Smartphone-integrated molecularly imprinted sensor with convolutional neural networks for on-site detection of Norfloxacin.

We developed a portable sensing platform that integrates a dual-emission molecularly imprinted fluorescent sensor (Ir-NOR@MIP-CdSe@ZnS) with deep learning for the rapid, on-site detection of Norfloxacin (NOR). The Ir-NOR-probe was incorporated with CdSe/ZnS quantum dots, the molecularly imprinted sensor, constructed by integrating the Ir-NOR probe with CdSe/ZnS quantum dots, achieves highly sensitive detection of NOR through a Ligand to Metal Charge Transfer (LMCT) mechanism. Under optimized experimental conditions, with a linear detection range of 0.010-15.0 mg/L, a detection time (5 min) and excellent anti-interference capability and long-term stability. By leveraging smartphone-captured fluorescence images, we developed and optimized four Convolutional Neural Network (CNN) models. These models autonomously extract concentration-related features, surpassing conventional Red, Green, Blue(RGB) analysis and effectively correcting for environmental interference. A miniaturized portable fluorescence sensing platform was constructed using 3D printing technology. The optimized CNN models enabled quantitative detection of NOR in biological samples achieving a tissue detection limit as low as 0.53 μg/kg. The recoveries of spiked samples ranged from 88.0% to 110.0%, with relative standard deviations (RSD) below 3.5%. The results indicate that this method provides technical support for antibiotic monitoring in aquatic product safety assurance.