Efficient identification of antibiotic components in three-dimensional excitation-emission matrix fluorescence spectra via deep learning

Antibiotic contamination has emerged as a critical environmental issue due to its adverse effects on ecosystems and public health. Rapid and precise antibiotic detection is essential for mitigating environmental risks. This study introduces a novel deep learning framework for analyzing three-dimensional excitation-emission matrix (3D-EEM) fluorescence spectra to achieve efficient and accurate antibiotic identification. Unlike conventional spectral unmixing techniques such as PARAFAC, our deep learning approach eliminates sample size constraints and labor-intensive preprocessing while significantly enhancing feature extraction capabilities. We established a comprehensive fluorescence database comprising 1632 spectral samples from 65 antibiotic groups spanning sulfonamides, fluoroquinolones, and tetracyclines. Through rigorous comparative analysis of five state-of-the-art deep learning models (YOLOv11, YOLOv8, YOLOv5, RTDETR, and Faster R-CNN), we demonstrated that YOLOv11 achieves superior performance with 98.66 % precision, 98.87 % mAP, and an F1-score of 0.9832. Notably, YOLOv11 exhibited exceptional robustness in identifying fluoroquinolones even under conditions of severe spectral overlap. Further validation across varied component datasets confirmed YOLOv11's exceptional versatility, maintaining 98.7 % precision, 98.9 % mAP, and 0.983 F1-score even when tasked with complex multi-component identification challenges. As the pioneering application of YOLO architecture in antibiotic detection, this work successfully overcomes critical technical limitations inherent to traditional spectral analysis methodologies, substantially reducing data processing time and operational complexity. The framework developed herein provides a valuable reference for deep learning-assisted 3D-EEM analysis and represents a significant advancement toward real-time environmental monitoring systems that offer unprecedented efficiency, accuracy, and intelligence for antibiotic detection in aquatic environments.