Compared with traditional Raman spectroscopy, ultra-violet (UV) Raman spectroscopy has been used in many fields, with many advantages such as high sensitivity, solar blindness, and high safety for human eyes. When a 266 nm UV laser is used as the light source, the Raman spectrum and fluorescence spectrum may partially overlap, which affects the accurate acquisition of Raman spectrum characteristics. Considering this problem, the morphology and polynomial fitting algorithms are combined to subtract the fluorescence backgrounds of UV Raman spectra. This method integrates the morphology that preserves the spectral features and the simple and effective polynomial fitting algorithm to realize accurate subtraction of fluorescence backgrounds in UV Raman spectra. To verify the effectiveness of the method, we perform baseline corrections on the simulated spectra of four different backgrounds and compare this method with the traditional methods. The results reveal that the proposed method has obvious advantages in accuracy and obtains a better baseline correction effect compared with existing baseline correction algorithms. Furthermore, this method is used to perform baseline corrections on the measured spectra of potassium nitrate samples with different substrates obtained by the UV Raman spectroscopy setup. The results indicate that this method can obtain pure Raman spectra for different substrate backgrounds, which can provide more accurate spectral information for subsequent spectral analysis.