Estimation of potato leaf chlorophyll content using machine learning and UAV-based multimodal data fusion with Boruta-SHAP feature optimization

Accurate estimation of leaf chlorophyll content (LCC) is critical for monitoring crop physiological status and supporting precision management. This study proposes a multimodal regression framework that integrates unmanned aerial vehicle (UAV)-based hyperspectral and RGB imagery with ground-measured LCC data for estimating potato LCC. Experimental data were collected at two sites across three key growth stages: S1 (tuber formation), S2 (tuber bulking), and S3 (starch accumulation). The dataset included hyperspectral reflectance, RGB-derived texture and color features, and the Heat–Water Index (HWI). Hyperspectral preprocessing techniques included fractional-order derivative (FOD), continuous wavelet transform (CWT), and multiplicative scatter correction (MSC). A hybrid Boruta-SHAP algorithm was employed for feature selection, and three machine learning models—Random Forest (RF), Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost)—were evaluated. Results showed that hyperspectral preprocessing significantly enhanced model performance, with the FOD-CWT-MSC combination improving R² by 55.6% compared with raw spectra. The feature selection strategy reduced redundancy while increasing the average R² by 23.9%. The optimized XGBoost model incorporating multimodal features achieved the best performance, with a test R² of 0.866 and an RMSE of 4.28 μg/cm². These findings underscore the value of integrating multimodal remote sensing data with spectral preprocessing strategies for improving potato LCC estimation under the conditions of this study.