Journal Article

Oil Species Identification Based on the Fluorescence Spectroscopic Analysis Using the Excitation-Emission Matrix and Transfer Learning

Xu, Qintuan, Xie, Ming, Li, Ying

2024

Overview

Oil pollutants pose significant threats to marine and terrestrial ecosystems, necessitating the effective methods of oil species identification for the emergence responses of oil spill incidents. This study employs the excitation-emission matrix (EEM) fluorescence spectroscopy to capture and analyse the spectral characteristics of various oil species at different thicknesses. Some data augmentation techniques, including data smoothing and denoising, are introduced in this study to expand the dataset and enhance data quality. The methodology of transfer learning, which significantly reduces training time and improves model accuracy by sharing parameters, is adopted in this study. The enhancement of transfer learning method is examined using several typical deep learning networks. It is found that the implementation of transfer learning not only reduces the number of trainable parameters, but also improves identification accuracies by leveraging shared parameters, which makes it more efficient and accurate than building models from scratch. The proposed methodology enhances the capability of identifying petroleum pollutants using deep learning method and provides a new perspective on the advancement of oil spill monitoring technology.

• Publisher: Springer Nature B.V
• Subject: Accuracy / Data analysis / Data augmentation / Data smoothing / Deep learning / Emission analysis / Emission spectroscopy / Emissions / Excitation spectra / Fluorescence / Fluorescence spectroscopy / Introduced species / Marine ecosystems / Marine pollution / Oil pollution / Oil spills / Parameter identification / Parameters / Petroleum / Pollutants / Species identification / Terrestrial ecosystems / Threat evaluation / Transfer learning