Based on the strong fluorescence property of polycyclic aromatic hydrocarbons benzo［k］ fluoranthene (BkF), a fluorescence detection system is constructed and ten BkF methanol solution samples with different concentrations are prepared to analyze their fluorescence characteristics. In order to proceed qualitative and quantitative analyses well, the improved empirical mode decomposition (EMD) threshold method combined with mathematical morphology is applied to denoising the fluorescence spectrum, which is compared with the EMD threshold denoising method. The results show that the proposed method can smooth the denoised fluorescence spectra preferably, make the linear correlation coefficient between the fluorescence intensity and the sample concentration high and reach 0.99746, make the signal-to-noise ratio increase, and make the mean square error between the original signal and the denoised signal decrease from 0.0053 to 0.0012. The proposed method possesses a remarkable denoising effect, which effectively improves the analytical precision of spectra and provides a new method for preprocessing fluorescence spectra.