Answers are grounded in this paper and cite their sources.
Tobacco leaf starch content is an important biochemical indicator for assessing leaf maturity and industrial processability, exerting a crucial influence on tobacco quality formation and curing characteristics. This study proposes a physiology-aware spectral feature abstraction framework based on Multi-metric Clustering Analysis (MMCA), integrating multi-metric band evaluation with clustering analysis and combining UAV hyperspectral data with machine learning to achieve accurate and robust estimation of tobacco leaf starch content. Five feature importance metrics, namely correlation coefficient, grey relational analysis, variable importance in projection, mutual information, and Shapley Additive Explanations (SHAP), were applied to identify sensitive bands (695–705 nm). Unlike approaches that rely on a single sensitivity criterion predominantly confined to the red edge region, MMCA adopts a multi metric guided clustering strategy to construct a joint evaluation framework over the full spectral domain and extract cluster center bands (487, 564, 701, 735, and 792 nm) using cosine distance with average linkage, thereby reducing spectral redundancy while preserving biochemical sensitivity. Among the four machine learning algorithms, MMCA-KRR showed the best performance, with a test R² of 0.86 and an RMSE of 4.24%. Compared with conventional band selection methods such as SPA, CARS, and GA, MMCA provided competitive predictive accuracy while using a compact and interpretable set of representative wavelengths. This research demonstrates that integrating multi-metric evaluation clustering with nonlinear machine learning provides a robust and interpretable framework for high-accuracy estimation of tobacco leaf starch content using UAV hyperspectral data.
No takes yet. Share an insight, caveat, or question.
Fan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: