An adaptive threshold denoising algorithm based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) is proposed in this paper. The noisy signal is decomposed into several modal components(IMF) by CEEMDAN algorithm. Based on the sample entropy theory, the adaptive selection of high frequency components in IMF components is realized, and the noise figure in high frequency components is located according to the different correlation between noise and useful information or the original signal. The main noise interval is selected by energy entropy, and the noise coefficient variance of the main noise interval in high frequency components is used as the threshold. Threshold denoising of high-frequency components is carried out to further remove noise and retain useful information in high-frequency. Finally, high-frequency components and low-frequency components separated from signal-noise are reconstructed. The denoising of synthetic and actual seismic signals is processed separately and compared with conventional denoising algorithms. Data simulation and experimental results show that when the signal-to-noise ratio of the original signal is 0.5 dB, the signal-to-noise ratio obtained by the conventional and improved algorithms is 4.55 dB and 9.97 dB respectively, which indicates significant improvement in the signal-to-noise ratio, achieving the purpose of random noise suppression and realizing the self-adaptive selection of high-frequency components and the re-extraction of useful information from high-frequency components.