Anti-aliasing seismic data regularization method based on deep learning based on wave-atom transformation

Abstract

The invention belongs to the field of earth science and technology, and specifically relates to a deep learning anti-aliasing seismic data regularization method based on wave-atom transformation. The steps of the method are 1. training data set preparation; 2. wave-atom domain sample label preparation; 3. . Network input and label setting; 4. Deep learning network model G structure setting; 5. Loss function setting; 6. Network model training; 7. Seismic data regularization test. According to the good distribution characteristics of seismic data in the wave atomic domain, a deep convolutional neural network model for joint learning of the spatial domain and the wave atomic domain is established, and the seismic data is regularized by combining the characteristics of the spatial domain and the wave atomic domain. The training evaluation index of the model is adopted Space domain, wave atom domain error and f-k domain error jointly constrain the regularization error, feed back to adjust the network parameters, and improve the accuracy and generalization ability of the seismic data regularization network model.

Description

Technical field: [0001] The invention belongs to the technical field of earth sciences, and in particular relates to a deep learning anti-aliasing seismic data regularization method based on wave-atom transformation. Background technique: [0002] Today, with the rapid development of big data and artificial intelligence, the problems of traditional seismic data rule methods are expected to be solved by new technologies. The purpose of seismic exploration is to obtain accurate imaging of subsurface structures. Ideally, the sampling of seismic wavefields should be regular and dense. It is technically and computationally feasible to use modern instrumentation equipment to perform dense sampling in a specific time and space, but since the cost of field data acquisition accounts for more than 80% of the entire seismic exploration cost, from an economic point of view Considering that seismic data is often sparsely sampled in the spatial direction, this results in less data being ...

AI Technical Summary

Problems solved by technology

However, the problem with this method is that in the process of solving the regularized ill-posed inverse problem, the optimization method with sparse constraints is basically used, and this method is greatly affected by the sparse prior of the data, resulting in unstable results.

The above-mentioned existing methods solve the problem that the traditional seismic data reconstruction algorithm needs to meet the limitation of Nyquist sampling theorem; solve the problem of difficult selection of sparse bases for seismic data reconstruction using compressed sensing algorithm, but do not solve the problem of insufficient feature extraction in the spatial domain

[0006] In the current seismic data regularization method, the regularization technology based on deep learning is currently a hot spot that is widely concerned, and it can also achieve better results than traditional methods. Basically, the spatial convolution method is used to obtain the characteristic information of the spatial domain, while ignoring the good characteristics of seismic data in some transformation domains, and in the definition of the training evaluation index function, most of the single distance is used to calculate the error. Consider errors from the perspective of spectrum, such as Chinese patent [CN201910599289.6]

These defects will cause the reconstruction quality to fail to achieve ideal results in complex terrain, and even aliasing may occur, so that the regularization results are too smooth and the imaging effect is not ideal

Images