the test set for suppressing the seismic noise by applying the deep learning method are both from the same data set, so that the generalization of the model is limited. In order to solve the generalization problem, the design idea of the network structure is to fuse a double residual block on the basis of the Unet network so as to enhance the capture capability of the network on random noise. The invention is established on an end-to-end coding-decoding network structure, takes noisy seismic data as input, and extracts essential characteristics of random noise by a plurality of convolution layers and residual blocks to form codes; and then a plurality of deconvolution layers and residual blocks form decoding, and the output of the network is the seismic data after noise suppression. Compared with the existing seismic data denoising method, the method has the advantages of obvious generalization, effective suppression of random noise and effective signal protection due to the fact that the double residual blocks are fused to carry out secondary digestion learning on the extracted random noise features and the intrinsic features of the noise are more fully learned.
Description
Technical Field
The invention relates to the technical field of data noise suppression, in particular to earthquake random noise suppression.
Background
The traditional seismic data denoising method comprises f-k domain filtering, f-x domain denoising, wavelet transformation, curvelet transformation, discrete cosine transformation and the like. The method is widely applied to the field of seismic data denoising, but the problems of insufficient denoising capability, effective signal damage and the like still exist.
In recent years, with the development of deep learning techniques, researchers have proposed a method for denoising seismic data by using the deep learning technique. Different from the traditional denoising method, the deep learning belongs to the category of statistical learning, and the statistical learning can be used for learning the essential characteristics of effective signals and noise according to a noise sample and fitting to obtain a model capable of classifying the effective signals and the noise. Due to the advantage of statistical learning, the documents ' Wangyuqing, Luwenkhaki, Liujin forest, and the like, ' seismic random noise suppression [ J ] based on data augmentation and CNN, geophysical newspaper, 2019,62(1):421 and 433 ' propose a method for random noise suppression of seismic data based on a Unet convolution neural network and obtain a certain effect, but the problem is that a training set and a testing set in a post-stack seismic data testing experiment are from the same data set, and the generalization of a model is limited. The document "Ma J. deep Learning for orienting Random and Coherence Noise Simultaneous [ C ]/80 th EAGE Conference and inhibition 2018.2018" proposes that the method for denoising seismic data based on DnCNN can suppress Gaussian Noise, but the data sets used for training and testing are synthetic data, and do not necessarily achieve good Noise suppression effect in the actual seismic data set. The document "Liu J, Lu W, Zhang P. Random Noise extension Using volumetric Neural Networks [ C ]//80th EAGE Conference and inhibition 2018.2018" proposes a seismic Random Noise suppression method based on the Unet Convolutional Neural network, but the data sets are all from synthetic data, and the generalization is limited in the actual seismic data.
Disclosure of Invention
A data noise suppression method based on a residual block full convolution neural network is characterized by comprising the following steps:
step 1: making a training set and a test set of seismic data, taking the seismic data with different noise levels and containing noise levels and the seismic data without noise as the training set, and taking the other seismic data different from the data of the training set as the test set;
step 2: designing an end-to-end network structure for coding and decoding, wherein the coding process is composed of 5 groups of double residual blocks with different scales, wherein 4 groups of residual blocks are composed of 5 convolution layers and 1 pooling layer, 1 group of residual blocks are composed of 5 convolution layers, the decoding process is symmetrical to the coding process and is composed of 4 groups of double residual blocks with different scales, wherein 3 groups of residual blocks are composed of 1 deconvolution layer and 5 convolution layers, and 1 group of residual blocks is composed of 1 deconvolution layer and 6 convolution layers and is fused with the characteristics extracted in the corresponding coding stage;
and step 3: training a network and storing a network model;
and 4, step 4: adjusting parameters and selecting a final ideal model;
and 5: and (3) suppressing the random noise of the data by using the ideal denoising model obtained by training, and outputting the seismic data after the random noise suppression.
The method for suppressing data noise based on the full convolution neural network of residual blocks as claimed in claim 1, wherein the 5 sets of dual residual blocks of the encoding part in step 2 have a scale size of 256 × 256, 128 × 128, 64 × 64, 32 × 32, 16 × 16, respectively, and the 4 sets of dual residual blocks of the decoding part have a scale size of 32 × 32, 64 × 64, 128 × 128, 256 × 256, respectively.
