As Low-Probability-of-Intercept (LPI) signal detection algorithm based on cyclic spectrum theory has a very large amounts of calculation, this paper proposes an optimized algorithm, Fast Fourier Transform(FFT) accumulation, to update the software program. The signal processing of the original software is analyzed, and the new algorithm based on a parallel signal processing structure is proposed. The new algorithm derives the next FFT sampling signal sequence by the previous FFT sampling signal sequence in a sliding window. By reducing the repeated sampling points, the new algorithm reduces the amount of calculation. In view of the symmetry of the four quadrant of the cyclic spectrum, the quadrant to be calculated is decreased. Moreover, by using one-dimensional search peak detection, the amount of calculation is reduced further. Theoretical derivation demonstrates that the complexity and calculation amounts of the proposed algorithm are reduced obviously. Finally, through verifying the Linear Frequency Modulation (LFM) and Binary Phase Shift Keying(BPSK) signals, it indicates that the calculation time is reduced to about 1/25 that of the original algorithm. In addition, the amplitude values of signals in the optimized algorithm are reduced less than 10%, which not affects the estimation of parameter of a cyclic spectrum.