Adaptive multiscale optical spatial filtering via a polarization-tunable phase-change micropore array

As artificial intelligence and machine vision tasks evolve toward high throughput and low latency, all-optical analog signal processing demonstrates tremendous potential. However, the optical responses of most conventional static structures are fixed, rendering dynamic reconfiguration challenging. Dynamic devices based on conventional phase-change materials typically only achieve bright/dark field switching due to high extinction losses. Herein, a tunable composite micropore array is proposed based on the low-loss phase-change material Sb₂S₃. We achieve optical reconfiguration between spatial high-pass/low-pass filtering across the crystalline and amorphous states, while simultaneously unveiling a physical mechanism for the continuous tuning of spatial bandwidth via the polarization state. In the crystalline state, polarization modulation enables high-contrast extraction of microscopic features and background suppression of macroscopic structures, thereby achieving the recognition of multiscale targets with adaptive contrast. In the amorphous state, an all-optical trade-off between image smoothing and denoising performance is accomplished. This study elevates optical analog signal processing from a simple stacking of single functionalities to what we believe to be a novel stage of multiscale processing and recognition.