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

?

無人機(jī)遙感監(jiān)測作物病蟲害脅迫方法與最新研究進(jìn)展

2022-05-30 10:48楊國峰何勇馮旭萍李禧堯張金諾俞澤宇
關(guān)鍵詞:無人機(jī)數(shù)據(jù)處理深度學(xué)習(xí)

楊國峰 何勇 馮旭萍 李禧堯 張金諾 俞澤宇

摘要:病蟲害是作物生產(chǎn)面臨的主要脅迫之一。近年來,隨著無人機(jī)產(chǎn)業(yè)的快速發(fā)展,無人機(jī)農(nóng)業(yè)遙感因其圖像空間分辨率高、數(shù)據(jù)獲取時(shí)效性強(qiáng)和成本低等特點(diǎn),在作物病蟲害脅迫監(jiān)測應(yīng)用中發(fā)揮了重要作用。本文首先介紹了利用無人機(jī)遙感監(jiān)測作物病蟲害脅迫的相關(guān)背景;其次對目前無人機(jī)遙感監(jiān)測作物病蟲害脅迫中的常用方法進(jìn)行了概述,主要探討無人機(jī)遙感監(jiān)測作物病蟲害脅迫的數(shù)據(jù)獲取方式和數(shù)據(jù)處理方法;之后從可見光成像遙感、多光譜成像遙感、高光譜成像遙感、熱紅外成像遙感、激光雷達(dá)成像遙感和多遙感融合與對比六個(gè)方面重點(diǎn)綜述了近期國內(nèi)外無人機(jī)遙感監(jiān)測作物病蟲害脅迫的研究進(jìn)展。最后提出了無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究與應(yīng)用中尚未解決的關(guān)鍵技術(shù)問題與未來的發(fā)展方向。本文為把握無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究熱點(diǎn)、應(yīng)用瓶頸、發(fā)展趨勢提供借鑒和參考,以期助力中國無人機(jī)遙感監(jiān)測作物病蟲害脅迫更加標(biāo)準(zhǔn)化、信息化、精準(zhǔn)化和智能化。

關(guān)鍵詞:無人機(jī);遙感監(jiān)測;病蟲害脅迫;數(shù)據(jù)獲取;數(shù)據(jù)處理;深度學(xué)習(xí);多遙感融合

中圖分類號:S-435???????????????? 文獻(xiàn)標(biāo)志碼:A??????????????? 文章編號:SA202201008

引用格式:楊國峰, 何勇, 馮旭萍, 李禧堯, 張金諾, 俞澤宇.無人機(jī)遙感監(jiān)測作物病蟲害脅迫方法與最新研究進(jìn)展[J].智慧農(nóng)業(yè)(中英文), 2022, 4(1):1-16.

YANG Guofeng, HE Yong, FENG Xuping, LI Xiyao, ZHANG Jinnuo, YU Zeyu. Methods and new research prog‐ress of remote sensing monitoring of crop disease and pest stress using unmanned aerial vehicle[J]. Smart Agricul‐ture, 2022, 4(1):1-16.(in Chinese with English abstract)

1 引言

病蟲害的爆發(fā)會影響農(nóng)業(yè)生產(chǎn),造成作物產(chǎn)量和質(zhì)量的嚴(yán)重下降[1]。近年來,中國病蟲害的發(fā)生和傳播呈現(xiàn)增多態(tài)勢,各省份發(fā)生率都有所增加,導(dǎo)致病蟲害的防治更加困難[2]。目前,中國作物病蟲害監(jiān)測主要依靠植保部門、植保站等人員實(shí)地調(diào)查取樣,不僅費(fèi)時(shí)費(fèi)力、準(zhǔn)確性低、時(shí)效性差、區(qū)域覆蓋小,而且受人為因素影響,無法實(shí)現(xiàn)大范圍全覆蓋的作物病蟲害脅迫監(jiān)測。

遙感技術(shù)基于其精準(zhǔn)、快速、大面積、無破壞等特點(diǎn),在作物病蟲害脅迫監(jiān)測領(lǐng)域已經(jīng)顯示出了獨(dú)特優(yōu)勢。遙感技術(shù)是指運(yùn)用不與探測目標(biāo)接觸的遙感傳感器,收集記錄目標(biāo)物輻射、反射、散射的電磁波信息,進(jìn)而獲取目標(biāo)物的特征、性質(zhì)及其變化的綜合探測技術(shù)[3]。當(dāng)前,遙感監(jiān)測方式可大致分為天基、空基和地基三種方式。其中,天基監(jiān)測如衛(wèi)星遙感監(jiān)測,容易受氣象條件影響、重訪周期較長且空間分辨率低;地基監(jiān)測如固定監(jiān)測站,監(jiān)測范圍有限、成本很高且難以大面積設(shè)立[4]。而空基中的無人機(jī)( Un ‐ manned Aerial Vehicle ,UAV )遙感監(jiān)測具備運(yùn)行成本低、靈活度高、采集迅速、覆蓋面廣等特點(diǎn),可獲得更高的空間、時(shí)間和光譜分辨率的影像,彌補(bǔ)了傳統(tǒng)病蟲害監(jiān)測過程中存在的缺陷,被認(rèn)為是遙感監(jiān)測作物病蟲害脅迫的有效手段[5]。

近年來,隨著以無人機(jī)飛行平臺和數(shù)據(jù)獲取傳感器等為代表的硬件不斷成熟,以機(jī)器學(xué)習(xí)乃至深度學(xué)習(xí)等為代表的數(shù)據(jù)處理技術(shù)快速發(fā)展,基于無人機(jī)的遙感監(jiān)測技術(shù)逐漸成為低空遙感領(lǐng)域的重要監(jiān)測方式[6]。在作物病蟲害脅迫遙感監(jiān)測領(lǐng)域,基于無人機(jī)的遙感監(jiān)測技術(shù)已是一個(gè)重要且熱門的研究方向,并且基于感知、決策、執(zhí)行的無人機(jī)技術(shù)也已成為智慧農(nóng)業(yè)的關(guān)鍵核心技術(shù)。

本文圍繞無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究,首先在監(jiān)測數(shù)據(jù)獲取和處理兩部分梳理了無人機(jī)作物病蟲害遙感監(jiān)測方法;然后從可見光成像遙感、多光譜成像遙感、高光譜成像遙感、熱紅外成像遙感、激光雷達(dá)成像遙感和多遙感融合與對比方面分析了國內(nèi)外無人機(jī)遙感監(jiān)測作物病蟲害脅迫的研究方法與進(jìn)展;最后提出了無人機(jī)遙感監(jiān)測作物病蟲害脅迫中尚未解決的關(guān)鍵技術(shù)問題與未來的發(fā)展方向。

2 無人機(jī)遙感監(jiān)測作物病蟲害脅迫方法

2.1概述

目前,基于無人機(jī)的遙感監(jiān)測技術(shù)已在作物病蟲害脅迫領(lǐng)域被廣泛應(yīng)用與研究。當(dāng)作物受到病蟲害脅迫時(shí)通常在不同光譜波段上表現(xiàn)出吸收和反射特性的變化,即為作物病蟲害脅迫的光譜響應(yīng)[7]。作物由于病蟲害脅迫受損會引起色素、形態(tài)、結(jié)構(gòu)等改變,通??梢酝ㄟ^提取其光譜響應(yīng)特征并加以分析處理,實(shí)現(xiàn)對病蟲害脅迫的精準(zhǔn)、快速、無損監(jiān)測[8]。

無人機(jī)遙感監(jiān)測作物病蟲害脅迫是以無人機(jī)為遙感監(jiān)測平臺,利用搭載的各種傳感器獲取目標(biāo)作物的遙感影像、視頻、點(diǎn)云等數(shù)據(jù),通過對數(shù)據(jù)的處理、挖掘和建模來獲取作物病蟲害脅迫信息。監(jiān)測方法大致可分為兩類:(1) 單一遙感監(jiān)測方法,主要通過無人機(jī)搭載相應(yīng)傳感器進(jìn)行作物病蟲害脅迫數(shù)據(jù)獲取、處理及分析;(2)綜合遙感監(jiān)測方法,主要利用無人機(jī)遙感監(jiān)測技術(shù)與地面人工調(diào)查取樣等方式綜合進(jìn)行作物病蟲害脅迫數(shù)據(jù)獲取、處理及分析。根據(jù)實(shí)際監(jiān)測情況的不同選擇單一或綜合無人機(jī)遙感監(jiān)測作物病蟲害脅迫方法,以實(shí)現(xiàn)作物病蟲害信息的精準(zhǔn)獲取和高效動態(tài)監(jiān)測,為作物病蟲害科學(xué)防治提供支撐。其中,主要涉及到以無人機(jī)飛行平臺和機(jī)載傳感器為代表的無人機(jī)遙感監(jiān)測硬件系統(tǒng),以無人機(jī)測繪攝影測量等專業(yè)處理軟件和數(shù)據(jù)處理分析相關(guān)算法或模型為代表的無人機(jī)遙感監(jiān)測軟件系統(tǒng)。

2.2監(jiān)測數(shù)據(jù)獲取方式

2.2.1 無人機(jī)飛行平臺

無人機(jī)是利用無線電遙控設(shè)備和自備的程序控制裝置操縱的不載人飛行器[9]。大致可分為多旋翼、固定翼、單旋翼(直升機(jī))和混合翼(垂直起降固定翼)幾種,如圖1所示。在使用無人機(jī)遙感監(jiān)測作物病蟲害脅迫時(shí),關(guān)注的重點(diǎn)是無人機(jī)載荷、續(xù)航時(shí)間、飛行高度、監(jiān)測精度和空間分辨率等。如表1所示為不同的無人機(jī)飛行平臺。

由于遙感監(jiān)測數(shù)據(jù)獲取過程中飛行平臺的選擇對獲取的數(shù)據(jù)質(zhì)量有影響,因此選擇的飛行平臺應(yīng)具備可操控性、高穩(wěn)定性和飛行持久性等特點(diǎn),以獲取質(zhì)量較好的數(shù)據(jù)。目前,多旋翼無人機(jī)因具有航速姿態(tài)可調(diào)、飛行穩(wěn)定、能夠定點(diǎn)懸停等優(yōu)勢,適用于定點(diǎn)重復(fù)獲取多尺度、高分辨率的作物病蟲害脅迫數(shù)據(jù),在遙感監(jiān)測作物病蟲害脅迫研究與應(yīng)用中最為廣泛。

