Microseism events near the Ordovician limestone aquifer generally had the characteristics of weak energy, low signal-to-noise ratio, and large background field interference. Traditional filtering methods were difficult to effectively retain deep weak signals while denoising. In order to solve this problem, the microseism data denoising method based on wavelet analysis was systematically studied. Based on the theoretical basis of continuous wavelet transform and discrete wavelet transform, the application principle of wavelet multi-scale decomposition in time-frequency analysis of microseism signals was expounded. Taking the deep microseism data of No.182703 Face in Wutongzhuang Coal Mine as the research object, the denoising performance of different wavelet basis functions was compared and analyzed. The db6 wavelet basis was determined as the optimal basis function, and the accuracy of event recognition after denoising was 92.8%. After wavelet soft threshold denoising, the P-wave arrival time error of deep events was reduced from 12.5 ms to 5.2 ms, the signal-to-noise ratio was increased by 12.5 dB, and the positioning error was reduced from 14.5 m to 5.8 m, which was 60% higher than that of the conventional filtering method. The research showed that the wavelet denoising method was significantly better than the traditional filtering method in the processing of deep low signal-to-noise ratio microseism data in Wutongzhuang Mine, which provided an effective signal enhancement method for deep monitoring of mine water disaster.