Acoustic Inversion via Causal Wave Simulation and ML

This project implements a full pipeline for acoustic inversion using wave propagation, machine learning, and causal graph reasoning.

---

🧠 What It Does

• Simulates acoustic wave propagation in heterogeneous 2D media
• Generates pressure field p(t,x,y) and integrated hologram
• Trains a direct model to simulate holograms from medium parameters
• Trains an inverse model to reconstruct c(x,y) and ρ(x,y) from holograms
• Builds causal and backtrace graphs from the wave tensor
• Validates medium reconstruction without ground truth

---

📦 Key Files

File	Description
TestVoln.py	Simulates the wave using finite difference
generate_dataset.py	Generates synthetic data samples
train_model_multisample.py	Trains the direct CNN
inverse_model_full.py	Trains and infers the inverse model
inference.py	Runs inference without ground truth
build_graph.py	Builds full causal graph
build_backtrace_graph.py	Builds reduced backtrace graph
*.pt	Trained models
*.png, *.gif	Visualizations
*.graphml	Graph files viewable in Gephi

---

📊 Visual Outputs

• wave_propagation.gif: Simulated wave animation
• train_results.png: Side-by-side ground truth and prediction
• inferred_fields.png: Inferred c(x,y) and ρ(x,y) without GT
• backtrace_graph.png: Final causal front

---

🔧 Requirements

pip install torch numpy matplotlib networkx tqdm
 License
MIT / CC-BY 4.0 — Open for academic and non-commercial use.