国产日韩欧美一区二区三区三州_亚洲少妇熟女av_久久久久亚洲av国产精品_波多野结衣网站一区二区_亚洲欧美色片在线91_国产亚洲精品精品国产优播av_日本一区二区三区波多野结衣 _久久国产av不卡

?

基于無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)的冬小麥產(chǎn)量估算對(duì)比

2019-02-20 13:31陶惠林馮海寬楊貴軍楊小冬苗夢(mèng)珂吳智超翟麗婷
關(guān)鍵詞:數(shù)碼影像冬小麥生育期

陶惠林,馮海寬,楊貴軍,楊小冬,苗夢(mèng)珂,5,吳智超,5,翟麗婷,5

基于無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)的冬小麥產(chǎn)量估算對(duì)比

陶惠林1,2,3,4,馮海寬1,3,4※,楊貴軍1,3,4,楊小冬1,3,4,苗夢(mèng)珂1,3,4,5,吳智超1,3,4,5,翟麗婷1,3,4,5

(1. 農(nóng)業(yè)農(nóng)村部農(nóng)業(yè)遙感機(jī)理與定量遙感重點(diǎn)實(shí)驗(yàn)室,北京農(nóng)業(yè)信息技術(shù)研究中心,北京 100097; 2. 安徽理工大學(xué)測(cè)繪學(xué)院,淮南 232001; 3. 國(guó)家農(nóng)業(yè)信息化工程技術(shù)研究中心,北京 100097; 4. 北京市農(nóng)業(yè)物聯(lián)網(wǎng)工程技術(shù)研究中心,北京 100097;5. 河南理工大學(xué)測(cè)繪與國(guó)土信息工程學(xué)院,焦作 454000)

作物產(chǎn)量準(zhǔn)確估算在農(nóng)業(yè)生產(chǎn)中具有重要意義。該文利用無(wú)人機(jī)獲取冬小麥挑旗期、開花期和灌漿期數(shù)碼影像和高光譜數(shù)據(jù),并實(shí)測(cè)產(chǎn)量。首先利用無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)分別提取數(shù)碼影像指數(shù)和光譜參數(shù),然后將數(shù)碼影像指數(shù)和光譜參數(shù)與冬小麥產(chǎn)量作相關(guān)性分析,挑選出相關(guān)性較好的9個(gè)指數(shù)和參數(shù),最后以選取的數(shù)碼影像指數(shù)和光譜參數(shù)為建模因子,通過MLR(multiple linear regression,MLR)和RF(random forest,RF)對(duì)產(chǎn)量進(jìn)行估算。結(jié)果表明:數(shù)碼影像指數(shù)和光譜參數(shù)與實(shí)測(cè)產(chǎn)量均有很強(qiáng)的相關(guān)性。利用數(shù)碼影像指數(shù)和光譜參數(shù)通過MLR和RF構(gòu)建的產(chǎn)量估算模型均在灌漿期表現(xiàn)精度最高,在灌漿期,數(shù)碼影像指數(shù)和光譜參數(shù)構(gòu)建的MLR模型2和NRMSE分別為0.71、12.79%,0.77、10.32%。對(duì)模型對(duì)比分析可知,以光譜參數(shù)為因子的MLR模型精度較高,更適合用于估算冬小麥產(chǎn)量。利用無(wú)人機(jī)遙感數(shù)據(jù),通過光譜參數(shù)建立的MLR模型能夠快速、方便地對(duì)作物進(jìn)行產(chǎn)量預(yù)測(cè),并可以根據(jù)不同生育期的產(chǎn)量估算模型有效地對(duì)作物進(jìn)行監(jiān)測(cè)。

