Smart three-in-one detection platform: Chemical sensing, image recognition, and machine learning for rapid identification of tetracycline antibiotics.

The hidden allergic reactions and cumulative toxic effects caused by antibiotic residues in aquatic products have become significant public health concerns. Among these, tetracycline antibiotics (TCs) pose considerable ecological risks and health threats due to their high molecular stability and prolonged environmental persistence. To address this challenge, this study developed a guanosine-5-monophosphate (GMP)-Tb/rhodamine B (RhB) multi-proportion ratiometric fluorescence probe array, integrated with computer vision and an intelligent color mapping recognition framework, to construct a comprehensive and intelligent solution for antibiotic identification. Specifically, the "You Only Look Once" version 5 s (YOLOv5s) model was utilized to automatically detect the reaction areas on the 96-well plate and extract RGB feature data. These features were then analyzed using k-Nearest Neighbor(KNN), support vector machines (SVM), and random forest(RF) classifiers for multi-dimensional modeling. Finally, the optimized model was integrated into a web application built using the Flask framework. Experimental results demonstrated that the system achieved an overall recognition accuracy of 94.4 % for TC, OTC, and CTC antibiotics within a concentration range of 0-50 μM; notably, the optimal KNN classifier achieved an F1-score of 1.00 on the validation set. This study successfully established a three-in-one analytical paradigm that integrates chemical sensing, image recognition, and intelligent decision-making, thereby providing an expandable technical platform for the rapid screening of environmental pollutants.