Compared with the prior seismic data denoising method, the method has the following advantages:
(1) the RUnet (namely, the residual block full convolution neural network) can not only effectively suppress random noise, but also protect effective signals;
(2) compared with the Unet convolution neural network, the innovation point of the invention is that the extracted random noise features are subjected to secondary digestion learning due to the fusion of the residual block, and the intrinsic features of the noise are more fully learned, so that the method has obvious advantages in generalization.
Drawings
FIG. 1 is a flow chart of the random noise suppression of seismic data according to the present invention, in which noisy seismic data with different noise levels are used as a training set of a training model, a RUnet network is built, the training model is selected, an ideal model is selected according to a peak signal-to-noise ratio and a signal-to-noise ratio index, a test set is input into the ideal model, and an output is data after the random noise suppression of the seismic;
FIG. 2 is a diagram of a fused double residual block structure in a network structure of the present invention, xi-2An input representing a residual block; f (x)i-2) Representing the noise characteristics extracted by the two convolutional layers; x is the number ofiRepresenting the output characteristic value of the residual block, i.e. xi＝f(xi-2)+xi-2(ii) a CONV denotes a convolutional layer.
FIG. 3 is a diagram of an original Unet convolutional neural network structure;
FIG. 4 is a diagram of a RUnet convolutional neural network structure of the present invention, fusing a dual residual block on the basic structure of the original Unet convolutional neural network to perform secondary feature learning on noise features, wherein the dotted line part in the diagram is the dual residual block;
FIG. 5 is an example of denoising implemented in the present invention: a is original data of a training set; b is training set noise data; c is the denoising result of the RUnet in the same data set; d is noise data of the test set; e is the test result of the RUnet in different data sets; f is the noise removed by RUnet.
Detailed Description
In order to effectively remove random noise in seismic data, the RUnet convolutional neural network denoising model is provided, and comprises the following steps.
Step 1: taking the sum of the seismic data containing noise with different noise levels and the preprocessed three-dimensional post-stack seismic data as a training set, and specifically comprising the following steps of:
(1) selecting 256 seismic data slices from the Parihaka post-stack three-dimensional seismic data volume, wherein sampling points are 256;
(2) respectively adding 20%, 25% and 30% of Gaussian random noise to seismic data, and taking the Gaussian random noise and corresponding preprocessed seismic data together as a training set, wherein the noise-added seismic data is used as input, the preprocessed seismic data is used as a label, and the sample size is 900;
step 2: the network as a whole comprises an encoding process and a decoding process. The encoding process is composed of 5 groups of residual blocks, each group of residual blocks is composed of 5 convolutional layers and 1 pooling layer, input data of [256 × 256] dimension is encoded into [16 × 16] dimension characteristic information, the size of a convolutional kernel is set to be 3 × 3, and the step size is set to be 1. And each time of the residual block operation, the size of the feature map is compressed to 1/2 of the last operation, and correspondingly, the number of channels of the feature map is 2 times of that of the last residual block operation, so that the feature information is ensured not to be lost. The feature decoding process is composed of 4 sets of residual blocks, each of which is composed of 1 deconvolution layer and 5 convolution layers, and upsamples the [16 × 16] dimensional feature information generated by the encoding process into [256 × 256] dimensional output data. In contrast to the encoding process, each time a residual operation is performed, the size of the feature map is up-sampled by 2 times of the previous residual operation, and the number of channels of the feature map becomes 1/2 of the previous residual operation. And adding the feature map of the corresponding position of the coding part into the feature map of the decoding part to fuse the feature information of different scales. The final output is completed by a convolution layer with convolution kernel size of 1 × 1 and step length of 1 and an activation function tanh, and the function of the layer is similar to that of a full connection layer;
and step 3: inputting the training set obtained in the step 1 into the network model built in the step 2 through a queue, measuring the distance between a real value and a label value output by the network by adopting error back propagation and using a mean square error loss function, adjusting the weight between neurons by using a random gradient descent algorithm to minimize the loss function, judging the network denoising effect through quantitative peak signal-to-noise ratio, signal-to-noise ratio and qualitative visual perception, and storing each parameter of the network model after obtaining the optimal denoising effect;
the mean square error formula is:
where y is the true value of the network output,
the smaller the mean square error is, the closer the true value and the predicted value are represented, and the better the learning effect of the network on the training set is;
and 4, step 4: selecting a plurality of seismic data samples from the same three-dimensional data volume as a test set, inputting the seismic data samples into the network model obtained in the step 3, judging the network denoising effect through quantitative peak signal-to-noise ratio, signal-to-noise ratio and qualitative visual perception, returning to the step 3 to continue training or adjusting parameters from a new training network if the network denoising effect does not meet the requirements until the network model meets the requirements, and storing the final ideal network model;
and 5: and removing noise of the Kerry seismic data volume by using the stored network model, and outputting the noise-removed seismic data.
