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

?

Sentinel-2影像和BP神經(jīng)網(wǎng)絡(luò)結(jié)合的小麥條銹病監(jiān)測方法

2019-11-11 06:33黃林生黃文江葉回春趙晉陵馬慧琴
農(nóng)業(yè)工程學(xué)報 2019年17期
關(guān)鍵詞:條銹病植被指數(shù)波段

黃林生,江 靜,,黃文江,,葉回春,趙晉陵,馬慧琴,阮 超,

農(nóng)業(yè)信息與電氣技術(shù)

Sentinel-2影像和BP神經(jīng)網(wǎng)絡(luò)結(jié)合的小麥條銹病監(jiān)測方法

黃林生1,江 靜1,3,黃文江1,2,3※,葉回春2,3,趙晉陵1,馬慧琴3,阮 超1,3

(1. 安徽大學(xué)農(nóng)業(yè)生態(tài)大數(shù)據(jù)分析與應(yīng)用技術(shù)國家地方聯(lián)合工程研究中心,合肥 230601;2. 三亞中科遙感研究所,海南 572029;3. 中國科學(xué)院遙感與數(shù)字地球研究所,數(shù)字地球重點實驗室,北京 100094)

選用包含紅邊等多種不同波段信息的多光譜衛(wèi)星數(shù)據(jù),為區(qū)域尺度上展開作物病害監(jiān)測研究提供更加豐富有效的信息,相比于常規(guī)的寬波段衛(wèi)星遙感影像,搭載紅邊波段的Sentinel-2影像對作物病害脅迫更加敏感,能顯著提高模型精度。該文以陜西省寧強縣小麥條銹病為研究對象,基于Sentinel-2影像共提取了26個初選特征因子:3個可見光波段反射率(紅、綠、藍)、1個近紅外波段反射率、3個紅邊波段反射率、14個對病害敏感的寬波段植被指數(shù)和5個紅邊植被指數(shù)。結(jié)合K-Means和ReliefF算法篩選病害敏感特征,最終篩選出3個寬波段植被指數(shù),包括:增強型植被指數(shù)(enhanced vegetation index,EVI)、結(jié)構(gòu)加強色素指數(shù)(structure intensive pigment index,SIPI)、簡單比值植被指數(shù)(simple ratio index,SR),2個紅邊波段植被指數(shù):歸一化紅邊2植被指數(shù)(normalized red-edge2 index,NREDI2)、歸一化紅邊3植被指數(shù)(normalized red-edge3 index,NREDI3)。利用BP神經(jīng)網(wǎng)絡(luò)方法(back propagation neural network,BPNN),分別以寬波段植被指數(shù)和寬波段植被指數(shù)結(jié)合紅邊波段指數(shù)作為輸入變量構(gòu)建小麥條銹病嚴(yán)重度監(jiān)測模型,對比2種模型的監(jiān)測精度。結(jié)果顯示,基于寬波段植被指數(shù)結(jié)合紅邊波段植被指數(shù)的監(jiān)測模型的總體精度達到83.3%,Kappa系數(shù)0.73,優(yōu)于僅基于寬波段植被指數(shù)特征所建監(jiān)測模型的精度73.3%,Kappa系數(shù)0.58。說明紅邊波段能夠為病害監(jiān)測提供有效信息,采用寬波段植被指數(shù)和紅邊波段植被指數(shù)相結(jié)合的方法能夠有效提高作物病蟲害監(jiān)測模型精度。

遙感;算法;病害;Sentinel-2紅邊;小麥;條銹病;BP神經(jīng)網(wǎng)絡(luò)

0 引 言

小麥條銹?。ǎ┦菤鈧鞑『?,具有發(fā)病廣、流行性強、發(fā)病概率高的特點,是影響小麥減產(chǎn)的主要病害之一。小麥?zhǔn)芎髣t會引起葉片早枯,成穗數(shù)降低,減產(chǎn)嚴(yán)重[1]。傳統(tǒng)的病蟲害監(jiān)測主要依靠地面調(diào)查,雖可信性高,但費時費力且較難滿足在大區(qū)域監(jiān)測的要求。遙感技術(shù)的飛速進步為作物病害監(jiān)測提供了更多的可能性,能更加精確、及時地了解作物病蟲害發(fā)生和發(fā)展的時空變化狀況,這對病害科學(xué)防控具有重大意義[2]。

