CN115639605B - Automatic identification method and device for high-resolution fault based on deep learning - Google Patents
Automatic identification method and device for high-resolution fault based on deep learning Download PDFInfo
- Publication number
- CN115639605B CN115639605B CN202211331692.9A CN202211331692A CN115639605B CN 115639605 B CN115639605 B CN 115639605B CN 202211331692 A CN202211331692 A CN 202211331692A CN 115639605 B CN115639605 B CN 115639605B
- Authority
- CN
- China
- Prior art keywords
- fault
- model
- quality
- seismic
- seismic image
- 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 108
- 238000013135 deep learning Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 126
- 238000012795 verification Methods 0.000 claims abstract description 61
- 230000006872 improvement Effects 0.000 claims abstract description 49
- 238000004088 simulation Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims description 28
- 238000002310 reflectometry Methods 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 20
- 238000004590 computer program Methods 0.000 claims description 9
- 230000008485 antagonism Effects 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 8
- 238000007796 conventional method Methods 0.000 abstract description 3
- 238000013528 artificial neural network Methods 0.000 description 25
- 238000010586 diagram Methods 0.000 description 16
- 230000004913 activation Effects 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 8
- 230000015572 biosynthetic process Effects 0.000 description 7
- 238000005755 formation reaction Methods 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 7
- 230000001427 coherent effect Effects 0.000 description 6
- 238000009826 distribution Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 208000010392 Bone Fractures Diseases 0.000 description 3
- 206010017076 Fracture Diseases 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000007480 spreading Effects 0.000 description 3
- 238000003892 spreading Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 238000010521 absorption reaction Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000000151 deposition Methods 0.000 description 2
- 230000008021 deposition Effects 0.000 description 2
- 238000003708 edge detection Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 239000004215 Carbon black (E152) Substances 0.000 description 1
- 208000013201 Stress fracture Diseases 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 229930195733 hydrocarbon Natural products 0.000 description 1
- 150000002430 hydrocarbons Chemical class 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides an automatic identification method and device of high-resolution faults based on deep learning, comprising the following steps: acquiring a seismic image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels. Compared with the conventional method, the method can obtain a cleaner and clearer fault probability map, and the predicted fault position is more accurate.
Description
Technical Field
The invention relates to the technical field of geophysical exploration, in particular to an automatic high-resolution fault identification method and device based on deep learning.
Background
Seismic imaging is an important piece of geophysical data that provides continuous, wide-range subsurface geologic information. Experienced geophysicists can infer age of depositions of formations, analyze depositional environments and structural movements from seismic images, and assist in hydrocarbon exploration. Subsurface faults are an important geological structure that controls the deposition of the basin, the migration of fluids in the subsurface and the distribution of minerals. The identification of subsurface faults by seismic images is a common approach, but because of the large scale of seismic exploration and the complex data, manually interpreting faults is a laborious and highly subjective task. Thus, with the development of computer technology, computer-aided automatic tomographic recognition is becoming more popular.
Patent application CN202010188216.0 discloses a three-attribute fusion fault identification method based on dominant frequency of seismic data, firstly, frequency division processing is carried out on full-frequency-band seismic data, dominant frequency volume data are selected to manufacture coherence, dip angle and azimuth angle attribute volumes, and finally, three attributes are displayed in an HIS superposition mode to obtain fusion volumes for fault interpretation. The fault identification capability is improved through frequency division and image fusion technologies, but a great deal of effort is required to select dominant frequency bodies, a great deal of calculation is required for subsequent attribute body calculation, and the selection of parameters influences the subsequent interpretation accuracy.
Patent application CN201910854606.4 discloses a coherence enhancement fault identification method based on seismic analysis channels under horizon constraint, which improves the traditional coherence algorithm by introducing Hilbert transformation, histogram equalization and other means, has higher noise resistance, but prevents the limitation of a coherence body, so that multiple solutions are easy to generate and small faults cannot be clearly depicted.
Patent application CN201710186136.X discloses an automatic identification method for an earthquake coherent body image fault based on an AdaBoost algorithm, which realizes automatic identification of an earthquake coherent body earthquake image, improves the accuracy of fault layer identification, but needs to adjust the size of a data block when characteristic information is mined, and increases the space occupation rate of data.
Patent application CN202110946552.1 discloses a seismic fault identification method based on deep learning semantic segmentation, which realizes rapid and accurate fault identification by means of strong fitting capacity of a deep convolutional neural network, but the identification accuracy is affected by the signal-to-noise ratio of seismic data.
The analysis technologies such as body attribute, layer structure attribute, absorption attenuation attribute and the like are adopted by 2016 years Huang Cheng and the like to optimize the seismic attribute body, the fault layer system of a research area is effectively predicted in stages and layers, and the combination application of multiple methods reduces the multiple solutions in the fault layer identification process. 2017 Zhong Weijun et al comprehensively utilizes technical means such as coherent bodies, variance bodies, structural curvature bodies, multi-window inclination angle scanning bodies, structural guide filtering, coherent energy gradient bodies, ant body tracking, edge detection and the like, forms a fault identification technical flow represented by an inclination angle guide technology, and identifies faults with different levels and different periods. Zhang Rui et al in 2017 propose a frequency division ant tracking technology by means of generalized S transformation and wavelet transformation analysis technology, and the method can better suppress noise interference, is clearer and more accurate in fault identification, and can identify deep minor faults which are difficult to identify by conventional full-frequency band data. In 2017 Li Quanhe Tong Liqing, complex fracture is identified through seismic attribute optimization combination, firstly, fracture spreading trend is controlled based on guided coherence body attribute, and then, spatial spreading features of three-level and four-level fracture and micro-fracture are finely delineated by seismic attribute bodies such as dip angle body, azimuth angle body and curvature body which are preferably fused to highlight local details. In 2017, sun Zhenyu et al construct a fault layer identification model by taking the seismic attribute as the input of a support vector machine, the accuracy reaches 98%, the influence of human subjective factors is reduced, and the interpretation period is shortened. In 2020 Liu Mojin and salted sea dragon, coherent body and amplitude, coherent body and maximum positive curvature are image fused, and the respective construction information is fully utilized, so that the capability of identifying small faults is enhanced, and the reliability of fault plane combination, plane spreading characteristics and interrelationships is improved. 2022, hou Juntao et al applied the automatic fault extraction technique to frequency division coherence based on guided filtering, and achieved finer and more accurate fault characterization results by suppressing noise. Ding Changwei et al propose the reduction of seismic attributes by using information value in 2022, optimize XGBoost parameters in combination with an improved Bayesian optimization algorithm to further improve the accuracy of small fault seismic interpretation, and the method effectively solves the problem of unbalanced distribution of small fault samples and has certain anti-interference capability. Liu Naihao et al introduced edge detection technology in deep learning in 2022 and optimized the network structure according to seismic data and fault characteristics, proposed an improved HED network using seismic fault interpretation, which was more accurate and better continuous for fault prediction. In 2022, zhang Zheng et al combine deep residual error network with transfer learning and apply to fault recognition, and on the basis of a synthetic data training model, use a small amount of actual fault samples to perform transfer learning, enhance generalization capability of the network and improve fault recognition performance.
