In order to overcome the pseudo-modal problem of adaptive noise fully integrated empirical modal decomposition (CEEMDAN) in processing signals, the improved adaptive noise fully integrated empirical modal decomposition (ICEEMDAN) and discrete wavelet transform (DWT) denoising algorithm for heart sound signals was investigated. The denoised heart sound signal (PCG) is decomposed into wavelet coefficients at different frequency scales using DWT and the high-frequency coefficients are processed with adaptive thresholding; each wavelet subband is further decomposed using ICEEMDAN to obtain multiple empirical modal functions (IMFs) associated with the heart sound features; and the processed IMFs are reconstructed into the denoised heart sound signal. Through experimental verification, the proposed algorithm has obvious advantages in improving the quality of the heart sound signal and preserving the heart sound features, and has higher efficiency and accuracy than the traditional wavelet denoising and EMD-like methods.