Melissa Muszelewski, Johannes Münderlein, Cosima Birkmaier opencampus.sh Scientific Machine Learning Winter term 2025/2026
https://github.com/theDatascientista/ML-Microscpy_imaging_PINN
The idea of this project was based on the publication "Untrained, physics-informed neural networks for structured illumination microscopy" by Zachary Burns and Zhaowei Liu to train a PINN for a better microscopy image processing solution, regarding quality improvement and automation.
Creation of a Physics-informed neural network for microscopy imaging, based on partial differential equations related to image processing.
| Model | Loss | Comments |
|---|---|---|
| CNN | MSE = 0.0054 | 1000 Epochs |
| Supervised | MSE = 0.ooo12 | 1500 Epochs |
| Frequency Domain | Frequency Loss = 0.080636 | 30 Epochs |
| Frequency Domain (Combined with Spatial) | Frequency Loss = 0.080636 | 111 Epochs |
