CN115936990A - Synchronous processing method and system for multi-scale super-resolution and denoising of seismic data - Google Patents
Synchronous processing method and system for multi-scale super-resolution and denoising of seismic data Download PDFInfo
- Publication number
- CN115936990A CN115936990A CN202211586531.4A CN202211586531A CN115936990A CN 115936990 A CN115936990 A CN 115936990A CN 202211586531 A CN202211586531 A CN 202211586531A CN 115936990 A CN115936990 A CN 115936990A
- Authority
- CN
- China
- Prior art keywords
- conv
- resolution
- seismic
- relu
- network
- 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.)
- Granted
Links
- 230000001360 synchronised effect Effects 0.000 title claims description 12
- 238000003672 processing method Methods 0.000 title claims description 10
- 238000000034 method Methods 0.000 claims abstract description 24
- 238000005070 sampling Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000000007 visual effect Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 230000004913 activation Effects 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 claims description 2
- 238000010348 incorporation Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000011176 pooling Methods 0.000 claims description 2
- 238000005215 recombination Methods 0.000 claims description 2
- 230000006798 recombination Effects 0.000 claims description 2
- 230000006641 stabilisation Effects 0.000 claims description 2
- 238000011105 stabilization Methods 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 235000019587 texture Nutrition 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000019580 granularity Nutrition 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a method and a system for synchronously processing multi-scale super-resolution and denoising of seismic data, wherein the method comprises the following steps: carrying out down-sampling on the high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set; initializing a seismic reconstruction network, wherein the seismic reconstruction network comprises an autoencoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network; learning the seismic reconstruction network by using a multi-scale deep supervised learning strategy; and inputting the new seismic image into the trained seismic enhancement network to obtain a de-noised high-resolution seismic image and complete the reconstruction of seismic data. The invention has the following beneficial effects: the high-frequency information of the seismic image can be reserved, random noise is suppressed, geological structure information is effectively highlighted, the visual effect is more consistent with a human eye system, and the efficiency and accuracy of subsequent seismic interpretation can be effectively improved.
Description
Technical Field
The invention relates to the technical field of new generation information, in particular to a seismic data multi-scale super-resolution and denoising synchronous processing method and system.
Background
In the field of reservoir exploration, the geometrical shapes of fault and fracture networks play an important role in reservoir formation and migration of oil and gas, so that the fault and fracture networks in seismic images are accurately identified, and the method has important significance in finding oil and gas reservoirs. The traditional seismic image fault identification is carried out in an artificial mode, and a fault line is drawn on a seismic profile map manually by experienced geologists according to local characteristics of faults and the combination of the geological structure and the characteristics of a whole exploration area, so that a complete fault plane is obtained finally. The disadvantages of this method are also evident: the efficiency is low, and the identification accuracy rate greatly depends on the experience and knowledge of geological experts.
In recent years, with the continuous development of artificial intelligence technology, seismic image interpretation technology based on deep learning is also emerging, and a great deal of research shows that a model constructed based on a deep learning method can achieve good performance in the problem of seismic fault identification, but the seismic interpretation model is generally limited by low resolution and strong noise data. Due to the limitation of seismic acquisition and processing, the problems of low resolution and noise damage of field seismic data often exist, which brings huge challenges to subsequent seismic interpretation, so that super-resolution reconstruction and denoising are performed on the seismic data, and geological information with rich details is very necessary and worthy to recover.
Disclosure of Invention
The invention provides a seismic data multi-scale super-resolution and denoising synchronous processing method which can at least solve one of the technical problems.
In order to realize the purpose, the invention adopts the following technical scheme:
a method for synchronously processing multi-scale super-resolution and denoising of seismic data comprises the following steps,
s1, performing down-sampling on a high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set;
s2, initializing a seismic reconstruction network, wherein the seismic reconstruction network comprises a self-encoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network;
s3, learning the seismic reconstruction network by using a multi-scale depth supervised learning strategy;
and S4, inputting the new seismic image into the trained seismic enhancement network to obtain a de-noised high-resolution seismic image and complete the reconstruction of seismic data.
In another aspect, the present invention also discloses a computer readable storage medium storing a computer program, which when executed by a processor causes the processor to perform the steps of the method as described above.