2.2.2 機(jī)載傳感器

無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究中的機(jī)載傳感器類型主要包括多光譜相機(jī)、高光譜相機(jī)、數(shù)碼相機(jī)、熱紅外相機(jī)、激光雷達(dá)等[18],如圖2所示。無人機(jī)遙感監(jiān)測作物病蟲害脅迫的機(jī)載傳感器通常為光學(xué)、光電學(xué)和熱力學(xué)傳感器,少部分機(jī)載傳感器屬于聲學(xué)等領(lǐng)域。常用機(jī)載傳感器及其測量指標(biāo)、優(yōu)勢和劣勢如表2所示。因此,利用無人機(jī)獲取病蟲害脅迫數(shù)據(jù)時(shí),需要依據(jù)地域特征、病蟲害爆發(fā)程度和作物種類等情況選擇合適的傳感器。

2.2.3 數(shù)據(jù)獲取流程

無人機(jī)遙感監(jiān)測作物病蟲害脅迫的數(shù)據(jù)獲取流程是保證每次飛行能正常操作且安全準(zhǔn)確獲取監(jiān)測數(shù)據(jù)的重要流程,無人機(jī)數(shù)據(jù)獲取的質(zhì)量和數(shù)量對后續(xù)處理分析的結(jié)果有重要影響。對自主作業(yè)模式(全球定位系統(tǒng)模式,以實(shí)現(xiàn)精確懸停、指點(diǎn)飛行、規(guī)劃航線等操作)下無人機(jī)遙感監(jiān)測作物病蟲害脅迫方法來說,主要有以下步驟:

(1) 飛行前期準(zhǔn)備。確認(rèn)飛行任務(wù)區(qū)域及申請空域;查詢地理、天氣環(huán)境信息;選擇并調(diào)試飛行與地面設(shè)備(無人機(jī)飛行平臺、機(jī)載傳感器、遙控器、導(dǎo)航等)以及檢查電量、是否能正常工作等;是否攜帶其他設(shè)備,如輻射定標(biāo)板等。

(2) 正式飛行前準(zhǔn)備?,F(xiàn)場組裝、調(diào)試、連接飛行與地面設(shè)備;根據(jù)任務(wù)區(qū)域地形、作物病蟲害、續(xù)航、載荷等情況,設(shè)計(jì)飛行任務(wù)方案,如起降點(diǎn)、航線、高度、架次、重疊率等[19,20]。

(3) 飛行作業(yè)執(zhí)行。實(shí)時(shí)關(guān)注無人機(jī)飛行平臺的速度、位置、電量、電壓、任務(wù)時(shí)間等飛行情況,監(jiān)督飛行時(shí)穩(wěn)定、安全作業(yè),必要時(shí)可以手動接管飛行。

(4) 飛行作業(yè)結(jié)束。自主返航或操控返航;返航完畢可關(guān)閉飛行與地面設(shè)備電源;回收飛行與地面設(shè)備,讀取儲存卡數(shù)據(jù)或在飛行作業(yè)時(shí)通過地面設(shè)備實(shí)時(shí)獲取遙感監(jiān)測數(shù)據(jù)。

2.3監(jiān)測數(shù)據(jù)處理方法

如何從無人機(jī)遙感監(jiān)測作物病蟲害脅迫獲取的大量數(shù)據(jù)中高效提取表型特征十分重要,并且很大程度決定處理分析的結(jié)果。表型特征主要包括光譜特征、紋理特征、顏色特征、形狀特征和生理特征等。目前,對這些表型特征的分析處理是無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究的熱點(diǎn)[21,22]?,F(xiàn)階段主流的無人機(jī)遙感監(jiān)測作物病蟲害脅迫數(shù)據(jù)處理的方法可以大致分為統(tǒng)計(jì)分析方法和機(jī)器學(xué)習(xí)方法兩類。

目前,監(jiān)測數(shù)據(jù)處理方法的流程主要為:遙感影像的格式調(diào)整、清洗、預(yù)處理、拼接、校正、特征提取、特征選擇、設(shè)計(jì)方法模型、評價(jià)指標(biāo)與調(diào)優(yōu)等。對于遙感影像格式調(diào)整一般使用機(jī)載傳感器配套軟件處理,對于遙感影像拼接和校正等則使用無人機(jī)測繪攝影測量軟件,如Metashape軟件(Agisoft公司)、 PIX4Dmapper軟件 ( Pix4D 公司) 和Inpho軟件 ( Trimble 公司)等。

2.3.1 統(tǒng)計(jì)分析方法

在無人機(jī)遙感監(jiān)測作物病蟲害脅迫相關(guān)研究中,常用描述統(tǒng)計(jì)、相關(guān)分析、回歸分析、判別分析、方差分析和聚類分析等統(tǒng)計(jì)分析方法。

通常,借助上述方法使用植被的光譜特征進(jìn)行作物病蟲害脅迫遙感監(jiān)測。通過對光譜曲線進(jìn)行分析,可以發(fā)現(xiàn)不同作物病蟲害脅迫的光譜曲線變化特征。一般基于特定光譜波段、波段計(jì)算與組合以及植被指數(shù)(Vegetation Index ,VI )等方法進(jìn)行光譜特征提取。因?yàn)椴煌∠x害脅迫對作物生長造成的影響程度不一致,所以特定光譜波段更有利于對不同病蟲害脅迫進(jìn)行監(jiān)測。光譜波段經(jīng)運(yùn)算組合后可以得到反映植被生長狀況、植被覆蓋度等有相關(guān)意義的值,即 VI 。VI 已廣泛用來定性和定量評價(jià)植被覆蓋及其生長活力,常用于作物病蟲害脅迫監(jiān)測的 VI 主要由2~3個(gè)波段構(gòu)成[23]。利用 VI進(jìn)行回歸或相關(guān)分析可以建立遙感監(jiān)測數(shù)據(jù)與作物生長信息的反演模型,即經(jīng)驗(yàn)?zāi)P?。由于?jīng)驗(yàn)?zāi)P托枰罅康膶?shí)測數(shù)據(jù)為基礎(chǔ),而實(shí)測數(shù)據(jù)的精度如何,很大程度取決于實(shí)測數(shù)據(jù)的測量精度;經(jīng)驗(yàn)?zāi)P痛嬖趨^(qū)域適用性的限制,常常在實(shí)測數(shù)據(jù)采集的區(qū)域模型的適用性較高,在其他區(qū)域適用性較低;地表粗糙度的變化經(jīng)驗(yàn)?zāi)P蜔o法考慮。目前,由于植被參數(shù)的遙感反演物理模型具有因果關(guān)系和數(shù)學(xué)物理基礎(chǔ)[24,25],因而正成為遙感反演作物病蟲害脅迫研究的主要方向,但主要集中在輻射傳輸模型[26,27],尚無幾何光學(xué)模型和混合模型等在作物病蟲害脅迫中研究與應(yīng)用。

2.3.2 機(jī)器學(xué)習(xí)方法

當(dāng)前,無人機(jī)遙感監(jiān)測作物病蟲害脅迫的數(shù)據(jù)處理方法主要集中于機(jī)器學(xué)習(xí)方法。機(jī)器學(xué)習(xí)最基本的做法是使用算法來解析遙感監(jiān)測數(shù)據(jù),然后對真實(shí)世界中的病蟲害脅迫做出決策和預(yù)測。機(jī)器學(xué)習(xí)傳統(tǒng)的算法包括決策樹、聚類、貝葉斯分類、隨機(jī)森林( Random Forest ,RF )、支持向量機(jī)(Support Vector Machine , SVM )、 k-近鄰(k-Nearest Neighbor ,kNN )算法等。從學(xué)習(xí)方法上劃分,機(jī)器學(xué)習(xí)算法可以分為監(jiān)督學(xué)習(xí)、無監(jiān)督學(xué)習(xí)、半監(jiān)督學(xué)習(xí)、集成學(xué)習(xí)、深度學(xué)習(xí)和強(qiáng)化學(xué)習(xí)等[28]。

在機(jī)器學(xué)習(xí)傳統(tǒng)的算法中,如何最大限度地在數(shù)據(jù)中提取有用的特征以供算法和模型使用至關(guān)重要。當(dāng)作物受到一定程度病蟲害脅迫后,患病蟲害的作物外部形態(tài)(葉面積、株高、顏色等)與內(nèi)部生理均會發(fā)生較為明顯的改變。因此,除了使用無人機(jī)遙感監(jiān)測獲取作物病蟲害脅迫影像的光譜特征之外,顏色特征(如顏色直方圖、顏色熵、顏色矩、顏色聚合向量等)、紋理特征(如局部二值模式、灰度直方圖、灰度共生矩陣、小波變換等)和形狀特征(如傅立葉變換、形狀不變矩、小波輪廓描述符等)等也常被提取使用。此外,還可以結(jié)合實(shí)地調(diào)查取樣獲取作物病蟲害脅迫下的株高、葉綠素含量、生物量和含水量等各項(xiàng)指標(biāo)。特別地,作物受病蟲害脅迫的環(huán)境(溫度、濕度、海拔、土壤含水量、養(yǎng)分等)也會對病蟲害脅迫有影響,也是需要重點(diǎn)關(guān)注與開展長期性、周期性的研究。

近年來,隨著數(shù)據(jù)的高速增長、算力的迅猛增強(qiáng)、算法的完善成熟,深度學(xué)習(xí)逐漸在無人機(jī)遙感監(jiān)測作物病蟲害脅迫領(lǐng)域得以大量應(yīng)用[29]。使用傳統(tǒng)的機(jī)器學(xué)習(xí)方法提取特征往往需要進(jìn)行手工提取或特征工程,并且需要大量的時(shí)間優(yōu)化,而深度學(xué)習(xí)減少人為設(shè)計(jì)特征的過程,將自動學(xué)習(xí)與任務(wù)相關(guān)的特征即特征學(xué)習(xí)融入建立模型的過程。不過,使用深度學(xué)習(xí)方法面臨需要大量訓(xùn)練數(shù)據(jù)和性能較高的計(jì)算機(jī)硬件支持,以及更優(yōu)的模型等主要限制[30]。盡管如此,相關(guān)研究表明使用深度學(xué)習(xí)方法比機(jī)器學(xué)習(xí)傳統(tǒng)的算法能獲得更好的分類、檢測、識別和分割性能,因此未來也需要不斷深入研究與改進(jìn)。同時(shí),隨著高光譜機(jī)載傳感器和深度學(xué)習(xí)技術(shù)的不斷發(fā)展,利用深度學(xué)習(xí)技術(shù)可以充分挖掘高光譜影像的潛在特征,實(shí)現(xiàn)更好地對各種作物病蟲害脅迫進(jìn)行監(jiān)測。

3 無人機(jī)遙感監(jiān)測作物病蟲害脅迫最新研究進(jìn)展

本節(jié)主要針對2019―2021年無人機(jī)遙感監(jiān)測作物病蟲害脅迫相關(guān)研究,從可見光成像遙感、多光譜成像遙感、高光譜成像遙感、熱紅外成像遙感、激光雷達(dá)成像遙感和多遙感融合與對比方面進(jìn)行綜述。

