一种基于无人机高光谱影像的土壤墒情检测新方法

土壤含水量(SMC)是生物地球化学和大气耦合过程的关键变量,在干旱区农业、生态和环境中扮演着重要角色。相较于星载遥感系统,无人机(UAV)具有可控性强、分辨率高等特点从而被广泛应用,为中小尺度地表参量的快速监测提供新的遥感平台。机载高光谱传感器的引入,为UAV遥感系统提供了高维海量、纳米级的数据源。然而基于UAV高光谱数据的研究并未深度挖掘,也尚未形成一个标准的技术方案。该研究立足于新疆维吾尔自治区典型农业区,利用6种预处理方案,包括一阶导数(FDR),二阶导数(SDR),连续体去除(CR)、吸光度(A)、吸光度一阶(FDA)和吸光度二阶(SDA),对所获取的UAV高光谱数据进行处理。在此背景下构建4种类型的适宜光谱指数:差值型指数(DI),比值型指数(RI),归一化型指数(NDI)和垂直型指数(PI),并从光谱机理上讨论指数的合理性。最后利用梯度提升回归树(GBRT)、随机森林(RF)和eXtreme Gradient Boosting(XGBoost)算法,以28个最适光谱指数为独立变量建立SMC估算模型,并通过不同集成学习算法的重要性对变量进行排序,从线性和非线性的角度对所构建光谱指数的效果进行考量评价。结果表明:(1)预处理和最适光谱指数能有效地消除了大气干扰和土壤背景,其中预处理A突出更多的光谱信息,PI相关性显著;(2)通过分析比较相关性系数(r)和集成学习算法的重要性,发现A_PI(|r|=0.773)是最适光谱指数,在线性和非线性关系中均有较优的表现;(3)在3种基于集成学习的SMC预测模型中,XGBoost估算模型效果拔群(R 2 val=0.926,RMSEP=1.943和RPD=2.556),其预测值的统计学特征与实测值的最为接近。3种模型效果排序为:XGBoost>RF>GBRT。综上所述,基于UAV高光谱影像,结合不同预处理和光谱指数,为低空遥感监测土壤墒情提出新的方案。该研究的方案具有潜在的高精度,是检测干旱区SMC的有效方法,针对快速易行地监测地表属性提供了崭新视角。相关结果为干旱区精准农业、生态系统给予更好的管理和保护策略。

A New Method for Predicting Soil Moisture Based on UAV Hyperspectral Image

Soil moisture content(SMC)is a key factor in biogeochemical and atmospheric coupling processes.It plays an important role in areas such as agriculture,ecology and environment in arid region.Compared to the spaceborne remote sensing system,UAV platform with hyperspectral sensors possess higher spatial resolution and maneuverability.With UAV(Unmanned Aerial Vehicle)being increasingly popular,it offers brand new platform of remote sensing.This platform realizes the goal that quickly and quantificationally monitor object in the area.Moreover,hyperspectral sensors contribute to remote sensing when they enrich high dimensional and nanoscale data source.However,there still lacks a standardized research scheme for estimation of UAV by hyperspectral Remote Sensing.In this study,we obtained UAV hyperspectral image from a typically dry-farming region lying in Xinjiang Uygur Autonomous Region.Hyperspectral image was pretreated using six methods of pretreatment,including first-derivative(FDR),second-derivative(SDR),continuum removal(CR),absorbance(A),first-derivative absorbance(FDA)and second-derivative absorbance(SDA).From pretreatment foundation,four types spectral indices were proposed containing the Difference Index(DI),the Ratio Index(RI),the Normalization Index(NDI)and the Perpendicular Index(PI).And the rationality of the spectral index was discussed from the spectral mechanism.Considering the superiority of ensemble learning algorithm rising in recent years,the SMC estimation model was constructed via Gradient Boosted Regression Tree(GBRT),Random Forest(RF)and eXtreme Gradient Boosting(XGBoost).In these models,28 appropriate spectral indices were used as independent variables and 70 SMC measured values as response variables.Spectral indices were ranked via importance based on ensemble learning model analyzed and compared to make a more comprehensive evaluation.The result indicated that:(1)atmospheric disturbance and soil background were eliminated effectively throughvarious pretreatment schemes and spectral indices.Pretreatment scheme A highlighted more spectral information and PI correlation was significant.(2)Optimum spectral index was A_PI(|r|=0.773)that the ranking of importance ranks first,and the correlation coefficient|r|is the highest,and it had excellent performance in both linear and nonlinear relationships.(2)XGBoost prediction model was outstanding in three ensemble learning models,and it yielded the highest R 2 val,the lowest RMSP and the best RPD(R 2 val=0.926,RMSEP=1.943 and RPD=2.556).The ranking of the predictive performance was XGBoost>RF>GBRT.This proved that this scheme was effective in digital mapping in arid region.In conclusion,there is potential high accuracy for UAV imagery based on hyperspectral imagery.This study afforded an effective method for predicting SMC in arid regions,and it provided a new perspective for quickly and easily monitoring object attributes and it proposed an alternative solution for predicting soil moisture.Ultimately,our program is supporting better management and conservation strategies for precision agriculture and ecosystems in arid regions.