近些年來,在作物病害監(jiān)測研究中廣泛地引進及應(yīng)用遙感技術(shù),而遙感技術(shù)研究方法與內(nèi)容也在不斷的改進和創(chuàng)新中。其中高光譜影像兼有高的空間和光譜分辨率,在研究中得到廣泛應(yīng)用。高光譜技術(shù)能探測植被光譜曲線在某些特定波段上的細(xì)節(jié)相應(yīng)信息,Zhang等[3]利用連續(xù)小波分析方法區(qū)分小麥病害(白粉病、條銹?。┖拖x害(蚜蟲),將Fisher線性判別分析用于構(gòu)建區(qū)分模型,總體精度較高。Zheng等[4]在冠層尺度上,利用小麥條銹病監(jiān)測最佳的3波段光譜指數(shù),將光化學(xué)反射指數(shù)(photochemical reflectance index,PRI)和花青素反射指數(shù)(anthocyanin reflectance index,ARI)分別用于不同發(fā)病階段的小麥條銹病監(jiān)測與識別中,并證明了其準(zhǔn)確性。Huang等[5]應(yīng)用航空高光譜的圖像,采用回歸分析建立了小麥條銹病嚴(yán)重度反演模型,并將病害監(jiān)測模型從冠層尺度擴展到了地塊尺度。盡管基于地面/航空等高光譜數(shù)據(jù)的作物病害監(jiān)測研究進展有效支撐了病害遙感應(yīng)用,但受其尺度小、利用率低且高成本等因素的限制,很難滿足大尺度作物病害監(jiān)測。多光譜遙感數(shù)據(jù)在可接受的空間分辨率下具有衛(wèi)星數(shù)量多、影像全、成本低等優(yōu)勢,適合于大區(qū)域作物病害監(jiān)測。近年來Landsat-8、GF-1、HJ-CCD、Worldview-2等遙感影像被成功應(yīng)用于作物病蟲害的監(jiān)測預(yù)測研究。如馬慧琴等[6]利用Landsat-8遙感影像與氣象數(shù)據(jù)結(jié)合實現(xiàn)了小麥白粉病的區(qū)域尺度預(yù)測的較高精度。Yuan等[7]基于Worldview-2衛(wèi)星影像數(shù)據(jù),通過Fisher線性判別分析構(gòu)建了小麥白粉病和蚜蟲的監(jiān)測模型。黃林生等[8]利用GF-1影像嘗試結(jié)合Relief-mRMR-GASVM模型有效提高了區(qū)域尺度上小麥白粉病的監(jiān)測精度。以上研究證明了多光譜衛(wèi)星數(shù)據(jù)在作物病害監(jiān)測研究中的潛力。與上述衛(wèi)星傳感器相比,Sentinel-2在保證相對較高的空間分辨率和高幅寬的同時還提供了豐富的紅邊信息,是唯一一個在紅邊范圍含有3個波段的衛(wèi)星傳感器[9],為作物長勢和脅迫區(qū)分提供有效數(shù)據(jù)源,為病害健康狀況的監(jiān)測提供了更豐富的信息。如Chemura等[10]重采樣Sentinel-2影像估計咖啡葉片上銹病發(fā)病的嚴(yán)重程度。Zheng等[11]嘗試通過高光譜數(shù)據(jù)模擬 Sentine-2傳感器的紅邊波段,并通過利用紅邊波段構(gòu)建的新植被指數(shù)實現(xiàn)了小麥條銹病的監(jiān)測。以上研究證明了Sentinel-2衛(wèi)星的紅邊波段在病蟲害監(jiān)測研究中的潛力。因此應(yīng)充分挖掘紅邊波段信息,為區(qū)域尺度上作物病害的監(jiān)測提供更多可操作的可能。

BP神經(jīng)網(wǎng)絡(luò)(back propagation neural network,BPNN)具有較強的非線性函數(shù)逼近能力,是神經(jīng)網(wǎng)絡(luò)應(yīng)用最廣泛的部分[12]。因此BPNN算法在趨勢預(yù)測、故障診斷、樣本分類等研究中均取得了較高的精準(zhǔn)度及應(yīng)用價值[13-15],同時近期在病害監(jiān)測及農(nóng)業(yè)發(fā)展的研究中也得到了普遍應(yīng)用。采用BPNN方法預(yù)測柑橘葉片含氮量[16]和臍橙果實可溶性固形物含量[17]取得了較好的效果,并在楓楊葉綠素含量光譜反演中得到較高的精度[18]。沈文穎等[19]采用BPNN構(gòu)建了小麥葉片白粉病反演模型,反演模型對小麥白粉病整個浸染期均具有較高的應(yīng)用性。Ma等[20]利用雙時相Landsat-8影像,開發(fā)一種SMOTE-BPNN平衡新訓(xùn)練數(shù)據(jù)集的方法,可生成區(qū)域小麥病蟲害分布圖,區(qū)分小麥白粉病和蚜蟲?;谝陨涎芯勘砻鰾PNN模型在病害的反演上有較高的應(yīng)用價值,將此方法應(yīng)用在區(qū)域尺度上的小麥條銹病嚴(yán)重度監(jiān)測上能取得較好的監(jiān)測精度。

根據(jù)以上分析,本文以陜西省漢中市寧強縣為研究區(qū),利用Sentinel-2影像反演得到與病害相關(guān)的寬波段及紅邊波段植被指數(shù)特征,通過K-Means結(jié)合ReliefF的方法進行病害敏感寬波段植被指數(shù)特征和紅邊波段指數(shù)特征的篩選,并分別以寬波段植被指數(shù)特征和寬波段植被指數(shù)結(jié)合紅邊波段指數(shù)作為輸入變量,采用BPNN算法建立小麥條銹病嚴(yán)重度監(jiān)測模型,并對比分析2種模型的優(yōu)越性。

1 材料與方法

1.1 研究區(qū)概況

研究區(qū)位于陜西省漢中市寧強縣(118°35′ 9.51′′ E~37°35′ 51.75′′ N)(圖1)。該區(qū)域地處秦嶺和巴山兩大山系的交匯地帶。寧強雨量充沛,空氣濕潤,氣候單一且環(huán)境適宜條銹病傳播[21],是小麥條銹病發(fā)生的較典型區(qū)域[22],因此適合利用遙感衛(wèi)星影像開展小麥條銹病嚴(yán)重度監(jiān)測。

