Dynamic Scattering Imaging in Ring Core Fiber via Enhanced Optical Neural Networks with Zernike-Vortex Modulation

Dynamic scattering in Ring Core fibers (RCF) poses fundamental limitations for high-precision imaging and broadband communications, where conventional all-optical neural networks (ONNs) remain constrained by rigid architectures and poor generalization across fiber geometries. Here, we introduce a space-Fourier domain-enhanced optical convolutional neural network (EOCNN) that enables speckle reconstruction in RCF platforms. The concept and implementation establish a new framework for complex photonic systems. Departing from static diffractive designs, our framework integrates Zernike phase kernels, which are optimized by improving the feature extraction of speckle. Critically, the single-stage architecture of EOCNN bypasses fiber-specific retraining, addressing a long-standing tradeoff between model versatility and optical system complexity. This work establishes a physics-aware machine learning paradigm for light-field control in disordered media, with direct implications for endoscopic imaging and optical signal processing.