Oil spills on the sea surface are seen relatively often with the development of the petroleum exploitation and
transportation of the sea. Oil spills are great threat to the marine environment and the ecosystem, thus the oil pollution in
the ocean becomes an urgent topic in the environmental protection. To develop the oil spill accident treatment program
and track the source of the spilled oils, a novel qualitative identification method combined Kernel Principal Component
Analysis (KPCA) and Least Square Support Vector Machine (LSSVM) was proposed. The proposed method adapt
Fourier transform NIR spectrophotometer to collect the NIR spectral data of simulated gasoline, diesel fuel and kerosene
oil spills samples and do some pretreatments to the original spectrum. We use the KPCA algorithm which is an extension
of Principal Component Analysis (PCA) using techniques of kernel methods to extract nonlinear features of the
preprocessed spectrum. Support Vector Machines (SVM) is a powerful methodology for solving spectral classification
tasks in chemometrics. LSSVM are reformulations to the standard SVMs which lead to solving a system of linear
equations. So a LSSVM multiclass classification model was designed which using Error Correcting Output Code
(ECOC) method borrowing the idea of error correcting codes used for correcting bit errors in transmission channels. The
most common and reliable approach to parameter selection is to decide on parameter ranges, and to then do a grid search
over the parameter space to find the optimal model parameters. To test the proposed method, 375 spilled oil samples of
unknown type were selected to study. The optimal model has the best identification capabilities with the accuracy of
97.8%. Experimental results show that the proposed KPCA plus LSSVM qualitative analysis method of near infrared
spectroscopy has good recognition result, which could work as a new method for rapid identification of spilled oils.
|