Oil pollutant identification based on excitation-emission matrix of UV-induced fluorescence and deep convolutional neural network.

Abstract

Identifying the types of oil pollutants in a spill event can help determine the source of spill and formulate the plan of emergency responses. Excitation-emission matrix (EEM), which is also called three-dimensional fluorometric spectra, includes abundant spectral information in the domain of excitation wavelength and can be potentially applied to identify oil types. UV-induced fluorometric experiments were conducted in this study to collect EEMs for five types of oil that are commonly used in maritime transportation. A deep convolutional neural network (CNN) model for oil types identification was built based on the classic VGG-16 model. According to the identification results, the model was able to provide a reasonable classification on the five types of oil used in the experiments. Additionally, a biased classification result was observed in the experiment: the model was able to provide the most accurate classification on 0W40 lubricant but encounters difficulty distinguishing between - 10# diesel and 92# gasoline. The potential reasons for this result and the approaches to improve the model were also discussed.