Oil spill has become one of the most serious problems in global environmental pollution and brings a serious threat to the marine ecological balance and human health. Therefore, it is of great importance to study efficient oil spill detection methods to protect the marine ecological environment. As three-dimensional fluorescence spectra technology has advantages of getting oil spill fingerprints, it has become an important analytical method in the field of oil spill identification. A satisfactory oil spill identification effect was obtained by combining 3D fluorescence spectra technology with the parallel factor (PARAFAC) analysis algorithm. The applicable concentration range for different oils should be determined before the implementation of PARAFAC algorithm. Besides, PARAFAC is sensitive to number of components. The selection of number of components directly affects qualitative and quantitative analysis results. The method of 3D fluorescence spectra technology combined with PARAFAC is limited in real sea surface oil spill due to above reasons. The composition of oil spill is extremely complex, in which each component not only has a uniform concentration linear range but also is affected by the fluorescence quenching. Due to different content of components, the three-dimensional fluorescence spectra of the oil spill sample (sample is not diluted) are quite different. Some algorithms (such as parallel factor analysis) that resolve the three-dimensional fluorescence spectra are no longer applicable. With the change of the type and content of the sample components, the change rule of the three-dimensional fluorescence spectra image characteristics is also obvious. Therefore, a novel detection method for oil spill based on 3D fluorescence spectra technology and digital image recognition is proposed in this paper. Firstly, three types of mixed oil samples were formulated. Each type of mixed oil was directly mixed with two types of five mineral oils (gasoline, diesel, jet fuel, engine oil, lubricating oil) at different volume ratios. The three-dimensional fluorescence spectral of samples were obtained by FS920 fluorescence spectrometer. The corresponding three-dimensional fluorescence derivative spectral grayscale image was obtained by preprocessing of derivation and graying. Then, the digital image features such as color, texture and shape of three-dimensional fluorescence derivative spectral grayscale image were extracted. Finally, the classification and quantitative models of samples were established by fisher discriminant and stepwise regression respectively. The classification model has good classification and recognition effect on three types of mixed oil samples. The linear correlation coefficient R of the quantitative model is greater than 0.99. The significance test p-value of the quantitative model is less than 0.05. The results show that the digital image characteristics three-dimensional fluorescence spectral can be effectively extracted by our method and used for the qualitative and quantitative analyses of oil samples. The study provides a simple and accurate identification method for sea surface oil spill.