Claims (2)
1. A data noise suppression method based on a residual block full convolution neural network is characterized by comprising the following steps:
step 1: making a training set and a test set of seismic data, taking the seismic data with different noise levels and containing noise and the seismic data without noise as the training set, and taking the other seismic data different from the data of the training set as the test set;
step 2: designing an end-to-end network structure for coding and decoding, wherein a coding process is composed of 5 groups of double residual blocks with different scales, wherein the first 4 groups of residual blocks are composed of 5 convolutional layers and 1 pooling layer, the 5 th group of residual blocks are composed of 5 convolutional layers, a decoding process is symmetrical to the coding process and composed of 4 groups of double residual blocks with different scales, wherein the first 3 groups of residual blocks are composed of 1 anti-convolutional layer and 5 convolutional layers, the last 1 group of residual blocks are composed of 1 anti-convolutional layer and 6 convolutional layers and are fused with characteristics extracted in a corresponding coding stage, the last 4 convolutional layers of the first 8 groups of residual blocks are in a form of connecting two double residual blocks in series, and the middle 4 convolutional layers of the 6 convolutional layers of the last 1 group of residual blocks are in a form of connecting two double residual blocks in series;
and step 3: training a network and storing a network model;
and 4, step 4: adjusting parameters and selecting a final ideal model;
and 5: and (3) suppressing the random noise of the data by using the ideal denoising model obtained by training, and outputting the seismic data after the random noise suppression.
2. The method as claimed in claim 1, wherein the 5 sets of dual residual blocks of the encoding part in step 2 have a scale size of 256 × 256, 128 × 128, 64 × 64, 32 × 32, 16 × 16, respectively, and the 4 sets of dual residual blocks of the decoding part have a scale size of 32 × 32, 64 × 64, 128 × 128, 256 × 256, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910608174.9A CN110221346B (en) | 2019-07-08 | 2019-07-08 | Data noise suppression method based on residual block full convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910608174.9A CN110221346B (en) | 2019-07-08 | 2019-07-08 | Data noise suppression method based on residual block full convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110221346A CN110221346A (en) | 2019-09-10 |
| CN110221346B true CN110221346B (en) | 2021-03-09 |
Family
ID=67812768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910608174.9A Active CN110221346B (en) | 2019-07-08 | 2019-07-08 | Data noise suppression method based on residual block full convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110221346B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110719240B (en) * | 2019-09-30 | 2022-03-18 | 长沙理工大学 | Channel estimation method and system based on deconvolution and denoising convolution |
| CN110992301A (en) * | 2019-10-14 | 2020-04-10 | 数量级(上海)信息技术有限公司 | Gas contour identification method |
| CN113012050B (en) * | 2019-12-18 | 2024-05-24 | 武汉Tcl集团工业研究院有限公司 | Image processing method and device |
| CN113034373A (en) * | 2019-12-24 | 2021-06-25 | 无锡祥生医疗科技股份有限公司 | Ultrasonic image denoising neural network, denoising method and ultrasonic diagnostic equipment |
| CN111368710B (en) * | 2020-02-27 | 2022-03-04 | 东北石油大学 | Seismic data random noise suppression method combined with deep learning |
| CN111627082B (en) * | 2020-05-21 | 2022-06-21 | 浙江大学 | PET image reconstruction method based on filtering back projection algorithm and neural network |
| CN111580161A (en) * | 2020-05-21 | 2020-08-25 | 长江大学 | Earthquake random noise suppression method based on multi-scale convolution self-coding neural network |
| CN111796326B (en) * | 2020-07-07 | 2022-11-22 | 中海石油(中国)有限公司 | Method and system for constructing sequence stratum framework based on Unet network |
| CN111708087A (en) * | 2020-07-09 | 2020-09-25 | 中国科学技术大学 | Method for suppressing seismic data noise based on DnCNN neural network |
| CN111882543B (en) * | 2020-07-29 | 2023-12-26 | 南通大学 | Cigarette filter stick counting method based on AA R2Unet and HMM |