3.1可見光成像遙感

使用基于無人機(jī)搭載可見光傳感器,如數(shù)碼相機(jī)與 RGB ( Red Green Blue )傳感器,可以獲得可見光影像(紅、藍(lán)、綠波段)。目前,可見光成像遙感在作物病蟲害脅迫領(lǐng)域已被廣泛應(yīng)用研究。

在利用統(tǒng)計(jì)分析方法方面,有研究提取紅綠藍(lán)波段的特征信息后,構(gòu)建 VI 和分類模型識別病蟲害脅迫程度。Nazir等[31]為監(jiān)測種植園生長的桉樹健康狀況,利用冠層葉片的可見光波段開發(fā)視覺大氣阻力指數(shù)(Visual Atmospheric Resis‐tance Index ,VARI ),并基于 VARI-green 將病蟲害脅迫程度分為四級,將 VARI-green 指數(shù)和與 NDVI 進(jìn)行相關(guān)性分析,平均相關(guān)系數(shù)為0.73。該研究展示了利用無人機(jī)和 RGB 相機(jī)對大規(guī)模工業(yè)林場作業(yè)區(qū)域進(jìn)行航空調(diào)查的實(shí)際應(yīng)用。 Dutta等[32]使用植被指數(shù)和 Otsu閾值分割方法對十字花科作物進(jìn)行早期病害檢測,利用在兩種不同的陽光條件下收集的 RGB 圖像,區(qū)分健康葉片、患病葉片和背景,研究表明需要根據(jù)作物類型調(diào)整植被指數(shù),以便通過圖像處理方法準(zhǔn)確識別作物的感染部位。

在使用機(jī)器學(xué)習(xí)方法方面,對于可見光成像遙感主要利用提取的可見光影像的顏色、紋理、光譜和形狀等特征開展如分類、識別、檢測等監(jiān)測任務(wù)[33]。在利用機(jī)器學(xué)習(xí)方法提取特征方面,Calou等[34]使用五種傳統(tǒng)機(jī)器學(xué)習(xí)分類算法來識別和量化香蕉黃斑病, SVM 算法取得了最佳分類性能。類似地,Dang等[35]基于一系列顏色特征進(jìn)行閾值檢測以對蘿卜枯萎病的嚴(yán)重程度進(jìn)行分類,并使用 k-means聚類算法實(shí)現(xiàn)將蘿卜區(qū)域與田地的其他區(qū)域(例如土地和地膜)分割。為評估無人機(jī)遙感監(jiān)測方法在小麥蠕孢葉枯病檢測方面的潛力,Huang 等[36]使用無人機(jī)進(jìn)行可見光影像采集,同時(shí)進(jìn)行地面調(diào)查,根據(jù)不同的病害脅迫程度建立了正常、輕度、中度和重度四種病害類別,構(gòu)建了基于 LeNet-5的卷積神經(jīng)網(wǎng)絡(luò),準(zhǔn)確區(qū)分了小麥蠕孢葉枯病的感染區(qū)和健康區(qū)。為量化非洲葉蟬對葡萄園的影響,并開發(fā)害蟲脅迫嚴(yán)重程度的專題圖,del-Campo-San等[37]使用 RGB 圖像構(gòu)建基于人工神經(jīng)網(wǎng)絡(luò)與幾何技術(shù)相結(jié)合的計(jì)算視覺算法,該算法與從 RGB 圖像生成的地理產(chǎn)品相結(jié)合,改進(jìn)了受影響植被、健康植被和地面的圖像分割效果,并通過消除土壤的影響提高了檢測的效果。Tetila等[38]使用無人機(jī)拍攝的大豆害蟲圖像進(jìn)行五種深度學(xué)習(xí)模型的評估,發(fā)現(xiàn)經(jīng)過微調(diào)訓(xùn)練的深度學(xué)習(xí)模型可以獲得更高的分類準(zhǔn)確率93.82%。此外,也展示了深度學(xué)習(xí)模型的性能優(yōu)于傳統(tǒng)的特征提取方法。

上述研究表明,采用較低成本的可見光成像遙感可以方便快捷地對作物病蟲害脅迫進(jìn)行監(jiān)測,并且也能取得不錯(cuò)的識別效果。開發(fā)基于可見光成像的無人機(jī)遙感監(jiān)測作物病蟲害脅迫系統(tǒng)能以更大經(jīng)濟(jì)優(yōu)勢進(jìn)行推廣普及,從而助力農(nóng)業(yè)高效生產(chǎn)。

3.2多光譜成像遙感

多光譜成像遙感是指用2個(gè)以上光譜通道傳感器進(jìn)行地物同步成像,接收和記錄目標(biāo)物反射輻射的電磁波信息被分成若干個(gè)窄波段的光束。在利用無人機(jī)搭載多光譜傳感器遙感監(jiān)測作物病蟲害脅迫時(shí),通常獲取不同時(shí)空、冠層等2~5個(gè)波段遙感影像開展相關(guān)研究。

在使用統(tǒng)計(jì)分析方法方面,國內(nèi)外研究通常在提取光譜特征后,再篩選、組合與作物病蟲害脅迫遙感監(jiān)測任務(wù)相關(guān)的 VI ,如歸一化植被指數(shù) (Normalized? Difference? Vegetation? Index,NDVI )、增強(qiáng)型植被指數(shù)和垂直植被指數(shù)等常用VI或提出新的植被指數(shù)如 Excess Near-Infrared和Excess Red-Edge [39],同時(shí)結(jié)合地面調(diào)查的方式人工采集相關(guān)數(shù)據(jù)?;谏鲜?VI ,主要通過指數(shù)統(tǒng)計(jì)相關(guān)性、方差分析、線性回歸、二元邏輯回歸等統(tǒng)計(jì)分析方法識別敏感指數(shù)以及結(jié)合多時(shí)空變化進(jìn)行小麥[40]、辣椒[41]、香蕉[42]、西瓜[43]、大豆[44] 和油棕[45]作物病蟲害的定量評估和分析。部分研究建立了不同的機(jī)器學(xué)習(xí)分類模型對梨[10]、檳榔[46,47]、柑橘[48,49]、花生[50]、檸檬[51]和馬鈴薯[52]等病蟲害發(fā)生情況進(jìn)行了有效地監(jiān)測,其中使用 SVM 和 RF 的方法占多數(shù),還有使用徑向基函數(shù) ( Radical Basis Function,RBF )和kNN等。部分研究測試不同的輸入波段組合,改進(jìn)分割算法 U-net 進(jìn)行小麥黃銹病檢測[53] 以及開發(fā)新的分割算法處理獲取的 RGB、CIR ( Color and Infrared )和 NDVI圖像,隨后使用線性判別分析對受木質(zhì)部難養(yǎng)菌(Xylella Fas‐tidiosa)影響的橄欖樹進(jìn)行分類[54]。一些研究將統(tǒng)計(jì)分析方法和機(jī)器學(xué)習(xí)方法相結(jié)合,如 Lan 等[11]先計(jì)算 VI ,然后使用主成分分析和 Auto ‐ Encoder 進(jìn)行相關(guān)分析和特征壓縮以發(fā)現(xiàn)潛在特征,之后比較了幾種機(jī)器學(xué)習(xí)分類算法以實(shí)現(xiàn)對柑橘黃龍病的監(jiān)測。Chivasa等[12]通過玉米品種對玉米條紋病毒的反應(yīng),使用基于無人機(jī)的多光譜遙感數(shù)據(jù)來提高作物表型分析效率,用 RF 評估無人機(jī)衍生的光譜和 VI ,再對無人機(jī)獲取的數(shù)據(jù)與人工玉米條紋病毒評分之間進(jìn)行相關(guān)性分析。馬云強(qiáng)等[55]通過深度學(xué)習(xí)技術(shù)定量反演云南切梢小蠹脅迫情況,結(jié)果顯示 NDVI與蟲害的危害程度呈負(fù)相關(guān)。

在僅使用機(jī)器學(xué)習(xí)技術(shù)方面,研究者使用傳統(tǒng)方法包括 SVM 、RF 、kNN、樸素貝葉斯和集成學(xué)習(xí)等對油棕靈芝屬病害、馬鈴薯晚疫病進(jìn)行分類[56,57],并且開發(fā)了新的針對棉花根腐病的無監(jiān)督 Plant-by-Plant 分類算法[58]。此外,還有研究使用改進(jìn)的 U-Net進(jìn)行小麥黃銹病病害區(qū)域分割研究[59]。

大量研究表明,通過使用統(tǒng)計(jì)分析方法和機(jī)器學(xué)習(xí)方法可以較為準(zhǔn)確地評估作物病蟲害脅迫的程度、類型和位置等。與可見光成像遙感相比,多光譜成像遙感能獲取更多的光譜信息。當(dāng)使用這些更多的信息時(shí)會使得監(jiān)測結(jié)果更為準(zhǔn)確有效。

3.3高光譜成像遙感

當(dāng)前,基于無人機(jī)搭載高光譜傳感器遙感監(jiān)測病蟲害脅迫的作物主要集中于小麥、水稻、柑橘等。與無人機(jī)搭載多光譜傳感器遙感監(jiān)測作物病蟲害脅迫的方法不同,高光譜影像分辨率高、數(shù)據(jù)量大,相鄰光譜波段相關(guān)性高,具有空間域和光譜域信息。

在使用統(tǒng)計(jì)分析方法方面,馬書英等[60]對板栗樹進(jìn)行高光譜遙感監(jiān)測,通過對冠層紅蜘蛛脅迫下的光譜特征與紅蜘蛛蟲害感染程度直接進(jìn)行相關(guān)性分析;郭偉等[61]通過分析棉花的冠層光譜特征,再基于敏感波段的比值導(dǎo)數(shù)值開發(fā)了蚜害脅迫程度的預(yù)測模型。上述研究使用光譜特征、敏感波長進(jìn)行統(tǒng)計(jì)分析以識別板栗樹紅蜘蛛蟲害與棉花蚜害的脅迫程度。相關(guān)研究在統(tǒng)計(jì)分析方法中還使用 VI ,與多光譜使用 VI類似,研究者基于多種 VI 、特征波長和冠層特征光譜參數(shù)等使用如 Fisher 判別分析法對毛竹剛竹毒蛾[62]、偏最小二乘回歸 (Partial Least SquaresRegression,PLSR )方法對小麥全蝕病[63]、線性回歸對玉米大斑病[64]、秩和檢驗(yàn)對水稻稻曲病[65]進(jìn)行了準(zhǔn)確分類和識別,或構(gòu)建由東亞飛蝗造成的蘆葦損失估計(jì)模型,用定量的方式評估和量化東亞飛蝗損害程度[66]。Liu 等[67] 和 Ma等[68]考慮到高光譜影像處理的特點(diǎn),在使用光譜帶、VI 的基礎(chǔ)上融入了紋理特征,構(gòu)建和改進(jìn)了 BP神經(jīng)網(wǎng)絡(luò)、SVM等機(jī)器學(xué)習(xí)來監(jiān)測小麥枯萎病;而 Guo等[69]構(gòu)建了基于 PLSR的不同感染時(shí)期的小麥黃銹病監(jiān)測模型。此外,大量研究使用如 RF 、RBF 、kNN等機(jī)器學(xué)習(xí)方法將不同的光譜特征,如 VI 、光譜特征參數(shù)等作為輸入量對水稻[70]、柑橘[71,72]、南瓜[73]病蟲害脅迫等進(jìn)行監(jiān)測,其中有研究使用地面車輛和無人機(jī)兩種測量平臺對小麥黃銹病的發(fā)展階段和病害嚴(yán)重程度進(jìn)行檢測和量化評估[74],也有研究通過逐步判別法識別和區(qū)分番茄病害程度的結(jié)果與 RBF和多層感知器的分類結(jié)果分別進(jìn)行比較[75,76]。

