Abstract:By circumventing the resolution limitations of optics, coherent diffractive imaging (CDI) and ptychography are making their way into scientific fields ranging from X-ray imaging to astronomy. Yet, the need for time consuming iterative phase recovery hampers real-time imaging. While supervised deep learning strategies have increased reconstruction speed, they sacrifice image quality. Furthermore, these methods' demand for extensive labeled training data is experimentally burdensome. Here, we propose an unsupervised physics-informed neural network reconstruction method, PtychoPINN, that retains the factor of 100-to-1000 speedup of deep learning-based reconstruction while improving reconstruction quality by combining the diffraction forward map with real-space constraints from overlapping measurements. In particular, PtychoPINN significantly advances generalizability, accuracy (with a typical 10 dB PSNR increase), and linear resolution (2- to 6-fold gain). This blend of performance and speed offers exciting prospects for high-resolution real-time imaging in high-throughput environments such as X-ray free electron lasers (XFELs) and diffraction-limited light sources.

What problem does this paper attempt to address?

The paper primarily focuses on addressing several key issues in the fields of Coherent Diffractive Imaging (CDI) and ptychography: 1. **The Challenge of Real-Time Imaging**: Traditional CDI techniques rely on time-consuming iterative phase retrieval algorithms, which limit their practicality in applications requiring high-speed imaging. 2. **Compromise in Image Quality**: While supervised learning methods can accelerate the CDI reconstruction process, these methods often come at the cost of sacrificing image quality. 3. **The Need for Training Data**: Existing deep learning-based methods require a large amount of labeled data to train models, which is very challenging and time-consuming in experiments. To address these issues, the paper proposes a new method called PtychoPINN, which is a Physics-Informed Neural Network (PINN) that combines the diffractive forward mapping with real-space constraints from overlapping measurements. This method significantly improves the quality of reconstructed images while maintaining the speed advantages brought by deep learning. Specifically, the key contributions of the PtychoPINN method include: - Providing an unsupervised training framework that utilizes diffractive forward mapping without the need for extensive labeled data. - Integrating additional physical constraint information from the ptychography setup, further enhancing the model's generalization ability and accuracy. - Explicitly addressing photon counting statistics (Poisson statistics), which enhances the model's robustness to real data noise. With these improvements, the PtychoPINN method not only enhances the resolution of reconstructed images (2 to 6 times linear resolution improvement) but also increases the peak signal-to-noise ratio (PSNR, typically a 10-decibel increase), offering exciting prospects for high-resolution real-time imaging in high-throughput environments, such as applications with X-ray Free Electron Lasers (XFELs) and diffraction-limited light sources.

Physics Constrained Unsupervised Deep Learning for Rapid, High Resolution Scanning Coherent Diffraction Reconstruction

Universal and High-Fidelity Resolution Extending for Fluorescence Microscopy Using a Single-Training Physics-Informed Sparse Neural Network

AI-enabled high-resolution scanning coherent diffraction imaging

Real-time 3D Nanoscale Coherent Imaging via Physics-aware Deep Learning

A Physics-Inspired Deep Learning Framework for an Efficient Fourier Ptychographic Microscopy Reconstruction under Low Overlap Conditions

Physics-guided Deep Learning-Based Real-Time Image Reconstruction of Fourier-domain Optical Coherence Tomography.

Snapshot Coherent Diffraction Imaging Via a Physics-embedded Untrained Neural Network

Physics to the Rescue: Deep Non-line-of-sight Reconstruction for High-speed Imaging

Deep Learning Coherent Diffractive Imaging

AutoPhaseNN: unsupervised physics-aware deep learning of 3D nanoscale Bragg coherent diffraction imaging

Accelerated deep self-supervised ptycho-laminography for three-dimensional nanoscale imaging of integrated circuits

Three-dimensional Coherent X-ray Diffraction Imaging via Deep Convolutional Neural Networks

DNF: Diffractive Neural Field for Lensless Microscopic Imaging

Unsupervised physics-informed deep learning-based reconstruction for time-resolved imaging by multiplexed ptychography

Quantitative Pure-Phase Object Reconstruction under Single-Shot Fourier Measurement Via Deep Learning

Reconstruction of nanoscale particles from single-shot wide-angle FEL diffractions patterns with physics-informed neural networks

Physics-Informed Ellipsoidal Coordinate Encoding Implicit Neural Representation for high-resolution volumetric wide-field microscopy

Physics-informed Deep Learning for Fringe Pattern Analysis

Disentangling multiple scattering with deep learning: application to strain mapping from electron diffraction patterns

Deep-learning real-time phase retrieval of imperfect diffraction patterns from X-ray free-electron lasers