| CN113156513B (en) * | 2021-04-14 | 2024-01-30 | 吉林大学 | Convolutional neural network seismic signal denoising method based on attention guidance |
| CN113687414B (en) * | 2021-08-06 | 2022-07-22 | 北京大学 | Data-augmentation-based seismic interbed multiple suppression method for convolutional neural network |
| CN113850269B (en) * | 2021-12-01 | 2022-03-15 | 西南石油大学 | Denoising method based on multi-branch selective kernel nested connection residual error network |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3616198A4 (en) * | 2017-04-24 | 2021-01-06 | Virginia Tech Intellectual Properties, Inc. | Radio signal identification, identification system learning, and identifier deployment |
| CN108010021B (en) * | 2017-11-30 | 2021-12-10 | 上海联影医疗科技股份有限公司 | Medical image processing system and method |
| CN108389201B (en) * | 2018-03-16 | 2020-06-30 | 北京推想科技有限公司 | Lung nodule benign and malignant classification method based on 3D convolutional neural network and deep learning |
| CN108550125B (en) * | 2018-04-17 | 2021-07-30 | 南京大学 | Optical distortion correction method based on deep learning |
| CN109840471B (en) * | 2018-12-14 | 2023-04-14 | 天津大学 | Feasible road segmentation method based on improved Unet network model |
| CN109903292A (en) * | 2019-01-24 | 2019-06-18 | 西安交通大学 | A kind of three-dimensional image segmentation method and system based on full convolutional neural networks |
-
2019
- 2019-07-08 CN CN201910608174.9A patent/CN110221346B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110221346A (en) | 2019-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110221346B (en) | Data noise suppression method based on residual block full convolution neural network | |
| CN110322499B (en) | Monocular image depth estimation method based on multilayer characteristics | |
| CN110782462A (en) | Semantic segmentation method based on double-flow feature fusion | |
| CN110007347B (en) | Deep learning seismic data denoising method | |
| CN110020684A (en) | A kind of image de-noising method based on residual error convolution autoencoder network | |
| CN109146944B (en) | Visual depth estimation method based on depth separable convolutional neural network | |
| CN111340046A (en) | Visual saliency detection method based on feature pyramid network and channel attention | |
| CN109272499A (en) | Non-reference picture quality appraisement method based on convolution autoencoder network | |
| CN108549106B (en) | Aliasing noise drawing method and device | |
| CN103903261B (en) | Spectrum image processing method based on partition compressed sensing | |
| CN109460815B (en) | Monocular vision depth estimation method | |
| CN105607122B (en) | A kind of earthquake texture blending and Enhancement Method based on full variation geological data decomposition model | |
| CN113935467B (en) | DAS (data acquisition system) well exploration data noise suppression method based on iterative multi-scale attention network | |
| CN107240136A (en) | A kind of Still Image Compression Methods based on deep learning model | |
| Helal et al. | Seismic data compression using deep learning | |
| CN113687414A (en) | Data-amplification-based seismic interbed multiple suppression method for convolutional neural network | |
| CN105513048A (en) | Sub-band-information-entropy-measure-based image quality evaluation method | |
| CN109508639B (en) | Road scene semantic segmentation method based on multi-scale porous convolutional neural network | |
| CN114460648A (en) | 3D convolutional neural network-based self-supervision 3D seismic data random noise suppression method | |
| CN112818777B (en) | Remote sensing image target detection method based on dense connection and feature enhancement | |
| CN111505706B (en) | Microseism P wave first arrival pickup method and device based on deep T-Net network | |
| CN114200520B (en) | Seismic data denoising method | |
| CN111181574A (en) | End point detection method, device and equipment based on multi-layer feature fusion | |
| CN109857979A (en) | A kind of log approximating method based on wavelet analysis and BP neural network | |
| CN115393733A (en) | Water body automatic identification method and system based on deep learning |
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 |