Engineering Phonon Anharmonicity in Lanthanide‐Doped Double Perovskites for Deep‐Learning‐Assisted Colorimetric Thermography

ABSTRACT Self‐trapped excitons (STEs) in lead‐free metal halide perovskites offer a powerful toolbox for next‐generation optical thermometry, yet tailoring their heat‐assisted detrapping dynamics to optimize luminescent thermal responses for high‐performance applications remains elusive. Herein, we report a portable and intelligent colorimetric thermography platform based on cation‐engineered double perovskites, integrating hue‐saturation‐value (HSV) color encoding with deep learning for accurate thermal imaging. By tailoring the trivalent cation sites, size‐dependent lattice hardness enables stronger electron‐phonon coupling and a lowered thermal activation barrier for de‐trapping. In indium‐based perovskites, pronounced fourth‐order phonon anharmonicity arising from high bond ionicity drives significant lattice thermal expansion and softened bond force constant, yielding an ultralow thermal activation energy of 45.08 meV and the strongest thermal quenching, while maintaining a photoluminescence quantum yield of 85.27%. Furthermore, we incorporate Er 3+ as a spectrally stable emitter to produce distinct thermochromic behavior, and propose an HSV‐based colorimetric thermography with robust color recognition. To further mitigate artifacts from ambient lighting and viewing‐angle variations, we employ a deep‐learning algorithm that delivers end‑to‑end thermal mapping with a high accuracy of 92.20%. This work establishes a materials‐by‐design framework for on‑demand tuning of STE thermal behavior and provides an advanced thermal sensing platform enhanced by artificial intelligence.