Gabor-Enhanced-PINN

Gabor-Enhanced Physics-Informed Neural Networks (PINNs) for Fast Simulations of Acoustic Wavefields

This repository contains the implementation of Gabor-Enhanced Physics-Informed Neural Networks (PINNs) for solving the Helmholtz equation efficiently, as presented in the paper:

"Gabor-Enhanced Physics-Informed Neural Networks for Fast Simulations of Acoustic Wavefields"

---

Overview

Traditional PINNs converge slowly for wavefield simulations due to their low-frequency bias. This work introduces an improved PINN framework that integrates explicit Gabor functions into the network architecture, significantly enhancing convergence speed and accuracy for solving the Scattered Helmholtz equation. The framework is designed to work with realistic velocity models and includes a Perfectly Matched Layer (PML).

The implementation supports training and validation using various benchmark models and provides pretrained models and visualization tools for streamlined experimentation and evaluation.

---

Features

• ✅ Implements both Gabor-enhanced PINN and standard PINN for comparison
• ✅ Supports multiple benchmark velocity models, including:
  • Simple layered model
  • Marmousi
  • Overthrust
• ✅ Includes PML to prevent wave reflections at domain boundaries
• ✅ Includes Positional encoding to better capture oscilatory wavefields
• ✅ Uses an exponentially decaying learning rate for improved convergence
• ✅ Provides training and validation datasets for all test models
• ✅ Includes pretrained models saved at multiple epochs for fast inference and benchmarking
• ✅ Offers visualization tools to plot:
  • Velocity models
  • Reference wavefields (finite-difference)
  • Predicted wavefields (PINN outputs)

---

Main Scripts

• Gabor_enhanced_PINN.py
  Main training script to train either Gabor-PINN or standard PINN on any of the supported velocity models.

• Inference_Plotting.py
  Utility script to load pretrained models and plot velocity profiles, reference wavefields, and model predictions for easy comparison.

Usage

• Use Gabor_enhanced_PINN.py to train on your chosen velocity model. You can specify training settings within the script.

• Use Inference_Plotting.py to:

  • Load a pretrained model (either Gabor-PINN or standard PINN)
  • Visualize the corresponding velocity model
  • Compare the predicted wavefield to the reference finite-difference solution

---

Installation

This code requires Python 3.8+ and the following dependencies:

pip install numpy matplotlib scipy tensorflow