A Spectral Domain Interpolation Framework for Robust Hyperspectral Feature Construction Under Spectral Uncertainty

Reliable characteristic spectrum construction is fundamental to hyperspectral target identification, spectral library development, and quantitative remote sensing applications. However, conventional mean characteristic spectra are often affected by spectral uncertainty arising from environmental heterogeneity, measurement noise, outliers, and spectral oscillations, reducing their representativeness. To address this issue, this study proposes an interpolation characteristic spectrum (ICS) method that introduces the First Law of Geography and spatial interpolation concepts into spectral space to construct characteristic spectra through weighted spectral-domain interpolation. The method was evaluated using UAV hyperspectral imagery and field ASD spectral measurements under different sample sizes and uncertainty conditions. A total of 896 comparative experiments were conducted between ICS and the conventional mean characteristic spectrum (MCS). The results show that within the experimental sample set, ICS achieves superior spectral representation performance in 69.28% of the experiments when the sample size is less than 35 and exhibits increasing advantages with increasing spectral uncertainty. Compared with MCS, ICS better preserved spectral morphology and feature amplitude information while reducing the influence of extreme values and spectral fluctuations. These findings indicate that ICS provides a robust approach for characteristic spectrum construction under spectral uncertainty and provides a certain theoretical and technical foundation for hyperspectral classification, target recognition and spectral library construction.