Dynamic imaging through random perturbed fibers via physics-informed learning

Lensless flexible multicore fiber (MCF) endoscopes have the capability of imaging beyond conventional endoscopes. The MCF provides a simple imaging solution when the object is adjacent to the fiber facet. In practice, an ideal fiber endoscope is effective for high-resolution imaging through perturbed fibers. However, different fiber states lead to different configurations which bring different scattering distributions. The traditional methods are limited by the field of view (FOV) and reconstruction capability of algorithm. In this paper, through the effective combination of the speckle-correlation theory and the deep learning (DL) method, we demonstrate a physics-informed DL method for imaging through perturbed fibers. With the speckle redundancy, the object imaging through perturbed fibers is reconstructed completely and accurately by training with only one configuration. And objects of different complexity can be reconstructed effectively. Furthermore, the approach is also effective for imaging through dynamic fiber and random length fibers. This method gives impetus to the development of a lensless fiber endoscope in practical scenes and provides an enlightening reference for using DL methods to solve fiber imaging problems.

Automated summary

In this paper, a physics-informed deep learning (DL) method for imaging through perturbed fibers is demonstrated by effectively combining speckle-correlation theory and DL method. With speckle redundancy, the object imaging through perturbed fibers can be completely and accurately reconstructed by training with only one configuration, and objects of different complexity can be effectively reconstructed. Furthermore, this approach is also effective for imaging through dynamic fibers and random length fibers. This method provides impetus to the development of lensless fiber endoscope in practical scenes and offers an enlightening reference for using DL methods to solve fiber imaging problems.