Abstract：In the present paper, spectral reconstruction of three-dimensional fluorescence spectra of oil was studied based on singular value division (SVD) of fluorescence excitation-emission matrix (EEM). Depending on oil components, three-dimensional (3-D) fluorescence spectra of oils can be seen as their “fingerprints”. Feature abstraction and selection of 3-D fluorescence spectra is important to oil identification. Statistic parameters such as the average, standard error, centroid, kurtosis, geometrical distribution as well as main-shaft slope selected, forming “apparent statistic feature” vector of 3-D fluorescence spectra of oils, are limited due to their roughness. And the apparent statistic feature vector can not support spectral reconstruction reversely. In this paper, with singular-value features abstracted from EEMs composing the feature chain, reversal spectral reconstruction can be realized, and its information loss can be estimated. In other words, a pan-gene series of 3-D fluorescence spectra consists of singular values and their corresponding accompanying vectors. By SVD operation on EEMs of dozens of oil samples in water, their singular values accompanied with corresponding vectors were calculated showing obviously gathering energy distribution. With singular values being properly cut, principal feature parameters were selected and combined with their accompanying vectors composing the pan-gene series (or singular-value pan-gene chain) and spectral reconstruction was reversely completed. A couple of spectra (original and reconstructed) of diesel as the typical oil in water were presented for comparison. It can be seen that there is no obvious difference between the reconstructed and original spectra. Certainly, information loss exists but in an scheduled extent. It is shown that the singular-value features of fluorescence excitation-emission matrix (EEM) represent spectral gathering energy, and the length-limited pan-gene chain has the ability to reconstruct three-dimensional fluorescence spectra. Practically, for common kinds of mineral oils such as crude, diesel, kerosene and lubricant, the preceding three principal parameters are enough to complete spectral reconstruction and oil identification，with every singular value being followed by its right accompanying vector respectively. This is meaningful for recognition of contaminating oil in water, and important to construct oil fluorescence database.
Key words：Oil in water;3-D fluorescence spectrum;Spectral reconstruction;Singular-value feature