無(wú)人機(jī);數(shù)碼影像;高光譜;冬小麥;產(chǎn)量;估算;多元線性回歸;隨機(jī)森林

0 引 言

在精準(zhǔn)農(nóng)業(yè)中,作物產(chǎn)量的準(zhǔn)確監(jiān)測(cè)對(duì)農(nóng)業(yè)管理方面有著重要的意義[1-3]。通過衛(wèi)星可以實(shí)現(xiàn)大范圍、大區(qū)域監(jiān)測(cè)[4]。而對(duì)于小尺度作物的監(jiān)測(cè),由于衛(wèi)星數(shù)據(jù)獲取時(shí)間長(zhǎng),受分辨率和氣象等多方面影響,所體現(xiàn)的作用不明顯[5-6]。無(wú)人機(jī)遙感具有較高的分辨率,操作便捷,能夠快速地進(jìn)行觀測(cè)[7-10],相比衛(wèi)星受限制條件少,在田間對(duì)作物進(jìn)行監(jiān)測(cè)能夠取得很好的效果。估算作物產(chǎn)量方面,根據(jù)平臺(tái)的不同,所獲取數(shù)據(jù)的方式各異,如利用衛(wèi)星、無(wú)人機(jī)或地面獲取。國(guó)內(nèi)外學(xué)者在作物產(chǎn)量估算方面做了大量研究,何亞娟等[11]根據(jù)SPOT衛(wèi)星遙感數(shù)據(jù)通過甘蔗的葉面積指數(shù)(leaf area index,LAI)對(duì)甘蔗的產(chǎn)量進(jìn)行了估算,并構(gòu)建LAI-產(chǎn)量估算模型取得較好效果。朱婉雪等[12]利用無(wú)人機(jī)對(duì)冬小麥的3個(gè)生育期進(jìn)行了觀測(cè),通過最小二乘分析方法建立不同植被指數(shù)的產(chǎn)量估算模型,各指數(shù)構(gòu)建的冬小麥估算模型中在灌漿期建模模型的效果最好,指數(shù)EVI2(enhanced vegetation index without a blue band,EVI2)建立的模型精度最高。Kefauver等[13]利用無(wú)人機(jī)RGB、多光譜、熱航空影像數(shù)據(jù),多影像組合數(shù)據(jù)構(gòu)建的多元線性模型可以對(duì)大麥產(chǎn)量進(jìn)行有效估算。Mengmeng等[14]利用低空無(wú)人機(jī)圖像數(shù)據(jù)得到的9種植被指數(shù)構(gòu)建了逐步回歸模型,表明小麥產(chǎn)量與可見帶差異植被指數(shù)、歸一化綠-藍(lán)差異指數(shù)、綠-紅比率指數(shù)和綠色植被指數(shù)相關(guān)。Gong等[15]通過無(wú)人機(jī)遙感數(shù)據(jù),探究出歸一化差異植被指數(shù)能夠很好地對(duì)產(chǎn)量進(jìn)行估算,估算的誤差小于13%。劉煥軍等[16]用航空高光譜數(shù)據(jù),研究出通過光譜一階微分參數(shù)構(gòu)建的產(chǎn)量預(yù)測(cè)模型精度優(yōu)于以歸一化植被指數(shù)、優(yōu)化土壤調(diào)節(jié)植被指數(shù)和二階微分構(gòu)建的估產(chǎn)模型。吳瓊等[17]利用地面高光譜數(shù)據(jù),驗(yàn)證出多生育期的估測(cè)效果優(yōu)于單個(gè)生育期。張松等[18]基于地面高光譜遙感數(shù)據(jù)將植被指數(shù)作為構(gòu)建產(chǎn)量估算模型的因子,構(gòu)建的估算模型孕穗期和抽穗期的精度較高,能夠很好地監(jiān)測(cè)冬小麥產(chǎn)量。以上研究從不同數(shù)據(jù)獲取方式,不同分析方法等方面估算作物產(chǎn)量,但利用無(wú)人機(jī)數(shù)碼和高光譜數(shù)據(jù)估算作物產(chǎn)量并進(jìn)行對(duì)比精度分析的研究還很少。本文通過無(wú)人機(jī)遙感平臺(tái),利用無(wú)人機(jī)數(shù)碼與高光譜數(shù)據(jù)分別估算冬小麥產(chǎn)量,以數(shù)碼影像指數(shù)和光譜參數(shù)為建模因子,通過機(jī)器學(xué)習(xí)多元線性回歸(MLR)和隨機(jī)森林(RF)2種方法分別構(gòu)建了基于無(wú)人機(jī)數(shù)碼和高光譜的冬小麥產(chǎn)量估算模型,探討2種遙感數(shù)據(jù)的估算精度、最優(yōu)模型和最佳估算生育期。本研究提供了一種基于無(wú)人機(jī)遙感平臺(tái)的科學(xué)技術(shù)方法,以便農(nóng)業(yè)管理者更好地監(jiān)測(cè)作物產(chǎn)量。

1 材料與方法

1.1 研究區(qū)概況與試驗(yàn)設(shè)計(jì)

試驗(yàn)所在地位于北京市昌平區(qū)小湯山鎮(zhèn)國(guó)家精準(zhǔn)農(nóng)業(yè)研究示范基地,處溫榆河沖積平原和燕山、太行山支脈的結(jié)合地帶,地理坐標(biāo)為40°00′~40°21′N,116°34′~117°00′E,在溫帶季風(fēng)區(qū),屬暖溫帶大陸性季風(fēng)氣候,地勢(shì)平坦,土壤肥沃,多種植小麥、玉米等,研究區(qū)位置如圖1所示。

圖1 研究區(qū)位置

此次試驗(yàn)在示范基地的田間進(jìn)行,試驗(yàn)田東西總長(zhǎng)度為84 m,南北總長(zhǎng)度為32 m,在冬小麥品種方面,選用了京9843(P1)和中麥175(P2)2個(gè)品種;在施肥肥料方面,不同類型處理的尿素施肥量分別為N1(不施用)、N2(195 kg/hm2)、N3(390 kg/hm2)和N4(585 kg/hm2);水分處理方面分別為W1(雨養(yǎng))、W2(正常水,675 m3/hm2)和W3(1.5倍正常水,1 012.5 m3/hm2)3種方式處理。該試驗(yàn)田16個(gè)小區(qū),3次重復(fù)處理,并在不同時(shí)期進(jìn)行監(jiān)測(cè),分別進(jìn)行了3個(gè)不同生育期試驗(yàn),試驗(yàn)田分布如圖2所示。

注:P是冬小麥品種,P1:京9843,P2: 中麥175;W是水分處理,W1、W2、W3分別為雨養(yǎng)、675、1 012.5 m3·hm-2;N為尿素施用,N1、N2、N3、N4分別為 0、195、390和585 kg·hm-2。

1.2 數(shù)據(jù)獲取及處理

1.2.1 地面數(shù)據(jù)獲取與處理

在冬小麥的成熟后,在每個(gè)試驗(yàn)小區(qū)隨機(jī)取1 m2區(qū)域進(jìn)行調(diào)查,通過收獲1 m2的籽粒并折算成14%含水率下的產(chǎn)量計(jì)算實(shí)測(cè)產(chǎn)量,共獲得48個(gè)小區(qū)的產(chǎn)量數(shù)據(jù)。

1.2.2 無(wú)人機(jī)數(shù)碼影像與高光譜數(shù)據(jù)獲取與處理

本次試驗(yàn)采用了八旋翼無(wú)人機(jī),機(jī)身凈質(zhì)量4.2 kg,載物質(zhì)量6 kg,單臂長(zhǎng)386 mm,續(xù)航時(shí)間15~20 min,無(wú)人機(jī)獲取數(shù)據(jù)時(shí)要求天氣晴朗少云,時(shí)間在 10:00-14:00之間,飛行高度為80 m,攜帶的傳感器是數(shù)碼相機(jī)和成像光譜儀,其主要參數(shù)如表1所示,獲取的影像見圖3。分別在冬小麥的挑旗期(2015年4月26號(hào))、開花期(5月13號(hào))、灌漿期(5月22號(hào))3個(gè)不同生育期進(jìn)行數(shù)碼和高光譜影像數(shù)據(jù)獲取。

