
A method using neural networks to detect or reconstruct polarization states from photon data.Neural network polarimetry is a computational technique that uses machine learning to infer the polarization state of light from a very small number of detected photons. Traditional polarimetry requires measuring light intensity across multiple filter orientations, which demands many photons for statistical accuracy. When only a handful of photons are available—for example, from a faint astronomical source or in a fast-moving measurement—the reconstruction becomes noisy or impossible with classical methods. Neural networks, trained on simulated or calibrated data, can recognize subtle patterns in sparse photon counts and recover the polarization information that would otherwise be lost.
The application of neural networks to polarimetry emerged in the 2010s, driven by advances in deep learning and the need for faster, more sensitive polarimetric measurements in fields like biomedical imaging and astronomy. No single inventor is credited; it developed through contributions from multiple research groups.
The method typically involves a neural network that takes as input the raw detection events (e.g., photon counts in different polarization channels) and outputs the estimated Stokes parameters that describe the polarization state. The network is trained on a large dataset of simulated photon detections with known polarization states, learning the statistical mapping from sparse data to the underlying polarization. Once trained, it can process new measurements in real time, effectively 'seeing the invisible' by extracting information from noise.
It is used in biomedical imaging to analyze tissue structure from scattered light, in remote sensing (e.g., lidar) to distinguish water from ice clouds, and in astronomy to measure polarization of faint objects like neutron stars or exoplanets.
Neural network polarimetry can reconstruct polarization from as few as a few dozen photons, where classical methods would require thousands.
| A | variable | Polarization vector — polarization vector |
| Δ | variable | Amplitude of the field — amplitude volt per metre |
| e_x | variable | Unit vector along x-axis — unit vector |
| e_y | variable | Unit vector along y-axis — unit vector |
| i | operator | Complex number with square -1 — imaginary unit |













