Deep Learning‐Powered Perovskite Paper Sensors for Ultrasensitive on‐Site Mercury Detection

Abstract Rapid, instrument‐free detection of mercury ions (Hg 2+ ) is critical for addressing environmental emergencies, yet current methods face limitations in poor portability and sensitivity. Herein, a deep learning‐optimized paper‐based kit that incorporates controllable turn‐off fluorescence of metal halide perovskite for instrument‐free time‐critical Hg 2+ detection by the dynamic quenching mechanism and ensures visual discernment at 2 ppb, achieving a 1000‐fold sensitivity improvement over commercial strips is presented. Furthermore, a smartphone app powered by the deep learning model (MobileViT) further optimizes sensitivity, achieving ultrasensitive quantification (0.11 ppb limit of detection) across a 0.2–200 ppb linear range (R 2 = 0.987) with 98% overall accuracy, and the whole detection is completed only within 5 min. This approach of instrument‐free time‐critical Hg 2+ detection anticipates a new paradigm for rapid and on‐site Hg 2+ monitoring in environmental and industrial settings.