表1 數(shù)碼相機(jī)和UHD185成像光譜儀的主要參數(shù)

無(wú)人機(jī)數(shù)碼影像處理主要利用俄羅斯Agisoft LLC公司的Agisoft PhotoScan Professional軟件進(jìn)行影像拼接處理,過程如下:1)將無(wú)人機(jī)數(shù)碼影像和飛行時(shí)的位置和姿態(tài)數(shù)據(jù)導(dǎo)入;2)生成密集點(diǎn)云;3)建立網(wǎng)格;4)生成空間紋理;5)生成冬小麥材料的無(wú)人機(jī)數(shù)碼高清正射影像。

無(wú)人機(jī)高光譜數(shù)據(jù)進(jìn)行處理過程具體如下:

1)無(wú)人機(jī)高光譜影像校正和拼接。對(duì)于影像的校正,需要把無(wú)人機(jī)高光譜影像遙感影像像元亮度值DN(digital number,DN)值轉(zhuǎn)化為地表反射率,在進(jìn)行影像拼接處理的時(shí)候借鑒Turner等[19]的研究成果,將獲取的影像和位置數(shù)據(jù),利用UHD185自帶的處理軟件和俄羅斯Agisoft LLC公司的Agisoft PhotoScan 軟件進(jìn)行影像拼接,經(jīng)過影像的校正和拼接生成了挑旗期、開花期和灌漿期的高光譜正射影像,每個(gè)生育期的影像中有125個(gè)波段,每個(gè)波段間隔4 nm,波段范圍為454~950 nm。

2)提取光譜反射率。在進(jìn)行反射率提取時(shí),為了更好地控制區(qū)域矢量問題,過程主要在Arcgis軟件中完成,在軟件中根據(jù)每個(gè)小區(qū)的面積繪制出30個(gè)矢量,每個(gè)小區(qū)編號(hào),繪制出的矢量面積總和等于每個(gè)小區(qū)的矢量面積,每個(gè)生育期影像有48個(gè)小區(qū),得到1 440個(gè)矢量數(shù)據(jù),再通過IDL程序?qū)?yīng)小區(qū)編號(hào)計(jì)算出每個(gè)小區(qū)的光譜反射率的平均值,并以平均值為該小區(qū)的光譜反射率,獲取不同生育期的各小區(qū)光譜反射率。

圖3 無(wú)人機(jī)數(shù)碼和高光譜影像

1.3 研究方法

1.3.1冬小麥估算模型

本文選取了多元線性回歸(multiple linear regression,MLR)和隨機(jī)森林(random forest,RF)2種分析方法,分別建立不同植被指數(shù)與冬小麥產(chǎn)量的估算模型,模型為常用的統(tǒng)計(jì)模型,MLR使用的前提是自變量和因變量有著很好的相關(guān)性,能夠通過多個(gè)自變量來(lái)預(yù)測(cè)因變量。 RF的算法是基于bootstrap取樣的一種機(jī)器學(xué)習(xí)算法,通過將樣本放回抽樣的方法進(jìn)行多次取樣,建立決策樹模型,進(jìn)行決策樹組合預(yù)測(cè),分類效果越好,說明構(gòu)建的模型精度越高。

1.3.2 數(shù)碼影像指數(shù)選取

利用生成的無(wú)人機(jī)數(shù)碼高清正射影像,綠色植物反射綠光和吸收紅和藍(lán)光的特征,從數(shù)碼高清正射影像中提取出冬小麥試驗(yàn)田各小區(qū)的DN值,將影像中的紅、綠、藍(lán)光這3個(gè)通道的DN值進(jìn)行歸一化處理,得到、、,計(jì)算公式見表2。

根據(jù)前人的研究成果,選取11種植被指數(shù),加上紅綠藍(lán)以及歸一化后的數(shù)碼影像指數(shù),總共17個(gè)數(shù)碼影像指數(shù),用來(lái)對(duì)冬小麥的產(chǎn)量進(jìn)行估算研究,具體如表2所示。

1.3.3 光譜參數(shù)選取

植被指數(shù)被廣泛應(yīng)用于農(nóng)業(yè)作物監(jiān)測(cè),本文根據(jù)前人的研究成果,選擇8個(gè)典型的高光譜植被指數(shù)。另外Aasen等[27]發(fā)現(xiàn)紅邊區(qū)域波段可以有效監(jiān)測(cè)作物長(zhǎng)勢(shì),故文中挑選了4個(gè)紅邊參數(shù),具體光譜參數(shù)如表3所示。

表2 數(shù)碼影像指數(shù)

表3 光譜參數(shù)

1.3.4 模型精度驗(yàn)證

文中根據(jù)冬小麥產(chǎn)量估算模型的構(gòu)建,選取決定系數(shù)(coefficient of determination,2)、均方根誤差(root mean squared error,RMSE)和標(biāo)準(zhǔn)均方根誤差(normalized root mean squared error,NRMSE)3個(gè)指標(biāo)作為模型精度的驗(yàn)證。

2 結(jié)果與分析

2.1 無(wú)人機(jī)數(shù)碼影像指數(shù)和光譜參數(shù)與產(chǎn)量的相關(guān)性分析

將數(shù)碼影像指數(shù)和光譜參數(shù)與冬小麥實(shí)測(cè)產(chǎn)量數(shù)據(jù)進(jìn)行相關(guān)性分析,結(jié)果見表4。

表4 無(wú)人機(jī)數(shù)碼影像指數(shù)和光譜參數(shù)與冬小麥實(shí)測(cè)產(chǎn)量相關(guān)性分析

注:*表示在0.05水平上顯著,**表示在0.01水平上顯著。

Note:*represents significant at 0.05 level, ** represents significant at 0.01 level.

