Multi-scale adaptive signal denoising for smart road infrastructure through pelican optimized variational mode decomposition-improved wavelet threshold synergy

Asphalt pavement health monitoring signals are susceptible to environmental and system noise interference, hampering the accurate capture of structural response characteristics. Previous denoising methods have insufficient adaptability to non-linear and non-stationary signals, failing to meet the high-precision monitoring demands of smart roads and autonomous driving systems. This research proposes a collaborative denoising framework integrating the pelican optimization algorithm (POA), variational mode decomposition (VMD), and improved wavelet threshold (IWT). A POA-VMD parameter optimization model based on minimizing multi-scale permutation entropy is constructed, achieving adaptive optimization of mode number K and penalty factor α . A multi-criteria mechanism utilizing cosine similarity, Pearson correlation coefficient, and variance contribution rate identifies noise-dominated components for subsequent wavelet refinement. Validation on simulated signals across noise intensities achieves signal-to-noise ratios of 13.90−27.26 dB and root mean square errors of 0.1763−0.8209, outperforming comparative methods while preserving peak characteristics. Full-scale pavement test reveals progressive enhancement with structural depth, achieving noise mode reductions of 25.42%, 39.86%, and 52.57% in upper, middle, and lower layers respectively, with signal energy ratios maintained at 89.9%−98.2%. Flexible base pavements exhibit superior performance compared to semi-rigid structures. This research has provided a novel method of sensing signals quality assurance for constructing safer, more reliable, and smarter road infrastructure systems and autonomous driving traffic systems.