圖1 研究區(qū)概況

1.2 數(shù)據(jù)獲取

研究數(shù)據(jù)主要分為兩部分,遙感影像和小麥條銹病野外調(diào)查數(shù)據(jù)。遙感數(shù)據(jù)為Sentinel-2衛(wèi)星遙感影像(表 1)。根據(jù)研究區(qū)天氣狀況,選擇與地面調(diào)查時間相近且影像質(zhì)量較高的衛(wèi)星影像,即2018年5月12日的Sentinel-2影像數(shù)據(jù)。小麥條銹病實地調(diào)查數(shù)據(jù)于2018年5月12日在寧強縣實地調(diào)查獲得。該區(qū)域為寧強縣小麥種植集中的地塊,較適合衛(wèi)星影像處理,且此區(qū)域小麥條銹病發(fā)病程度均勻,比其他區(qū)域更典型。為匹配Sentinel-2影像分辨率,選取20 m×20 m的地塊開展調(diào)查,使用全球定位系統(tǒng)(global position system,GPS)記錄樣本中心點經(jīng)緯度,在每塊樣地采用五點調(diào)查法進行調(diào)查[6],每點調(diào)查面積為1/m2,即在每個調(diào)查樣方中取5個對稱的點,每點隨機取20株小麥。根據(jù)國家農(nóng)作物病害調(diào)查和預(yù)測規(guī)則(GB/T 15795-2011),嚴(yán)重度用分級法表示,共設(shè)為8級,分別用1%、5%、10%、20%、40%、60%、80%、100% 表示,對處于等級之間的病情則取其接近值,嚴(yán)重度低于1%,按1% 記。采用公式(1)計算病情指數(shù)DI[23]。

總共獲得30個野外調(diào)查點,為避免過多嚴(yán)重度等級從而增加監(jiān)測的難度,將難區(qū)分的等級合并為一級。即將發(fā)病嚴(yán)重程度分為健康(Ⅰ,DI≤5%),輕發(fā)(Ⅱ,5%<DI≤20%),重發(fā)(Ⅲ, DI>20%),總共3個等級構(gòu)建監(jiān)測模型。

表1 Sentinel-2衛(wèi)星基本參數(shù)

1.3 數(shù)據(jù)處理

首先,對Sentinel-2遙感影像預(yù)處理,包括輻射定標(biāo)、大氣校正等。主要是基于SNAP(sentinel application platform software)應(yīng)用平臺軟件進行。依據(jù)研究區(qū)的作物種植類型及物候歷[24],利用分類決策樹[25]的方法提取小麥種植區(qū)域,經(jīng)地面調(diào)查點對該分類進行驗證,小麥面積提取的總體精度達90% 以上,滿足后續(xù)分析的精度要求。接著,基于預(yù)處理后的遙感影像提取波段反射率及對小麥條銹病比較相關(guān)的寬波段及紅邊波段植被指數(shù)。本文提取了Sentinel-2的4個波段反射率特征(紅、綠、藍、近紅外)和3個紅邊波段反射率(B5、B6、B7),14個寬波段植被指數(shù)以及5個紅邊植被指數(shù)共26個特征因子作為條銹病監(jiān)測模型的初選特征因子。表2列舉出了各植被指數(shù)的具體名稱及計算公式。

1.4 建模特征選擇

模型構(gòu)建時篩選出對病害發(fā)生較敏感的特征變量,可提高小麥條銹病嚴(yán)重度分類精度。合適的特征選擇方法可以有效去除不相關(guān)變量和冗余變量,提升模型的性能。K-Means算法是一種常見的聚類算法[43],可通過聚類分析提高特征之間聚類精度,但該算法對初始中心的選取要求較高,若初始中心選擇不合適時會影響聚類過程及效果。ReliefF算法是一種特征權(quán)重算法(feature weighting algorithms),根據(jù)植被指數(shù)特征及病害嚴(yán)重度的相關(guān)性給予特征相應(yīng)的權(quán)重,對病害相關(guān)性強的特征賦予更高的權(quán)重[8]。因此,為了避免K-Means算法進行特征篩選時初始中心選取不恰當(dāng)對結(jié)果的影響,本文采用K-Means結(jié)合ReliefF算法的方式進行最優(yōu)特征的選取。首先通過ReliefF算法計算出各個特征對小麥條銹病嚴(yán)重度發(fā)生的權(quán)重。由于該算法在運行過程中隨機選擇樣本,因此會導(dǎo)致結(jié)果權(quán)重的誤差,因此本文采取多次平均的方法,將主程序運行20次后的平均結(jié)果作為各個特征的最終權(quán)重。之后,以權(quán)重值最大的特征作為K-Means算法的初始中心,并按照特征權(quán)重從高到底的順序依次進行聚類,若該特征對聚類精度的貢獻為正,則保留該特征,否則去除該特征,最后將聚類精度最高的特征組合作為最終的模型輸入變量。

表2 選取的植被指數(shù)計算公式

注:NIR為近紅外波段;R為紅波段;R為綠波段; R為藍波段;Re1為紅邊波段1;Re2為紅邊波段2;Re3為紅邊波段3。

Note:NIRis near-infrared band;Ris red band;Ris green band;Ris blue band;Re1is red edge band 1;Re2is red edge band 2;Re3is red edge band 3.

1.5 監(jiān)測模型的構(gòu)建方法

通過BPNN來構(gòu)建小麥條銹病的遙感監(jiān)測模型。BPNN是一種信號前向傳遞、誤差反向回饋的有監(jiān)督的神經(jīng)網(wǎng)絡(luò),具有自學(xué)學(xué)習(xí)能力的優(yōu)勢[44]。小麥條銹病的發(fā)病嚴(yán)重度與特征因子的關(guān)系是一個非線性問題,而BPNN具有處理復(fù)雜非線性函數(shù)的能力。

