Mineral Oil Identification Based On3D Fluorescence Spectra Technology

Water pollution caused by kinds of accidents in the process of oil exploitation and transportation has become a common concern issue all over the world. After accidents happened, in order to trace pollution sources, distinguish accident responsibility and provide guidance for subsequent cleanup work, it is necessary to identify the types of spilling oil. In this paper, based on3D fluorescence spectra technology, different features extracted using principal component analysis, wavelet transform and SIFT(Scale Invariant Feature Transform) from3D fluorescence spectra, accomplishing mineral oil identification.By analyzing the property of the main fluorescence components in mineral oil, the feasibility and unique advantages of3D fluorescence spectra in mineral oil identification has been proved. Obtains several common kinds of mineral oil’s3D fluorescence spectra in different solvents, and provides reference for the following feature extraction by analyzing the effect of concentration and solvent on spectra feature changes.When reconstructing the spectra using principal component analysis, root mean square error is employed to restrain spectra energy loss. Based on this principal, the principal component features is extracted. Because of principal component features missing minor feature, this paper combines wavelet transform and principal component analysis together exacting WT-PCA feature that including main and minor components eigenvalues, and further optimizing WT-PCA features from the perspective of energy distribution. BP neural network identification results show the recognition rate of WT-PCA feature and optimized WT-PCA feature higher than principal component feature’s.So, WT-PCA feature does better than principal component feature in mineral oil identification.SIFT features of3D fluorescence spectra fingerprint in different solvents is extracted, and then build SIFT feature set for every mineral oil. According to the match rate with different SIFT feature sets, unknown mineral oil can be identified. This method breaks the limitation of solvent effect on mineral oil identification in some degree and riches the mineral oil identification means.