The above methods all contribute to automatic identification of the earthquake fault to a certain extent, but the methods lack precision analysis of fault identification, and lack analysis of identification effect when the quality of multi-seismic data is low, for example, the seismic data contains a large amount of random noise, and the resolution of the seismic data is low. However, the interference of random noise caused by the noise of the seismic data collected in the field and the low resolution caused by the absorption of the high frequency components of the seismic waves by the underground medium are unavoidable, especially in deep and ultra-deep exploration. These reasons lead to the poor versatility of the currently proposed methods of fault identification.
Therefore, how to avoid the lower accuracy and lower reliability of fault automatic identification at lower seismic data quality is still a problem to be solved by those skilled in the art.
Disclosure of Invention
The invention provides an automatic fault identification method based on deep learning, which is used for solving the problems of lower precision and lower reliability of automatic fault identification when the quality of seismic data is lower in the prior art.
The invention provides an automatic identification method of a high-resolution fault based on deep learning, which comprises the following steps:
Acquiring a seismic image to be identified;
inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image;
The quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation;
Inputting the lifted seismic image into a fault identification model, and outputting a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
The invention provides an automatic identification method of high-resolution faults based on deep learning, which further comprises the following steps:
Inputting the seismic image to be identified into the fault identification model, and outputting a direct identification result of the seismic image;
And comparing the fault probability map with the seismic image direct identification result.
According to the automatic identification method of the high-resolution fault based on deep learning provided by the invention, the first training set and the first verification set are generated through random parameter simulation, and the method specifically comprises the following steps:
Generating a stratum reflectivity model according to a preset rule;
Generating stratum reflectivity models of different types through random parameter control;
convolving any stratum reflectivity model with low-frequency Rake wavelets and adding seismic data of random noise to perform double downsampling to obtain a sample low-quality seismic image;
and convolving any stratum reflectivity model with the high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag.
According to the automatic identification method of the high-resolution fault based on deep learning, the network structure of the quality improvement model in the training process generates a countermeasure network, the countermeasure network comprises a generator and a discriminator, wherein the generator comprises a layer jump connection and a plurality of residual connections, and the discriminator comprises a convolution layer and seven convolution blocks.
According to the automatic identification method of the high-resolution fault based on deep learning, in the training process of the quality improvement model, the network parameter corresponding to the highest peak signal-to-noise ratio of the quality improvement model on the verification set is the optimal model parameter, wherein the peak signal-to-noise ratio is the similarity between the reconstructed high-quality seismic image and the true high-quality seismic image of the antagonism network.
According to the automatic identification method of the high-resolution fault based on the deep learning, a model network structure used in the training process of the fault identification model is a U-net network, the U-net network comprises an encoding branch and a decoding branch, the encoding branch consists of four convolution operations and downsampling, and the decoding branch consists of four upsampling and convolution operations.
According to the automatic fault identification method based on deep learning, the weight given to the pixel representing the fault in the loss function of the fault identification model training process is larger than the weight given to the pixel representing the non-fault.
The invention also provides an automatic identification device of the high-resolution fault based on deep learning, which comprises the following steps:
the acquisition unit is used for acquiring the seismic image to be identified;
the quality improving unit is used for inputting the seismic image to be identified into a quality improving model and outputting an improved seismic image;
The quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation;
The fault identification unit is used for inputting the lifted seismic image into a fault identification model and outputting a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the automatic identification method of the deep learning-based high-resolution fault.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method for automatically identifying high resolution faults based on deep learning as described in any of the above.
According to the automatic identification method and the device for the high-resolution fault based on the deep learning, provided by the invention, the seismic image to be identified is obtained; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels. The method realizes that a cleaner and clearer fault probability map is obtained, and the predicted fault position is more accurate.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an automatic identification method of a high-resolution fault based on deep learning;
fig. 2 is a schematic structural diagram of an automatic recognition device of a high-resolution fault based on deep learning;
FIG. 3 is a block diagram of a neural network for seismic data quality enhancement provided by the present invention;
FIG. 4 is a schematic diagram of a training process of the seismic data quality enhancement neural network provided by the invention;
FIG. 5 is a block diagram of a seismic fault identification neural network provided by the invention;
FIG. 6 is a schematic diagram of a training process of a fault identification neural network provided by the invention;
FIG. 7 is a schematic diagram of an end-to-end workflow of intelligent identification of high resolution faults provided by the present invention;
FIG. 8 is a schematic diagram of the application of the present invention to field real seismic data and a comparison of the present invention with a conventional implementation process;
Fig. 9 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Because the problems of lower accuracy and lower reliability of fault automatic identification at lower seismic data quality are common in the prior art. The automatic recognition method of the deep learning-based high resolution fault of the present invention is described below with reference to fig. 1 to 9. Fig. 1 is a schematic flow chart of an automatic identification method of a high-resolution fault based on deep learning, which is provided by the invention, as shown in fig. 1, and the method comprises the following steps:
At step 110, a seismic image to be identified is acquired.
