Robust imaging through multimode fibers: from conventional to machine learning-assisted

Multimode fiber (MMF) imaging has demonstrated great potential in fields such as biomedicine, industrial inspection, and optical communications, owing to its high spatial resolution, ultra-compact form factor, and flexible light transmission properties. However, MMFs are highly sensitive to external disturbances such as temperature fluctuations, mechanical perturbations, and fiber bending, which can lead to significant variations in the transmission matrix (TM). These changes severely degrade image reconstruction quality and system stability. To address these challenges, a variety of disturbance-robust strategies have been proposed, ranging from traditional optical compensation techniques to intelligent modeling approaches based on deep learning. This review systematically summarizes recent advances in robust imaging through MMFs, focusing on three main categories: (1) traditional optics-based approaches involving structural design, phase conjugation, and TM estimation; (2) reconstruction techniques utilizing structured illumination and compressive sensing; and (3) machine learning-assisted strategies based on neural networks, further subdivided into experimental methodologies and network architecture designs. In addition, this review outlines robust imaging solutions under diverse perturbation scenarios, including temperature fluctuations, polarization changes, and noise interference, and provides a comparative analysis of representative models and their performance. Finally, we discuss the future prospects and potential applications of robust MMF imaging in real-world scenarios such as flexible endoscopy, high-speed imaging, and intelligent sensing, offering a theoretical foundation and research reference for advancing its practical deployment.