According to the technical scheme, the seismic data multi-scale super-resolution and denoising synchronous processing method has the following beneficial effects:
in the sampling stage on the network, a multi-scale reconstruction sub-network is introduced, a multi-scale depth supervision learning strategy is adopted, the characteristics of the seismic image in different scales are fully learned, and the characteristics under different receptive fields and granularities are integrated, so that the characteristic diagram generated by the decoder can contain richer semantic information, has more texture information and can describe a finer geological structure. The invention can retain the high-frequency information of the seismic image, inhibit random noise, effectively highlight the geological structure information, has a visual effect more in line with a human eye system, and can effectively improve the efficiency and accuracy of subsequent seismic interpretation.
Drawings
FIG. 1 is a schematic diagram of the method of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention.
As shown in fig. 1, the method for synchronously processing multi-scale super-resolution and denoising of seismic data in this embodiment includes the following steps,
s1, performing down-sampling on a high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set;
s2, initializing an earthquake reconstruction network, wherein the earthquake reconstruction network comprises a self-encoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network;
s3, learning the seismic reconstruction network by using a multi-scale depth supervised learning strategy;
and S4, inputting the new seismic image into the trained seismic enhancement network to obtain a de-noised high-resolution seismic image and complete the reconstruction of seismic data.
The following is a detailed description:
as shown in fig. 1, the present invention is described by taking the maximum down-sampling order K =4 as an example. The embodiment discloses a method and a system for synchronously processing multi-scale super-resolution and denoising of seismic data, which comprises the following steps S1 to S4:
s1, downsampling the high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set, wherein the method specifically comprises the following steps:
s11, for seismic image
m and n are the length and height of the seismic image, and subscripts hr, sr and lr represent high resolution, super resolution and low resolution; high resolution seismic image I hr Downsampling to obtain a low-resolution seismic image I lr ＝Down(I hr Θ), down represents a downsampling function, based on the downsampling function>
For the downsampling parameters, K =0,1,2,3, \ 8230, where K denotes the downsampling order, and/or>
Represents->
K =0 means no downsampling, i.e. downsampling
S12, a sheet I hr A corresponding group
Forming a training sample pair with independent identically distributed noise>
As an input to the network>
Supervised learning labels as a network at K =0,1,2,3, \ 8230;, K multi-scale output;
s2, initializing an earthquake reconstruction network, wherein the earthquake reconstruction network comprises a self-encoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network, and the earthquake reconstruction network specifically comprises the following parts:
s21, defining a seismic reconstruction network, specifically as follows:
the connection relationships of the three sub-networks are as follows:
the self-encoder sub-network is one of artificial neural networks, an encoder extracts features, a decoder reconstructs data, and a self-encoder is a U-shaped encoder-decoder network and comprises K downsampling blocks and corresponding K upsampling blocks; k downsample blocks of
Comprises the following components:
wherein
Is->
Characteristic map of the output, based on the comparison of the characteristic map and the reference value>
Is->
Is greater than or equal to>
K upsampling modules are
Comprising:
wherein, the first and the second end of the pipe are connected with each other,
is->
Characteristic map of the output, based on the comparison of the characteristic map and the reference value>
Is the combination of the encoder and decoder outputs on the k-order scale; conc is an abbreviation for Concatenate, indicating incorporation; for the case of K = K, ->
Is->
The parameters of (1);
super-resolution subnetwork
Obtaining a 0-order super-resolution seismic image->
Wherein theta is s Is a super-resolution subnetwork parameter;
k multi-scale reconstruction sub-networks
Comprising:
wherein the content of the first and second substances,
is->
The parameters of (1);
the three sub-networks are embodied as follows:
the coding part of the self-coder sub-network consists of 4 down-sampling modules, and the shallow to deep forms are respectively as follows:
Conv(1,64,3)+BN+ReLU+Conv(64,64,3)+BN+ReLU+Max
Conv(64,128,3)+BN+ReLU+Conv(128,128,3)+BN+ReLU+Max
Conv(128,256,3)+BN+ReLU+Conv(256,256,3)+BN+ReLU+Max
Conv(256,512,3)+BN+ReLU+Conv(512,512,3)+BN+ReLU+Max
conv () represents convolution operation, wherein 3 parameters in Conv are the number of input channels, the number of output channels and the side length of a square convolution kernel respectively, the convolution step length of Conv is 1, the filling is 1 so as to ensure that the size of a feature graph after convolution is unchanged, BN represents batch normalization, reLU is an activation function, max represents maximum pooling, and the kernel size is 2 multiplied by 2 and the step length is 2;