Specifically, the seismic image to be identified is obtained, the image quality of the seismic data image which is usually obtained is not high due to the restriction of the field acquisition environment, and the image quality refers to the resolution and the image signal-to-noise ratio, namely, the resolution of the seismic data image which is acquired in the field is lower, the image signal-to-noise ratio is also lower, and the noise ratio is larger.
Step 120, inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image;
The quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation.
Specifically, a low-quality seismic image to be identified is firstly input into a quality improvement model for quality improvement, the output improved seismic image has higher resolution and larger signal-to-noise ratio than an image before quality improvement, the quality improvement model is obtained by training based on a large number of sample low-quality seismic images and corresponding high-quality seismic image labels, training data in the training process are divided into a training set and a verification set, after the training set is completed, the verification set is used for verification to select a model with optimal network parameters, the low-quality seismic image and the high-quality seismic image are required to be described, the low-quality seismic image and the high-quality seismic image refer to images with image resolution lower than a first threshold value and image resolution higher than a second threshold value respectively, and the image signal-to-noise ratio is lower than a third threshold value and higher than a fourth threshold value respectively, and the first threshold value, the second threshold value, the third threshold value and the fourth threshold value are set based on application scenes. In addition, the first training set and the first verification set in the embodiment of the invention are generated through random parameter simulation, so that manual labeling is not needed, and the labor cost is saved.
130, Inputting the lifted seismic image into a fault identification model, and outputting a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
Specifically, after the quality of the seismic image to be identified is improved, the seismic image to be identified is input into a fault identification model, the model outputs a corresponding fault probability map, the fault identification model is obtained by training based on a large number of sample seismic images and corresponding fault probability icon labels, training data and the quality improvement model are also divided into a training set and a verification set in the training process, and after the training set is trained, the verification set is used for verification to select a model with optimal network parameters.
The method provided by the invention comprises the steps of obtaining a seismic image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels. The method realizes that a cleaner and clearer fault probability map is obtained, and the predicted fault position is more accurate.
Based on the above embodiment, the method further includes:
Inputting the seismic image to be identified into the fault identification model, and outputting a direct identification result of the seismic image;
And comparing the fault probability map with the seismic image direct identification result.
Specifically, after the two layers of input and output of the quality improvement model and the fault identification model are completed, comparing the obtained fault probability map with a seismic image direct identification result obtained by directly inputting the seismic image to be identified into the fault identification model, wherein the experimental result can find that the fault probability map obtained by the quality improvement model is clearer at a fault display position and the fault display effect is better.
Based on the above embodiment, in the method, the first training set and the first verification set are generated through random parameter simulation, and specifically include:
Generating a stratum reflectivity model according to a preset rule;
Generating stratum reflectivity models of different types through random parameter control;
convolving any stratum reflectivity model with low-frequency Rake wavelets and adding seismic data of random noise to perform double downsampling to obtain a sample low-quality seismic image;
and convolving any stratum reflectivity model with the high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag.
Specifically, training data is generated by random parametric simulation. First, a reflectivity model with horizontal layers is randomly generated. Random vertical perturbations are then added to the horizontal layer model to simulate the formation dip. Next, gaussian perturbations are randomly added to the model to simulate the pleat formation. Next, the true fault distribution is simulated by adding a volumetric vector field perturbation to the model. To this end, different types of formation reflectivity models may be generated by random parameter control. And finally, convoluting the reflectivity model with the low-frequency Rake wavelet and adding the seismic data with random noise to perform double downsampling to obtain a low-quality sample seismic image, and convoluting the reflectivity model with the high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag.
Based on the above embodiment, in the method, the network structure of the quality improvement model in the training process generates a countermeasure network, and the generated countermeasure network includes a generator and a discriminator, where the generator includes a layer-jump connection and a plurality of residual connections, and the discriminator includes a convolution layer and seven convolution blocks.
Specifically, the construction of the neural network builds a network for improving the quality of the seismic data based on the current popular generation countermeasure network. The network structure comprises a generator and a arbiter. The network structure of the generator comprises a layer jump connection and a plurality of residual connections. First, shallow features of low quality seismic images are extracted by a convolution layer and parameterized ReLU (prime) activation function. Five convolved blocks with residual connections are then used to extract deep features. And then, splicing and fusing the shallow layer and the deep layer features through layer jump connection. The sub-pixel convolution layer is used to double up-sample the signature to obtain an output image of the same size as the high quality seismic image. Finally, a layer of convolution layers and a Tanh activation function are used to predict high quality seismic images. The network structure of the discriminant is simpler than the generator because it only needs to determine the probability that the input image is a true high quality seismic image. The arbiter network comprises one convolutional layer and seven convolutional blocks. Each convolution block consists of a convolution layer, a batch normalization layer (BN) and a leak ReLU activation function. Finally, two fully connected layers (Dense) and a Sigmoid activation function are performed to predict the probability that the input image is a true high quality seismic image. The antagonism learning through the generator and discrimination helps the generator to generate a more realistic high quality seismic image.
Based on the above embodiment, in the method, in the training process of the quality enhancement model, the network parameter corresponding to the highest peak signal-to-noise ratio of the quality enhancement model on the verification set is the optimal model parameter, where the peak signal-to-noise ratio is the similarity between the reconstructed high-quality seismic image and the true high-quality seismic image of the generated countermeasure network.
