2025 新生 DL-FWI 培训

摘要: 本贴给出 8 次讨论式培训的提纲, 每次培训 1 小时.

1. Basic concepts
   1.1 Sesmic data processing – regular process
   1.2 Full waveform inversion
   1.3 Deep learning full waveform inversion

2. Network structure
   2.1 InversionNet
   Encoder-decorder structure
   2.2 FCNVMB
   UNet
   2.3 DDNet
   Multi-task

3. Network components
   3.1 Deformable convolution
   3.2 Spatial attention module

4. Loss function
   4.1 Pixel: L1 and L2
   4.2 Boundary

5. Training
   5.1 Curriculum learning
   5.2 Transfer learning
   5.3 Domain adaptation

6. Observation system

7. Problem statement
   7.1 1D CMP based inversion
   7.2 2D velocity inversion

8. Chellenges
   8.1 Data distribution
   Transfer learning
   8.2 Data size
   Too big
   8.3 Dataset size
   Few field data
   8.4 Interpretability
   PINN: Physics informed neural network