在使用機(jī)器學(xué)習(xí)方法方面,鄧小玲等[77]對光譜進(jìn)行預(yù)處理和特征工程后,采用連續(xù)投影算法提取對作物病蟲害脅迫程度或類別分類貢獻(xiàn)最大的特征波長組合,以基于全波段或特征波段建立多個(gè)機(jī)器學(xué)習(xí)分類模型。雖然傳統(tǒng)機(jī)器學(xué)習(xí)的方法希望提取具有代表性的光譜特征,但是對于高光譜影像的空間特征過去往往在處理的過程中被自動丟棄,從而不可避免地減少了信息量,并且限制了作物病蟲害脅迫相關(guān)任務(wù)的深入研究[78-80]。當(dāng)前,已有研究構(gòu)建了能同時(shí)利用光譜和空間信息的深度學(xué)習(xí)模型對小麥黃銹病進(jìn)行遙感監(jiān)測[81],以充分發(fā)揮高光譜影像的潛力。目前,有研究基于光譜和時(shí)間特征,利用 RF 構(gòu)建水稻稻曲病監(jiān)測模型,發(fā)現(xiàn)水稻稻曲病的侵染面積隨著時(shí)間的推移呈擴(kuò)大趨勢,符合水稻稻曲病的自然發(fā)展規(guī)律[82]。也有研究獲取高光譜影像的敏感光譜特征和波段,以及通過灰度共生矩陣提取紋理特征,并將光譜和紋理特征融合以檢測小麥赤霉病[83]。

現(xiàn)有研究表明,與可見光成像遙感和多光譜成像遙感相比,高光譜成像遙感具有連續(xù)光譜、更多波段和數(shù)據(jù)量更大等特點(diǎn),因此很多研究人員能實(shí)現(xiàn)更好的作物病蟲害脅迫遙感監(jiān)測效果。此外,隨著數(shù)據(jù)處理方法的發(fā)展,基于深度學(xué)習(xí)的方法已能更好地利用高光譜圖像開展相關(guān)研究并取得相較于傳統(tǒng)機(jī)器學(xué)習(xí)方法更好的效果。但是,如何更好地挖掘并使用高光譜遙感圖像中病蟲害光譜變化特征,仍然是實(shí)現(xiàn)遙感監(jiān)測大面積作物病蟲害脅迫的重點(diǎn)和難點(diǎn)。

3.4熱紅外成像遙感

作物病蟲害脅迫進(jìn)行熱紅外( Thermal Infra‐ red ,TIR )成像遙感主要是針對溫度的差異進(jìn)行分析。然而,由于遙感監(jiān)測時(shí)易受風(fēng)、云和雨等惡劣天氣因素的影響,當(dāng)前 TIR成像遙感監(jiān)測作物病蟲害脅迫面臨巨大的挑戰(zhàn)。目前,TIR成像遙感監(jiān)測在作物病蟲害脅迫方面的研究相對較少。研究人員在利用 TIR傳感器遙感監(jiān)測作物病蟲害脅迫時(shí),通常主要監(jiān)測作物冠層溫度的差異,從而對健康和受病蟲害脅迫的作物進(jìn)行分類。

在使用統(tǒng)計(jì)分析方法方面,F(xiàn)rancesconi等[84]將 TIR 、RGB影像和地面測量(穗的溫度、光合效率和赤霉病原體的分子鑒定)相結(jié)合,采用方差分析和主成分分析以及計(jì)算 VI 對小麥進(jìn)行赤霉病監(jiān)測,研究顯示了基于無人機(jī)的 TIR和 RGB 影像在小麥和其他谷類作物應(yīng)對病蟲害脅迫時(shí)進(jìn)行田間表型分析的潛力。陳欣欣等[85]將 TIR和無人機(jī)模擬平臺相結(jié)合,基于冠層和葉片兩個(gè)尺度,采用判別分析、單因素方差分析和相關(guān)性分析等方法可以對菌核病侵染油菜的過程進(jìn)行監(jiān)測。

從目前研究看,熱紅外成像遙感具備其獨(dú)特優(yōu)勢,取得了較好的成果。但是,由于熱紅外成像遙感容易受到環(huán)境因素影響,未來需要研發(fā)適用性更強(qiáng)的熱紅外傳感器,以及設(shè)計(jì)相應(yīng)的監(jiān)測方法。值得一提的是,融合 TIR成像遙感數(shù)據(jù)和其他成像遙感數(shù)據(jù)能更好地提升遙感監(jiān)測的效果。

3.5激光雷達(dá)成像遙感

激光雷達(dá)(Light Detection and Ranging ,Li‐DAR )成像遙感方法監(jiān)測作物病蟲害脅迫,依據(jù)激光反射強(qiáng)度來分析作物的病蟲害程度,一般提取作物水平以及垂直結(jié)構(gòu)的冠層信息,如株高、生物量等。此外,激光雷達(dá)還可以獲取蟲群飛行方向和運(yùn)動軌跡。當(dāng)前,適用于無人機(jī)的機(jī)載激光雷達(dá)價(jià)格較貴(通常需5萬~15萬人民幣),使用激光雷達(dá)獲取點(diǎn)云等數(shù)據(jù)的方法和處理算法與模型還有待改進(jìn)。使用無人機(jī)搭載激光雷達(dá)在遙感監(jiān)測病蟲害脅迫方面的研究大都應(yīng)用在林業(yè),且多與其他遙感影像數(shù)據(jù)融合進(jìn)行病蟲害脅迫的監(jiān)測。

在使用統(tǒng)計(jì)分析方法方面,在針對松樹常見的松針紅斑?。≧ed Band Needle Blight), Lin等[86]研究使用基于無人機(jī)的熱紅外成像檢測松針紅斑病引起的冠層溫度升高,并探討成像時(shí)間和天氣條件對檢測的影響,發(fā)現(xiàn)冠層溫度降低與病害水平之間存在統(tǒng)計(jì)學(xué)顯著相關(guān)性,這可能與病害引起的針頭損傷癥狀有關(guān),即細(xì)胞完整性喪失、壞死和最終干燥;樹冠溫度的標(biāo)準(zhǔn)偏差表現(xiàn)出微弱但具有統(tǒng)計(jì)學(xué)意義的相關(guān)性;PLSR 中冠層溫度降低和冠溫標(biāo)準(zhǔn)差的組合進(jìn)一步改善了觀察結(jié)果與估計(jì)病害程度的關(guān)系。

在使用機(jī)器學(xué)習(xí)方法方面,Briechle等[87]使用3D 深度神經(jīng)網(wǎng)絡(luò)PointNet++和激光雷達(dá)數(shù)據(jù)與多光譜影像對多種樹種(松樹、樺樹、榿木)和帶有樹冠的枯立木進(jìn)行了分類,驗(yàn)證了3D 深度神經(jīng)網(wǎng)絡(luò)對于多種樹種和枯立木的分類前景。松材線蟲病是對森林的全球破壞性威脅,在中國對森林造成了極大的破壞。Savian等[88]將松材線蟲病感染分為五個(gè)階段(綠色、早期、中期、重度和灰色),使用 RF 算法估計(jì)高光譜成像數(shù)據(jù)、LiDAR數(shù)據(jù)及其組合預(yù)測松材線蟲病的感染階段,發(fā)現(xiàn)高光譜數(shù)據(jù)在預(yù)測松材線蟲病感染階段的分類準(zhǔn)確率高于 LiDAR ,并且它們的組合具有最佳精度,此外還證明 LiDAR 數(shù)據(jù)比高光譜數(shù)據(jù)具有更高的死樹識別能力。類似地,為準(zhǔn)確評估早期監(jiān)測松梢甲蟲的松林枝條損傷率,Yu 等[89] 同樣基于無人機(jī)的高光譜成像數(shù)據(jù)和激光雷達(dá)數(shù)據(jù)融合檢測害蟲,并通過類似的方法評估松梢甲蟲造成的損失。研究證實(shí),如果結(jié)合高光譜數(shù)據(jù)和激光雷達(dá)數(shù)據(jù),在單個(gè)樹級別準(zhǔn)確預(yù)測枝條損傷率的可能性很高,并且三維輻射傳輸模型可以確定來自激光雷達(dá)的三維樹冠陰影。

目前的研究已經(jīng)取得了一定的成果,但是相關(guān)算法、模型仍然具有局限性,有待進(jìn)一步研究與應(yīng)用。未來,需要重點(diǎn)研發(fā)更低成本的激光雷達(dá)與配套的算法、模型,并且能夠更好地在復(fù)雜場景下精確成像。

3.6多遙感融合與對比

將可見光、多光譜、高光譜、TIR 、雷達(dá)等遙感數(shù)據(jù)進(jìn)行融合與對比研究已逐漸成為熱點(diǎn),具有良好的研究前景。

在融合方面,研究人員使用上述多種傳感器同步或異步獲取作物病蟲害脅迫信息,采取統(tǒng)計(jì)學(xué)習(xí)或機(jī)器學(xué)習(xí)方法對其進(jìn)行處理分析。對于使用機(jī)器學(xué)習(xí)方法的研究,Smigaj等[90] 使用 K-means 、Ward分層聚類等遙感方法成功預(yù)測獼猴桃藤衰退綜合征;Kerkech等[91,92]通過構(gòu)建深度學(xué)習(xí)的分割模型檢測葡萄病害,與直接使用單一來源的遙感影像相比,融合多源遙感影像的方法能獲得更好的檢測結(jié)果。