從表4中可知,數(shù)碼影像指數(shù)與冬小麥實(shí)測(cè)產(chǎn)量均呈現(xiàn)極顯著相關(guān)(<0.01);光譜參數(shù)和冬小麥實(shí)測(cè)產(chǎn)量相關(guān)性大部分也達(dá)到極顯著水平。相比不同生育期,挑旗期的無(wú)人機(jī)數(shù)碼影像指數(shù)均與冬小麥實(shí)測(cè)產(chǎn)量呈現(xiàn)極顯著相關(guān),其中指數(shù)的相關(guān)系數(shù)絕對(duì)值最大,為0.671;而光譜參數(shù)中,除光譜參數(shù)紅邊振幅表現(xiàn)顯著相關(guān)外,剩余的參數(shù)均表現(xiàn)出極顯著水平,相關(guān)性最好的是TCARI/OSAVI,相關(guān)系數(shù)為0.744;開花期,數(shù)碼影像指數(shù)與實(shí)測(cè)產(chǎn)量相關(guān)性均表現(xiàn)出極顯著相關(guān),指數(shù)的相關(guān)性系數(shù)絕對(duì)值最大,為0.737;光譜參數(shù)中TCARI和最小振幅表現(xiàn)無(wú)顯著相關(guān),其余的光譜參數(shù)都表現(xiàn)出極顯著相關(guān),其中表現(xiàn)效果最好的是參數(shù)SR,相關(guān)系數(shù)是0.795;灌漿期,數(shù)碼影像指數(shù)與實(shí)測(cè)產(chǎn)量相關(guān)性均極顯著相關(guān),相關(guān)性絕對(duì)值最高的指數(shù)為,為0.747;對(duì)于光譜參數(shù)而言,僅TCARI無(wú)顯著相關(guān),其余參數(shù)均極顯著相關(guān),表現(xiàn)最好的參數(shù)是TCARI/OSAVI,相關(guān)系數(shù)達(dá)到0.800。

2.2 基于無(wú)人機(jī)數(shù)碼影像數(shù)據(jù)的冬小麥產(chǎn)量估算精度分析和驗(yàn)證

根據(jù)無(wú)人機(jī)數(shù)碼影像指數(shù)與冬小麥產(chǎn)量相關(guān)性分析結(jié)果,按相關(guān)性強(qiáng)弱,在不同的生育期和利用高光譜數(shù)據(jù)估算產(chǎn)量一樣分別挑選出9個(gè)相關(guān)系數(shù)較大的數(shù)碼影像指數(shù),選取的數(shù)碼影像指數(shù)作為冬小麥產(chǎn)量估算模型的因子通過MLR和RF來(lái)構(gòu)建模型,得到了挑旗期、開花期與灌漿期的MLR和RF模型的各評(píng)價(jià)指標(biāo),結(jié)果如表5所示。

表5 建模集不同生育期的數(shù)碼影像指數(shù)估算冬小麥產(chǎn)量的精度分析

根據(jù)表5,MLR模型的精度在不同生育期均明顯優(yōu)于RF模型,且2種模型的估算精度均表現(xiàn)為灌漿期最高,挑旗期最低。其中MLR模型最佳2是0.71(RMSE=730.66kg/hm2,NRMSE=12.79%),RF模型最佳2是0.57(RMSE=894.16kg/hm2,NRMSE=15.65%),說明MLR模型優(yōu)勢(shì)較為明顯。

分別對(duì)冬小麥的3個(gè)生育期構(gòu)建的產(chǎn)量估算模型進(jìn)行驗(yàn)證,獲得不同生育期的驗(yàn)證分析結(jié)果,見圖4和圖5。MLR和RF模型驗(yàn)證的估算效果和建模效果保持一致,從挑旗期到灌漿期,MLR和RF模型均表現(xiàn)效果逐漸增強(qiáng),至灌漿期,達(dá)到最佳,MLR和RF驗(yàn)證2分別是0.77、0.52,NRMSE分別達(dá)到13.56%、17.22%,驗(yàn)證效果較好。

圖4 驗(yàn)證集數(shù)碼影像的MLR方法預(yù)測(cè)產(chǎn)量與實(shí)測(cè)產(chǎn)量對(duì)比

圖5 驗(yàn)證集數(shù)碼影像的RF方法預(yù)測(cè)產(chǎn)量與實(shí)測(cè)產(chǎn)量對(duì)比

2.3 基于無(wú)人機(jī)高光譜數(shù)據(jù)的冬小麥產(chǎn)量估算精度分析和驗(yàn)證

利用建模分析方法MLR和RF構(gòu)建冬小麥挑旗期、開花期、和灌漿期的產(chǎn)量估算模型,根據(jù)光譜參數(shù)的相關(guān)性分析結(jié)果,挑選出相關(guān)性絕對(duì)值較大的前9個(gè)光譜參數(shù)作為估算冬小麥產(chǎn)量的自變量,保證了自變量中同時(shí)包含植被指數(shù)和紅邊參數(shù)。挑旗期、開花期和灌漿期取重復(fù)1和重復(fù)2區(qū)域數(shù)據(jù),共32個(gè)樣本作建模集,結(jié)果如表6所示。

從表6可知,從挑旗期到灌漿期,對(duì)于MLR和RF模型,模型的估算精度逐漸提高,擬合效果越來(lái)越好,其中MLR模型最佳2是0.77,NRMSE是10.32%;RF模型最佳2是0.61,NRMSE是14.79%,MLR模型估算精度在不同生育期均優(yōu)于RF模型。

為了驗(yàn)證建模集估算效果,將重復(fù)3區(qū)域數(shù)據(jù)(16個(gè)樣本)進(jìn)行模型驗(yàn)證,結(jié)果見圖6和圖7。隨著生育期推移,驗(yàn)證集2逐漸增大,RMSE和NRMSE逐漸減小,結(jié)果與建模集效果保持一致,說明驗(yàn)證的效果比較穩(wěn)定。另外,與數(shù)碼影像指數(shù)估算的結(jié)果一致,MLR和RF模型驗(yàn)證精度也是灌漿期最高,開花期次之,挑旗期精度最低。

