Answers are grounded in this paper and cite their sources.
Soil total nitrogen (TN) is indispensable for crop development, and its rapid, non-destructive, and precise determination is vital for optimizing agricultural yields. As a rapid, cost-effective, and non-destructive technique, Near-infrared spectroscopy (NIRS) offers significant potential for efficient soil TN determination. However, variations in soil parent material induced by geographic, climatic, and cultivation factors often lead to substantial spectral heterogeneity, which limits the generalization performance of NIRS-based detection models. To address this challenge, we develop Transfer-Soil-Spectra, a dual-stage spectral feature alignment framework that synergizes maximum mean discrepancy and contrastive representation learning to enhance model robustness across diverse soil types. A comprehensive evaluation was conducted, encompassing staged experiments, comparative analyses, and multi-scenario applications. In staged experiments, the dual-stage strategy enabled both global and fine-grained alignment of spectral principal components, yielding more stable and informative features. For instance, when combined with the LightGBM algorithm, the method achieved a predictive performance of R² = 0.87 ± 0.02, RMSEP = 0.27 ± 0.01 g/kg, and RPD = 2.81 ± 0.19 after the first-stage alignment. In the comparative analysis, Transfer-Soil-Spectra produced more compact feature distributions and effectively captured spectral bands associated with N–H and C–H functional groups. Finally, multi-scenario validation demonstrated the adaptability of Transfer-Soil-Spectra across soils from different regions as well as within the same region but differing in soil use type or soil occurrence. Beyond soil nitrogen detection, the framework is readily transferable to other spectroscopic applications, highlighting its broad potential for large-scale precision agriculture engineering.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: