Surveillance of SF6 decomposition products is significant for detection of partial discharge (PD) in gas insulation switchgear (GIS). As a basis in on-site detection and diagnosis, PD early-warning aims to quickly find the abnormalities using a simple and cheap device. In this paper, SO2 is chosen as a feature product and detected through ultraviolet spectroscopy. The derivative method is employed for baseline correction and spectral enhancement. The standard gases of the main decomposition products are qualitatively and quantitatively detected. Then decomposition experiments with different defects are designed to further verify the feasibility. As a stable decomposition product under PD, SO2 is proved to be applicable for PD early-warning in the field. By selecting the appropriate wavelength range, namely 290-310 nm, ultraviolet derivative spectroscopy is sensitive enough to the trace SO2 in the decomposed gas and the interference of other products can be avoided. Fast Fourier transform could be used for feature extraction in qualitative detection. Concentrations of SO2 and other by-products increase with increasing discharge time and could be affected by the discharge energy and PD type. Ultraviolet detection based on SO2 is effective for PD early-warning but the threshold should still be carefully selected in practice.