表6 建模集不同生育期的光譜參數(shù)估算冬小麥產(chǎn)量的精度分析

圖6 驗(yàn)證集高光譜數(shù)據(jù)的MLR方法預(yù)測(cè)產(chǎn)量與實(shí)測(cè)產(chǎn)量對(duì)比

圖7 驗(yàn)證集高光譜數(shù)據(jù)的RF方法預(yù)測(cè)產(chǎn)量與實(shí)測(cè)產(chǎn)量對(duì)比

2.4 精度對(duì)比

分別利用無(wú)人機(jī)數(shù)碼和高光譜影像數(shù)據(jù)對(duì)冬小麥產(chǎn)量進(jìn)行估算,構(gòu)建了基于2種遙感數(shù)據(jù)的冬小麥不同生育期的MLR和RF模型。利用相同方法,不同數(shù)據(jù)構(gòu)建產(chǎn)量估算模型,基于無(wú)人機(jī)高光譜數(shù)據(jù)構(gòu)建的產(chǎn)量模型效果均優(yōu)于基于無(wú)人機(jī)數(shù)碼數(shù)據(jù)的模型,通過無(wú)人機(jī)高光譜和無(wú)人機(jī)數(shù)碼構(gòu)建的最優(yōu)模型建模2分別為0.77、0.71,RMSE分別646.67 kg/hm2、730.66 kg/hm2,NRMSE分別為10.32%、12.79%。對(duì)比不同方法建立的2種估算模型,相同數(shù)據(jù)下MLR構(gòu)建的模型優(yōu)于RF模型?;跓o(wú)人機(jī)數(shù)碼影像構(gòu)建的MLR與RF模型挑旗期2分別為0.48、0.24,NRMSE分別為17.24%,21.60%;開花期2分別是0.69、0.56,NRMSE分別為13.18%、 15.85%;灌漿期2分別為0.71、0.57,NRMSE分別為12.79%、15.65%?;跓o(wú)人機(jī)高光譜數(shù)據(jù)構(gòu)建的MLR與RF挑旗期2、NRMSE分別為0.64、14.33%,0.37、19.01%,開花期2、NRMSE分別是0.75、11.10%,0.58、15.37%,灌漿期2、NRMSE分別為0.77、10.32%,0.61、14.79%。綜合模型的評(píng)價(jià)指標(biāo)和模型的適用性,MLR方法模型更適合產(chǎn)量估算,且利用無(wú)人機(jī)高光譜數(shù)據(jù)估算產(chǎn)量效果較好。

3 討 論

人工估產(chǎn)所進(jìn)行的工作量大、過程繁瑣且很容易受到人為和其他因素的干擾。從而影響估算精度,利用無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)進(jìn)行估算可以避免這些問題。數(shù)碼相機(jī)和高光譜的空間分辨率較高,在進(jìn)行取樣時(shí)所取的大小為1 m2,區(qū)域內(nèi)冬小麥的生長(zhǎng)較為一致,數(shù)據(jù)具有代表性,然后在通過無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)進(jìn)行產(chǎn)量估算。

所選取的數(shù)碼影像指數(shù)和光譜參數(shù)大部分都與冬小麥產(chǎn)量呈現(xiàn)極顯著相關(guān)(<0.01),在冬小麥不同生育期,數(shù)碼影像指數(shù)與光譜參數(shù)和產(chǎn)量相關(guān)性存在差異,這與不同生育期時(shí)波段敏感程度有關(guān),挑旗期,波段敏感性不強(qiáng),數(shù)碼影像指數(shù)和光譜參數(shù)的相關(guān)性絕對(duì)值較小,隨著冬小麥生長(zhǎng),開花期和灌漿期波段敏感性逐漸增強(qiáng),與產(chǎn)量相關(guān)性絕對(duì)值逐漸變大,因此,數(shù)碼影像指數(shù)與光譜參數(shù)均與小麥產(chǎn)量有著較好的相關(guān)性??傮w來(lái)說,小麥產(chǎn)量與灌漿期相關(guān)性最好,開花期其次,而挑旗期最差。

將數(shù)碼影像指數(shù)和光譜參數(shù)通過MLR和RF方法進(jìn)行冬小麥產(chǎn)量估算,發(fā)現(xiàn)在不同時(shí)期模型的效果不同,模型精度最高的時(shí)期是灌漿期,精度最差的是挑旗期。從擬合性、一致性和精度來(lái)看,MLR模型效果均強(qiáng)于RF模型,模型精度表現(xiàn)為灌漿期高于開花期,開花期高于挑旗期,2種模型的估算結(jié)果和冬小麥生長(zhǎng)規(guī)律較為一致,隨著生育期推移,越靠近生育期后期冬小麥的產(chǎn)量越趨于穩(wěn)定,而此時(shí)進(jìn)行估算的產(chǎn)量結(jié)果能夠很好地反映冬小麥的實(shí)測(cè)產(chǎn)量,所以在灌漿期構(gòu)建冬小麥產(chǎn)量估算模型效果是最好的。

盡管數(shù)據(jù)源不同,但建立的估算模型均表現(xiàn)出模型MLR精度高于RF,這是由于RF適合數(shù)據(jù)較多的樣本,在樣本量少的情況下表現(xiàn)波動(dòng)。另外,相比數(shù)碼相機(jī),高光譜數(shù)據(jù)分辨率高,波段較多,能夠較全面地反映冬小麥信息,綜合來(lái)說,通過MLR方法以光譜參數(shù)為因子的估算模型較優(yōu)。

