In order to solve the problem of modal aliasing in Empirical Mode Decomposition (EMD) and the fact that windowed interpolation FFT cannot accurately detect signals with transient components, a harmonic detection method based on generalized S-transform mutation recognition is proposed. First, the generalized S-transform is used to obtain the modular time-frequency matrix of the signal, and the signal is divided into steady-state signals and signals with transient components based on the energy continuity of the larger value frequency band in the modular time-frequency matrix. Then, the harmonic and interharmonic parameters for the two kinds of signals are calculated using the Hanning windowed trispectral interpolation FFT and the modified HHT based on the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). Finally, the verification of steady-state, amplitude mutation and frequency mutation signals shows that the method can distinguish the steady-state signals and signals with transient components adaptively, achieve high detection accuracy for both signals, and has good practicability. This work is supported by Natural Science Foundation of Guangdong Province (No. 2018A030310354).