Research on the optimal modeling path for inversion of Pb content in rice leaves based on hyperspectral data of ground objects and machine learning and cross-scale remote sensing monitoring.

Real-time and accurate monitoring of heavy metal concentrations in rice is essential for ensuring food safety and supporting the safe utilization of contaminated agricultural land. Hyperspectral technology offers advantages such as rapid, nondestructive, cost-effective, and environmentally friendly monitoring. This study developed multiple models for estimating Pb content in rice leaves, compared their performance, and identified stable modeling pathways for continuous regional monitoring using hyperspectral remote sensing data. A meta-analysis was conducted to assess how different spectral preprocessing methods and modeling strategies affect model performance. The results indicated that the combination of feature band selection algorithms and modeling methods substantially affected model performance. Among the evaluated strategies, the model constructed using Competitive Adaptive Reweighted Sampling (CARS) and Partial Least Squares (PLS) showed the most consistent performance while maintaining a high level of estimation accuracy. In contrast, the combination of Whale Optimization Algorithm (WOA) and PLS achieved the highest validation estimation accuracy (R2 = 0.7452). When applied to GF-5A hyperspectral data, the selected models achieved regional-scale predictions with R2 values greater than 0.55. These findings provide a technical basis for regional monitoring of heavy metal contamination in crops using hyperspectral remote sensing.