估產(chǎn)過程中受土壤背景和陰影光譜的影響[12],直接通過單個(gè)植被指數(shù)估算產(chǎn)量,而不考慮不同生長(zhǎng)期光譜影響,估產(chǎn)模型適用性不高[18]。另外,利用無(wú)人機(jī)數(shù)碼和高光譜估產(chǎn),可以對(duì)比不同傳感器的估產(chǎn)性能,以便選取更合適的估產(chǎn)方式。冬小麥不同生長(zhǎng)期光譜受影響大小不同,如何更好地利用植被指數(shù)消除外界因素干擾提高估產(chǎn)精度,需要進(jìn)一步研究。另外,本文僅利用一年冬小麥數(shù)據(jù)分析,未來(lái)將針對(duì)長(zhǎng)時(shí)間序列數(shù)據(jù)進(jìn)行深入研究。

4 結(jié) 論

本文基于無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)對(duì)不同生育期冬小麥產(chǎn)量進(jìn)行估算,研究結(jié)論如下:

針對(duì)不同生育期的數(shù)碼影像指數(shù)和光譜參數(shù),模型MLR擬合性和精度均高于RF。就生育期而言,表現(xiàn)最好的均是灌漿期,基于光譜參數(shù)的MLR模型2、RMSE和NRMSE分別為0.77、646.67 kg/hm2、10.32%,RF模型的2、RMSE和NRMSE分別是0.61、838.99kg/hm2、14.79%;基于數(shù)碼影像指數(shù)的MLR模型2、RMSE和NRMSE分別是0.71、730.66kg/hm2、12.79%,而RF模型分別為0.57,894.16kg/hm2,15.65%。

[1] 任建強(qiáng),陳仲新,唐華俊,等. 基于遙感信息與作物生長(zhǎng)模型的區(qū)域作物單產(chǎn)模擬[J]. 農(nóng)業(yè)工程學(xué)報(bào),2011,27(8):257-264.

Ren Jianqiang, Chen Zhongxin, Tang Huajun, et al.Regional crop yield simulation based on crop growth model and remote sensing data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2011, 27(8): 257-264. (in Chinese with English abstract)

[2] Mueller N D, Gerber J S, Johnston M, et al. Closing yield gaps through nutrient and water management[J]. Nature, 2012, 490(7419): 254-257.

[3] Neumann K, Verburg P H, Stehfest E , et al. The yield gap of global grain production: A spatial analysis[J]. Agricultural Systems, 2010, 103(5): 316-326.

[4] Du M, Noguchi N. Multi-temporal monitoring of wheat growth through correlation analysis of satellite images, unmanned aerial vehicle images with ground variable[J]. IFAC PapersOnLine, 2016, 49(16): 5-9.

[5] 蘇偉,侯寧,李琪,等. 基于Sentinel-2遙感影像的玉米冠層葉面積指數(shù)反演[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào),2018,49(1):151-156.

Su Wei, Hou Ning, Li Qi, et al. Retrieving leaf area index of corn canopy based on sentinel-2 remote sensing image[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018, 49(1): 151-156.(in Chinese with English abstract)

[6] Yu N, Li L, Schmitz N, et al. Development of methods to improve soybean yield estimation and predict plant maturity with an unmanned aerial vehicle based platform[J]. Remote Sensing of Environment, 2016(187): 91-101.

[7] 張春蘭,楊貴軍,李賀麗,等. 基于隨機(jī)森林算法的冬小麥葉面積指數(shù)遙感反演研究[J]. 中國(guó)農(nóng)業(yè)科學(xué),2018,51(5):855-867.

Zhang Chunlan, Yang Guijun, Li Heli, et al.Remote sensing inversion of leaf area index of winter wheat based on random forest algorithm[J]. Scientia Agricultura Sinica, 2018, 51(5): 855-867. (in Chinese with English abstract)

[8] Yue Jibo, Feng Haikuan, Jin Xiuliang, et al. A comparison of crop parameters estimation using images from UAV-mounted snapshot hyperspectral sensor and high-defifinition digital camera[J]. Remote Sensing, 2018(10): 1138-1162.

[9] 常瀟月,常慶瑞,王曉凡,等. 基于無(wú)人機(jī)高光譜影像玉米葉綠素含量估算[J]. 干旱地區(qū)農(nóng)業(yè)研究,2019,37(1):66-73.

Chang Xiaoyue, Chang Qingrui, Wang Xiaofan, et al.Estimation of maize leaf chlorophyll contents based on UAV hyperspectral drone image[J]. Agricultural Research in the Arid Areas. 2019, 37(1): 66-73. (in Chinese with English abstract)

[10] Watanabe K, Guo W, Arai K, et al. High-throughput phenotyping of sorghum plant height using an unmanned aerial vehicle and its application to genomic prediction modeling[J]. Frontiers in Plant Science, 2017, 8:421-432.

[11] 何亞娟,潘學(xué)標(biāo),裴志遠(yuǎn),等. 基于SPOT遙感數(shù)據(jù)的甘蔗葉面積指數(shù)反演和產(chǎn)量估算[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào),2013,44(5):226-231.

He Yajuan, Pan Xuebiao, Pei Zhiyuan, et al. Estimation of LAI and yield of sugarcane based on SPOT remote sensing data[J]. Transactions of the Chinese Society for Agricultural Machinery, 2013, 44(5): 226-231. (in Chinese with English abstract)

[12] 朱婉雪,李仕冀,張旭博,等. 基于無(wú)人機(jī)遙感植被指數(shù)優(yōu)選的田塊尺度冬小麥估產(chǎn)[J]. 農(nóng)業(yè)工程學(xué)報(bào),2018,34(11):78-86.

Zhu Wanxue, Li Shiji, Zhang Xubo, et al. Estimation of winter wheat yield using optimal vegetation indices from unmanned aerial vehicle remote sensing[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(11): 78-86. (in Chinese with English abstract)

[13] Kefauver S C, Vicente, Vergara-Díaz O, et al. Comparative UAV and field phenotyping to assess yield and nitrogen use efficiency in hybrid and conventional barley[J]. Frontiers in Plant Science, 2017, 8 : 1733-1748.

