Abstract: Online monitoring of loose parts in pressurized water reactor coolant system is critical to the safety of nuclear power plants, but the monitoring signals are always interfered by the strong background noise generated by flow-induced vibration and equipment operations. In order to enhance the impact feature, a signal-noise separation and feature extraction method based on variational mode decomposition (VMD) and wavelet packet transform (WPT) was proposed. Firstly, the impact signal with noise was decomposed into a series of intrinsic mode functions (IMFs) with different frequency components, and the number of IMFs was determined by the correlation coefficients between the IMFs. Then, a measurement index termed weighted kurtosis index was constructed by kurtosis index and correlation coefficient, and a new signal with strong impact components was reconstructed by the IMFs selected based on the weighted kurtosis index. Finally, the new signal was further denoised by the WPT algorithm. The proposed method was used to denoise the signals generated by the simulation and impact experiment, and the impact components are successfully separated, which verifies the effectiveness of the method.