Specifically, by alternately optimizing the generator and the arbiter, the generator is to spoof the arbiter to generate a more realistic high quality seismic image, while the arbiter is optimized to promote the ability to discriminate between the image generated by the generator and the realistic high quality image. The Adam optimizer is used to update the generator and arbiter network parameters. The peak signal-to-noise ratio is used for representing the similarity between the reconstructed image and the original real high-quality seismic image, the optimal model is determined by checking the change of the peak signal-to-noise ratio of the model on the training set and the verification set along with the increase of the iteration times, and the network parameters corresponding to the highest peak signal-to-noise ratio of the model on the verification set are saved.
Based on the above embodiment, in the method, the model network structure used in the training process of the fault identification model is a U-net network, and the U-net network includes an encoding branch and a decoding branch, where the encoding branch is composed of four convolution operations and downsampling, and the decoding branch is composed of four upsampling and convolution operations.
Specifically, the fault recognition is regarded as an image semantic segmentation task, and each sampling point in the image to be recognized can be classified into two types, namely a fault and a non-fault. The U-net commonly used in semantic segmentation is used as a fault identification network, and is divided into an encoding branch and a decoding branch, wherein the encoding branch consists of four convolution operations and downsampling, and the decoding branch consists of four upsampling and convolution operations. The layer jump connection fuses the up-sampled feature map of the decoding branch with the feature map of the encoding branch after convolution. Finally, a convolution layer with a convolution kernel size of 1×1 and a Sigmoid activation function are used to output a predicted tomographic probability map, which is the same size as the input image to be identified.
Based on the above embodiment, in the method, the weight given to the pixel representing the fault in the loss function of the fault identification model training process is greater than the weight given to the pixel representing the non-fault.
In particular, since positive samples (tomograms) are far less than negative samples (non-tomograms), there is a strong sample imbalance in tomogram segmentation. To solve this problem, using focal loss, pixels representing faults are given higher weights as follows:
Where N represents the number of pixels in the fault probability map, y i represents the true binary label (0 represents non-fault, 1 represents fault), and p i represents the predicted probability of the fault identification network. The weight of the loss value of the positive and negative samples is used for adjusting the weight of the sample learning difficulty. According to our experimental experience, in this example, α=0.9, γ=2 is set.
The automatic recognition device of the high-resolution fault based on the deep learning provided by the invention is described below, and the automatic recognition device of the high-resolution fault based on the deep learning described below and the automatic recognition method of the high-resolution fault based on the deep learning described above can be correspondingly referred to each other.
Fig. 2 is a schematic structural diagram of the automatic recognition device for deep learning-based high resolution faults, which is provided by the present invention, as shown in fig. 2, the device includes an acquisition unit 210, a quality improvement unit 220 and a fault recognition unit 230, wherein,
The acquiring unit 210 is configured to acquire a seismic image to be identified;
the quality improving unit 220 is configured to input the seismic image to be identified into a quality improving model, and output an improved seismic image;
The quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation;
The fault identification unit 230 is configured to input the lifted seismic image into a fault identification model, and output a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
The device provided by the invention obtains the earthquake image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels. The method realizes that a cleaner and clearer fault probability map is obtained, and the predicted fault position is more accurate.
Based on the above embodiment, the apparatus further includes a comparing unit, configured to:
Inputting the seismic image to be identified into the fault identification model, and outputting a direct identification result of the seismic image;
And comparing the fault probability map with the seismic image direct identification result.
Based on the above embodiment, in the apparatus, the first training set and the first verification set are generated through random parameter simulation, and specifically include:
Generating a stratum reflectivity model according to a preset rule;
Generating stratum reflectivity models of different types through random parameter control;
convolving any stratum reflectivity model with low-frequency Rake wavelets and adding seismic data of random noise to perform double downsampling to obtain a sample low-quality seismic image;
and convolving any stratum reflectivity model with the high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag.
Based on the above embodiment, in the apparatus, the network structure of the quality improvement model in the training process generates a countermeasure network, and the generated countermeasure network includes a generator and a discriminator, where the generator includes a layer-jump connection and a plurality of residual connections, and the discriminator includes a convolution layer and seven convolution blocks.
Based on the above embodiment, in the device, in the training process of the quality enhancement model, the network parameter corresponding to the highest peak signal-to-noise ratio of the quality enhancement model on the verification set is the optimal model parameter, where the peak signal-to-noise ratio is the similarity between the reconstructed high-quality seismic image and the true high-quality seismic image of the generated countermeasure network.
Based on the above embodiment, in the device, the model network structure used in the training process of the fault identification model is a U-net network, and the U-net network includes an encoding branch and a decoding branch, where the encoding branch is composed of four convolution operations and downsampling, and the decoding branch is composed of four upsampling and convolution operations.
Based on the above embodiment, in the apparatus, the weight given to the pixel representing the fault in the loss function of the fault recognition model training process is greater than the weight given to the pixel representing the non-fault.
Based on the above embodiments, the present invention provides a method for enhancing resolution and noise reduction of a seismic image based on synthetic seismic data utilization generation countermeasure network, comprising the steps of:
S1, constructing a training data set, constructing a seismic data quality improvement neural network and training and verifying the neural network;
S2, constructing a training data set, constructing a fault identification neural network, and training and verifying the neural network;
S3, establishing a high-resolution fault automatic identification end-to-end workflow based on the seismic data quality improvement neural network and the fault identification neural network;
S4, directly sending the field real seismic data into a fault identification network to carry out fault identification and comparing the fault identification result by the method provided herein, and verifying the effectiveness of the method provided herein.
