In response to the problem that the extraction accuracy of existing harmonic detection methods is easily affected by noises, a novel harmonic detection method based on iteration symplectic geometry mode decomposition combined with total least squares and estimation of signal parameters with rotation invariant technique (TLS-ESPRIT) is proposed. First, the initial symplectic geometric components are obtained by constructing a trace matrix of harmonic signals and combining the matrix transformation. Second, based on dynamic time warping, the initial symplectic geometric components with a high similarity are linearly combined, and the residual components are eliminated. Third, the normalized mean absolute error is calculated to obtain the independent harmonic components. Finally, the TLS-ESPRIT algorithm is used to obtain the parameters of different harmonic and inter-harmonic components. Through an analysis of simulation and measured data, the superiority of the proposed method when it is applied in harmonic and inter-harmonic analysis under a noisy environment is verified.