Polarization-Multiplexed Optical Neural Network for Simultaneous Amplitude Reconstruction and Edge Detection

Integrating amplitude retrieval and edge detection in optical neural network (ONN)-driven metasurfaces advances intelligent photonic devices by addressing multifunctional design challenges. Conventional electronic architectures face fundamental limitations in processing speed and energy efficiency when handling optical information streams, whereas ONNs exploit light's parallelism for real-time physical-layer computation. In this work we propose a polarization-multiplexed metasurface that synergistically combines both tasks optically, eliminating digital post-processing. The framework employs trainable nanostructures to modulate light across object, metasurface, and imaging planes. Simulation results validate high-fidelity amplitude reconstruction and edge extraction through polarization switching, enabling seamless dual-function operation. The design achieves robust polarization isolation and energy efficiency, surpassing hybrid optoelectronic systems. An end-toend training pipeline bridges computational optimization and nanofabrication, ensuring scalability. Future work targets broadband operation, nonlinear integration, and 3D photonic architectures. This platform pioneers intelligent optical computing for applications in machine vision, sensing, and autonomous navigation, overcoming critical barriers in multifunctional metasurface technology.