基于全波段高光谱的冬小麦生长参数估算方法比较

Comparison of Estimation Methods for Growth Parameters of Winter Wheat Based on Full-Band Hyperspectral Data

利用高光谱数据监测作物生长情况具有无损和高效的特点,是现代农业的发展方向。为了简化高光谱数据处理流程,直接利用原始的高光谱反射率完成从建模到估算作物生长参数的全过程,应用于作物长势的实时监测。本文利用偏最小二乘回归(partial least squares regression,PLSR)、支持向量回归(support vector regression,SVR)和前馈神经网络(feedforward neural network,FNN)3种方法,利用全波段高光谱数据分别对冬小麦多个关键生育期(拔节、孕穗、扬花和乳熟期)生长参数(地上部生物量、叶面积指数、全氮含量和叶绿素浓度)进行了估算。比较3种方法的建模及估测效果,发现对于建模集数据,SVR对上述生长参数4个生育期的估测结果R2均值为0.89~0.98,MAPE为1.70%~7.53%,对于验证集数据,R2均值为0.90~0.94,MAPE为4.04%~7.46%,拟合优度和估测精度均超过PLSR和FNN,是估算方法中利用全波段光谱反射率估测冬小麦生长参数的最佳方案。随着无人机载高光谱技术成熟,SVR方法能够用于处理航拍获取的大范围田间高光谱信息,简便快捷地进行建模与参数反演,实时反映作物生长状态。

Hyperspectral data monitoring in crop growth status has the characteristics of nondestructive and efficient, which will be the developing direction of modern agriculture. In order to simplify the hyperspectral data processing procedure and apply to the real-time monitoring of crop growth, the original hyperspectal reflectance was directly used to complete the whole process from modeling to estimating the crop growth patameters and three methods including partial least square regression(PLSR), support vector regression(SVR) and feed forward neural network(FNN) were used to estimate the growth parameters(aboveground biomass, leaf area index, total nitrogen content and chlorophyll concentration) of winter wheat at multiple key growth stages(jointing, booting, anthesis and milk ripe stage) based on full-band hyperspectral data, respectively. The modeling and estimation performances of the three methods were also compared. For the SVR method, the mean R2 of the above growth parameters during the four growth periods was between 0.89 and 0.98, while MAPE was between 1.70% and 7.53% in calibration set and the mean R2 of the above growth parameters during the four growth periods was between 0.90 and 0.94, while MAPE was in between 4.04% and 7.46% in validation set. Both the interpretation of parameters and the estimation accuracy were higher than PLSR and FNN methods, which illustrated that the SVR method performed best in estimating winter wheat growth parameters using all-band spectral reflectance in the tested methods. With the maturity of UAV-based hyperspectral technology, the SVR method can be used to process a wide range of field hyperspectral information obtained fromaerial photography, facilitate modeling and parameter inversion, and reflect crop growth status in time.