基于多源特征融合与堆叠集成学习的冬油菜叶面积指数估测研究

Estimation of Winter Rapeseed Leaf Area Index Based on Multi-source Feature Fusion and Stacking Ensemble Learning

摘要:

叶面积指数(Leaf area index,LAI)是评价作物生长状况与作物单产的关键生物学参数。针对基于无人机多光谱数据构建的植被指数特征在高LAI 区易饱和的问题,本研究以甘蓝型冬油菜(Brassica napus L.)为研究对象,获取冬油菜苗期、蕾薹期、花期和角果期的多光谱图像,融合提取的11个光谱特征(Spectral features,SFs)、40 个纹理特征(Texture features,TFs)以及地面实测株高(Plant height,PH),结合随机森林(Random forest,RF)、梯度提升决策树(Gradient boosting decision tree, GBDT)、支持向量机回归(Support vector regression,SVR)以及K近邻(K-nearestneighbors regression,KNN)算法,构建了基于堆叠集成学习(Stacking ensemble learning,SEL)的冬油菜LAI估测框架,实现LAI精准估测。结果表明:SFs、TFs与PH的多源特征融合能显著提高模型估测精度;在全生育期尺度下,SEL模型精度最优,R2达到0.93,MAE和RMSE分别为0.27和0.33;LAI空间分布图能够清晰揭示4个生育期不同施肥、密度及播期处理下冬油菜长势的空间异质性。研究成果可为冬油菜生长的精准监测提供决策支持。

Abstract:

Leaf area index (LAI) is a crucial biological parameter for evaluating crop growth status and predicting yield. Aiming at the common problem of spectral saturation in high LAI regions when using vegetation indices derived from unmanned aerial vehicle (UAV) multispectral data, winter rapeseed (Brassica napus L.) was taken as the research object. Based on multispectral images acquired at the seedling, bolting, flowering, and podding stages, a total of 11 spectral features (SFs), 40 texture features (TFs), and ground-measured plant height (PH) were fused systematically. By integrating random forest (RF), gradient boosting decision tree (GBDT), support vector regression (SVR), and K-nearest neighbors (KNN) algorithms, a stacked ensemble learning (SEL) framework was constructed to achieve efficient and precise LAI inversion. The results indicated that the multi-source feature fusion of SFs, TFs, and PH significantly improved the estimation accuracy and generalization ability of the model. At the scale of the whole growth period, the SEL model exhibited the highest accuracy, the coefficient of determination (R2) was 0.93, the mean absolute error (MAE) was 0.27, and the root mean square error (RMSE) was 0.33. The LAI spatial distribution maps clearly revealed the spatial heterogeneity of winter rapeseed growth under different fertilization, density, and sowing date treatments during the four growth stages. These research findings can provide valuable decision support for the field precision monitoring and efficient cultivation management of winter rapeseed growth.