For the problem of oil spilling detection at night, a detection method based on band difference index of high spectral and ultraviolet(UV) LED induction is proposed. Through hyperspectral imaging collector, hyperspectral images of crude oil, emulsion oil and seawater obtained under UV LED and halogen lamp are collected synchronously. The band difference index is extracted based on radiation values of 33 bands (400~720 nm)to act as the identification feature of oil spilling. In the respect of feature optimization, effective wavelengths are chosen by methods of plus L reduce R and Fisher. The band features are extracted by multidimensional scaling, principal component analysis (PCA) and independent component analysis. An radial basis function support vector machine model is established to recognize the results. The results show that the band mean recognition rates of oil spilling detection method based on UV-LED and hyperspectral band difference index are higher than halogen lamps by 6.01% and 8.17%. Therefore, UV-LED light source is more suitable for accurate detection of oil spilling and emulsified oil testing at night. The optimized method by Fisher as an effective wavelength selection and through the PCA feature combination effect have almost same recognition rate. Recognition rates of 3 bands characteristics under UV light attain to 85.89% and 87.02% , while the accuracy of oil spilling reaches 100% by 12 characteristics. Effective wavelengths (400~420 nm) extracted by Fisher are more suitable for actual online oil spill detection. The model of the UV induced by hyperspectral imaging sea spilling identification built provides a rapid identification method for the sea oil spills detection at night.