无人机高光谱波段选择的叶面积指数反演

叶面积指数(LAI)是评价作物长势和作物产量的重要参数.为有效利用高光谱信息,优选出最佳波段进而构建新型双波段指数来提高LAI估测精度,以冬小麦为研究对象,获取冬小麦孕穗期无人机高光谱数据和实测地面LAI数据,开展冬小麦LAI反演研究.首先采用连续投影算法(SPA)、最佳指数法(OIF)以及逐波段组合法(E)分别进行无...

Leaf Area Index Estimation Based on UAV Hyperspectral Band Selection

Leaf area index (LAI) is an important parameter to evaluate crop condition and crop yield. In order to effectively utilize hyperspectral information and improve the estimation accuracy of LAI, the best band was selected, and the new two-band vegetation indexes were constructed. In this study, winter wheat was taken as the research object, the UAV hyperspectral data and ground LAI data were obtained at the booting stage. First, the successive projection algorithm (SPA), optimum index factor (OIF), and each band combination method (E) were used to screen the best band of UAV hyperspectral data, and then the selected best bands were constructed into the new two-band vegetation indexes (VI_OIF，VI_SPA，VI_E). Then, the new two-band vegetation indexes and the conventional two-band vegetation indexes (VI_F) constructed were compared and analyzed for correlation with LAI. Finally, support vector regression (SVR), partial least square (PLSR) and random forest for regression (RFR) were used to construct LAI estimation models. Meanwhile, comparing with the estimation accuracy of the conventional two-band vegetation indexes, the feasibility of LAI estimation was verified by the optimal regression model of the best new two-band vegetation indexes. The results were as follows: (1) The newly constructed two-band vegetation indexes VI_OIF, VI_SPA, VI_E and VI_F correlated with LAI were all at the significant level of 0.05, VI_SPA and VI_E correlated (r>0.65), among which RSI_SPA and RSI_E had the highest correlation coefficient with LAI (r>0.71) ; (2) The accuracy of LAI estimation of winter wheat based on SVR model, PLSR model and RFR model constructed by VI_OIF, VI_SPA, VI_E and VI_F were compared and analyzed. It was found that the VI_SPA_PLSR model had the highest accuracy and the best predictive ability, whose coefficient of determination (R2) and root mean square error (RMSE) were 0.75 and 0.90, respectively. The research results can provide technical support and theoretical reference for the band selection of UAV hyperspectral data and winter wheat LAI estimation.