With the increasing severity of water environmental pollution， there is an urgent need for rapid and accurate detection and identification of organic pollutants in water. Three-dimensional fluorescence spectroscopy technology， which provides rich spectral information about pollutants， has become a hot topic in the research of pollutant identification and source tracing in water bodies. The current methods mainly focus on deep learning based spectral data analysis， which requires a large amount of spectral data and is difficult to promote on site. This paper utilized three-dimensional Excitation-Emission Matrix （3D-EEM） data and proposed a method for multi-classification identification and precise component fitting of water bodies based on a combination of two-dimensional Gabor wavelets and Support Vector Machine （SVM）. This method effectively extracted texture features and peak positions of three-dimensional fluorescence spectra， which improved the efficiency of water sample component analysis. Here blank subtraction and Delaunay triangle interpolation were used to reduce background noise and scattering interference in spectral data， and spectral fluctuation interference was suppressed by extending the Savitzky-Golay smoothing approach. Subsequently， texture feature information of 3D-EEM data and global information of three-dimensional fluorescence peaks were extracted using two-dimensional Gabor wavelets and fluorescence peak extraction methods. Finally， an EEM_MSVM model based on MSVC and CF_MSVR was constructed to achieve high-accuracy classification identification and component prediction of water pollutants. Experimental results show that the classification accuracy for water body types is 97.6%. In terms of component prediction， the Root Mean Square Error （RMSE） loss is only 5.3， with a correlation coefficient of 0.94. This effectively achieves accurate classification of typical water bodies and analysis of their components.