本研究中用于小麥條銹病監(jiān)測模型的BPNN結(jié)構(gòu)如圖2所示,BPNN網(wǎng)絡(luò)由3層組成,分別為輸入層、輸出層和隱藏層。BPNN網(wǎng)絡(luò)訓(xùn)練是一個不斷訓(xùn)練數(shù)據(jù),調(diào)整權(quán)重和閾值使網(wǎng)絡(luò)誤差減少到最小值或停在預(yù)設(shè)值的過程。圖中輸入層x為輸入的特征因子,為輸入變量的個數(shù),即為通過ReliefF結(jié)合K-Means篩選的2組特征集。隱藏層中的a( j),=1,2,3,即3個隱藏層,為每層的神經(jīng)元個數(shù),隱藏層神經(jīng)元的個數(shù)由經(jīng)驗公式確定[45],3層的神經(jīng)元個數(shù)分別為{10,10,3},激活函數(shù)分別為{‘logsig’,‘logsig’,‘logsig’}。學(xué)習(xí)規(guī)則采用traingdx(梯度下降自適應(yīng)學(xué)習(xí)率訓(xùn)練函數(shù)),該方法能夠自適應(yīng)調(diào)整學(xué)習(xí)率,極大加速收斂速度,增加穩(wěn)定性,提高速度與精度[46]。設(shè)置最大迭代次數(shù)為5000,訓(xùn)練的目標(biāo)誤差為0.000001。輸出層中的神經(jīng)元代表監(jiān)測的小麥條銹病嚴(yán)重等級(健康、輕發(fā)、重發(fā))。

注:x1~ xi代表輸入的特征變量;an( j)中,j代表隱藏層的層數(shù),n代表每層神經(jīng)元的個數(shù);y代表監(jiān)測的小麥條銹病嚴(yán)重等級。

2 結(jié)果與分析

2.1 特征變量的選取

圖3給出了通過ReliefF計算20次平均后各個特征的權(quán)重分布的降序排列結(jié)果。從圖3中可以看出,SIPI為權(quán)重值最高的特征,即與病害最相關(guān)的特征,因此以SIPI為K-Means聚類的起始中心。為了減少運算量,本文依據(jù)各個特征權(quán)重值降序的變化情況,只對權(quán)重值排在前10的特征即與病害最相關(guān)的10個特征依次進行K-Means聚類分析,篩選出聚類精度最高的特征組合。表3為各個特征的組合聚類精度。最終篩選出3個寬波段植被指數(shù)EVI、SIPI、SR及2個紅邊波段植被指數(shù)NREDI2和NREDI3用于小麥條銹病嚴(yán)重度監(jiān)測模型的構(gòu)建。

圖3 基于ReliefF算法的不同特征權(quán)重平均值

表3 基于K-Means算法的各個特征組合聚類精度

2.2 模型的評估與驗證

研究采用2018年5月12日的條銹病的地面調(diào)查數(shù)據(jù)對模型監(jiān)測結(jié)果進行評價。因研究所用實地野外調(diào)查點數(shù)量較少,則采用留一交叉法進行監(jiān)測結(jié)果的精度驗證。分別以寬波段植被指數(shù)特征集及寬波段和紅邊植被指數(shù)特征結(jié)合的特征集作為輸入變量,通過BPNN方法構(gòu)建2種監(jiān)測小麥條銹病嚴(yán)重程度的模型。各監(jiān)測方法所得監(jiān)測結(jié)果的漏分誤差、錯分誤差、總體精度、Kappa系數(shù)見表4。

分析2種模型的監(jiān)測情況發(fā)現(xiàn),寬波段植被指數(shù)加紅邊植被指數(shù)特征的監(jiān)測模型精度比僅有寬波段植被指數(shù)特征的模型總體精度提高10個百分點,達到83.3%,Kappa系數(shù)為0.73。從模型的漏分和錯分情況來看,2種模型均表現(xiàn)為把輕發(fā)地塊分到健康或者重發(fā)地塊的情況較為嚴(yán)重,但總體比較,無論是健康地塊還是病害浸染地塊,加紅邊波段指數(shù)特征的監(jiān)測模型的錯分誤差和漏分誤差都低于僅基于寬波段植被指數(shù)特征構(gòu)建的監(jiān)測模型。從病理的角度分析,小麥?zhǔn)軛l銹病菌浸染后破壞了葉片的結(jié)構(gòu),導(dǎo)致紅邊波段產(chǎn)生較大響應(yīng)。綜合以上結(jié)果可表明與僅有傳統(tǒng)寬波段指數(shù)特征構(gòu)建的模型相比,寬波段植被指數(shù)特征結(jié)合紅邊指數(shù)特征的方法能夠為病害監(jiān)測提供更豐富的信息。因此加入紅邊波段特征后能更加全面的反映小麥的長勢及發(fā)病情況,有效改善監(jiān)測模型精度。

