To address the challenge that existing denoising methods struggle to effectively extract microseismic signals from fractured rock masses in mines, a combined algorithm based on improved adaptive noise ensemble empirical mode decomposition (ICEEMDAN) and improved wavelet thresholding (IWT) was proposed. Firstly, the algorithm decomposed the noisy microseismic signal using ICEEMDAN to obtain the instrumental modal functions (IMF_s). Secondly, noise-containing components were filtered based on correlation coefficients, and the IWT algorithm was used for denoising. Finally, the denoised microseismic signal was obtained through wavelet transformation. At the same time, the accuracy of the algorithm was verified using both microseismic simulation data and real-world data. The denoising performance of the ICEEMDAN-IWT combined algorithm was quantitatively compared with that of other mainstream denoising algorithms, using the signal-to-noise ratio (SNR), normalized cross-correlation coefficient (NCC), and mean square error (MSE) as evaluation indicators. The results show that the ICEEMDAN-IWT combined algorithm achieves the best denoising performance, with an SNR of 28.06 dB, an NCC of 0.999 2, and an MSE of 0.014 45. After processing with the ICEEMDAN-IWT combined algorithm, the denoised waveform exhibits significantly improved characteristics while fully preserving the original phase information. However, there is still room for improvement in the denoising performance for waveform data with high amplitude ratios. Future research can further optimize the algorithm performance by incorporating intelligent methods such as deep learning.