Oil spill pollution is a typical form of environmental pollution in todays era of rapid development, which harms biodiversity and human safety through multiple channels. Therefore, given the composition and characteristics of oil pollutants, it is particularly critical to improve the ecological environment and ensure the steady development of the economy and society by using multi-method cross-fusion to detect them in real-time, accurately and efficiently. Three-dimensional fluorescence spectroscopy is widely used in the substance detection field with fluorescence characteristics with its advantages of high detection accuracy, good real-time performance, simple operation and small interference. Three-dimensional fluorescence spectroscopy combined with a support vector machine and other algorithms have achieved good results in material classification and identification and concentration prediction, but there are still defects, such as slow convergence speed and easy fall into local optimum. A new method for the classification and identification of oil pollutants was proposed by combining a three-dimensional fluorescence spectrum with a support vector machine algorithm ( IGOA-SVM ) optimized by an improved grasshopper algorithm. Firstly, with 0.1 mol·L-1 sodium dodecyl sulfate as a solvent, 0# diesel oil, 95# gasoline and kerosene were prepared into 20 and 18 mixed samples of 0# diesel oil and 95# gasoline, 0# diesel oil and kerosene, and 20 mixed samples of three components. Half of each was taken as a training set and a test set. The fluorescence data of the mixed solution were collected by an F-7000 fluorescence spectrometer. Matlab analyzed the standard solution of the three oils and the mixed solution. It was found that the fluorescence spectra had different degrees of overlap within a certain range, and it could not be accurately identified by spectral detection alone. Finally, the grasshopper optimization algorithm is improved by combining chaotic initialization, elite optimization, and differential evolution algorithms. The fluorescence peak data in the excitation wavelength 270 nm and emission wavelength 270~450 nm are extracted as the input value of training. With three kinds of classification labels as output, the data are input into the grasshopper optimization algorithm support vector machine (GOA-SVM), particle swarm optimization support vector machine (PSO-SVM) and genetic algorithm optimization support vector machine (GA-SVM) for training. The IGOA-SVM model is superior to GOA-SVM, PSO-SVM and GA-SVM in convergence speed, stability and ability to jump out of local optimum, providing a new idea for accurately identifying oil contaminants.