Seismic data texture feature reconstruction method based on deep learning

A technology of seismic data and texture features, applied in neural learning methods, image data processing, biological neural network models, etc., can solve the problems of image texture damage and cannot be restored, achieve a small number of layers, restore texture feature information, and improve peak value Effects of Signal-to-Noise Ratio and Structural Similarity

Pending Publication Date: 2021-08-17
HEBEI UNIV OF TECH
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This reconstruction method is based on image super-resolution of a single image, but there are certain problems in the reconstruction based on a single image: during tra

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Seismic data texture feature reconstruction method based on deep learning
  • Seismic data texture feature reconstruction method based on deep learning
  • Seismic data texture feature reconstruction method based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0086] Step 1, in this embodiment, the size is 128 × 128; the low-resolution seismic image missing traces is randomly missing 30% traces, that is, images with a random sampling rate of 70% (such as image 3 shown). The original seismic image HR (such as Figure 5 shown) and the reference image Ref (such as Figure 4 shown).

[0087] Step 2. In order to facilitate the training of the seismic data reconstruction network based on texture migration (referred to as the seismic data reconstruction network), the low-resolution missing channel seismic image is first down-sampled (in this embodiment, it is down-sampled by 4 times), and the low-resolution image is generated. LR, and then perform the same multiple (4 times) upsampling of the low-resolution image LR to make it an upsampled low-resolution image LR_sr with the same size as the reference image Ref (this embodiment is a double-cubic upsampled low-resolution image);

[0088] Considering that the resolutions of the low-reso...

Embodiment 2

[0095] Step 1, in this embodiment, the size is 128×128; the low-resolution missing channel seismic images are images with random sampling rates of 50%, 60%, 80% and 90%.

[0096] Same as Steps 1-7 of Example 1.

[0097] Step 8. The test set is input into the trained seismic data reconstruction network. After texture extraction and texture migration, the high-resolution reconstruction image SR is reconstructed, and the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) are used as evaluation Indicators are evaluated.

[0098] Depend on Figure 9 It can be seen that as the sampling rate increases, the reconstruction results are better. The image with higher sampling rate can find more similar texture feature information when calculating the similarity with the texture feature information of the reference image, but when the sampling rate is too large, the reconstruction result is less improved.

[0099] What is not mentioned in the present invention is applica...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a seismic data texture feature reconstruction method based on deep learning. According to the method, the texture extraction network is adopted, the shallow convolutional network is used for training, and the texture extraction network continuously updates own parameters along with the training process, so that the texture extraction network can extract the most appropriate texture feature information. The similarity ri, j between every two feature blocks of one feature block in the up-sampling low-resolution image feature map Q and one feature block in the down-sampling and up-sampling reference image feature map K is respectively calculated by adopting a normalized inner product method, transfer learning is carried out by calculating the similarity through blocks, and texture transfer is carried out by using an attention mechanism. And adversarial loss and perception loss are added to the loss function part. According to the method, parameters can be automatically updated, other prior information is not needed, a complex texture feature structure can be learned, the problem of spatial aliasing is effectively avoided, and clear high-resolution seismic data can be quickly reconstructed.

Description

technical field [0001] The technical solution of the present invention relates to the field of seismic data reconstruction, in particular to a method for reconstructing seismic data texture features based on deep learning. Background technique [0002] With the continuous development of science and technology, oil and gas exploration has become more mature, and the exploration of oil and natural gas has become more and more intensive. However, in the process of data collection for exploration, it will be limited by factors such as the collection environment and collection cost. Problems such as mountains, rivers, and hardware instruments will cause bad sectors in the collection, so that the collected data is irregular and incomplete. Therefore, it is of great practical significance to interpolate and reconstruct the collected seismic data to obtain high-resolution seismic data so that it has more realistic geophysical information. [0003] Traditional seismic interpolation ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06T7/40G06T3/40G06K9/46G06K9/62G06N3/04G06N3/08
CPCG06T7/40G06T3/4053G06N3/08G06T2207/20081G06V10/44G06N3/045G06F18/22
Inventor 顾军华李一凡贾永娜沈晓宁王国伟吴杰
Owner HEBEI UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap