利用无人机多光谱估算小麦叶面积指数和叶绿素含量

利用无人机遥感的方式进行农作物长势监测是目前精准农业、智慧农业发展的重要方向,为了探究无人机多光谱反演小麦叶面积指数(Leaf Area Index,LAI)和叶绿素含量的模型估算潜力,该研究在3个飞行高度(30、60、120 m)采集多光谱影像,通过使用全波段差值光谱指数(Difference Spectral Index,DSI)、比值光谱指数(Ratio Spectral Index,RSI)、归一化光谱指数(Normalized Spectral Index,NDSI)和经验植被指数与地面实测数据进行相关性分析,获得不同高度下的光谱指数与LAI和叶绿素含量的关系模型及其决定系数,以决定系数为依据分别构建多元逐步回归、偏最小二乘回归和人工神经网络模型,分析不同飞行高度无人机多光谱反演小麦冠层LAI和叶绿素含量SPAD(Soil and Plant Analyzer Development)值的精度。结果表明:1)30 m高度下,绿-红比值光谱指数与小麦LAI的相关性最高,相关系数为0.84;60 m高度下,红-蓝比值光谱指数与小麦叶绿素含量的相关性最高,相关系数为0.68;2)在60 m高度下,经验植被指数与小麦LAI和叶绿素含量的相关性较好,最大相关系数分别为0.77和0.50;3)利用偏最小二乘回归反演小麦LAI的精度最高,决定系数为0.732,均方根误差为0.055;利用人工神经网络模型反演小麦叶绿素含量的精度最高,决定系数为0.804,均方根误差0.135。该研究成果可为基于无人机平台的高通量作物监测提供理论依据,并为筛选无人机多光谱波段实现作物长势参数快速估测提供应用参考。

Estimation of the leaf area index and chlorophyll content of wheat using UAV multi-spectrum images

Monitoring the crop growth by using Unmanned Aerial Vehicle (UAV) based remote sensing technique is one of important directions for the development of precision and smart agriculture in China. In recent years, the development of UAV technology has greatly promoted the timely and rapid acquisition and long-term dynamic monitoring of agricultural and forestry ecological environment elements such as crop vegetation, water and soil. Compared with the data acquisition methods of satellite remote sensing and aerial remote sensing, UAV has the advantages of flexibility, convenience, low data acquisition cost and high image resolution. UAV remote sensing image is gradually becoming the main data source for the development of intelligent agriculture and forestry. In order to explore the inversion potential of Leaf Area Index (LAI) and chlorophyll content (SPAD) of wheat from UAV multi-spectral images, the multispectral images at three levels of flight altitudes (30, 60 and 120 m) by using the DJI Phantom4-M UAV platform which integrated five multispectral sensors (blue, green, red, red edge and near infrared) and TimeSync time synchronization system were collected to achieve centimeter-level positioning accuracy with more than 2 million pixel resolution, in Yuanyang wheat breading based, Xinxiang City, Henan Province. Based on the collected multispectral images, four different kinds of spectral indexes including: DSI (Difference Spectral Index), Ratio Spectral Index (Ratio Spectral Index), Normalized Spectral Index (NDSI) and empirical vegetation Index (EDVI) were used to compute the wheat canopy LAI and chlorophyll content (SPAD). The correlation analysis between different spectral index from different height UAV images and in-situ measured LAI and SPAD data were applied to select the optimal spectral index at different height. The Multiple Linear Stepwise Regression (MLSR), Partial Least Squares Regression (PLSR) and Artificial neural network (BP) model were constructed respectively for estimation of LAI and SPAD values. The experimental result showed that: 1) At 30 m height, the correlation coefficient between the green-red ratio spectral index and wheat LAI was the highest, with the value of 0.84. At the height of 60 m, the correlation coefficient between red-blue ratio spectral index and wheat chlorophyll content was the highest, with the value of 0.68. 2) At the height of 60 m, the correlation between empirical vegetation Index (EDVI) and LAI and chlorophyll content of wheat were both good, and the maximum correlation coefficients were 0.77 and 0.5, respectively. 3) The accuracy of wheat LAI inversion using partial least squares regression was the highest, with a determination coefficient of 0.732 and a root mean square error of 0.055. The accuracy of chlorophyll content inversion using artificial neural network model is the highest, the determination coefficient is 0.804, and the root mean square error is 0.135. This study provides a theoretical basis for high-throughput crop monitoring based on UAV platform, and provides an application reference for selecting UAV multi-spectral bands to achieve rapid estimation of crop growth parameters.