2.3 小麥條銹病嚴(yán)重度監(jiān)測

利用研究區(qū)2018年5月12日遙感影像數(shù)據(jù),以單個像元為基本處理單元,采用K-Means算法與ReliefF算法相結(jié)合的方式篩選出3個寬波段植被指數(shù)EVI、SIPI、SR和2個紅邊波段指數(shù)NREDI2、NREDI3,分別以寬波段植被指數(shù)為特征集(EVI、SIPI、SR)、寬波段植被指數(shù)結(jié)合紅邊植被指數(shù)為特征集(EVI、SIPI、SR、NREDI2、NREDI3)作為BPNN方法的2組輸入變量構(gòu)建模型,并對研究區(qū)小麥病害進行監(jiān)測,得到小麥條銹病嚴(yán)重程度的空間分布情況(如圖4所示)。從監(jiān)測結(jié)果分布圖中可以看出,2種模型監(jiān)測結(jié)果中條銹病空間分布整體趨勢是一致的,即東南地區(qū)發(fā)病較嚴(yán)重,病害浸染面積較多,且呈現(xiàn)整片區(qū)域連續(xù)分布,健康與發(fā)病區(qū)域分布較為均勻。但2種模型監(jiān)測結(jié)果在細(xì)節(jié)和發(fā)病程度上差異較大。圖4a重發(fā)麥區(qū)明顯低于圖4b,而輕發(fā)麥區(qū)所占比例較大。圖4a在采樣點的區(qū)域分布中,發(fā)病麥區(qū)面積分布較為零散且比健康麥區(qū)面積少,而圖4b東部地區(qū)發(fā)病較為嚴(yán)重,北部較輕。健康、輕發(fā)、重發(fā)麥區(qū)面積比例分布合理。為更明確地顯示2種模型監(jiān)測結(jié)果之間的差異,表5列出了實地調(diào)查,寬波段植被指數(shù)模型、寬波段植被指數(shù)結(jié)合紅邊波段指數(shù)模型的健康、輕發(fā)、重發(fā)3種不同浸染狀況下的小麥面積百分比。從表中數(shù)據(jù)來看,圖4a和圖4b條銹病發(fā)生面積相差不大,分別為47.9%和49.1%,但圖 4a中輕發(fā)麥區(qū)所占比例為35.3%,遠(yuǎn)大于圖4b的20.5%。且圖4a的重發(fā)麥區(qū)僅占12.6%,遠(yuǎn)小于圖4b的28.6%。圖4b與實地調(diào)查的病害浸染比例更相近。結(jié)合圖4模型結(jié)果中病害的空間分布情況和表5病害浸染統(tǒng)計情況來看,圖寬波段+紅邊波段植被指數(shù)模型整體更符合實際情況,對小麥嚴(yán)重度的區(qū)分能力要優(yōu)于寬波段植被指數(shù)模型,更能合理的反映小麥條銹病真實發(fā)病情況。

表4 BPNN監(jiān)測模型的總體驗證結(jié)果

圖4 BPNN模型監(jiān)測小麥條銹病嚴(yán)重度空間分布圖

表5 各模型病害浸染比例統(tǒng)計

3 結(jié) 論

本文基于Sentinel-2遙感數(shù)據(jù)建立了小麥條銹病的嚴(yán)重度監(jiān)測模型,通過K-Means 結(jié)合ReliefF算法的方式篩選出3個寬波段指數(shù)特征EVI、SIPI、SR和2個紅邊波段指數(shù)特征NREDI2、NREDI3作為模型的輸入變量,采用BPNN方法構(gòu)建條銹病的2種監(jiān)測模型,對陜西寧強縣的小麥條銹病發(fā)生嚴(yán)重度進行監(jiān)測,且對2種數(shù)據(jù)所構(gòu)建模型的結(jié)果進行了比較分析。結(jié)果表明:采用寬波段植被指數(shù)結(jié)合紅邊波段植被指數(shù)特征作為輸入變量的BPNN模型的監(jiān)測效果優(yōu)于僅以寬波段指數(shù)特征作為輸入變量的模型,其總體精度達到83.3%。與常規(guī)的寬波段植被指數(shù)特征模型相比,寬波段植被指數(shù)特征與紅邊波段植被指數(shù)特征結(jié)合更能全面反映小麥的長勢及病害光譜信息的變化,使模型在輸入?yún)?shù)中融合了更多的有效信息,對小麥條銹病更敏感,有效提高了小麥條銹病嚴(yán)重度監(jiān)測模型的精度,進一步加深了實際監(jiān)測和病害防治中的可靠性。

實地調(diào)查數(shù)據(jù)的質(zhì)量對模型的精度有較大影響,本研究在野外調(diào)查開展中因各種不可控因素的影響,采樣數(shù)量較少,僅獲取了陜西寧強縣的小范圍地面調(diào)查數(shù)據(jù),因此模型的通用性有待提高和驗證。另外,本文所選特征全部為遙感數(shù)據(jù),未選擇其他可能影響小麥條銹病嚴(yán)重度的數(shù)據(jù),因此所構(gòu)建的模型精度必會存在一定誤差。在今后的研究中盡可能多融合各類數(shù)據(jù),構(gòu)建一個融入多源數(shù)據(jù)的小麥條銹病嚴(yán)重度監(jiān)測模型,從而提高小麥條銹病嚴(yán)重度監(jiān)測的精度。

[1] 黃文江. 作物病害遙感監(jiān)測機理與應(yīng)用[M]. 北京:中國農(nóng)業(yè)科學(xué)技術(shù)出版社,2009.

[2] 王利民,劉佳,楊福剛,等. 基于GF-1/WFV數(shù)據(jù)的冬小麥條銹病遙感監(jiān)測[J]. 農(nóng)業(yè)工程學(xué)報,2017,33(5):153-160. Wang Limin, Liu Jia, Yang Fugang, et al. Remote sensing monitoring of winter wheat stripe rust based on GF-1/WFV data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2017, 33(20): 153-160. (in Chinese with English abstract)

[3] Zhang J, Wang N, Yuan L, et al. Discrimination of winter wheat disease and insect stresses using continuous wavelet features extracted from foliar spectral measurements[J]. Biosystems Engineering, 2017, 162: 20-29.

[4] Zheng Qiong, Huang Wenjiang, Cui Ximin, et al. Identification of wheat yellow rust using optimal three-band spectral indices in different growth stages[J]. Sensors, 2019, 19(1):35.

[5] Huang W, Lamb D W, Niu Z, et al. Identification of yellow rust in wheat using in-situ spectral reflectance measurements and airborne hyperspectral imaging[J]. Precision Agriculture, 2007, 45 (8): 187-197.

[6] 馬慧琴,黃文江,景元書,等. 遙感與氣象數(shù)據(jù)結(jié)合預(yù)測小麥灌漿期白粉病[J]. 農(nóng)業(yè)工程學(xué)報,2016,32(9):165-172. Ma Huiqin, Huang Wenjiang, Jing Yuanshu, et al. Wheat powdery mildew forecasting in filling stage based on remote sensing and meteorological data [J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2016, 32(9): 165-172. (in Chinese with English abstract)

[7] Yuan L, Bao Z, Zhang H, et al. Habitat monitoring to evaluate crop disease and pest distributions based on multi-source satellite remote sensing imagery[J]. Optik-International Journal for Light and Electron Optics, 2017, 145: 66-73.

