基于无人机高光谱影像的水稻叶片磷素含量估算

为快速获取水稻叶片磷素含量信息，采用无人机搭载高光谱成像仪获取水稻冠层高光谱影像，并采样检测叶片磷素含量（质量分数）（Leaf phosphorus content, LPC）。分析了水稻LPC在无人机高光谱影像上的光谱特征，使用连续投影算法提取对磷素敏感的特征波长，通过任意波段组合构建并筛选与磷素高度相关的光谱指数，基于特征波长反射率和光谱指数建立水稻LPC的估算模型，利用最佳模型对高光谱影像进行反演填图，得到LPC空间分布信息。结果表明：全生育期内LPC与462~718 nm范围内光谱反射率显著负相关，负相关最大处相关系数达到-0.902;LPC的特征波长为670、706、722、846 nm，基于特征波长、使用偏最小二乘回归建立的LPC估算模型精度最高，验证R2达到0.925，RMSE为0.027%;在任意波段组合构建的3种类型的光谱指数中，NDSI(R498,R606)、RSI(R498,R606)和DSI(R498,R586)与LPC的相关性最高，相关系数分别为0.913、0.915和0.938;基于3个光谱指数、使用神经网络构建的LPC估算模型精度较高，验证R2为0.885，RMSE为0.029%;对各生育期水稻LPC空间分布的反演结果与实测数据相一致，说明利用无人机高光谱遥感可以实现田间水稻LPC的快速无损监测。

Estimation of Rice Leaf Phosphorus Content Using UAV-based Hyperspectral Images

In order to rapidly learn the rice canopy phosphorus content in the field, an imaging spectrometer (Cubert S185) mounted on a UAV was used to acquire the hyperspectral images of rice canopy in an experimental field and the leaves of each plot were sampled for leaf phosphorus content (LPC) measurement in the laboratory. The spectral features of the LPC in the UAV hyperspectral images were analyzed. The characteristic wavelengths of LPC were selected using the successive projections algorithm (SPA). Three spectral indices which were normalized difference spectral index (NDSI), ratio spectral index (RSI) and difference spectral index (DSI), were calculated by combing each two bands. The correlation analysis was performed between LPC and each spectral index in order to screen the most related spectral indices. LPC estimation models were built based on the spectral reflectance of the characteristic wavelength and the spectral indices using multiple linear regression (MLR), partial least squares regression (PLSR), support vector regression (SVR) and artificial neural network (ANN). The rice LPC distribution maps of each growth stage were made by computing the hyperspectral images pixel-by-pixel using the best LPC estimation model. The results showed that the LPC had significant negative correlations with the spectral reflectance within the range of 462~718 nm and the highest correlation coefficient reached -0.902. By using SPA, 670 nm, 706 nm, 722 nm and 846 nm were chosen as the characteristic wavelengths of LPC. The LPC estimation model which was built based on the four characteristic wavelengths using PLSR method achieved the highest accuracy and the validation R2 value reached 0.925 and the RMSE was 0.027%. Among all the spectral indices, NDSI(R498,R606), RSI(R498, R606), and DSI(R498,R586) had the highest correlation with LPC and the correlation coefficients were 0.913, 0.915 and 0.938, respectively. The validation R2 values of the ANN models based on the three spectral indices was 0.885 and the RMSE was 0.029%. The predicted LPC values derived from the LPC distribution map of each growth stage were consistent with the measured values. Therefore, the UAV-based hyperspectral remote sensing technology could provide a rapid and non destructive method to monitor the phosphorus status of rice leaves on the field scale.