Deep Learning Fluorescence Imaging Based on CNN-BiLSTM Hybrid Neural Network for Anti-Counterfeiting Application

KY3F10:Yb3+, Tm3+ upconversion microparticles (UCMPs) with varying Mn2+ doping concentrations were synthesized via a hydrothermal method. Under 980 nm laser excitation, the sample with 3 mol% Mn2+ doping demonstrated markedly enhanced luminescence performance, exhibiting a significant intensity increase compared to undoped samples. The as-synthesized UCMPs were successfully incorporated into an anti-counterfeiting ink. Target information was encrypted using a hash function to generate a QR code, which was then screen-printed onto substrate materials. Under 980 nm laser irradiation, the printed QR code exhibited visible blue fluorescence with high stability, confirming its anti-counterfeiting capability. Furthermore, an image recognition system for anti-counterfeiting, based on a hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) architecture, was developed on the Matlab platform. The system achieved 100% recognition accuracy for the luminescent QR code patterns, providing valuable insights for the development of deep learning-based image anti-counterfeiting technologies.