the decoding part of the self-encoder subnetwork consists of 1 feature mapping layer and 4 up-sampling modules, and the shallow-to-deep forms are respectively as follows:
Conv(512,1024,3)+BN+ReLU+Conv(1024,1024,3)+BN+ReLU+SubPix
Conv(512,1024,3)+BN+ReLU+Conv(1024,512,3)+BN+ReLU+SubPix
Conv(256,512,3)+BN+ReLU+Conv(512,256,3)+BN+ReLU+SubPixConv(128,256,3)+BN+ReLU+Conv(256,256,3)+BN+ReLU+SubPix
the convolution step length of the 4 up-sampling module output channels is respectively 256, 128, 64 and 64, conv is 1, the sub-pixel convolution layer is filled with 1, subPix, the up-sampling is completed by utilizing a channel recombination mode, and if the multiple of the up-sampling is 2, the number of the channels is reduced by 4 times after passing through the SubPix;
the form of the super-resolution subnetwork is specifically as follows:
Conv(64,128,9)+ReLU+SubPix+Conv(32,16,1)+ReLU+Conv(16,1,5)
wherein the step length of the three convolution layers is 1, and the filling values are 4, 0 and 2 respectively so as to ensure that the size of the characteristic diagram is unchanged;
the multi-scale reconstruction sub-network consists of 4 convolutional layers, and the form is as follows for the 1 st to 4 th scales respectively:
Conv(128,32,9)+ReLU+Conv(32,16,1)+ReLU+Conv(16,1,5)
Conv(128,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
Conv(256,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
Conv(512,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
wherein, the filling values of the three convolution layers are respectively 4, 0 and 2;
s22, defining a loss function of the 0 th order scale, which is specifically as follows:
defining the high resolution reconstruction loss, and specifically for a single image, the following steps are carried out:
wherein | 1 Is a norm of 1, and N is the total number of pixel points of the seismic image;
the convergence of training can be accelerated, but the texture details of the generated super-resolution image are too smooth, so that structural similarity SSIM is introduced, an SSIM function is based on the assumption that structural information is extracted when human eyes watch the image, a model is more sensitive to the local structural change of the image, high-frequency information such as the edge and the details of the image is reserved, and the reconstructed super-resolution image is more suitable for a human visual system;
defining a high-resolution structural similarity loss, specifically for a single image as follows:
SSIM(x,y)＝[l(x,y) α ·[c(x,y) β ·[s(x,y) γ
where x, y are the two images to be evaluated, α, β, γ are the corresponding weights, and 0.15, 0.29, 0.31, l (x, y), c (x, y), s (x, y) can be taken to represent three measurements between x and y, i.e., the brightness or amplitude, contrast, and texture in the seismic image
Wherein, mu x ,μ y Representing the mean, σ, of the images x, y x ,σ y Representing the standard deviation, σ, of the images x, y xy Is the covariance between the images x and y, c 1 ,c 2 ,c 3 Is three constants, can be set as c 1 ＝10 -4 ，
Finally, a loss function of 0 th order scale is obtained, and for a single image, the loss function is:
where a is the weight coefficient of the loss function, which can be generally set to 0.6;
s23, defining a loss function of the kth scale, wherein k is greater than 0, and the method specifically comprises the following steps:
for the
Reconstructed image with resolution +>
Focusing on whether the reconstructed image looks more like an original seismic image or not rather than focusing on the matching degree of the pixel points;
defining a k-order perceptual loss, for a single image, namely:
where φ represents a certain layer of output of the pre-trained VGG network, N φ Representing the total number of pixel points of the output image of the layer;
define the k-order lattice loss, for a single image, i.e.:
wherein N is g Representing the total number of elements of a Gram matrix, namely a Gram matrix;
defining a k-order smoothing loss, for a single image, namely:
wherein (, j) represents the location of a pixel point in the image; the addition of the smoothing loss term can enable the edge of the generated seismic image to be clearer, further eliminate noise in the seismic image and improve the performance of subsequent seismic interpretation;
finally, a loss function of the kth order scale is obtained, k >0, and for a single image, namely:
wherein, γ 1 And gamma 2 Contributions for balancing the three loss terms, which may be set to 5 and 0.4;
s24, initializing the seismic reconstruction network, wherein random initialization can be adopted;
s3, learning the seismic reconstruction network by using a multi-scale depth supervised learning strategy, which specifically comprises the following steps:
pre-training the whole network to wait
After stabilization, add->
To the k-th>Training the 0-order scale network until the model converges; the model is optimized during training using an Adam optimizer, where the first order moment estimate is set to an exponential decay rate of 0.9, the second order moment estimate is set to an exponential decay rate of 0.99, the epsilon parameter is used to prevent division by zero during implementation and is therefore set to a very small number, e.g., 10 -8 (ii) a The learning rate is initialized to 10 -4 When the loss tends to be smooth, the learning rate is reduced to 10 -5 ；
And S4, inputting the new seismic image into the trained seismic enhancement network to obtain a de-noised high-resolution seismic image and complete the reconstruction of seismic data.