在對比方面,Moriya 等[93] 比較不同遙感監(jiān)測方法的效果,對于25波段傳感器獲取的影像和3波段傳感器獲取的影像,結(jié)果表明,使用25波段傳感器獲取的影像檢測柑橘樹脂病的性能更好,因?yàn)槠渚哂懈叩臄?shù)據(jù)維度和更詳細(xì)的光譜信息從而可以更準(zhǔn)確地檢測感染柑橘樹脂病的樹木。當(dāng)比較可見光和多光譜對水稻紋枯病的檢測效果時(shí),盡管多光譜傳感器的檢測結(jié)果更準(zhǔn)確,但是可見光傳感器更加經(jīng)濟(jì)且易使用[94]。Dang等[95]檢測蘿卜枯萎病時(shí),對于可見光數(shù)據(jù)集,采用線性譜聚類超像素算法分割田塊圖像,然后通過新構(gòu)建的RadRGB模型對不同蘿卜、土壤和地膜區(qū)域繼續(xù)分類,此外還從近紅外數(shù)據(jù)集構(gòu)建NDVI圖高精度檢測不同階段蘿卜枯萎病。盡管與識別可見光數(shù)據(jù)集中蘿卜枯萎病的復(fù)雜算法相比近紅外方法更簡單,但它需要近紅外這種特殊類型的傳感器。

當(dāng)前研究表明,相比于單一來源的監(jiān)測數(shù)據(jù),融合更多源的遙感監(jiān)測數(shù)據(jù)將能實(shí)現(xiàn)優(yōu)勢互補(bǔ),為作物病蟲害脅迫監(jiān)測提供更全面信息,從而實(shí)現(xiàn)更好的監(jiān)測效果。未來,需要繼續(xù)深入探索數(shù)據(jù)融合在作物病蟲害脅迫中的研究與應(yīng)用,構(gòu)建通用性強(qiáng)、精度高的遙感監(jiān)測算法和模型。

4 問題與展望

目前,利用無人機(jī)遙感監(jiān)測病蟲害脅迫的相關(guān)研究與應(yīng)用還處于初級階段,不僅在無人機(jī)飛行平臺和機(jī)載傳感器的研發(fā)、應(yīng)用和管理方面存在問題,而且在遙感監(jiān)測病蟲害脅迫數(shù)據(jù)的獲取、處理和應(yīng)用方面也具有極大的提升空間。

4.1無人機(jī)性能亟待優(yōu)化

目前針對遙感監(jiān)測所使用的無人機(jī)飛行平臺主要存在穩(wěn)定性不足、續(xù)航時(shí)間較短、易受外界干擾和載荷不足等問題[18]。未來需要進(jìn)一步開發(fā)穩(wěn)定性強(qiáng)、續(xù)航時(shí)間長和載荷大的無人機(jī)飛行平臺。對于無人機(jī)機(jī)載傳感器,為滿足不同的遙感監(jiān)測任務(wù)需求,無人機(jī)可以搭載相應(yīng)的傳感器。然而,現(xiàn)有的無人機(jī)機(jī)載傳感器無法完全適應(yīng)復(fù)雜的外部環(huán)境[96],所獲取的作物病蟲害脅迫數(shù)據(jù)質(zhì)量往往由于環(huán)境的不同而存在差異,并且由于無人機(jī)平臺載荷不足,往往搭載傳感器的重量和數(shù)量等有限[97]。因此,研發(fā)低成本、輕量化和模塊化以及適用性更強(qiáng)的機(jī)載傳感器具有重要意義。

在使用無人機(jī)遙感監(jiān)測作物病蟲害脅迫時(shí),為實(shí)現(xiàn)精準(zhǔn)、經(jīng)濟(jì)、普適等目的需要綜合考慮飛行的任務(wù)、環(huán)境、天氣等因素,從而選擇合適的無人機(jī)飛行平臺和搭載的傳感器。特別地,在提高遙感監(jiān)測任務(wù)的無人機(jī)自身安全性的基礎(chǔ)上,還需進(jìn)一步完善與遵循空中交通管理機(jī)制,施行無人機(jī)空域管理以實(shí)現(xiàn)統(tǒng)一規(guī)劃,合理、充分、有效利用[98]。同時(shí),對于無人機(jī)操作人員需要具備安全飛行意識,遵守當(dāng)?shù)胤煞ㄒ?guī),選擇安全的飛行環(huán)境,預(yù)防無人機(jī)潛在危險(xiǎn)等[99]。

4.2遙感監(jiān)測數(shù)據(jù)獲取困難

無人機(jī)遙感監(jiān)測作物病蟲害脅迫容易受大風(fēng)、陰雨等惡劣天氣影響,同時(shí)采集數(shù)據(jù)時(shí)對太陽光照有較高的要求。大部分無人機(jī)飛行任務(wù)的操作較復(fù)雜且過度依賴于人工設(shè)置[100],制約了其在作物病蟲害脅迫遙感監(jiān)測中的廣泛應(yīng)用。當(dāng)前研究已逐漸從單塊單次作物遙感監(jiān)測變?yōu)槎鄩K多次的連續(xù)監(jiān)測,而這加大了遙感監(jiān)測數(shù)據(jù)獲取的任務(wù)工作量,同時(shí)為監(jiān)測數(shù)據(jù)獲取與處理帶來挑戰(zhàn)。

另外,在開展無人機(jī)遙感監(jiān)測作物病蟲害脅迫研究時(shí),實(shí)驗(yàn)人員通常需要自行攜帶輻射校正板等校正設(shè)備以便后續(xù)校正操作后獲得反射率值等數(shù)據(jù)。此外,過去大部分研究使用單一來源的遙感監(jiān)測數(shù)據(jù),難以全面反應(yīng)整體信息。隨著傳感器的輕型化和無人機(jī)載荷及續(xù)航時(shí)間的增加,已逐步實(shí)現(xiàn)多源數(shù)據(jù)同步遙感監(jiān)測作物病蟲害脅迫信息[4,6, 18]。未來,如何獲取更多的遙感監(jiān)測信息仍然需要深入研究,如獲取空間結(jié)構(gòu)數(shù)據(jù)與光譜成像對應(yīng)數(shù)據(jù)、光譜數(shù)據(jù)與相應(yīng)的環(huán)境數(shù)據(jù),以及空天地(衛(wèi)星、無人機(jī)和地面)一體化立體監(jiān)測數(shù)據(jù)等。值得一提的是,越來越多開源的更大型、更多源、覆蓋更廣的遙感監(jiān)測數(shù)據(jù)庫、數(shù)據(jù)集和數(shù)據(jù)平臺等正在不斷涌現(xiàn)[101, 102],將為相關(guān)研究與應(yīng)用提供數(shù)據(jù)基礎(chǔ)。

4.3遙感監(jiān)測數(shù)據(jù)處理復(fù)雜

通常研究與應(yīng)用人員需要設(shè)計(jì)開發(fā)相應(yīng)算法或使用相關(guān)軟件才能實(shí)現(xiàn)對無人機(jī)遙感監(jiān)測數(shù)據(jù)的拼接、解析和生成處方圖等操作,其中部分算法和特定軟件針對特定應(yīng)用而開發(fā)[53,58,66,81]。隨著利用無人機(jī)遙感監(jiān)測作物病蟲害脅迫時(shí)間增加,以及空間和光譜分辨率提高,需要解決無人機(jī)遙感監(jiān)測獲取的海量數(shù)據(jù)的處理問題。滯后的遙感監(jiān)測數(shù)據(jù)解譯將無法及時(shí)指導(dǎo)病蟲害的防治,導(dǎo)致無法實(shí)現(xiàn)病蟲害快速、精準(zhǔn)、高效地防治。特別地,為實(shí)現(xiàn)時(shí)空實(shí)時(shí)感知、周期實(shí)時(shí)監(jiān)測、要素實(shí)時(shí)評估,當(dāng)前利用空天地一體化立體監(jiān)測技術(shù)開展作物病蟲害脅迫監(jiān)測的綜合研究與應(yīng)用較少且具有巨大潛力[103]。

未來,須不斷完善數(shù)據(jù)處理方法,應(yīng)設(shè)計(jì)開發(fā)出適用性更強(qiáng)、適用面更廣的數(shù)據(jù)處理算法或軟件以提高數(shù)據(jù)處理的準(zhǔn)確性;利用更多源的監(jiān)測數(shù)據(jù)提取更全面綜合的作物病蟲害脅迫特征;縮短數(shù)據(jù)處理時(shí)間,使用基于5G 通訊網(wǎng)絡(luò)和邊緣計(jì)算設(shè)備以解決數(shù)據(jù)傳輸與數(shù)據(jù)及時(shí)處理的問題[104-107],更加及時(shí)、精準(zhǔn)地監(jiān)測病蟲害發(fā)生和危害程度。

4.4遙感監(jiān)測結(jié)果適用局限

由于存在作物的物候階段、種植區(qū)域與類型、生育期、病蟲害脅迫的監(jiān)測時(shí)間、氣候變化等影響,目前大部分算法或模型僅適用于對應(yīng)研究,而無法具備很好的穩(wěn)定性、普適性和通用性,往往由于時(shí)間和空間的局限性而嚴(yán)重制約其大面積應(yīng)用與推廣[10,56,61,90]。例如,在單次的無人機(jī)遙感監(jiān)測病蟲害脅迫中實(shí)現(xiàn)很高的識別率,但并不能保證在其他時(shí)刻通過無人機(jī)獲取的遙感監(jiān)測病蟲害脅迫數(shù)據(jù)能得到同樣的識別率。

因此,未來需要對作物病蟲害脅迫狀態(tài)進(jìn)行持續(xù)監(jiān)測,總結(jié)各種作物病蟲害脅迫的類型和數(shù)據(jù)特征,深化對無人機(jī)遙感監(jiān)測作物病蟲害脅迫數(shù)據(jù)的認(rèn)識。通過建立適用性更強(qiáng)的無人機(jī)遙感監(jiān)測作物病蟲害脅迫算法或模型,從而構(gòu)建無人機(jī)遙感監(jiān)測作物病蟲害脅迫的方法庫以推動病蟲害遙感監(jiān)測領(lǐng)域的發(fā)展。

參考文獻(xiàn):

[1] CARVAJAL-YEPES M, CARDWELL K, NELSON A,et al. A global surveillance system for crop diseases[J]. Science, 2019, 364(6447):1237-1239.

[2] WANG C, WANG X, JIN Z, et al. Occurrence of croppests and diseases has largely increased in China since 1970[J]. Nature Food, 2021, 3:57-65.

[3] GOETZ A F H, VANE G, SOLOMON J E, et al. Imag‐ing spectrometry for earth remote sensing[J]. Science, 1985, 228(4704):1147-1153.

[4] WEISS M, JACOB F, DUVEILLER G. Remote sens‐ing for agricultural applications: A meta-review[J]. Re‐ mote Sensing of Environment, 2020, 236: ID 111402.

[5] RADOGLOU-GRAMMATIKIS? P,? SARIGIANNIDISP, LAGKAS T, et al. A compilation of UAV applica‐tions for precision agriculture[J]. Computer Networks, 2020, 172: ID 107148.

[6] SISHODIA R P, RAY R L, SINGH S K. Applicationsof remote sensing in precision agriculture: A review[J]. Remote Sensing, 2020, 12(19): ID 3136.

