When we use wavelet transform to solve the denoising problem in pulsar noise signal, the choice of threshold and construction of threshold function determine the denoising effect. Herein, we analyze the distribution characteristics of noise wavelet decomposition coefficients by combining the wavelet transform properties. We construct a nonlinear two-parameter threshold function by combining the structural characteristics of the soft threshold and hard threshold functions. Furthermore, we use particle swarm algorithm to optimize the parameter size to change the position and bending degree of the threshold function adaptively, thereby obtaining a good threshold function denoising model. Based on the analysis on the variation in the noise wavelet decomposition coefficient with decomposition layer, the unified threshold selection strategy is improved. Then, a gradient attenuation factor is introduced to construct a threshold selection method based on the noise mean square error of each decomposition layer. The experimental results show that, compared to the traditional wavelet domain denoising method, the proposed method significantly improves the signal-to-noise ratio, peak signal-to-noise ratio, and peak-to-bit error of X-ray pulsar noise signals, supporting the new ideas of X-ray pulsar signal denoising.