[8] 黃林生,阮超,黃文江,等. 基于GF-1遙感影像和relief-mRMR-GASVM模型的小麥白粉病監(jiān)測[J]. 農(nóng)業(yè)工程學(xué)報,2018,34(15):167-175. Huang Linsheng, Ruan Chao, Huang Wenjiang, et al. Monitoring of wheat powdery mildew based on GF-1 remote sensing image and relief-mRMR-GASVM model[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(15): 167-175. (in Chinese with English abstract)

[9] Hedley J, Roelfsema C, Koetz B, et al. Capability of the Sentinel 2 mission for tropical coral reef mapping and coral bleaching detection[J]. Remote Sensing of Environment, 2012, 120(6): 145-155.

[10] Chemura A, Mutanga O, Dube T. Separability of coffee leaf rust infection levels with machine learning methods at Sentinel-2 MSI spectral resolutions[J]. Precision Agriculture, 2016(10): 1-23.

[11] Zheng Qiong, Huang Wenjiang, Cui Ximin, et al. New spectral index for detecting wheat yellow rust using Sentinel-2 multispectral imagery[J]. Sensors, 2018, 18(3): 868-887.

[12] Wang L, Zeng Y, Chen T, et al. Back propagation neural network with adaptive differential evolution algorithm for time series forecasting[J]. Expert Syst Appl, 2015, 42: 855-863.

[13] 沈澤君,楊文元. 粒計算思維下的BP神經(jīng)網(wǎng)絡(luò)在金融趨勢預(yù)測中的應(yīng)用[J]. 小型微型計算機系統(tǒng), 2019,40(3):527-532. Shen Zejun, Yang Wenyuan. Application of BP neural network under grain computing thinking in financial trend prediction[J]. Small Computer Systems, 2019, 40(3): 527-532. (in Chinese with English abstract)

[14] 俞瑋捷,劉光宇. 基于BP神經(jīng)網(wǎng)絡(luò)的光伏系統(tǒng)故障診斷方法[J]. 杭州電子科技大學(xué)學(xué)報:自然科學(xué)版,2018,38(4):56-61,93. Yu Yujie, Liu Guangyu. Fault diagnosis method of photovoltaic system based on BP neural network[J]. Journal of Hangzhou dianzi University, 2018, 38(4): 56-61, 93. (in Chinese with English abstract)

[15] 高斌,趙鵬飛,盧昱欣,等. 基于BP神經(jīng)網(wǎng)絡(luò)的血液熒光光譜識別分類研究[J]. 光譜學(xué)與光譜分析,2018,38(10):154-161. Gao Bin, Zhao Pengfei, Lu Yuxin, et al. Classification of blood fluorescence spectral recognition based on BP neural network[J]. Spectroscopy and Spectral Analysis, 2018, 38(10): 154-161. (in Chinese with English abstract)

[16] 李金夢,葉旭君,王巧男,等. 高光譜成像技術(shù)的柑橘植株葉片含氮量預(yù)測模型[J]. 光譜學(xué)與光譜分析,2014,34(1):212-216. Li Jinmeng, Ye Xujun, Wang Qiaonan, et al. Development of prediction models for determining N content in citrus leaves based on hyperspectral imaging technology[J]. Spectroscopy and Spectral Analysis, 2014, 34(1): 212-216. (in Chinese with English abstract)

[17] Liu Yande, Sun Xudong, Ouyang Aiguo. Nondestructive measurement of soluble solid content of navel orange fruit by visible-NIR spectrometric technique with PLSR and PCA-BPNN[J]. LWT- Food Science and Technology, 2010, 43(4): 602-607.

[18] 李文敏,魏虹,李昌曉,等. 基于高光譜參數(shù)的楓楊葉綠素含量估算模型優(yōu)化[J]. 林業(yè)科學(xué),2014,50(4):55-59. Li Wenmin, Wei Hong, Li Changxiao, et al. Optimization of a model for estimating pterocarya stenoptera chlorophyll concentration with hyperspectral parameters[J]. Scientia Silvae Sinicae, 2014, 50(4): 55-59. (in Chinese with English abstract)

[19] 沈文穎,李映雪,馮偉,等. 基于因子分析-BP神經(jīng)網(wǎng)絡(luò)的小麥葉片白粉病反演模型[J]. 農(nóng)業(yè)工程學(xué)報,2015,31(22):183-190. Shen Wenying, Li Yingxue, Feng Wei, et al. Inversion model of wheat leaf powdery mildew based on factor analysis-BP neural network[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(22): 183-190. (in Chinese with English abstract)

[20] Ma Huiqin, Huang Wenjiang, Jing Yuanshu, et al. Integrating growth and environmental parameters to discriminate powdery mildew and aphid of winter wheat using bi-temporal Landsat-8 imagery[J]. Remote Sensing, 2019, 11(7): 846.

[21] 彭昌家,馮禮斌,白體坤,等. 小麥條銹病發(fā)生流行趨勢及其成因探討[J]. 農(nóng)學(xué)學(xué)報,2015,5(5):39-47. Peng Changjia, Feng Libin, Bai Tikun, et al. The prevalence trend of wheat stripe rust and its genesis[J]. Acta Agriculturala Sinica, 2015, 5(5): 39-47. (in Chinese with English abstract)

[22] 謝飛舟. 陜西省小麥條銹病流行規(guī)律進一步研究[D].楊凌: 西北農(nóng)林科技大學(xué),2011. Xie Feizhou. Further Research on the Epidemic Law of Wheat Stripe Rust in Shaanxi Province[D]. Yangling: Northwest A&F University, 2011. (in Chinese with English abstract)

[23] 小麥條銹病測報調(diào)查規(guī)范:GB/T 15795-2011[S].

