Discriminative scale space tracker (DSST) is a generic scale estimation approach with excellent performance. Nevertheless, it does not exploit the powerful discriminability of multiple kernels. This paper will enhance the DSST to the multiple kernel version, SSMT, which effectively harnesses discriminative power spectrums of different features to improve the scale estimation performance. Extensive experimental results demonstrate that SSMT obtains a significant performance improvement on image sequences with scale variation of three object tracking benchmarks: OTB-2013 (+3.7\% in AUC), OTB-2015 (+3.9\% in AUC) and UAV (+3.5\% in AUC)