Ratiometric copper nanocluster fluorescence probe coupled with deep learning for intelligent recognition of tetracycline antibiotics in food.

In this study, a ratiometric fluorescence sensing strategy combined with deep learning was developed for the intelligent detection of tetracycline antibiotics (TCs) in food matrices. Blue-emitting and saffron yellow-emitting copper nanoclusters (Cu NCs) were synthesized using bovine serum albumin (BSA) and 2,3,5,6-tetrafluorothiophenol (TFTP) as ligands, respectively, and employed to construct a dual-emission ratiometric fluorescence probe (BSA/TFTP@Cu NCs). The probe exhibited distinct fluorescence response patterns toward tetracycline (TC), chlortetracycline (CTC) and doxycycline (DOX), achieving a detection limit as low as 9.26 nmol L-1. Under 302 nm UV illumination, visually distinguishable fluorescence color variations were observed for the three TCs. To enable intelligent analysis, a modified ResNet50 model incorporating multi-task learning and a progressive three-stage training strategy was developed. The model simultaneously achieved TCs classification and concentration prediction, attaining 100% classification accuracy and a concentration prediction accuracy exceeding 98% (R2 = 0.989). In the analysis of spiked milk, egg and honey samples, the recoveries of TCs ranged from 90.48% to 107.93% for the fluorescence method and 90.28% to 98.53% for the deep learning method. This study provides a novel strategy for the rapid, accurate and intelligent detection of antibiotic residues in food safety monitoring.