The embodiment of the invention can keep the high-frequency information of the seismic image, inhibit random noise, effectively highlight the geological structure information, have visual effect more in line with a human eye system, and effectively improve the efficiency and accuracy of subsequent seismic interpretation.
In yet another aspect, the present invention also discloses a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of any of the methods described above.
In yet another aspect, the present invention also discloses a computer device comprising a memory and a processor, the memory storing a computer program, the computer program, when executed by the processor, causing the processor to perform the steps of any of the methods as described above.
In a further embodiment provided by the present application, there is also provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the steps of any of the methods of the above embodiments.
It is understood that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, the example and the beneficial effects of the related contents can refer to the corresponding parts in the method.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
All possible combinations of the technical features in the above embodiments may not be described for the sake of brevity, but should be considered as being within the scope of the present disclosure as long as there is no contradiction between the combinations of the technical features.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (6)
1. A synchronous processing method of multi-scale super-resolution and denoising of seismic data is characterized by comprising the following steps,
s1, performing down-sampling on a high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set;
s2, initializing an earthquake reconstruction network, wherein the earthquake reconstruction network comprises a self-encoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network;
s3, learning the seismic reconstruction network by using a multi-scale depth supervised learning strategy;
and S4, inputting the new seismic image into the trained seismic enhancement network to obtain a de-noised high-resolution seismic image and complete the reconstruction of seismic data.
2. The seismic data multi-scale super-resolution and denoising synchronous processing method according to claim 1, wherein: s1, down-sampling the high-resolution seismic image to obtain a low-resolution seismic image, and constructing a training sample set, wherein the method specifically comprises the following steps:
s11, for seismic image
m and n are the length and height of the seismic image, and subscripts hr, sr and lr represent high resolution, super resolution and low resolution; high resolution seismic image I hr Downsampling to obtain a low-resolution seismic image I lr ＝Down(I hr Θ), down represents a downsampling function, based on the downsampling function>
For the down-sampling parameters, K =0,1,2,3, \ 8230, K denotes the down-sampling order,
represents->
K =0 means no downsampling, i.e. downsampling
S12, a sheet I hr A corresponding group
Forming a training sample pair with independent identically distributed noise>
As an input to the network>
The supervised learning label is output in K =0,1,2,3, \ 8230;, K multi-scale as a network.
3. The seismic data multi-scale super-resolution and denoising synchronous processing method according to claim 2, wherein: s2, initializing a seismic reconstruction network, wherein the seismic reconstruction network comprises an autoencoder sub-network, a super-resolution sub-network and a multi-scale reconstruction sub-network, and the specific steps are as follows:
s21, defining a seismic reconstruction network, specifically as follows:
the connection relationships of the three sub-networks are as follows:
the self-encoder sub-network is one of artificial neural networks, an encoder extracts features, a decoder reconstructs data, and a self-encoder is a U-shaped encoder-decoder network and comprises K downsampling blocks and corresponding K upsampling blocks; k downsampling blocks of
Comprising:
wherein
Is->
Output characteristic map>
Is->
Is greater than or equal to>
K upsampling modules are->
Comprises the following components:
wherein the content of the first and second substances,
is->
Characteristic map of the output, based on the comparison of the characteristic map and the reference value>
Is the combination of the encoder and decoder outputs on the k-order scale; conc is an abbreviation for Concatenate, indicating incorporation; for the case of K = K, ->
Is->
The parameters of (a);
super-resolution subnetwork
Obtaining a 0-order super-resolution seismic image->
Wherein theta is s Is a super-resolution subnetwork parameter; />
K multi-scale reconstruction sub-networks
Comprises the following components:
wherein, the first and the second end of the pipe are connected with each other,
