Considering the challenges of extracting features from weak friction vibration signals in reciprocating sliding friction pairs, this study proposes a de-noising algorithm that integrates an improved wavelet thresholding method with mathematical morphology. First, the improved wavelet threshold function is used to optimize the classical wavelet denoising method for preliminary denoising. Subsequently, mathematical morphological filtering is applied to further eliminate residual noise and anomalous components from the signal, ensuring the preservation of its essential features. To validate the effectiveness of the proposed algorithm, both simulation data and experimental data from friction pairs were separately utilized for testing. The simulation results confirm that the proposed algorithm outperforms empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), and wavelet thresholding and morphological filtering algorithms (used independently) in key evaluation metrics, including minimum mean square error (MSE), signal-to-noise ratio (SNR), and normalized cross-correlation (NCC). The denoising performance for real data was evaluated through spectral analysis. The results demonstrate that the algorithm effectively reduces noise in friction vibration signals, improving the extraction of weak signal features and preserving their integrity and accuracy.