[14] Mengmeng D, Noboru N. Monitoring of wheat growth status and mapping of wheat yield’s within-field spatial variations using color images acquired from UAV-camera system[J]. Remote Sensing, 2017, 9(3): 289-303.

[15] Gong Yan, Duan Bo, Fang Shenghui, et al. Remote estimation of rapeseed yield with unmanned aerial vehicle (UAV) imaging and spectral mixture analysis[J]. Plant Methods, 2018, 14(1): 1-14.

[16] 劉煥軍,康苒,Ustin S,等. 基于時(shí)間序列高光譜遙感影像的田塊尺度作物產(chǎn)量預(yù)測(cè)[J]. 光譜學(xué)與光譜分析,2016,36(8):2585-2589.

Liu Huanjun, Kang Ran,Ustin S, et al.Study on the prediction of cotton yield within field scale with time series hyperspectral imagery[J]. Spectroscopy and Spectral Analysis, 2016, 36(8): 2585-2589. (in Chinese with English abstract)

[17] 吳瓊,齊波,趙團(tuán)結(jié),等. 高光譜遙感估測(cè)大豆冠層生長(zhǎng)和籽粒產(chǎn)量的探討[J]. 作物學(xué)報(bào),2013,39(2):309-318.

Wu Qiong, Qi Bo, Zhao Tuanjie, et al.A tentative study on utilization of canopy hyperspectral reflectance to esti-mate canopy growth and seed yield in soybean[J]. Acta Agronomica Sinica, 2013, 39(2): 309-318. (in Chinese with English abstract)

[18] 張松,馮美臣,楊武德,等. 基于高光譜植被指數(shù)的冬小麥產(chǎn)量監(jiān)測(cè)[J]. 山西農(nóng)業(yè)科學(xué),2018,46(4):572-575.

Zhang Song, Feng Meichen, Yang Wude, et al.Monitoring of winter wheat yield based on hyperspectral vegetation index[J]. Shanxi Agricultural Science, 2018, 46(4): 572-575. (in Chinese with English abstract)

[19] Turner D, Lucieer A, Wallace L. Direct georeferencing of ultrahigh-resolution UAV imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(5): 2738-2745.

[20] 何彩蓮,鄭順林,萬(wàn)年鑫,等. 馬鈴薯光譜及數(shù)字圖像特征參數(shù)對(duì)氮素水平的響應(yīng)及其應(yīng)用[J]. 光譜學(xué)與光譜分析,2016,36(9):2930-2936.

He Cailian, Zheng Shunlin, Wan Nianxin, et al. Potato spectrum and the digital image feature parameters on the response of the nitrogen level and its application[J]. Spectroscopy and Spectral Analysis, 2016, 36(9): 2930-2936. (in Chinese with English abstract)

[21] Zhou X, Zheng H, Xu X Q, et al. Predicting grain yield in rice using multi-temporal vegetation indices from UAV-based multispectral and digital imagery[J]. Isprs Journal of Photogrammetry & Remote Sensing, 2017(130):246-255.

[22] Gitelson A A, Kaufman Y J, Stark R, et al. Novel algorithms for remote estimation of vegetation fraction[J]. Remote Sensing of Environment, 2002, 80(1): 76-87.

[23] Bendig J, Yu K, Aasen H, et al. Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley[J]. International Journal of Applied Earth Observation & Geoinformation, 2015(39): 79-87.

[24] Kataoka T, Kaneko T, Okamoto H, et al. Crop growth estimation system using machine vision[C] IEEE/ASME International Conference on Advanced Intelligent Mechatronics, Aim, 2003: 1079-1083.

[25] Torres-Sánchez J, Pena J M, De Castro A, et al. Multi-temporal mapping of the vegetation fraction in early-season wheat fields using images from UAV[J]. Computers and Electronics in Agriculture, 2014(103): 104-113.

[26] Francesco C,Leonardo D,Donatella G, et al. Estimation of canopy attributes in beech forests using true colour digital images from a small fixed-wing UAV[J]. International Journal of Applied Earth Observation and Geoinformation, 2016(47): 60-68.

[27] Aasen H, Burkart A, Bolten A, et al. Generating 3D hyperspectral information with lightweight UAV snapshot cameras for vegetation monitoring: From camera calibration to quality assurance[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015,108: 245-259.

[28] Chaoyang W, Zheng N, Quan T, et al. Estimating chlorophyll content from hyperspectral vegetation indices: Modeling and validation[J]. Agricultural and Forest Meteorology, 2008, 148(8/9): 1230-1241.

[29] Aparicio N, Villegas D, Araus J L, et al. Relationship between growth traits and spectral vegetation indices in durum wheat[J]. Crop Science Madison, 2002, 42: 1547-1555.

[30] Qi J, Chehbouni A, Huete AR, et al. A modified soil adjusted vegetation index[J]. Remote Sensing of Environment, 1994, 48(2): 119-126.

[31] Baret F, Guyot G , Major D J. TSAVI: A vegetation index which minimizes soil brightness effects on LAI and APAR estimation[C]12th Canadian Symposium on Remote Sensing Geoscience and Remote Sensing Symposium. 1989: 1355-1358.

[32] Gong P, Pu R, Biging G S, et al. Estimation of forest leaf area index using vegetation indices derived from hyperion hyperspectral data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2003, 41(6): 1355-1362.

[33] 馮偉,朱艷,姚霞,等. 利用紅邊特征參數(shù)監(jiān)測(cè)小麥葉片氮素積累狀況[J]. 農(nóng)業(yè)工程學(xué)報(bào),2009,25(11):194-201.