[7] PINTER JR P J, HATFIELD J L, SCHEPERS J S, etal. Remote? sensing? for? crop? management[J]. Photo‐grammetric Engineering & Remote Sensing, 2003, 69(6):647-664.

[8] WHITE J W, ANDRADE-SANCHEZ P, GORE M A,et? al. Field-based? phenomics? for? plant? genetics? re‐ search[J]. Field Crops Research, 2012, 133:101-112.

[9] WATTS A C, AMBROSIA V G, HINKLEY E A. Un‐manned aircraft systems in remote sensing and scientif‐ic research: Classification and considerations of use[J]. Remote Sensing, 2012, 4(6):1671-1692.

[10] BAGHERI? N. Application? of? aerial? remote? sensingtechnology? for? detection? of fire? blight? infected? pear trees[J]. Computers? and? Electronics? in? Agriculture, 2020, 168: ID 105147.

[11] LAN Y, HUANG Z, DENG X,? et? al. Comparison? ofmachine learning methods for citrus greening detection on UAV multispectral images[J]. Computers and Elec‐tronics in Agriculture, 2020, 171: ID 105234.

[12] CHIVASA? W,? MUTANGA? O,? BIRADAR? C. UAV-based multispectral phenotyping for disease resistanceto? accelerate? crop? improvement? under? changing? cli‐mate conditions[J]. Remote Sensing, 2020, 12(15): ID2445.

[13] CHIVASA W, MUTANGA O, BURGUENO? J. UAV-based high-throughput phenotyping to increase predic‐tion and selection accuracy in maize varieties under ar‐tificial MSV inoculation[J]. Computers and Electronicsin Agriculture, 2021, 184: ID 106128.

[14] SUGIURA R, NOGUCHI N, ISHII K. Remote-sensingtechnology? for? vegetation? monitoring? using? an? un‐manned? helicopter[J]. Biosystems? Engineering, 2005,90(4):369-379.

[15] BERNI J A J, ZARCO-TEJADA P J, SUAREZ L, et al.Thermal and narrowband multispectral remote sensingfor vegetation monitoring from an unmanned aerial ve‐hicle[J]. IEEE Transactions on Geoscience and RemoteSensing, 2009, 47(3):722-738.

[16] CORCOLES J I, ORTEGA J F, HERNANDEZ D, etal. Estimation of leaf area index in onion (Allium cepaL.) using? an? unmanned? aerial? vehicle[J]. BiosystemsEngineering, 2013, 115(1):31-42.

[17] WAHAB I, HALL O, JIRSTROM M. Remote sensingof yields: Application of UAV imagery-derived NDVIfor estimating maize vigor and yields in complex farm ‐ing systems in Sub-Saharan Africa[J]. Drones, 2018, 2(3): ID 28.

[18] YAO H,? QIN R,? CHEN X. Unmanned? aerial vehiclefor remote sensing applications—A review[J]. RemoteSensing, 2019, 11(12): ID 1443.

[19] SAARI? H,? PELLIKKA? I,? PESONEN? L,? et? al. Un‐manned Aerial Vehicle (UAV) operated? spectral cam ‐era system for forest and agriculture applications[C]//Remote Sensing for Agriculture, Ecosystems, and Hy‐drology XIII. International Society for Optics and Pho‐tonics, Prague, Czech Republic: SPIE, 2011, 8174: ID81740H.

[20] SADEQ? H? A. Accuracy? assessment? using? differentUAV image overlaps[J]. Journal of Unmanned VehicleSystems, 2019, 7(3):175-193.

[21] YANG G, LIU J, ZHAO C, et al. Unmanned aerial ve‐hicle remote sensing for field-based crop phenotyping:Current? status? and? perspectives[J]. Frontiers? in? PlantScience, 2017, 8: ID 1111.

[22] XIE C, YANG C. A review on plant high-throughputphenotyping traits using UAV-based? sensors[J]. Com ‐puters? and? Electronics? in Agriculture, 2020, 178: ID105731.

[23] XUE J, SU B. Significant remote sensing vegetation in‐dices: A review? of developments? and? applications[J].Journal of Sensors, 2017, 2017: ID 1353691.

[24] GITELSON A, ARKEBAUER T, VI?AA, et al. Evalu‐ating? plant? photosynthetic? traits? via? absorption? coef‐ficient? in? the? photosynthetically? active? radiation? re‐gion[J]. Remote? Sensing? of Environment, 2021, 258:ID 112401.

[25] HAUSER? L? T,? TIMMERMANS? J,? WINDT? NVANDER, et al. Explaining discrepancies between spectral and? in-situ? plant? diversity? in? multispectral? satellite earth observation[J]. Remote Sensing of Environment, 2021, 265: ID 112684.

[26] HORNERO?? A,?? HERN?NDEZ-CLEMENTE?? R,NORTH P R J, et al. Monitoring the incidence of Xylel‐ la fastidiosa infection in olive orchards using ground- based evaluations, airborne imaging spectroscopy and Sentinel-2 time? series? through 3-D? radiative? transfer modelling[J]. Remote? Sensing? of Environment, 2020, 236: ID 111480.

[27] PIGNATTI S, CASA R, LANEVE G, et al. Sino – EUearth? observation? data to? support the monitoring? and management of agricultural resources[J]. Remote Sens‐ing, 2021, 13(15): ID 2889.

[28] LEE J H,? SHIN J, REALFF M? J. Machine? learning:Overview of the recent progresses and implications for the process systems engineering field[J]. Computers & Chemical Engineering, 2018, 114:111-121.

[29] LIU J, XIANG J, JIN Y, et al. Boost precision agricul‐ture with unmanned aerial vehicle remote sensing and edge intelligence: A survey[J]. Remote Sensing, 2021, 13(21): ID 4387.

[30] VOULODIMOS A,? DOULAMIS N,? DOULAMIS A,et? al. Deep? learning? for? computer vision: A brief re‐ view[J]. Computational Intelligence and Neuroscience, 2018, 2018: ID 7068349.

[31] NAZIR M N M M, TERHEM R, NORHISHAM A R,et al. Early monitoring of health? status of plantation- grown eucalyptus pellita at large spatial scale via visi‐ble? spectrum? imaging? of? canopy? foliage? using? un‐ manned? aerial? vehicles[J]. Forests, 2021, 12(10): ID1393.

[32] DUTTA K, TALUKDAR D, BORA S S. Segmentationof unhealthy leaves in cruciferous crops for early dis‐ ease detection using vegetative indices and Otsu thresh‐olding? of aerial? images[J]. Measurement, 2022, 189:ID 110478.

[33] MAES W H, STEPPE K. Perspectives for remote sens‐ing with unmanned aerial vehicles in precision agricul‐ture[J]. Trends in Plant Science, 2019, 24(2):152-164.

[34] CALOU V B C, TEIXEIRA A D S, MOREIRA L C J,et al. The use of UAVs in monitoring yellow sigatoka in? banana[J]. Biosystems? Engineering, 2020, 193:115-125.

[35] DANG L M, HASSAN S I, SUHYEON I, et al. UAVbased? wilt? detection? system? via? convolutional? neuralnetworks[J]. Sustainable? Computing: Informatics? andSystems, 2020, 28: ID 100250.

[36] HUANG H, DENG J, LAN Y, et al. Detection of hel‐minthosporium leaf blotch disease based on UAV imag‐ery[J]. Applied Sciences, 2019, 9(3): ID 558.

[37] DEL-CAMPO-SANCHEZ? A,? BALLESTEROS? R,HERNANDEZ-LOPEZ D, et al. Quantifying the effectof Jacobiascalybica? pest? on? vineyards? with? UAVsby? combining? geometric? and? computer? vision? tech‐niques[J]. PLoS One, 2019, 14(4): ID e0215521.

[38] TETILA E? C, MACHADO B B, ASTOLFI? G,? et? al.Detection? and? classification? of? soybean? pests? usingdeep? learning? with? UAV? images[J]. Computers? andElectronics in Agriculture, 2020, 179: ID 105836.

[39] PADUA L, MARQUES P, MARTINS L, et al. Monitor‐ing of chestnut trees using machine learning techniquesapplied? to? UAV-based? multispectral? data[J]. RemoteSensing, 2020, 12(18): ID 3032.

[40] SU J, LIU C, HU X, et al. Spatio-temporal monitoringof wheat yellow? rust using? UAV multispectral? imag‐ery[J]. Computers and electronics in agriculture, 2019,167: ID 105035.

[41] ATSHAN L A, BROWN P, XU C, et al. Early? detec‐tion? of? disease? infection? in? chilli? crops? using? sen‐sors[C]// International? Horticultural? Congress,? Istan‐bul, Turkey: ISHS, 2018:263-270.

[42] YE H, HUANG W, HUANG S, et al. Recognition ofbanana fusarium wilt based on UAV remote sensing[J].Remote Sensing, 2020, 12(6): ID 938.

[43] KALISCHUK M, PARET M L, FREEMAN J H, et al.An improved crop scouting technique incorporating un‐manned aerial vehicle – assisted multispectral crop im‐aging? into? conventional? scouting practice? for? gummystem blight in watermelon[J]. Plant Disease, 2019, 103(7):1642-1650.

[44] MARSTON Z P D, CIRA T M, HODGSON E W, et al.Detection of stress induced by soybean aphid (Hemip‐tera: Aphididae) using multispectral imagery from un‐manned aerial vehicles[J]. Journal of Economic Ento‐mology, 2020, 113(2):779-786.

[45] VIERA-TORRES? M,? SINDE-GONZALEZ? I,? GIL-DOCAMPO? M,? et? al. Generating the baseline? in theearly detection of bud rot and red ring disease in oilpalms by geospatial technologies[J]. Remote? Sensing,2020, 12(19): ID 3229.

[46]趙晉陵, 金玉, 葉回春, 等.基于無人機(jī)多光譜影像的檳榔黃化病遙感監(jiān)測[J].農(nóng)業(yè)工程學(xué)報(bào), 2020, 36(8):54-61.

ZHAO J, JIN Y, YE H, et al. Remote sensing monitor‐ing of areca yellow leaf disease based on UAV multi- spectral images[J]. Transactions of the CSAE, 2020, 36(8):54-61.

[47] LEI S, LUO J, TAO X, et al. Remote sensing detectingof yellow leaf disease of arecanut based on UAV multi‐ source? sensors[J]. Remote? Sensing, 2021, 13(22): ID4562.

[48] DADRASJAVAN F,? SAMADZADEGAN F, POURA ‐ZAR? S? H? S,? et? al. UAV-based multispectral? imagery for? fast? citrus? greening? detection[J]. Journal? of Plant Diseases and Protection, 2019, 126(4):307-318.

[49] CHANG A, YEOM J, JUNG J, et al. Comparison ofcanopy shape and vegetation indices of citrus trees de‐ rived from UAV multispectral images for characteriza‐tion? of? citrus? greening? disease[J]. Remote? Sensing, 2020, 12(24): ID 4122.