is->
The parameters of (1);
the three sub-networks are embodied as follows:
the coding part of the self-coder subnetwork consists of 4 down-sampling modules, and the shallow to deep forms are respectively as follows:
Conv(1,64,3)+BN+ReLU+Conv(64,64,3)+BN+ReLU+Max
Conv(64,128,3)+BN+ReLU+Conv(128,128,3)+BN+ReLU+Max
Conv(128,256,3)+BN+ReLU+Conv(256,256,3)+BN+ReLU+Max
Conv(256,512,3)+BN+ReLU+Conv(512,512,3)+BN+ReLU+Max
wherein Conv () represents convolution operation, wherein 3 parameters in Conv are the number of input channels, the number of output channels and the side length of a square convolution kernel respectively, the convolution step size of Conv is 1, the filling is 1 to ensure that the size of a feature graph after convolution is unchanged, BN represents batch normalization, reLU is an activation function, max represents maximum pooling, and the kernel size is 2 multiplied by 2 and the step size is 2;
the decoding part of the self-encoder subnetwork consists of 1 feature mapping layer and 4 up-sampling modules, and the shallow-to-deep forms are respectively as follows:
Conv(512,1024,3)+BN+ReLU+Conv(1024,1024,3)+BN+ReLU+SubPix
Conv(512,1024,3)+BN+ReLU+Conv(1024,512,3)+BN+ReLU+SubPix
Conv(256,512,3)+BN+ReLU+Conv(512,256,3)+BN+ReLU+SubPix
Conv(128,256,3)+BN+ReLU+Conv(256,256,3)+BN+ReLU+SubPix
the convolution step length of the 4 up-sampling module output channels is respectively 256, 128, 64 and 64, conv is 1, the filling is 1, subPix represents a sub-pixel convolution layer, the up-sampling is completed by utilizing a channel recombination mode, and if the multiple of the up-sampling is 2, the number of the channels is reduced by 4 times after passing through the SubPix;
the form of the super-resolution subnetwork is specifically as follows:
Conv(64,128,9)+ReLU+SubPix+Conv(32,16,1)+ReLU+Conv(16,1,5)
wherein the step length of the three convolution layers is 1, and the filling values are 4, 0 and 2 respectively so as to ensure that the size of the characteristic diagram is unchanged;
the multi-scale reconstruction sub-network consists of 4 convolutional layers, and the form is as follows for the 1 st to 4 th scales respectively:
Conv(128,32,9)+ReLU+Conv(32,16,1)+ReLU+Conv(16,1,5)
Conv(128,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
Conv(256,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
Conv(512,64,9)+ReLU+Conv(64,16,1)+ReLU+Conv(16,1,5)
wherein, the filling values of the three convolution layers are respectively 4, 0 and 2;
s22, defining a loss function of 0 th order scale, specifically as follows:
defining the high resolution reconstruction loss, and specifically for a single image, the following steps are carried out:
wherein | 1 Is a norm of 1, and N is the total number of pixel points of the seismic image;
the convergence of training can be accelerated, but the texture details of the generated super-resolution image are too smooth, so that structural similarity SSIM is introduced;
defining a high-resolution structural similarity loss, specifically for a single image as follows:
SSIM(x,y)＝[l(x,y)] α ·[c(x,y)] β ·[s(x,y)] γ
where x, y are the two images to be evaluated, and α, β, γ are the corresponding weights, it is possible to take 0.15, 0.29, 0.31, l (x, y), c (x, y), s (x, y) to represent the three measurements between x and y, i.e. the brightness or amplitude, contrast and structure in the seismic image
Wherein, mu x ,μ y Representing the mean, σ, of the images x, y x ,σ y Representing the standard deviation, σ, of the images x, y xy Is the covariance between the images x and y, c 1 ,c 2 ,c 3 Is three constants set as c 1 ＝10 -4 ，
Finally, a loss function of 0 th order scale is obtained, and for a single image, namely:
wherein a is a weight coefficient of the loss function;
s23, defining a loss function of the kth scale, wherein k is greater than 0, and the method comprises the following specific steps:
for the
Reconstructed image with resolution +>
Not focusing on the matching degree of the pixel points, but focusing on the reconstructed image from the visual point of viewWhether to resemble more of the original seismic image;
defining a k-order perceptual loss, for a single image, namely:
where φ represents a certain layer of output of the pre-trained VGG network, N φ Representing the total number of pixel points of the layer of output image;
define the k-order lattice loss, for a single image, i.e.:
wherein N is g Represents the total number of elements of the Gram matrix, gram being Gram;
defining a k-order smoothing loss, for a single image, namely:
wherein (, j) represents the location of a pixel point in the image;
finally, a loss function of the kth order scale is obtained, k is greater than 0, and for a single image, namely:
wherein, gamma is 1 And gamma 2 Contributions for balancing the three loss terms, set to 5 and 0.4;
and S24, initializing the seismic reconstruction network, and adopting random initialization.