Feng Wei, Zhu Yan, Yao Xia, et al. Monitoring nitrogen accumulation in wheat leaf with red edge characteristics parameters[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2009, 25(11): 194-201. (in Chinese with English abstract)

Comparison of winter wheat yields estimated with UAV digital image and hyperspectral data

Tao Huilin1,2,3,4, Feng Haikuan1,3,4※, Yang Guijun1,3,4, Yang Xiaodong1,3,4, Miao Mengke1,3,4,5, Wu Zhichao1,3,4,5, Zhai Liting1,3,4,5

(1.100097,; 2.,232001,; 3.100097,; 4.100097,; 5.454000,)

Accurate estimation of crops yield is of great significance in agricultural production and has a strong guiding significance for agricultural managers. It is necessary to use an effective technical means to estimate the yield of field crops quickly and accurately. Taking winter wheat in Xiaotangshan National Precision Agricultural Research Demonstration Base as the research object, this study compared the performance of unmanned aerial vehicle (UAV) digital image and hyperspectral data in winter wheat yield estimation. The field surveys and campaigns were conducted in three typical winter wheat growth stages including flagging, flowering and filling stages. The digital images and hyperspectral data were respectively acquired by digital camera and Cubert UHD 185 Firefly imaging spectrometer, which were mounted on a UAV platform. The wheat yield data were collected during harvest. Firstly, the typical digital image indices and hyperspectral parameters were extracted from UAV digital image and hyperspectral data, respectively. Then the correlation analyses between wheat measured yield and digital image indices and hyperspectral parameters were carried out. Nine digital image indices and hyperspectral parameters with high correlation were selected for each growth stages, respectively. The selected digital image indices and hyperspectral parameters were used as modeling factors and the yield were estimated bymultiple linear regression (MLR) and random forest (RF), and the models constructed by the two remote sensing data were compared to optimize the remote sensing data and model. The results showed that the digital image indices and hyperspectral parameters had significant correlation with the wheat measured yield. Among them, the correlation of the best index of different growth stages was the reflectance of the red and the best hyperspectral parameter of the three growth stages were transformed chlorophyll absorption reflectance index optimized soil adjusted vegetation index (TCARI/OSAVI), simple ratio vegetation (SR), and TCARI/OSAVI, respectively. Through the digital image indices, analyzing the effect of the modeling set, the accuracy of the MLR model was significantly better than the RF model in different growth stages,and the estimation accuracy of the two models was the highest during the filling stage and the lowest during the flagging stage.The best2of the MLR model was 0.71 (RMSE = 730.66 kg/hm2, NRMSE = 12.79%), and the best2of the RF model was 0.57 (RMSE = 894.16 kg/hm2, NRMSE = 15.65%), indicating that the advantages of the MLR model were more obvious.MLR and RF model verification effect and modeling effect remain the same. The performance of MLR and RF models had gradually increased to the filling stage to achieve the best. NRMSE reached 13.56% and 17.22%, respectively.The yield effect was estimated based on the spectral index. For MLR and RF models, the accuracy of model modeling was gradually improved, and the fitting effect was getting better and better. Among them,the best2of the MLR model was 0.77 and the NRMSE was 10.32%; the best2of the MLR model was 0.61, NRMSE was 14.79%, the estimation accuracy of MLR model was better than RF model in different growth stages.As the growth stage progresses, the verification2gradually increased, and RMSE and NRMSE gradually decreased. This result was consistent with the effect of the modeling set, indicating that the validation effect was relatively stable. So using UAV hyperspectral remote sensing data, the estimation model of winter wheat yield established by the MLR method can quickly and easily predict the yield of crops, and can effectively monitor the growth of crops and the performance of yield estimation models in different growth stages.

UAV; digital image; hyperspectral; winter wheat; yield; estimation; partial least squares; random forest

陶惠林,馮海寬,楊貴軍,楊小冬,苗夢(mèng)珂,吳智超,翟麗婷. 基于無(wú)人機(jī)數(shù)碼影像和高光譜數(shù)據(jù)的冬小麥產(chǎn)量估算對(duì)比[J]. 農(nóng)業(yè)工程學(xué)報(bào),2019,35(23):111-118.doi:10.11975/j.issn.1002-6819.2019.23.014 http://www.tcsae.org

Tao Huilin, Feng Haikuan, Yang Guijun, Yang Xiaodong, Miao Mengke, Wu Zhichao, Zhai Liting. Comparison of winter wheat yields estimated with UAV digital image and hyperspectral data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(23): 111-118. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2019.23.014 http://www.tcsae.org

2019-07-18

2019-10-24

國(guó)家自然科學(xué)基金(41601346,41871333)

陶惠林,主要從事農(nóng)業(yè)定量遙感研究。Email:15755515505@163.com

馮海寬,助理研究員,主要從事農(nóng)業(yè)定量遙感研究。Email:fenghaikuan123@163.com

10.11975/j.issn.1002-6819.2019.23.014

S252

A

1002-6819(2019)-23-0111-08

猜你喜歡
數(shù)碼影像冬小麥生育期
2022年山西省冬小麥春季田間管理意見
大豆生育期組鑒定分組方法的比較研究
冬小麥田N2O通量研究
吉林水稻關(guān)鍵生育期延遲型/障礙型冷害時(shí)空變化*
冬小麥的秘密
不同生育期大豆品種氮素積累特性研究
不誤農(nóng)時(shí)打好冬小麥春管“第一仗”
基于遙感ET數(shù)據(jù)的遼寧地區(qū)典型農(nóng)作物生育期耗水規(guī)律分析
基于普通數(shù)碼影像的近景攝影測(cè)量技術(shù)研究
數(shù)碼影像證據(jù)在刑事訴訟中的運(yùn)用
布拖县| 原阳县| 广西| 永吉县| 刚察县| 渝中区| 仁化县| 成安县| 万安县| 陆丰市| 庄浪县| 扎鲁特旗| 宁德市| 乐昌市| 贡觉县| 板桥市| 二手房| 土默特左旗| 龙胜| 红河县| 平武县| 凉山| 普格县| 新干县| 武川县| 贵德县| 枣阳市| 方山县| 晋宁县| 齐齐哈尔市| 龙井市| 且末县| 固镇县| 三台县| 宁武县| 寿宁县| 九江县| 三门县| 昆山市| 台东县| 镇雄县|