Contour line-based image pattern recognition technique for the interpretation of fluorescence excitation-emission matrix map: qualitation and quantitation

Spectral retrieval method plays a crucial role in spectral analysis technology. Nevertheless, research in this area has not received the requisite advancement for fluorescence excitation-emission matrix (EEM) spectroscopy, which has been widely used in environmental monitoring and other fields. In this work, we developed a spectral retrieval method, the contour line-based image pattern recognition (CLIPR) technique, to interpret the EEM spectrum. A three-level retrieval mechanism, including the primary feature Lib1, qualitation Lib2 and quantitation Lib3 has been set up. Optimization efforts have been conducted on the accuracy and efficiency of noise removing, feature extraction and image retrieval methods. To demonstrate the technique, four model pollutants with very similar chemical structures and a metal-organic framework (MOF), MOF-74 (Zn), as a fluorescence probe providing more dimensional information were used. A total of 28 × 4 EEM map images were modeled and validated using the leave-one-out cross-validation method. Two batches with 5 × 4 samples each as unknown samples were selected to evaluate the performance of the CLIPR method. When taking the weight value set with 0.8, 0.1 and 0.1, for the validation sets with the concentration greater than 50 μM, the intra class similarity is above 79.61 % and the accuracy reached 98.81 % (83/84). The best quantitative result is PhNH 2 -S1, with the RMSE/True value 3.09 % and R 2 0.99. For the unknown sets, when Batch 1 greater than 35 μM and Batch 2 greater than 50 μM, the recognition accuracy reaches 100 %. The research results indicate that the model has a certain degree of stability and generalization and it is feasible to apply spectral retrieval method to interpret EEM information. We have taken a step forward in laying the foundation for large-scale reference data preparation in this field.