4. The seismic data multi-scale super-resolution and denoising synchronous processing method according to claim 3, wherein: s3, learning the seismic reconstruction network by using a multi-scale deep supervised learning strategy, which specifically comprises the following steps:
pre-training the whole network to wait
After stabilization, add->
To the k-th>Training the 0-order scale network until the model converges; the model was optimized during training using an Adam optimizer, where the first order moment estimate was set to an exponential decay rate of 0.9, the second order moment estimate was set to an exponential decay rate of 0.99, the epsilon parameter was used to prevent division by zero in the implementation; the learning rate is initialized to 10 -4 When the loss tends to be smooth, the learning rate is reduced to 10 -5 。
5. The seismic data multi-scale super-resolution and denoising synchronous processing method according to claim 4, wherein: epsilon parameter set to 10 -8 。
6. A computer-readable system, in which a computer program is stored which, when being executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211586531.4A CN115936990B (en) | 2022-12-07 | 2022-12-07 | Seismic data multi-scale super-resolution and denoising synchronous processing method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211586531.4A CN115936990B (en) | 2022-12-07 | 2022-12-07 | Seismic data multi-scale super-resolution and denoising synchronous processing method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115936990A true CN115936990A (en) | 2023-04-07 |
| CN115936990B CN115936990B (en) | 2023-11-17 |
Family
ID=86557024
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211586531.4A Active CN115936990B (en) | 2022-12-07 | 2022-12-07 | Seismic data multi-scale super-resolution and denoising synchronous processing method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115936990B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116401597A (en) * | 2023-06-09 | 2023-07-07 | 中国科学技术大学先进技术研究院 | Method for constructing seismic interpretation model, electronic equipment and readable storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| CN110580682A (en) * | 2019-09-16 | 2019-12-17 | 电子科技大学 | Countermeasure network seismic data super-resolution reconstruction method based on optimization generation |
| US20190383965A1 (en) * | 2017-02-09 | 2019-12-19 | Schlumberger Technology Corporation | Geophysical Deep Learning |
| US20210041596A1 (en) * | 2019-08-06 | 2021-02-11 | Exxonmobil Upstream Research Company | Petrophysical Inversion With Machine Learning-Based Geologic Priors |
| CN113269818A (en) * | 2021-06-09 | 2021-08-17 | 河北工业大学 | Seismic data texture feature reconstruction method based on deep learning |
| CN113706380A (en) * | 2021-08-20 | 2021-11-26 | 西安交通大学 | Method and system for generating countermeasure network based on weak supervision to improve seismic data resolution |
| CN114063168A (en) * | 2021-11-16 | 2022-02-18 | 电子科技大学 | Artificial intelligence noise reduction method for seismic signals |
| WO2022242029A1 (en) * | 2021-05-18 | 2022-11-24 | 广东奥普特科技股份有限公司 | Generation method, system and apparatus capable of visual resolution enhancement, and storage medium |
-
2022
- 2022-12-07 CN CN202211586531.4A patent/CN115936990B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190383965A1 (en) * | 2017-02-09 | 2019-12-19 | Schlumberger Technology Corporation | Geophysical Deep Learning |
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| US20210041596A1 (en) * | 2019-08-06 | 2021-02-11 | Exxonmobil Upstream Research Company | Petrophysical Inversion With Machine Learning-Based Geologic Priors |
| CN110580682A (en) * | 2019-09-16 | 2019-12-17 | 电子科技大学 | Countermeasure network seismic data super-resolution reconstruction method based on optimization generation |
| WO2022242029A1 (en) * | 2021-05-18 | 2022-11-24 | 广东奥普特科技股份有限公司 | Generation method, system and apparatus capable of visual resolution enhancement, and storage medium |
| CN113269818A (en) * | 2021-06-09 | 2021-08-17 | 河北工业大学 | Seismic data texture feature reconstruction method based on deep learning |
