<p>Wheel tread defects generate periodic impulse signals in railway tracks, and acoustic monitoring offers the advantage of rapid, long-distance detection. However, traditional wireless acoustic monitoring using Nyquist sampling techniques results in high power consumption due to the large volume of data requiring processing. This paper explores the potential of acoustic signals for monitoring wheel changes and proposes an adaptive sparse wavelet packet node selection algorithm. Leveraging cluster characteristics of wavelet packet nodes and the narrowband nature of impulse signals, the algorithm achieves significant compression ratios while preserving impulse characteristics even at low signal-to-noise ratios. Simulation results demonstrate the efficacy of the proposed method. Experimental validation using data from a worn wheel tread confirms its feasibility, showing that the method effectively attenuates data while retaining impulse features. Comparative analysis against discrete cosine transform and wavelet compression algorithms highlights the superiority of the proposed approach. The proposed method, with its low computational complexity and high compression ratio, effectively mitigates the high transmission cost associated with large acoustic datasets, making it suitable for resource-constrained environments typical of wireless sensor networks deployed in railway tracks.</p>