Seismic data random noise suppression method combining deep learning
A seismic data and random noise technology, applied in the field of earth science, can solve the problem of insufficient feature extraction of seismic data, and achieve the effect of improving generalization ability and convergence ability, and improving noise suppression effect.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0034] refer to figure 1 and figure 2 , a seismic data random noise suppression method of combined deep learning, characterized in that: the suppression method comprises the following steps:
[0035] Step 1: Training dataset preprocessing:
[0036] In order to combine the characteristics of the wave atomic domain, the minimum unit of the training sample is set to several slice data x with a size of 256×256 cut from the seismic data samples in the training set. The random noise of the seismic data is passed through the zero mean normal distribution Gaussian random noise simulation, the standard deviation of the noise is positively correlated with the standard deviation of the original seismic data, the noise standard deviation is defined as:
[0037]
[0038] M is the total number of slice time samples, N is the total number of slice seismic trace samples, t is the time sampling sequence number, s is the seismic trace record sequence number, u is the mean value of seismic...
Embodiment 2
[0054] refer to Figure 3-8 The training of the joint learning network model of this method mainly includes training data preprocessing, preparing sample labels in the frequency domain and air domain, designing the network structure of the joint learning model, designing the air domain-frequency domain joint loss function, training and saving the network model, There are six parts to test network model performance. First, the training data is preprocessed to obtain the original seismic slice data and the slice data containing random noise; secondly, the sample labels in the frequency domain and the air domain are prepared; then the network structure of the joint learning model is designed, which consists of three parallel deep Convolution network, two of which are the prediction network of wave atomic domain coefficient matrix, and one is the prediction network of spatial domain seismic data; then design the space domain-frequency domain joint loss function; then the training ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


