Adaptive Multi-Order Penalty and Dual-Driven Weighting: aisPLS Algorithm for Raman Baseline Correction with Weak Peak Preservation.

Baseline correction of Raman spectra is a critical step for achieving high-precision quantitative analysis. However, the presence of complex background noise, nonlinear baseline drift, and spectral peak distortion due to peak overlap in real spectral data severely limits the performance of conventional correction methods. To better preserve spectral details, this study proposes an improved penalized least squares method for Raman spectral baseline correction. Compared with common baseline correction approaches, the proposed method optimizes the iterative weight function through precise noise classification, significantly enhancing the algorithm's flexibility. The traditional single smoothing parameter is extended into a smoothing vector, and a classification strategy consistent with that of the penalty parameter is adopted, enabling synchronous optimization and coordinated adjustment of both during iteration. Furthermore, based on the physical constraints of Raman spectra, the algorithm eliminates non-physical solutions that may arise in traditional iterative processes, ensuring the fidelity of the corrected spectra. Experimental results demonstrate that the proposed method exhibits strong robustness under various noise conditions and significantly improves correction accuracy.