Specifically, in S1, three stages can be divided. The first stage is the creation of a training dataset using a supervised learning deep learning approach, requiring a large number of low quality seismic data and their corresponding tags (high quality seismic data) for training in order to build a mapping of low quality seismic data to high quality seismic data, however, in reality it is very difficult to collect such data. Thus, training data is generated by random parametric simulation. First, a reflectivity model with horizontal layers is randomly generated. Random vertical perturbations are then added to the horizontal layer model to simulate the formation dip. Next, gaussian perturbations are randomly added to the model to simulate the pleat formation. Next, the true fault distribution is simulated by adding a volumetric vector field perturbation to the model. To this end, different types of formation reflectivity models may be generated by random parameter control. And finally, convoluting the reflectivity model with the low-frequency Rake wavelet and adding the seismic data with random noise to perform double downsampling to obtain low-quality seismic data, and convoluting the reflectivity model with the high-frequency Rake wavelet to obtain corresponding high-quality seismic data. A large number and variety of seismic data pairs can be generated by different random parameter combinations. And taking low-quality seismic data as the input of the seismic data quality-improving neural network, and taking high-quality seismic data as a tag. The training set and the validation set are divided according to the ratio of 8:2, so that the preparation of training data is completed. The second stage is the construction of the neural network, and fig. 3 is a structural diagram of the neural network for improving the quality of the seismic data, which is constructed based on the current popular generation countermeasure network (as shown in fig. 3). The network structure comprises a generator and a arbiter. As shown in fig. 3 (a), the network structure of the generator comprises one layer jump connection and a plurality of residual connections. First, shallow features of low quality seismic images are extracted by a convolution layer and parameterized ReLU (prime) activation function. Five convolved blocks with residual connections (as shown in fig. 3 (c)) are then used to extract deep features. And then, splicing and fusing the shallow layer and the deep layer features through layer jump connection. The sub-pixel convolution layer is used to double up-sample the signature to obtain an output image of the same size as the high quality seismic image. Finally, a layer of convolution layers and a Tanh activation function are used to predict high quality seismic images. As shown in fig. 3 (b), the network structure of the arbiter is simpler than the generator because it only needs to determine the probability that the input image is a true high quality seismic image. The arbiter network comprises one convolutional layer and seven convolutional blocks (as in fig. 3 (d)). Each convolution block consists of a convolution layer, a batch normalization layer (BN) and a leak ReLU activation function. Finally, two fully connected layers (Dense) and a Sigmoid activation function are performed to predict the probability that the input image is a true high quality seismic image. The antagonism learning through the generator and discrimination helps the generator to generate a more realistic high quality seismic image. The third stage is the training and verification of the neural network, the training to generate the countermeasure network is divided into the training of the generator and the training of the arbiter. The loss function of the generator is defined as the sum of the mean square error loss plus the perceptual loss and the antagonistic loss.
The loss to the generator is defined as follows:
lG＝lMSE+0.6×lVGG+0.1×ladv
Where lMSE denotes the MSE loss at the pixel level, lVGG denotes the VGG loss, and ladv denotes the contrast loss.
LMSE calculates the pixel-by-pixel mean square error of the generated image and its corresponding real image, the calculation formula is as follows:
Where W and H represent the dimensions of the low resolution noise seismic image ILR and t represents the resolution improvement factor from ILR to IHR.
VGG networks have proven to have powerful feature extraction capabilities. And the quality of the production image is improved by calculating the difference between the generated image and the real image in the feature extraction space of the VGG network. VGG loss functions are defined based on a pre-trained 16-layer VGG network (VGG 16). The Euclidean distance between the feature map obtained by the j-th convolution (after the activation function) of the high-resolution noiseless seismic image generated by the generator and the corresponding real image IHR before the i-th layer maximum pooling layer in the VGG16 network is calculated. The Euclidean distance is used as VGG loss to measure the similarity of the reconstructed image and the real image in the VGG network characteristic representation space, and the calculation formula is as follows:
Where phi i,j denotes the operation of extracting the feature map of the jth convolutional layer (after activating the function) before the ith max-pooling layer, and W i,j and H i,j denote the size of the feature map of each feature extraction layer of the VGG network.
To fool the discriminant, the generator learns as much as possible the distribution of the true high resolution noiseless seismic image. Thus, ladv is defined in terms of the probability that the arbiter recognizes the reconstructed seismic image as a true high resolution noise-free image:
where N represents the number of low resolution noisy samples, representing the probability that the arbiter will treat the seismic image reconstructed by the generator as a true high resolution noiseless seismic image.
The loss of the arbiter is only countered. By alternately optimizing the generator and the discriminant, the generator is configured to spoof the discriminant to generate high quality seismic data that is more closely related to the true, while the discriminant is optimized to enhance the ability to discriminate between the image generated by the generator and the true, high quality image. The Adam optimizer is used to update the generator and arbiter network parameters. The peak signal-to-noise ratio is used for representing the similarity between the reconstructed image and the original image, fig. 4 is a schematic diagram of a training process of the seismic data quality improvement neural network provided by the invention, and as shown in fig. 4, an optimal model is determined by checking the change of the peak signal-to-noise ratio of the model on a training set and a verification set along with the increase of the iteration times, and network parameters corresponding to the highest peak signal-to-noise ratio of the model on the verification set are saved for the subsequent prediction task of the model.
In the present invention, the peak frequency range of the seismic data with high seismic quality given randomly is 30-55HZ, and the peak frequency range of the seismic data with low quality is 10-30HZ. While the signal to noise ratio of random noise in given low quality seismic data ranges from 3 to 13. For the division ratio of the training data, division is performed according to the criterion of 80% for training and 20% for verification. In other embodiments, other parameters may be used for simulation of the data and partitioning of the training data set.