| CN113706380A (en) * | 2021-08-20 | 2021-11-26 | 西安交通大学 | Method and system for generating countermeasure network based on weak supervision to improve seismic data resolution |
| CN114063168A (en) * | 2021-11-16 | 2022-02-18 | 电子科技大学 | Artificial intelligence noise reduction method for seismic signals |
Non-Patent Citations (8)
| Title |
|---|
| JINTAO LI ETC.: "Deep Learning for simultaneous seismic image super-resolution and denoising", SEG TECHNICAL PROGRAM EXPANDED ABSTRACTS 2020 * |
| S.MANCINI ETC.: "On the use of High-Resolution and Deep-Learning seismic Catalogs for ZShort-Term Earthquake forecasts:Potential Benefits and Current Limitations", JGR SOLID EARTH * |
| Z.F.LI;M.-A.MEIER ETC.: "机器学习应用于地震预警中的地震波判别研究", 世界地震译丛, no. 05 * |
| 刘旭跃等: "基于学习的地震剖面超分辨率重建方法研究", 物探化探计算技术, no. 01 * |
| 吕文君等: "基于标签传播的岩性预测半监督学习算法研究", 地球物理学进展, vol. 36, no. 02 * |
| 张文婷等: "Sentinel-1数据用于鲜水河断裂带震间形变监测", 国际地震动态, no. 08 * |
| 张文婷等: "卫星大地测量成像地震周期形变研究综述", 武汉大学学报· 信息科学版, vol. 47, no. 06 * |
| 邓梦迪等: "基于深度学习的地震剖面图像超分辨率重建", 计算机工程与设计, no. 08 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116401597A (en) * | 2023-06-09 | 2023-07-07 | 中国科学技术大学先进技术研究院 | Method for constructing seismic interpretation model, electronic equipment and readable storage medium |
| CN116401597B (en) * | 2023-06-09 | 2023-08-22 | 中国科学技术大学先进技术研究院 | Method for constructing seismic interpretation model, electronic equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115936990B (en) | 2023-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Canchumuni et al. | Recent developments combining ensemble smoother and deep generative networks for facies history matching | |
| CN110516095B (en) | Semantic migration-based weak supervision deep hash social image retrieval method and system | |
| US20180225807A1 (en) | Single-frame super-resolution reconstruction method and device based on sparse domain reconstruction | |
| CN111563902A (en) | Lung lobe segmentation method and system based on three-dimensional convolutional neural network | |
| CN112132959B (en) | Digital rock core image processing method and device, computer equipment and storage medium | |
| CN110599461B (en) | Thalamus function partitioning method based on subspace feature learning | |
| CN109214989A (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| CN110599502B (en) | Skin lesion segmentation method based on deep learning | |
| CN113256494B (en) | Text image super-resolution method | |
| CN113269818B (en) | Deep learning-based seismic data texture feature reconstruction method | |
| US20240029204A1 (en) | Image processing method, system, device and storage medium | |
| CN115936990A (en) | Synchronous processing method and system for multi-scale super-resolution and denoising of seismic data | |
| CN112215267B (en) | Hyperspectral image-oriented depth space spectrum subspace clustering method | |
| CN115457568B (en) | Historical document image noise reduction method and system based on generation countermeasure network | |
| CN116757955A (en) | Multi-fusion comparison network based on full-dimensional dynamic convolution | |
| CN107424119A (en) | A kind of super-resolution method of single image | |
| CN112541566B (en) | Image translation method based on reconstruction loss | |
| CN117593275A (en) | Medical image segmentation system | |
| CN112149802A (en) | Image content conversion method with consistent semantic structure | |
| CN116778016A (en) | MRI image reconstruction method, system and medium | |
| CN114708353B (en) | Image reconstruction method and device, electronic equipment and storage medium | |
| CN116029908A (en) | 3D magnetic resonance super-resolution method based on cross-modal and cross-scale feature fusion | |
| CN115205527A (en) | Remote sensing image bidirectional semantic segmentation method based on domain adaptation and super-resolution | |
| CN109581483A (en) | Processing Seismic Data and system based on rarefaction representation | |
| CN114565528A (en) | Remote sensing image noise reduction method and system based on multi-scale and attention mechanism |
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 |