基于无人机遥感影像的大豆叶面积指数反演研究

作物叶面积指数的遥感反演是农业定量遥感研究热点之一,利用无人机遥感监测系统获取农作物光谱信息精确反演叶面积指数对精准农业生产与管理意义重大。本研究以山东省嘉祥县一带的大豆种植区为试验区,设计以多旋翼无人机为平台同步搭载Canon Power Shot G16数码相机和ADC-Lite多光谱传感器组成的无人机农情监测系统开展试验,分别获取大豆结荚期和鼓粒期的遥感影像。使用比值植被指数(RVI)、归一化植被指数(NDVI)、土壤调整植被指数(SAVI)、差值植被指数(DVI)、三角植被指数(TVI)5种植被指数,结合田间同步实测叶面积指数(leaf area index,LAI)数据,采用经验模型法分别构建了单变量和多变量LAI反演模型,通过决定系数(R2)、均方根误差(RMSE)和估测精度(EA)3个指标筛选出最佳模型。研究表明,有选择性地分时期进行农作物的叶面积指数反演是必要的,鼓粒期作为2个生育期中大豆LAI反演的最佳时期,其NDVI线性回归模型对大豆LAI的解释能力最强,R2=0.829,RMSE=0.301,反演大豆LAI最准确,EA=85.4%,生成的鼓粒期大豆LAI分布图反映了当地当时大豆真实长势情况。因此,以多旋翼无人机为平台同步搭载高清数码相机和多光谱传感器组成的无人机农情监测系统对研究大豆叶面积指数反演是可行性,可作为指导精准农业研究的一种新方法。

Soybean leaf area index retrieval with UAV (unmanned aerial vehicle) remote sensing imagery

Leaf area index (LAI) is the main parameter that reflects the status of crop growth. Retrieval of LAI is among the main focuses of quantitative remote sensing in agriculture. Crop spectral information with fine spatial resolution obtained by an Unmanned Aerial Vehicle (UAV) remote sensing monitoring system is used for estimating leaf area, which is important for precision agricultural production and management. In our study, an agricultural UAV remote sensing monitoring system was established based on a multi-rotor UAV with both Canon PowerShot G16 digital camera and ADC-Lite multispectral sensor mounted on the same platform. Based on this system, imageries were acquired over a soybean experimental field in Jiaxiang County of Shandong Province at podding and seed-filling stages. Five vegetation indices ratio vegetation index (RVI), normalized difference vegetation index (NDVI), soil-adjusted vegetation index (SAVI), difference vegetation index (DVI) and triangle vegetation index (TVI)] were calculated from the data. Together with measured LAI, both the univariate and multivariate empirical models were calibrated for estimating LAI of soybean. The best LAI retrieving models were identified based on best combinations of coefficient of determination (R₂), root mean square error (RMSE) and estimated accuracy (E_A). It was noted that there was the need to choose the best crop growth period for retrieving LAI. LAI was estimable at higher accuracy at seed-filling than at podding stage. Linear regression model of NDVI most accurately explained retrieval of LAI of soybean, with R₂ = 0.829, RMSE = 0.301 and E_A = 85.4%. NDVI linear regression model was therefore recommended as the most legible model for estimating LAI of soybean at seed-filling stage in this study area. The model was also recommended for application in mapping the LAI of soybean at seed-filling stage. According to our validation data, LAI map well reflected real-world spatial distribution pattern of LAI in soybean fields. The established agricultural UAV remote sensing monitoring system provided novel insights in guiding precision agriculture applications and the corresponding retrieval models for studying the feasibility of retrieving LAI.