[24] 許青云,楊貴軍,龍慧靈,等. 基于MODIS NDVI多年時序數(shù)據(jù)的農(nóng)作物種植識別[J]. 農(nóng)業(yè)工程學(xué)報,2014,30(11):134-144. Xu Qingyun, Yang Guijun, Long Huiling, et al. Crop planting identification based on MODIS NDVI time- series data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(11): 134-144. (in Chinese with English abstract)

[25] Zhang J, Pu R, Yuan L, Wang J, et al. Monitoring powdery mildew of winter wheat by using moderate resolution multi-temporal satellite imagery[J]. PLoS ONE 2014, 9: e93107.

[26] Baret F, Guyot G. Potentials and limits of vegetation indices for LAI and APAR assessment[J]. Remote Sensing of Environment, 1991, 35(2/3): 161-173.

[27] Gamon J A, Surfus J S. Assessing leaf pigment content and activity with a reflectometer[J]. New Phytologist, 2010, 143(1): 105-117.

[28] Gitelson A A. Novel Algorithms for Remote Estimation of Vegetation Fraction[J]. Remote Sens Environ, 2002, 80(1): 76-87.

[29] Huete A, Didan K, Miura, et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices[J]. Remote Sensing of Environment, 2002, 83(1): 195-213.

[30] Person R S, Kudina L P. Discharge frequency and discharge pattern of human motor units during voluntary contraction of muscle[J]. Electroencephalogr Clin Neurophysiol, 1972, 32(5): 471-483.

[31] Merzlyak M N, Chivkunova O B, Gitelson A A, et al. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening[J]. Physiologia Plantarum, 2010, 106(1): 135-141.

[32] Zhao Chunjiang, Huang Wenjiang, Liu Liangyun, et al. Analysis of winter wheat stripe rust characteristic spectrum and establishing of inversion models[C]// IEEE International Geoscience & Remote Sensing Symposium. IEEE, 2004: 4318-4320.

[33] Verstraete M M, Pinty B, Myneni R B. Potential and limitations of information extraction on the terrestrial biosphere from satellite remote sensing[J]. Remote Sensing of Environment, 1996, 58(2): 201-214.

[34] Roujean J L, Fran?ois Marie Breon. Estimating PAR absorbed by vegetation from bidirectional reflectance measurements[J]. Remote Sensing of Environment, 1995, 51(3): 375-384.

[35] Yang C M, Cheng C H, Chen R K. Changes in Spectral characteristics of rice canopy infested with brown planthopper and leaffolder[J]. Crop Science, 2007, 47(1): 329-335.

[36] Geneviève Rondeaux, Steven M, Frédéric Baret. Optimization of soil-adjusted vegetation indices[J]. Remote Sensing of Environment, 1996, 55(2): 95-107.

[37] Huete A R. A soil-adjusted vegetation index (SAVI)[J]. Remote Sensing of Environment, 1988, 25(3): 295-309.

[38] Chen S F, Goodman J. An empirical study of smoothing techniques for language modeling[C]//Proceeding of the 34th annual meeting on Association for Computational Linguistics. Association for Computational Linguistics, 1996: 310-318.

[39] Kononenko I. Estimating attributes: Analysis and extensio reliefF[C]//MachingLearning: ECMIL-94. Springer Berlin Heideberg,1994: 71-182.

[40] Schell J A. Monitoring vegetation systems in the great plains with ERTS[J]. Nasa Special Publication, 1973, 351: 309-317.

[41] Gitelson A, Merzlyak M N. Quantitative estimation of chlorophyll-a using relectance spectra: Experiments with autumn chestnut and maple leaves[J]. Photochem Photobiol B Biol, 1994, 22: 247-252.

[42] Alfonso Fernández-Manso, Oscar Fernández-Manso, Quintano C. Sentinel-2A red-edge spectral indices suitability for discriminating burn severity[J]. International Journal of Applied Earth Observation & Geoinformation, 2016, 50: 170-175.

[43] 陳平華,黃輝,麥淼,等. 結(jié)合ReliefF和互信息的多標(biāo)簽特征選擇算法[J]. 廣東工業(yè)大學(xué)學(xué)報,2018(5):20-25,50. Chen Pinghua, Huang Hui, Mai Wei, et al. Multi-label feature selection algorithm combining ReliefF and mutual information[J]. Journal of Guangdong University of Technology, 2018(5): 20-25, 50. (in Chinese with English abstract)

[44] Wang L, Zeng Y, Chen T. Back propagation neural network with adaptive differential evolution algorithmfor time series forecasting[J]. Expert Syst Appl, 2015, 42, 855-863.

[45] 沈花玉,王兆霞,高成耀,等. BP神經(jīng)網(wǎng)絡(luò)隱含層單元數(shù)的確定[J]. 天津理工大學(xué)學(xué)報,2008,24(5):13-15. Shen Huauu, Wang Zhaoxia, Gao Chengyao, et al. Determining the number of BP neural network hidden layer units[J]. Journal of Tianjin University of Technology, 24(5): 13-15. (in Chinese with English abstract)

[46] Shi Yue, Zhao Xin, Zhang Bing, et al. Back propagation neural network (BPNN) prediction model and control strategies of methanogen phase reactor treating traditional Chinese medicine wastewater (TCMW)[J]. Biotechnol, 2009,144: 70-74.