[50] CHEN T, YANG W, ZHANG H, et al. Early detectionof bacterial wilt in peanut plants through leaf-level hy‐perspectral and unmanned aerial vehicle data[J]. Com ‐puters? and? Electronics? in Agriculture, 2020, 177: ID105708.

[51] HEIM R H J, WRIGHT I J, SCARTH P, et al. Multi‐spectral, aerial disease detection for myrtle rust (Aus‐tropucciniapsidii) on? a? lemon? myrtle? plantation[J]. Drones, 2019, 3(1): ID 25.

[52] LE?N-RUEDA W A,? LE?N? C,? CARO? S? G,? et? al.Identification? of diseases? and? physiological? disorders in? potato? via? multispectral? drone? imagery? using? ma‐ chine learning tools[J]. Tropical Plant Pathology, 2021:1-16.

[53] SU J, YI D,? SU B,? et? al. Aerial visual perception insmart farming: Field study of wheat yellow rust moni‐toring[J]. IEEE Transactions on Industrial Informatics, 2020, 17(3):2242-2249.

[54] DI NISIO A, ADAMO F, ACCIANI G, et al. Fast de‐tection of olive trees affected by xylella fastidiosa from UAVs? using? multispectral? imaging[J]. Sensors, 2020, 20(17): ID 4915.

[55]馬云強(qiáng), 李宇宸, 劉夢盈, 等.基于無人機(jī)多光譜影像的云南切梢小蠹危害監(jiān)測反演研究[J].西南農(nóng)業(yè)學(xué)報(bào), 2021, 34(9):1878-1884.

MA Y, LI Y, LIU M, et al. Harm monitoring and inver‐sion? study? on? Tomicusyunnanensis? based? on? multi- spectral? image? of unmanned? aerial? vehicle[J]. South‐ west China Journal of Agricultural Sciences, 2021, 34(9):1878-1184.

[56] IZZUDDIN M A, HAMZAH A, NISFARIZA M N, etal. Analysis of multispectral imagery from unmannedaerial vehicle (UAV) using object-based image analy‐sis for detection of Ganoderma disease in oil palm[J].Journal of Oil Palm Research, 2020, 32(3):497-508.

[57] RODR?GUEZ J, LIZARAZO I, PRIETO F, et al. As‐sessment of potato late blight from UAV-based multi‐spectral imagery[J]. Computers and Electronics in Ag‐riculture, 2021, 184: ID 106061.

[58] WANG? T,? THOMASSON? J A,? ISAKEIT T,? et? al. Aplant-by-plant method to identify and treat cotton rootrot based on UAV remote sensing[J]. Remote Sensing,2020, 12(15): ID 2453.

[59] ZHANG T, XU Z, SU J, et al. Ir-UNet: Irregular seg‐mentation u-shape network for wheat yellow rust detec‐tion by UAV multispectral? imagery[J]. Remote? Sens‐ing, 2021, 13(19): ID 3892.

[60]馬書英, 郭增長, 王雙亭, 等.板栗樹紅蜘蛛蟲害無人機(jī)高光譜遙感監(jiān)測研究[J].農(nóng)業(yè)機(jī)械學(xué)報(bào) , 2021, 52(4):171-180.

MA S, GUO Z, WANG S, et al. Hyperspectral remotesensing monitoring of Chinese chestnut red mite insectpests in UAV[J]. Transactions of the CSAM, 2021, 52(4):171-180.

[61]郭偉, 喬紅波, 趙恒謙, 等.基于比值導(dǎo)數(shù)法的棉花蚜害無人機(jī)成像光譜監(jiān)測模型研究[J].光譜學(xué)與光譜分析, 2021, 41(5):1543-1550.

GUO W, QIAO H, ZHAO H, et al. Cotton aphid dam ‐age monitoring using UAV hyperspectral data based onderivative? of ratio? spectroscopy[J]. Spectroscopy? andSpectral Analysis, 2021, 41(5):1543-1550.

[62]鄭蓓君, 陳蕓芝, 李凱, 等.高光譜數(shù)據(jù)的剛竹毒蛾蟲害程度檢測[J].光譜學(xué)與光譜分析 , 2021, 41(10):3200-3207.

ZHENG P, CHEN Y, LI K, et al. Detection of pest de‐gree? of? phyllostachys? Chinese? with? hyperspectraldata[J]. Spectroscopy and Spectral Analysis, 2021, 41(10):3200-3207.

[63]郭偉, 朱耀輝, 王慧芳, 等.基于無人機(jī)高光譜影像的冬小麥全蝕病監(jiān)測模型研究[J].農(nóng)業(yè)機(jī)械學(xué)報(bào) ,2019, 50(9):162-169.

GUO W, ZHU Y, WANG H, et al. Monitoring model ofwinter wheat take-all based on UAV hyperspectral im‐aging[J]. Transactions? of? the? CSAM, 2019, 50(9):162-169.

[64]梁輝, 何敬, 雷俊杰.無人機(jī)高光譜的玉米冠層大斑病監(jiān)測[J].光譜學(xué)與光譜分析 , 2020, 40(6):1965-1972.

LIANG? H,? HE? J,? LEI? J. Monitoring? of corn? canopyblight disease based on UAV hyperspectral method[J].Spectroscopy and Spectral Analysis, 2020, 40(6):1965-1972.

[65]桑佳茂, 陳豐農(nóng).基于光譜特征點(diǎn)秩和檢驗(yàn)的稻曲病發(fā)病程度檢測[J].光譜學(xué)與光譜分析 , 2021, 41(10):3214-3219.

SANG J, CHEN F. Detection of rice false smut grade degree based on the rank sum test of spectral feature points[J]. Spectroscopy and Spectral Analysis, 2021, 41(10):3214-3219.

[66] SONG P, ZHENG X, LI Y, et al. Estimating reed losscaused by Locusta migratoriamanilensis using UAV- based hyperspectral data[J]. Science of the Total Envi‐ronment, 2020, 719: ID 137519.

[67] LIU L, DONG Y, HUANG W, et al. Monitoring wheatFusarium? head? blight? using? unmanned? aerial? vehicle hyperspectral? imagery[J]. Remote? Sensing, 2020, 12(22): ID 3811.

[68] MA H, HUANG W, DONG Y, et al. Using UAV-basedhyperspectral imagery to detect winter wheat fusarium head blight[J]. Remote Sensing, 2021, 13(15): ID 3024.

[69] GUO A, HUANG W, DONG Y,? et? al. Wheat yellowrust detection using UAV-based hyperspectral technolo‐gy[J]. Remote Sensing, 2021, 13(1): ID 123.

[70]孔繁昌, 劉煥軍, 于滋洋, 等.高寒地區(qū)粳稻穗頸瘟的無人機(jī)高光譜遙感識別[J].農(nóng)業(yè)工程學(xué)報(bào) , 2020, 36(22):68-75.

KONG F, LIU H, YU Z, et al. Identification of japoni‐ ca rice panicle blast in alpine region by UAV hyper‐ spectral remote sensing[J]. Transactions of the CSAE, 2020, 36(22):68-75.

[71] DENG X, ZHU Z, YANG J, et al. Detection of citrusHuanglongbing? based? on? multi-input? neural? network model? of UAV? hyperspectral? remote? sensing[J]. Re‐ mote Sensing, 2020, 12(17): ID 2678.

[72] ABDULRIDHA? J,? BATUMAN? O,? AMPATZIDIS? Y.UAV-based? remote? sensing? technique? to? detect? citrus canker disease utilizing hyperspectral imaging and ma‐ chine? learning[J]. Remote? Sensing, 2019, 11(11): ID1373.

[73] ABDULRIDHA J, AMPATZIDIS Y,? ROBERTS? P,? etal. Detecting powdery mildew disease in squash at dif‐ferent? stages using UAV-based hyperspectral? imaging and? artificial? intelligence[J]. Biosystems? Engineering, 2020, 197:135-148.

[74] BOHNENKAMP D, BEHMANN J, MAHLEIN A K.In-field detection of yellow rust in wheat on the ground canopy? and UAV? scale[J]. Remote? Sensing, 2019, 11(21): ID 2495.

[75] ABDULRIDHA J, AMPATZIDIS Y, KAKARLA S C,et al. Detection of target spot and bacterial spot diseas‐ es in tomato using UAV-based and benchtop-based hy‐perspectral? imaging? techniques[J]. Precision? Agricul‐ture, 2020, 21(5):955-978.

[76] ABDULRIDHA J, AMPATZIDIS Y,? QURESHI? J,? etal. Laboratory and UAV-based identification and classi‐fication of tomato yellow leaf curl, bacterial spot, andtarget? spot? diseases? in? tomato? utilizing? hyperspectralimaging? and? machine? learning[J]. Remote? Sensing,2020, 12(17): ID 2732.

[77]鄧小玲, 曾國亮, 朱梓豪, 等.基于無人機(jī)高光譜遙感的柑橘患病植株分類與特征波段提取[J].華南農(nóng)業(yè)大學(xué)學(xué)報(bào), 2020, 41(6):100-108.

DENG X, ZENG G, ZHU Z, et al. Classification andfeature band extraction of diseased citrus plants basedon? UAV? hyperspectral? remote? sensing[J]. Journal? ofSouth? China? Agricultural? University, 2020, 41(6):100-108.

[78] MACDONALD S L, STAID M, STAID M, et al. Re‐mote hyperspectral imaging of grapevine leafroll-asso‐ciated virus 3 in cabernet sauvignon vineyards[J]. Com ‐puters? and? Electronics? in? Agriculture, 2016, 130:109-117.

[79] GARCIA-RUIZ F, SANKARAN S, MAJA J M, et al.Comparison of two aerial imaging platforms for identi‐fication? of? Huanglongbing-infected? citrus? trees[J].Computers? and? Electronics? in Agriculture, 2013, 91:106-115.

[80] MCCANN C, REPASKY K S, LAWRENCE R, et al.Multi-temporal mesoscale hyperspectral data of mixedagricultural and grassland regions? for anomaly detec‐tion[J]. ISPRS Journal of Photogrammetry and RemoteSensing, 2017, 131:121-133.

[81] ZHANG X, HAN L, DONG Y, et al. A deep? learning-based? approach? for? automated? yellow? rust? diseasedetection from high-resolution hyperspectral UAV im‐ages[J]. Remote Sensing, 2019, 11(13): ID 1554.

[82] AN G, XING M, HE B, et al. Extraction? of areas? ofrice false smut infection using UAV hyperspectral da‐ta[J]. Remote Sensing, 2021, 13(16): ID 3185.

[83] XIAO Y, DONG Y, HUANG W, et al. Wheat fusariumhead? blight? detection? using? UAV-based? spectral? andtexture? features? in? optimal? window? size[J]. RemoteSensing, 2021, 13(13): ID 2437.

