Deep Learning-Assisted RTP Sensor Array Based on a Melt-Injection Reaction for Visual Discrimination of Fluoroquinolone Antibiotics

Abstract

The widespread residue and structural similarity of fluoroquinolone antibiotics (FQs) pose a critical requirement for analytical methods capable of both sensitive detection and accurate discrimination of FQ subtypes. Herein, we report a deep learning-assisted room-temperature phosphorescence (RTP) sensor array for FQs constructed via a simple melt-injection reaction. By incorporation of FQ subtypes into a urea–formaldehyde (UF) matrix, composites (UF@FQs) exhibiting tunable, ultralong RTP emission were produced. Photophysical studies reveal that the UF matrix provides a rigid protective environment that suppresses nonradiative decay and activates the weak phosphorescence of FQs through confinement and hydrogen-bonding interactions. This matrix not only amplifies the afterglow intensity but also prolongs the emission lifetime, generating fingerprint-like optical responses for the FQ subtypes. Leveraging these cross-reactive signals, we fabricated a sensor array to discriminate four different FQs and their mixtures. The array was further integrated with a deep learning model capable of visually discriminating FQs directly from afterglow images. The platform demonstrates excellent selectivity against common interferents and achieves satisfactory recovery in meat samples. This work presents a straightforward strategy for transforming nonemissive analytes into bright afterglow signatures, offering a powerful tool for on-site, high-throughput, and intelligent discrimination of antibiotics.