Wheat yellow rust monitoring method based on Sentinel-2 image and BPNN model

Huang Linsheng1, Jiang Jing1,3, Huang Wenjiang1,2,3※, Ye Huichun2,3, Zhao Jinling1, Ma Huiqin3, Ruan Chao1,3

(1.,230601,; 2.,572029,; 3.,100094,)

Wheat yellow rust is a deadly disease of winter wheat and its timely and accurate detection at regional scale is critical to ameliorate infectious yield loss and safeguard wheat production. With the development in remote sensing technology, satellite imagery with high spatial resolution and high revisiting frequency has become increasingly accessible. Remote sensing data has a unique advantage over traditional method in detecting crop diseases and controlling their spreading, including simple operation, real-time detection, high spatiotemporal resolution and targeting specific-disease, especially the multispectral satellite imagery which covers a wide range of wave bands and provides rich information related to crop diseasesat regional scale. Compared to conventional broad band satellite imagery, the Sentinel-2 is a sensor with three wave bands within the edge of the red light which are sensitive to crop diseases. In this study, a Sentinel-2 image acquired in May 12, 2018 was used to extract 26 characteristic variables related to wheat yellow rust, including 3 visible bands (blue, green and red) reflectance variables, one near infrared band, 3 red-edge bands, 14 broad-bands and 5 red-edge vegetation indices. An approach combining K-means and ReliefF algorithm is proposed to filter these features. We first used the RelieF algorithm to calculate the weight of each feature and filter out 10 features most relevant to the disease. The feature with maximum weight was then taken as the initial center of the K-Means algorithm, and other features were added to form a cluster in descending order of their weight. The combination of features with the highest clustering accuracy was taken as the final input variable to the model. The optimal features, including enhanced vegetation index (EVI), structure intensive pigment index (SIPI), simple ratio index (SR), normalized red-edge2 index(NREDI2), normalized red-edge3 index (NREDI3), three wide-band vegetation indices and 2 red edge band vegetation indices were fed into the yellow rust severity monitoring model as input. The back propagation neural network (BPNN) method was a widely used nonlinear artificial neural network and can learn, implicitly, the relationships between inputs and outputs via a multi-layered network. Network training is a process of continual readjustment of weights and thresholds by reducing the network error to a pre-sent minimum or pre-set training steps. We used BPNN to calculate severity of wheat yellow rust (healthy, slight, sever) inNingqiang County, Shaanxi province, by using the broad-band vegetation indices and the red-edge band vegetation indices as inputs. The results showed that the BPNN model considering broad-band and red-edge vegetation indices as inputs worked better than model using only a single broad-band vegetation indices, improving accuracy by more than 10 percent point and commission accuracy and kappa coefficient reached by 83.3% and 0.73, respectively. The BPNN model includes more information in its input parameters, thereby improving the accuracy of detecting crop diseases. It is more suitable for detecting wheat yellow rust at regional scales and has a wide implication in monitoring and controlling crop diseases at regional scale.

remote sensing; algorithms; diseases; Sentinel-2 red-edge; wheat; yellow rust; BPNN

2019-04-17

2019-08-27

安徽省科技重大專項(16030701091);國家高層次人才特殊支持計劃(萬人計劃,黃文江);海南省萬人計劃配套項目(黃文江);安徽省高等學(xué)校自然科學(xué)研究重點項目(KJ2019A0030)。

黃林生,博士,副教授,研究方向為農(nóng)業(yè)遙感技術(shù)與應(yīng)用。Email:linsheng0808@163.com。

黃文江,博士,研究員,博士生導(dǎo)師。主要從事植被定量遙感機理和應(yīng)用研究。Email:huangwj@radi.ac.cn

10.11975/j.issn.1002-6819.2019.17.022

S512.1

A

1002-6819(2019)-17-0178-08

黃林生,江 靜,黃文江,葉回春,趙晉陵,馬慧琴,阮 超.Sentinel-2影像和BP神經(jīng)網(wǎng)絡(luò)結(jié)合的小麥條銹病監(jiān)測方法 [J]. 農(nóng)業(yè)工程學(xué)報,2019,35(17):178-185. doi:10.11975/j.issn.1002-6819.2019.17.022 http://www.tcsae.org

Huang Linsheng, Jiang Jing, Huang Wenjiang, Ye Huichun, Zhao Jinling, Ma Huiqin, Ruan Chao. Wheat yellow rust monitoring method based on Sentinel-2 image and BPNN model[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(17): 178-185. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2019.17.022 http://www.tcsae.org

猜你喜歡
條銹病植被指數(shù)波段
小麥條銹病田間為害損失的初步分析
最佳波段組合的典型地物信息提取
基于無人機圖像的草地植被蓋度估算方法比較
冬小麥SPAD值無人機可見光和多光譜植被指數(shù)結(jié)合估算
陜西安康:農(nóng)技專家開展小麥條銹病普防工作
小麥條銹病的防控技術(shù)
小麥條銹病持續(xù)控制的策略
基于植被指數(shù)選擇算法和決策樹的生態(tài)系統(tǒng)識別
基于PLL的Ku波段頻率源設(shè)計與測試
小型化Ka波段65W脈沖功放模塊
图木舒克市| 张家港市| 历史| 茂名市| 阳朔县| 阜阳市| 新平| 灌云县| 凤凰县| 东光县| 乐清市| 大姚县| 廊坊市| 夏津县| 景泰县| 孟州市| 当阳市| 四会市| 揭西县| 中方县| 东平县| 崇义县| 禹城市| 乾安县| 宁晋县| 开远市| 苏州市| 永胜县| 新泰市| 江达县| 奉贤区| 广元市| 盘锦市| 大理市| 二手房| 手机| 温州市| 合肥市| 蓝山县| 屏南县| 龙江县|