[84] FRANCESCONI S, HARFOUCHE A, MAESANO M,et al. UAV-based thermal, RGB imaging and gene ex‐pression? analysis? allowed? detection? of fusarium? headblight and gave new insights into the physiological re‐sponses to the disease in durum wheat[J]. Frontiers inPlant Science, 2021, 12: ID 551.

[85]陳欣欣, 劉子毅, 呂美巧, 等.基于熱紅外成像技術(shù)的油菜菌核病早期檢測研究[J].光譜學(xué)與光譜分析 ,2019, 39(3):730-737.

CHEN X, LIU Z, LYU M, et al. Diagnosis and moni‐toring of sclerotinia stem rot of oilseed rape using ther‐ mal? infrared? imaging[J]. Spectroscopy? and? Spectral Analysis, 2019, 39(3):730-737.

[86] LIN Q, HUANG H, WANG J, et al. Detection of pineshoot beetle (PSB) stress on pine forests at individual tree level using UAV-based hyperspectral imagery and lidar[J]. Remote Sensing, 2019, 11(21): ID 2540.

[87] BRIECHLE S, KRZYSTEK P, VOSSELMAN G. Clas‐sification of tree species and standing dead trees by fus‐ing UAV-based lidar data and multispectral imagery in the 3D deep neural network PointNet++[J]. ISPRS An‐nals of the Photogrammetry, Remote Sensing and Spa‐tial Information Sciences, 2020, 2:203-210.

[88] SAVIAN F, MARTINI M, ERMACORA P, et al. Pre‐diction? of the kiwifruit? decline? syndrome? in? diseased orchards by remote sensing[J]. Remote Sensing, 2020, 12(14): ID 2194.

[89] YU R, LUO Y, ZHOU Q, et al. A machine learning al‐gorithm? to? detect pine wilt? disease using UAV-based hyperspectral? imagery? and? LiDAR? data? at? the? tree level[J]. International Journal of Applied Earth Obser‐vation and Geoinformation, 2021, 101: ID 102363.

[90] SMIGAJ M, GAULTON R, SUAREZ J C, et al. Cano‐py temperature from an unmanned aerial vehicle as an indicator of tree stress associated with red band needle blight? severity[J]. Forest? Ecology? and? Management, 2019, 433:699-708.

[91] KERKECH? M,? HAFIANE A,? CANALS? R. VddNet:Vine disease detection network based on multispectral images? and? depth map[J]. Remote? Sensing, 2020, 12(20): ID 3305.

[92] KERKECH M, HAFIANE A, CANALS R. Vine dis‐ease detection in UAV multispectral images using opti‐mized image registration and deep learning segmenta‐tion? approach[J]. Computers? and Electronics? in Agri‐ culture, 2020, 174: ID 105446.

[93] MORIYA ? A S, IMAI N N, TOMMASELLI A M G,et al. Detection and mapping of trees infected with cit‐rus gummosis using UAV hyperspectral data[J]. Com ‐puters? and? Electronics? in Agriculture, 2021, 188: ID106298.

[94]趙曉陽, 張建, 張東彥, 等.低空遙感平臺下可見光與多光譜傳感器在水稻紋枯病病害評估中的效果對比研究[J].光譜學(xué)與光譜分析, 2019, 39(4):1192-1198.

ZHAO X, ZHANG J, ZHANG D,? et al. Comparison between? the? effects? of visible? light? and? multispectral sensor based on low-altitude remote? sensing platform in the evaluation of rice sheath blight[J]. Spectroscopy and Spectral Analysis, 2019, 39(4):1192-1198.

[95] DANG L, WANG H, LI Y, et al. Fusarium wilt of rad‐ish detection using RGB and near infrared images fromunmanned aerial vehicles[J]. Remote Sensing, 2020, 12(17): ID 2863.

[96] AWAIS M, LI W, CHEEMA M J M, et al. Remotelysensed? identification? of? canopy? characteristics? usingUAV-based imagery under unstable environmental con‐ditions[J]. Environmental? Technology & Innovation,2021, 22: ID 101465.

[97] FENG L, CHEN S, ZHANG C, et al. A comprehensivereview on recent applications of unmanned aerial vehi‐cle? remote? sensing? with? various? sensors? for? high-throughput plant phenotyping[J]. Computers and Elec‐tronics in Agriculture, 2021, 182: ID 106033.

[98] RUBIO-HERVAS J, GUPTA A, ONG Y S. Data-drivenrisk assessment and multicriteria optimization of UAVoperations[J]. Aerospace? Science? and? Technology,2018, 77:510-523.

[99] HUSSEIN M, NOUACER R, CORRADI F, et al. Keytechnologies? for? safe? and? autonomous? drones[J]. Mi‐croprocessors and Microsystems, 2021, 87: ID 104348.

[100]MUKHERJEE A, MISRA S, RAGHUWANSHI N S. Asurvey of unmanned aerial sensing solutions in preci‐sion agriculture[J]. Journal of Network and ComputerApplications, 2019, 148: ID 102461.

[101]黃文江, 師越, 董瑩瑩, 等.作物病蟲害遙感監(jiān)測研究進(jìn)展與展望[J].智慧農(nóng)業(yè), 2019, 1(4):1-11.

HUANG W, SHI Y, DONG Y, et al. Progress and pros‐pects of crop diseases and pests monitoring by remotesensing[J]. Smart Agriculture, 2019, 1(4):1-11.

[102] MIGNONI? M? E,? HONORATO A,? KUNST? R,? et? al.Soybean images dataset for caterpillar and Diabroticaspeciosa? pest? detection? and? classification[J]. Data? inBrief, 2021: ID 107756.

[103]OUHAMI M, HAFIANE A, ES-SAADY Y, et al. Com ‐puter vision, IoT and data fusion for crop disease detec‐tion using machine learning: A survey and ongoing re‐search[J]. Remote Sensing, 2021, 13(13): ID 2486.

[104] LI F, LIU Z, SHEN W, et al. A remote sensing and air‐borne edge-computing based detection system for pinewilt disease[J]. IEEE Access, 2021, 9:66346-66360.

[105]蘭玉彬, 鄧小玲, 曾國亮.無人機(jī)農(nóng)業(yè)遙感在農(nóng)作物病蟲草害診斷應(yīng)用研究進(jìn)展[J].智慧農(nóng)業(yè) , 2019, 1(2):1-19.

LAN Y, DENG X, ZENG G. Advances in diagnosis ofcrop diseases, pests and weeds by UAV remote sens‐ing[J]. Smart Agriculture, 2019, 1(2):1-19.

[106] ELNABTY I A, FAHMY Y, KAFAFY M. A survey onUAV placement optimization for UAV-assisted commu‐nication in 5G and beyond networks[J]. Physical Com ‐munication, 2021: ID 101564.

[107] MISHRA D, NATALIZIO E. A survey on cellular-con‐nected UAVs: Design challenges, enabling 5G/B5G in‐novations, and experimental advancements[J]. Comput‐er Networks, 2020, 182: ID 107451.

Methods and New Research Progress of Remote Sensing Monitoring of Crop Disease and Pest Stress UsingUnmanned Aerial Vehicle

YANG Guofeng1,2,3 , HE Yong1,2,3* , FENG Xuping1,2,3 , LI Xiyao1,2,3 , ZHANG Jinnuo1,2,3 , YU Zeyu1,2,3

(1. Huanan Industrial Technology Research Institute of Zhejiang University, Guangzhou 510700, China;

2. College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China;

3. The Key Laboratory of Spectroscopy Sensing, Ministry of Agriculture and Rural Affairs, Hangzhou 310058, China )

Abstract: Diseases and pests are main stresses to crop production. It is necessary to accurately and quickly monitor and control the stresses dynamically, so as to ensure the food security and the quality and safety of agricultural products, protect the ecologi‐cal environment, and promote the sustainable development of agriculture. In recent years, with the rapid development of the un‐ manned aerial vehicle (UAV) industry, UAV agricultural remote sensing has played an important role in the application of crop diseases and pests monitoring due to its high image spatial resolution, strong data acquisition timeliness and low cost. The rele‐vant background of UAV remote sensing monitoring of crop disease and pest stress was introduced, then the current methods commonly used in remote sensing monitoring of crop disease and pest stress by UAV was summarized. The data acquisition method and data processing method of UAV remote sensing monitoring of crop disease and pest stress were mainly discussed. Then, from the six aspects of visible light imaging remote sensing, multispectral imaging remote sensing, hyperspectral imaging remote sensing, thermal infrared imaging remote sensing, LiDAR imaging remote sensing and multiple remote sensing fusion and comparison, the research progress of remote sensing monitoring of crop diseases and pests by UAV worldwide was re‐ viewed. Finally, the unresolved key technical problems and future development directions in the research and application of UAV remote sensing monitoring of crop disease and pest stress were proposed. Such as, the performance of the UAV flight plat‐ form needs to be optimized and upgraded, as well as the development of low-cost, lightweight, modular, and more adaptable air‐ borne sensors. Convenient and automated remote sensing monitoring tasks need to be designed and implemented, and more re‐ mote sensing monitoring information can be obtained. Data processing algorithms or software should be designed and devel‐oped with greater applicability and wider applicability, and data processing time should be shortened by using 5G-based commu‐nication networks and edge computing devices. The applicability of the algorithm or model for UAV remote sensing monitoring of crop disease and pest stress needs to be stronger, so as to build a corresponding method library. We hope that this paper can help Chinese UAV remote sensing monitoring of crop diseases and pests to achieve more standardization, informatization, preci‐sion and intelligence.

Key words: unmanned aerial vehicle; remote sensing monitoring; diseases and pests stress; data acquisition; data processing; deep learning; multiple remote sensing fusion

猜你喜歡
無人機(jī)數(shù)據(jù)處理深度學(xué)習(xí)
認(rèn)知診斷缺失數(shù)據(jù)處理方法的比較:零替換、多重插補(bǔ)與極大似然估計(jì)法*
ILWT-EEMD數(shù)據(jù)處理的ELM滾動軸承故障診斷
MOOC與翻轉(zhuǎn)課堂融合的深度學(xué)習(xí)場域建構(gòu)
大數(shù)據(jù)技術(shù)在反恐怖主義中的應(yīng)用展望
深度學(xué)習(xí)算法應(yīng)用于巖石圖像處理的可行性研究
高職院校新開設(shè)無人機(jī)專業(yè)的探討
基于深度卷積網(wǎng)絡(luò)的人臉年齡分析算法與實(shí)現(xiàn)
一種適用于輸電線路跨線牽引無人機(jī)的飛行方案設(shè)計(jì)
基于希爾伯特- 黃變換的去噪法在外測數(shù)據(jù)處理中的應(yīng)用
基于POS AV610與PPP的車輛導(dǎo)航數(shù)據(jù)處理