模型与数据驱动联合的地震数据重构

复杂弱信号重构是地震数据处理中非常具有挑战性的问题。传统压缩感知与稀疏变换等模型驱动类方法在常规地震数据重构上取得了较好的效果，但常常混淆弱信号和噪声。近年来，基于数据驱动的深度神经网络方法快速发展，从大量数据中智能学习得到深层次特征，在处理复杂弱信号上有一定优势，但其样本构造及网络训练成本极高。如何有机结合模型驱动与数据驱动两类方法的优点仍是一个开放难题。本项目针对弱信号重构问题，结合压缩感知框架，研究弱信号的模型和数据联合约束方法。重点研究内容包括：1、地震数据重构的生成约束方法；2、构造一种新的判别约束；3、地震数据重构的判别约束方法；4、新型约束在弱信号重构中的应用。本项目立足应用数学、人工智能和地震勘探的学科交叉，深度开展弱信号数据重构理论研究，有助于降低环境因素对成像质量的影响、提高弱信号反演分辨率。

Reconstruction of complex and weak signals is a challenging problem in seismic data processing. Traditional model-driven methods such as compressed sensing and sparse transform have achieved satisfying results in conventional seismic data reconstruction, but have difficulties in distinguishing weak signals and noise. In recent years, data-driven methods, such as deep neural network, have developed rapidly. Deep neural network obtains deep level features from a large amount of data and has advantages in processing complex and weak signals. But the costs of training set construction and network training are extremely high. How to combine the advantages of model-driven and data-driven methods is still an open problem. This project aims at weak signal reconstruction by combining compressed sensing with the new constraints. This project focuses on the following subjects: 1. Generative constraint for data reconstruction; 2. Constructing a new discriminant constraint; 3. Discriminant constraint for data reconstruction; 4. The application of new constraints in weak signal reconstruction. This project is based on the interdisciplinary of applied mathematics, artificial intelligence and geophysics. We focus on complex and weak data reconstruction, aiming to reduce the negative effect of environmental factors and increase efficiency and accuracy of seismic exploration.