Harmonic detection is a vital problem in power system, but the accuracy of harmonic detection is affected by noise. In order to suppress noise, this paper proposes the improved singular value decomposition (SVD) to denoise. Then, the optimal number of decomposition modes of variational mode decomposition (VMD) is determined with the aim of minimizing the residual of signal energy after Schmidt orthogonalization. The amplitude and frequency of each harmonic are detected by the Hilbert transform. The simulation shows that the algorithm can accurately identify the information characteristics of each harmonic and inter-harmonic, and has the advantages of high accuracy and small error.