The wide application of deep learning algorithms in the field of partial discharge pattern recognition has largely made up for the shortcomings of conventional methods like difficult implementation and low efficiency to a large extent. However, in practice, achieving ideal recognition accuracy with pattern recognition algorithms remains challenging. This is mainly due to the complex and variable nature of partial discharge mechanisms in realworld settings. Additionally, the available training data for identifying defective discharges often includes external interference signals such as noise. This paper proposes a PRPS pattern denoising method based on CWDT (Continuous Discrete Wavelet Transform) by analyzing the problems in the features of the measured partial discharge PRPS patterns, combined with the 2D pixel characteristics of the PRPS patterns. The final results show that the proposed method can overcome the inability of conventional wavelet transform to be applied to 2D data and improve the impact of noise on the PRPS patterns to a large extent.