Specifically, in S2, it is also possible to divide into three stages. The first stage is the creation of a training dataset, using the dataset disclosed (Wu et al 2020) to prepare a training dataset and a validation dataset in an 8:2 ratio. Next, a second stage of fault identification neural network is built, and fault identification is regarded as an image semantic segmentation task, and each sampling point in the seismic image can be divided into two types, namely fault and non-fault. Fig. 5 is a schematic diagram of a seismic fault identification neural network provided by the present invention, where a U-net commonly used in semantic segmentation is used as a fault identification network, and the network structure of the U-net is shown in fig. 5, where the U-net is divided into a coding branch and a decoding branch, the coding branch is composed of four convolution operations and a downsampling operation, and the decoding branch is composed of four upsampling and convolution operations. The layer jump connection fuses the up-sampled feature map of the decoding branch with the feature map of the encoding branch after convolution. Finally, a convolution layer with a convolution kernel size of 1×1 and a Sigmoid activation function are used to output a predicted fault probability map, which is the same size as the input seismic image. The third stage is the training and validation of the fault identification network, as the positive samples (fault pixels) are far less than the negative samples (non-fault pixels), there is a strong sample imbalance in the fault semantic segmentation. To solve this problem, using focal loss, the loss function L fault of the fault identification network is defined as follows, given a higher weight to the pixels representing the fault:
Where N represents the number of pixels in the fault probability map, y i represents the true binary label (0 represents non-fault, 1 represents fault), and p i represents the predicted probability of the fault identification network. The weight of the loss value of the positive and negative samples is used for adjusting the weight of the sample learning difficulty. According to experimental experience, in the present invention, α=0.9, γ=2 is set.
During training, input seismic data is normalized to eliminate the influence of data distribution difference, and image flipping and rotating data augmentation operations are performed to enhance the robustness of the model. Adam optimizers are used to update parameters of the network. Fig. 6 is a schematic diagram of a training process of a fault identification neural network provided by the invention, wherein the training process of the fault identification neural network is shown in fig. 6, and network parameters corresponding to the minimum loss of a model on a verification set are saved for a subsequent fault prediction task of the model. In other embodiments, other optimizers, training parameters, loss functions, and data enhancement modes may be used.
Specifically, in S3, an end-to-end workflow for high-resolution fault identification of field seismic data is constructed. Fig. 7 is a schematic diagram of an end-to-end workflow of the intelligent identification of high-resolution faults, wherein the workflow is formed by connecting trained neural networks in S1 and S2 in series, an original seismic image (shown in fig. 7 (a)) is sent to a trained data quality improving network in S1 to improve resolution and inhibit random noise, high-quality seismic data (shown in fig. 7 (b)) is obtained, and then the data with improved data quality is sent to a fault identification network in S2 to perform fault automatic identification, so that a high-resolution fault identification result (shown in fig. 7 (c)) is obtained. The workflow effectively integrates two tasks of seismic data quality improvement and fault identification, and can directly obtain a high-precision fault identification result from an original low-frequency seismic image filled with random noise through the method.
Specifically, in S4, compared with the conventional method, the method directly sends the original seismic image into the fault recognition network to perform fault recognition, and compares the proposed result of performing quality improvement on the seismic image and then performing fault recognition to verify the effectiveness and the advancement of the proposed new method. Firstly, the collected field seismic data are directly sent to a fault identification network trained in the step S2 to carry out fault identification. And then, adopting the workflow proposed in the step S3 to perform quality improvement on the field seismic data first and then performing fault identification. Finally, comparing the fault recognition results of the two modes. In the present invention, fig. 8 is a schematic diagram of the application of the present invention to field real seismic data and the comparison with the conventional implementation process, wherein the quality of the original seismic image (as shown in fig. 8 (a)) is first improved by using the trained seismic data quality improvement network in S1. Fig. 8 (b) is a seismic image with improved quality, random noise is significantly suppressed compared to the original seismic image, and some high-frequency geological features, such as faults and thin layers, are easier to observe by the naked eye. Then, the seismic image with improved quality is fed to a trained fault recognition network, and a fault recognition result with high resolution is obtained (as shown in fig. 8 (d)). Comparing the result with the result (shown in fig. 8 (b)) of the conventional fault identification process (the original seismic image (shown in fig. 8 (a)) is directly fed to the fault identification network), the fault identified by the proposed high-resolution fault identification method is finer and cleaner, and the noise interference is less. In addition, when some adjacent faults are directly identified through the original image, the identification results are mixed together, and the faults are separated and well characterized through the identification results of the proposed method. More importantly, as the original seismic image has lower resolution and contains a lot of random noise, some small-scale faults are not recognized, and after the quality of the seismic data is improved, the small-scale faults are well detected by a fault recognition neural network. Further comparing FIG. 8 (d) with FIG. 8 (b), the method identifies faults that are sharper than conventional methods, which will provide more accurate fault location information, which is of great help to the subsequent accurate geologic modeling effort. These all demonstrate the superiority of the proposed new high resolution fault identification method over existing fault identification methods, especially for low quality seismic data under complex geological conditions.
The invention provides an end-to-end high-resolution fault automatic identification workflow based on a deep learning technology. The workflow includes two neural networks, one for quality improvement of the seismic image and the other for seismic fault identification. First, the original seismic image is input into a trained seismic data quality enhancement network to obtain a high quality seismic image. And then inputting the fault identification data into a trained fault identification network to obtain a fault identification result with high resolution. Because of the lack of a large amount of labeled seismic data for training and verification of neural networks in practice, a large amount of training data was obtained by stochastic parametric simulation techniques in order to overcome this problem. Then, the method is applied to field real seismic data, and the result shows that the method can obtain a high-precision fault identification result from an original seismic image under the condition of no manual annotation data. And the proposed method is compared with the conventional fault identification method, and the result shows that the fault identified by the proposed high-resolution fault identification method is finer and cleaner, and the noise interference is less. In addition, when some adjacent faults are directly identified through the original image, the identification results are mixed together, and the faults are separated and well characterized through the identification results of the proposed method. More importantly, as the original seismic image has lower resolution and contains a lot of random noise, some small-scale faults are not recognized, and after the quality of the seismic data is improved, the small-scale faults are well detected by a fault recognition neural network. This means that the method will provide more accurate fault location information, which is a great aid for the subsequent accurate geologic modeling work. These all demonstrate the superiority of the proposed new high resolution fault identification method over existing fault identification methods, especially for low quality seismic data under complex geological conditions.
