Deep network reparameterized full-waveform inversion from sequential to simultaneous sources with adjoint-state method

A novel framework integrating deep learning reparameterization with the adjoint-state method is proposed for simultaneous source inversion. An innovative deep network reparameterization technique for FWI processes is introduced, combined with a crosstalk-free source encoding strategy to enhance accuracy in simultaneous source inversion. 1. This repository provides a real field seismic dataset application case, containing 20 high-quality shot gathers acquired in the field, with a profile length of approximately 6 km. 2. The dataset is equipped with complete survey geometry files, initial velocity model and detailed explanatory documentation, which can be directly used to test and validate Full Waveform Inversion (FWI) methods. 3. Inversion results of four representative methods are provided, with one-click plotting supported. 4. We also release a deep learning-powered simultaneous-source inversion software package with full reproducibility. 5. The repository includes core deep learning Python scripts that implement simultaneous-source, multi-scale, and parameterized inversion, as well as an external dynamic-link library (DLL) for standalone forward modeling and gradient computation. 6. After configuring the environment via the provided environment file, the entire pipeline supports one-click execution.