Detection Of Marine Oil Spill Based On Fluorescence Matrix Spectroscopy

With the rapid economic development and a significant increase in productivity,offshore oil mining and transportation are becoming increasingly frequent.In this circumstance,oil spills have become a major threat that cannot be neglected in the marine transportation.The detection and identification of oil spill is a basis and key content to control offshore oil spill pollution,which has great scientific significance to the ocean environment improvement.Laser fluorescence remote sensing oil spill detection technology can adapt to the complexly changing ocean environment and achieve rapid detection and identification of oil spill.However,similar oil species cannot be accurately identified.Although fluorescence matrix spectroscopy detection methods can accurately distinguish different oil species,they are mostly accomplished using large fluorescence spectrophotometers,which can only be performed in a laboratory and are difficult to be used directly in the field and have the problem of low computational efficiency for the actual measured large-scale data.The current trend of oil spill detection equipment toward miniaturization and portability requires research of the technology that can be applied to field measurements.This thesis accomplishes the following work in terms of task allocation:(1)Based on the principle of offshore oil film fluorescence detection,a wavelengthtunable xenon lamp and a portable spectrometer were selected as the experimental light source and detector.The fluorescence spectra acquisition experiments of various oil slicks at different excitation wavelengths were designed and completed.The fluorescence spectra of the corresponding oil slicks were obtained after eliminating the influence of fluorescence interference factors present in the experimental data,and the three-dimensional fluorescence matrix spectra were constructed accordingly.(2)For the recognition and classification challenges of similar oil species fluorescence spectra,an optimal excitation wavelength selection method based on random forest algorithm is proposed.Firstly,the experimentally obtained fluorescence spectra are smoothed and noisereduced using Savitzky Golay filter,and further,the corresponding optimal classification excitation wavelengths are identified by comparing the accuracy of offshore oil film fluorescence spectra recognition under different excitation wavelengths.The ability of laserinduced fluorescence technique to identify similar refined oil is improved without increasing time and other costs.(3)To address the problem of low computational efficiency of traditional chemometric methods in the face of large-scale 3D fluorescence matrix spectral data,median filtering is used to smooth and reduce the noise of 3D fluorescence matrix spectra of offshore oil films,and a3 D fluorescence matrix spectral classification method based on convolutional neural network is further proposed to achieve rapid detection and identification of offshore oil films.This thesis designed and completed the experiments on the acquisition of fluorescence spectra of oil slicks under different excitation wavelengths,compared the classification ability of similar oil under the different excitation wavelengths of fluorescence spectra,and found out the best excitation wavelength for the classification of similar light oil,which provides the basis for the excitation wavelength selection of laser fluorescence oil spill monitoring equipment.A convolutional neural network-based oil slick fluorescence matrix spectral classification algorithm is proposed to achieve a fast and accurate classification of oil slicks and provide technical support for maritime oil spill pollution monitoring.