Fig. 9 illustrates a physical schematic diagram of an electronic device, as shown in fig. 9, which may include: processor 910, communication interface (Communications Interface) 920, memory 930, and communication bus 940, wherein processor 910, communication interface 920, and memory 930 communicate with each other via communication bus 940. The processor 910 may invoke logic instructions in the memory 930 to perform a deep learning based automatic identification method of high resolution faults, the method comprising: acquiring a seismic image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
Further, the logic instructions in the memory 930 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform the method for automatic identification of deep learning-based high resolution faults provided by the above methods, the method comprising: acquiring a seismic image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the above-provided method for automatic identification of deep learning-based high resolution faults, the method comprising: acquiring a seismic image to be identified; inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image; the quality improvement model is obtained after training a first training set and a first verification set which are constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels, and the first training set and the first verification set are generated through random parameter simulation; inputting the lifted seismic image into a fault identification model, and outputting a fault probability map; the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. An automatic identification method of high-resolution faults based on deep learning is characterized by comprising the following steps:
Acquiring a seismic image to be identified;
inputting the seismic image to be identified into a quality improvement model, and outputting an improved seismic image;
The quality improvement model is obtained after a first training set and a first verification set constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels are trained, a network structure of the quality improvement model in the training process generates an antagonism network, the antagonism network comprises a generator and a discriminator, wherein the generator comprises a layer jump connection and a plurality of residual connections, and the discriminator comprises a convolution layer and seven convolution blocks; the first training set and the first verification set are generated through random parameter simulation, and specifically comprise the following steps: generating a stratum reflectivity model according to a preset rule; generating stratum reflectivity models of different types through random parameter control; convolving any stratum reflectivity model with low-frequency Rake wavelets and adding seismic data of random noise to perform double downsampling to obtain a sample low-quality seismic image; convolving any stratum reflectivity model with a high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag;
Inputting the lifted seismic image into a fault identification model, and outputting a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
2. The automatic recognition method of deep learning-based high resolution faults of claim 1, further comprising:
Inputting the seismic image to be identified into the fault identification model, and outputting a direct identification result of the seismic image;
And comparing the fault probability map with the seismic image direct identification result.
3. The automatic recognition method of deep learning-based high-resolution faults according to claim 2, wherein in the training process of the quality improvement model, network parameters corresponding to the highest peak signal-to-noise ratio of the quality improvement model on a verification set are optimal model parameters, and the peak signal-to-noise ratio is the similarity between a reconstructed high-quality seismic image and a true high-quality seismic image of an antagonism network.
4. The automatic fault identification method based on deep learning according to claim 1, wherein the model network structure used in the training process of the fault identification model is a U-net network, the U-net network includes coding branches and decoding branches, the coding branches are composed of four convolution operations and downsampling, and the decoding branches are composed of four upsampling and convolution operations.
5. The method of claim 4, wherein the weight given to the pixels representing faults in the loss function of the fault recognition model training process is greater than the weight given to the pixels representing non-faults.
6. An automatic recognition device of high-resolution fault based on deep learning, which is characterized by comprising:
the acquisition unit is used for acquiring the seismic image to be identified;
the quality improving unit is used for inputting the seismic image to be identified into a quality improving model and outputting an improved seismic image;
The quality improvement model is obtained after a first training set and a first verification set constructed based on a plurality of sample low-quality seismic images and corresponding high-quality seismic image labels are trained, a network structure of the quality improvement model in the training process generates an antagonism network, the antagonism network comprises a generator and a discriminator, wherein the generator comprises a layer jump connection and a plurality of residual connections, and the discriminator comprises a convolution layer and seven convolution blocks; the first training set and the first verification set are generated through random parameter simulation, and specifically comprise the following steps: generating a stratum reflectivity model according to a preset rule; generating stratum reflectivity models of different types through random parameter control; convolving any stratum reflectivity model with low-frequency Rake wavelets and adding seismic data of random noise to perform double downsampling to obtain a sample low-quality seismic image; convolving any stratum reflectivity model with a high-frequency Rake wavelet to obtain a corresponding high-quality seismic image tag;
The fault identification unit is used for inputting the lifted seismic image into a fault identification model and outputting a fault probability map;
the fault identification model is obtained after training a second training set and a second verification set which are constructed based on a plurality of sample seismic images and corresponding fault probability icon labels.
7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method for automatic identification of deep learning based high resolution faults according to any of claims 1 to 5 when the program is executed by the processor.
