The power harmonic detection method based on fractional wavelet transform is an emerging research achievement, which can effectively solve the problem of noise interference in novel power system and improve the accuracy of harmonic detection. However, traditional discrete fractional wavelet transform is based on Mallat algorithm, and the signal detection accuracy will be affected by the downsampling operation. To overcome this problem, A trous algorithm with non-downsampling process is employed. Then, a novel method for harmonic detection based on fractional wavelet transform with A trous algorithm is proposed. In addition, the fourth order origin moment of fractional spectrum is used to determine the optimal fractional transform order, effectively reducing computational complexity. Experimental results show that the novel method can effectively achieve noise reduction and signal separation for steady state harmonics, short-term harmonics and time varying harmonics, and can achieve high-precision detection of the amplitude, frequency and positioning information of signal components.