Estimation of the Comprehensive High Photosynthetic-Efficiency Phenotypic Index in Winter Wheat Based on UAV Multimodal Remote Sensing Data.

Accurate monitoring of photosynthetic phenotypes is fundamental for breeding high photosynthetic-efficiency wheat cultivars, and UAV remote sensing provides an effective approach for their large-scale identification. However, single photosynthetic parameters are limited in comprehensively evaluating crop photosynthetic efficiency. This study integrated multiple photosynthetic phenotypic parameters using principal component analysis (PCA) and the CRITIC objective weighting method (PCA-CRITIC) to construct a Comprehensive High Photosynthetic-Efficiency Phenotypic Index (CHPPI) for winter wheat. Concurrently, multispectral vegetation indices (MSVI), RGB vegetation indices (RGBVI), texture features (TF), and their combination, Comprehensive Multispectral-Visible-Texture Features (CMVTF), were extracted from UAV multimodal remote sensing data. Based on these features, four feature selection methods, namely PCC, VIP, SPA, and UVE, were evaluated in combination with five machine learning algorithms, including KNN, DT, SVR, RF, and XGBoost, to precisely estimate the CHPPI across key growth stages. The results demonstrated that the CHPPI exhibited significant cultivar variations across growth stages. Cultivars SH06144 and Liangxing19 consistently maintained high photosynthetic efficiency levels. The fused feature set CMVTF derived from multimodal remote sensing data outperformed individual feature categories, and the feature subset selected by the UVE method most significantly enhanced model accuracy. Regarding algorithms, the ensemble-based XGBoost and RF models markedly surpassed traditional machine learning algorithms. Specifically, the CMVTF-UVE-XGBoost model achieved the optimal comprehensive performance, yielding an R