8. A non-transitory computer readable storage medium, having stored thereon a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the automatic recognition method of deep learning based high resolution faults as claimed in any of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211331692.9A CN115639605B (en) | 2022-10-28 | 2022-10-28 | Automatic identification method and device for high-resolution fault based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211331692.9A CN115639605B (en) | 2022-10-28 | 2022-10-28 | Automatic identification method and device for high-resolution fault based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115639605A CN115639605A (en) | 2023-01-24 |
| CN115639605B true CN115639605B (en) | 2024-05-28 |
Family
ID=84946907
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211331692.9A Active CN115639605B (en) | 2022-10-28 | 2022-10-28 | Automatic identification method and device for high-resolution fault based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115639605B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115880505B (en) * | 2023-03-08 | 2023-05-09 | 山东科技大学 | Low-order fault intelligent identification method for target edge detection neural network |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516740A (en) * | 2019-08-28 | 2019-11-29 | 电子科技大学 | A kind of fault recognizing method based on Unet++ convolutional neural networks |
| CN112102322A (en) * | 2020-09-16 | 2020-12-18 | 电子科技大学 | Fault identification method based on multi-mode U-Net |
| CN112241741A (en) * | 2020-08-25 | 2021-01-19 | 华中农业大学 | Self-adaptive image attribute editing model and method based on classified countermeasure network |
| CN112396558A (en) * | 2019-08-15 | 2021-02-23 | 株式会社理光 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN112435309A (en) * | 2020-12-07 | 2021-03-02 | 苏州深透智能科技有限公司 | Method for enhancing quality and resolution of CT image based on deep learning |
| CN112686822A (en) * | 2020-12-30 | 2021-04-20 | 成都信息工程大学 | Image completion method based on stack generation countermeasure network |
| CN113160575A (en) * | 2021-03-15 | 2021-07-23 | 超级视线科技有限公司 | Traffic violation detection method and system for non-motor vehicles and drivers |
| CN113538324A (en) * | 2020-09-18 | 2021-10-22 | 腾讯科技（深圳）有限公司 | Evaluation method, model training method, device, medium, and electronic apparatus |
| CN113902769A (en) * | 2021-08-18 | 2022-01-07 | 南方海洋科学与工程广东省实验室(广州) | Seismic fault identification method based on deep learning semantic segmentation |
| CN114707534A (en) * | 2022-03-07 | 2022-07-05 | 华北电力大学（保定） | Rolling bearing fault diagnosis method under small sample data |
| CN114898160A (en) * | 2022-06-02 | 2022-08-12 | 电子科技大学 | Intelligent fault identification method based on multiple tasks |
| CN115201902A (en) * | 2022-07-04 | 2022-10-18 | 中海石油(中国)有限公司 | Fault intelligent identification method and system based on deep learning |
-
2022
- 2022-10-28 CN CN202211331692.9A patent/CN115639605B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112396558A (en) * | 2019-08-15 | 2021-02-23 | 株式会社理光 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN110516740A (en) * | 2019-08-28 | 2019-11-29 | 电子科技大学 | A kind of fault recognizing method based on Unet++ convolutional neural networks |
| CN112241741A (en) * | 2020-08-25 | 2021-01-19 | 华中农业大学 | Self-adaptive image attribute editing model and method based on classified countermeasure network |
| CN112102322A (en) * | 2020-09-16 | 2020-12-18 | 电子科技大学 | Fault identification method based on multi-mode U-Net |
| CN113538324A (en) * | 2020-09-18 | 2021-10-22 | 腾讯科技（深圳）有限公司 | Evaluation method, model training method, device, medium, and electronic apparatus |
| CN112435309A (en) * | 2020-12-07 | 2021-03-02 | 苏州深透智能科技有限公司 | Method for enhancing quality and resolution of CT image based on deep learning |
| CN112686822A (en) * | 2020-12-30 | 2021-04-20 | 成都信息工程大学 | Image completion method based on stack generation countermeasure network |
| CN113160575A (en) * | 2021-03-15 | 2021-07-23 | 超级视线科技有限公司 | Traffic violation detection method and system for non-motor vehicles and drivers |
| CN113902769A (en) * | 2021-08-18 | 2022-01-07 | 南方海洋科学与工程广东省实验室(广州) | Seismic fault identification method based on deep learning semantic segmentation |
| CN114707534A (en) * | 2022-03-07 | 2022-07-05 | 华北电力大学（保定） | Rolling bearing fault diagnosis method under small sample data |
| CN114898160A (en) * | 2022-06-02 | 2022-08-12 | 电子科技大学 | Intelligent fault identification method based on multiple tasks |
| CN115201902A (en) * | 2022-07-04 | 2022-10-18 | 中海石油(中国)有限公司 | Fault intelligent identification method and system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于GEP多标记学习的图像超分辨率复原算法;汤嘉立;柳益君;杜卓明;;计算机应用研究;20170614(第06期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115639605A (en) | 2023-01-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11320551B2 (en) | Training machine learning systems for seismic interpretation | |
| Shi et al. | Development of subsurface geological cross-section from limited site-specific boreholes and prior geological knowledge using iterative convolution XGBoost | |
| CN109709603B (en) | Seismic horizon identification and tracking method and system | |
| AlRegib et al. | Subsurface structure analysis using computational interpretation and learning: A visual signal processing perspective | |
| CN107688201B (en) | RBM-based seismic prestack signal clustering method | |
| Van den Ende et al. | A self-supervised deep learning approach for blind denoising and waveform coherence enhancement in distributed acoustic sensing data | |
| CN110609320B (en) | Pre-stack seismic reflection pattern recognition method based on multi-scale feature fusion | |
| EA004551B1 (en) | Method for seismic facies interpretation using textural analysis and neutral networks | |
| CN112731522B (en) | Intelligent recognition method, device and equipment for seismic stratum and storage medium | |
| Xiao et al. | Conditioning of deep-learning surrogate models to image data with application to reservoir characterization | |
| CN115639605B (en) | Automatic identification method and device for high-resolution fault based on deep learning | |
| Wang et al. | Seismic stratum segmentation using an encoder–decoder convolutional neural network | |
| CN111915618A (en) | Example segmentation algorithm and computing device based on peak response enhancement | |
| Yang et al. | A multi-task learning method for relative geologic time, horizons, and faults with prior information and transformer | |
| Lyu et al. | A buried hill fault detection method based on 3D U-SegNet and transfer learning | |
| CN116068644B (en) | Method for improving resolution and noise reduction of seismic data by using generation countermeasure network | |
| Dhara et al. | Seismic image registration using multiscale convolutional neural networks | |
| Udegbe et al. | Big Data Analytics for Seismic Fracture Identification, Using Amplitude-Based Statistics | |
| Han et al. | Algorithm for Intelligent Recognition Low-Grade Seismic Faults Using Codec Target Edges | |
| Lomask | Seismic volumetric flattening and segmentation | |
| CN112464976A (en) | Salt dome identification method based on multi-attribute Unet | |
| Liu | Multilevel strategy for O-PCA-based history matching using mesh adaptive direct search | |
| Sun et al. | Enabling uncertainty quantification in a standard full-waveform inversion method using normalizing flows | |
| Zhou et al. | Graph-learning approach to combine multiresolution seismic velocity models | |
| Zhang et al. | Deep learning-aided image-oriented history matching of geophysical data |
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 |