As an important means to detect mine safety, ground sound monitoring technology is receiving more and more attention. In order to achieve de-noising of mine acoustic emission signals, a noise reduction method based on improved symplectic geometric mode decomposition (SGMD) combined with wavelet threshold (WT) is proposed to address the uncertainty of analysis results of SGMD. Firstly, the symplectic geometric modal components (SGCs) decomposed by SGMD are screened by setting the energy entropy increment and frequency correlation number threshold, and the active and noise components of the signal are obtained. The effective signal mode is denoised and reconstructed by wavelet threshold, and the original signal is denoised. In order to verify the effectiveness and robustness of the proposed method, the model is experimentally studied by using simulated signals and measured signals. The experimental results show that the proposed method is less timeconsuming, the highest SNR is 27.2 dB, and the smallest RMSE is 0.039. The noise reduction effect is obviously better than other noise reduction methods.