999精品在线视频,手机成人午夜在线视频,久久不卡国产精品无码,中日无码在线观看,成人av手机在线观看,日韩精品亚洲一区中文字幕,亚洲av无码人妻,四虎国产在线观看 ?

機(jī)器學(xué)習(xí)方法在植物表型分析中的應(yīng)用研究現(xiàn)狀

2025-08-03 00:00:00管思彤張兆旭林一鳴蘇培森黃思羅孟憲勇柳平增顏君
山東農(nóng)業(yè)科學(xué) 2025年6期
關(guān)鍵詞:表型機(jī)器光譜

中圖分類號:S126:Q94 文獻(xiàn)標(biāo)識號:A 文章編號:1001-4942(2025)06-0158-13

AbstractPlant phenotypes are products of the interaction between genotypes and environment,and are theexternal manifestation of plant life activities,which cover multiple levels and dimensions of plant characteristics,including morphology,physiology and biochemistry.Research on plant phenotypes is a key link in breeding,which is of great significance for revealing related mechanisms of plant life activities,cultivating high-yield,high-quality and efficient crop varieties,and realizing precise management of agricultural production.With the development and application of high-throughput plant phenotype collection technology,plant phenotype data present characteristics of high-dimensionality,multi-source, heterogeneity and dynamics, which bring new opportunities and challenges for plant phenotype analysis. Machine learning,as a powerful tool for data mining and knowledge discovery,can extract useful features and patterns from complex phenotype data,providing new ideas and methods for plant phenotype research.This paper systematically reviewed the application and progress of machine learning methods in plant phenotype research,focusing on their application in the analysis of plant morphological structure,stress resistance and biochemical components,as wellas in crop improvement and yield prediction.The problems and future development directions of machine learning methods in plant phenotype research were also discussed,aiming to provide beneficial references and inspiration for future research in this field.

KeywordsPlant phenotypic analysis; Machine learning; Crop improvement; Yield prediction

植物表型是植物在特定環(huán)境條件下展現(xiàn)出的形態(tài)、生理、生化特征[1],反映了植物基因圖譜的時序三維表達(dá),以及在不同地域和代際間的變化規(guī)律[2-3]。植物表型研究為探索植物的生命活動規(guī)律、揭示其適應(yīng)性機(jī)制、改良作物產(chǎn)量等性狀提供了重要基礎(chǔ)[4]。隨著高通量表型采集技術(shù)的飛速發(fā)展,多角度和多層次的植物表型數(shù)據(jù)不斷涌現(xiàn),為深入挖掘植物“表型-基因型”關(guān)系提供了豐富的數(shù)據(jù)資源。機(jī)器學(xué)習(xí)(machine learming,ML)是一種基于數(shù)據(jù)驅(qū)動的智能計算方法,可以從大量數(shù)據(jù)中提取有用信息[5],為植物表型研究提供了新的思路和方法[6]。早在2010年,ML方法就被應(yīng)用于從復(fù)雜的表型數(shù)據(jù)中篩選出與作物抗逆性、產(chǎn)量構(gòu)成因子及品質(zhì)指標(biāo)相關(guān)的關(guān)鍵特征。例如,Steinfath 等[7]利用偏最小二乘(partialleast squares,PLS)法從馬鈴薯(Solanum tuberos-um)塊莖中篩選出果糖、葡萄糖等化合物作為生物標(biāo)志物。Yang等[8]通過逐步多元回歸(step-wisemultipleregression,SMR)和支持向量機(jī)(supportvectormachine,SVM)算法處理高光譜( 350~2500nm )遙感圖像數(shù)據(jù),估算水稻(Oryzasativa)的葉面積指數(shù)(leafarea index,LAI)、葉片葉綠素密度(green leaf chlorophyll density,GLCD)等參數(shù)。此外,隨著組學(xué)數(shù)據(jù)的積累,ML與生物信息學(xué)方法的結(jié)合為探索植物表型與基因組、轉(zhuǎn)錄組、代謝組等組學(xué)數(shù)據(jù)之間的關(guān)系提供了幫助[。目前機(jī)器學(xué)習(xí)方法在植物表型分析中的應(yīng)用主要集中在植物形態(tài)結(jié)構(gòu)、脅迫抗性、生化組分、作物改良、產(chǎn)量預(yù)測等方面。因此,本文總結(jié)了近年來ML在這些方面的研究進(jìn)展,旨在為今后更深入地研究植物生長發(fā)育的分子機(jī)制、作物基因改良等提供一定的參考。

1 機(jī)器學(xué)習(xí)在植物形態(tài)結(jié)構(gòu)特征研究中的應(yīng)用

植物形態(tài)結(jié)構(gòu)的研究主要是根、莖、葉等器官形態(tài)結(jié)構(gòu)特征的研究,例如形狀、大小、顏色等可觀測屬性。應(yīng)用ML方法研究植物形態(tài)結(jié)構(gòu)特征的相關(guān)處理流程一般分為5步(圖1): ① 數(shù)據(jù)收集,利用照相機(jī)、傳感器等測量儀器收集植物的株高等表型數(shù)據(jù),以及通過測序分析得到的單核苷酸多態(tài)性(single nucleotide polymorphisms,SNPs)等基因型數(shù)據(jù); ② 數(shù)據(jù)處理,對收集的數(shù)據(jù)進(jìn)行清洗、缺失值處理、標(biāo)準(zhǔn)化、歸一化等操作,確保數(shù)據(jù)質(zhì)量; ③ 特征提取,根據(jù)任務(wù)需求提取相關(guān)特征;④ 模型訓(xùn)練,選擇合適的ML模型進(jìn)行優(yōu)化、訓(xùn)練和評估,調(diào)整參數(shù)提高泛化能力; ⑤ 性狀分析,將訓(xùn)練好的模型應(yīng)用于新樣本,進(jìn)行預(yù)測和推斷。本部分總結(jié)了機(jī)器學(xué)習(xí)方法在植物葉面積、果實特征、根系結(jié)構(gòu)(root system architecture,RSA)以及其他形態(tài)結(jié)構(gòu)特征研究中應(yīng)用的最新進(jìn)展

1.1 葉面積

葉面積是評估植物生長狀況的重要指標(biāo),而葉面積指數(shù)(LAI)是生態(tài)系統(tǒng)研究中描述植被冠層結(jié)構(gòu)的關(guān)鍵變量,直接影響到植被的光合作用效率、蒸騰作用效率、能量平衡狀態(tài),通常定義為單位地表面積上葉面積總和的一半。近年來,機(jī)器學(xué)習(xí)方法逐步應(yīng)用于LAI的預(yù)測,研究者們通過無人機(jī)、遙感衛(wèi)星等采集圖像數(shù)據(jù),提取圖像的植被指數(shù)或紋理特征作為LAI的預(yù)測變量,結(jié)合人工神經(jīng)網(wǎng)絡(luò)(artificial neural network,ANN)、反向傳播神經(jīng)網(wǎng)絡(luò)(back propagation neural network,BPNN)、隨機(jī)森林(random forest,RF)、集成學(xué)習(xí)(ensemblelearning)等方法,建立LAI與特征之間的回歸模型,已在花生(Arachis hypogaea)、玉米(Zeamays)等作物的LAI預(yù)測方面有所突破[10-12]。Sun等[13]通過提取玉米無人機(jī)圖像的差分紋理指數(shù)(difference texture index,DTI)、比值紋理指數(shù)(ratio texture index,RTI)、歸一化差分紋理指數(shù)(normalized difference texture index,NDTI)三種特征,采用SVM算法進(jìn)行訓(xùn)練預(yù)測LAI,發(fā)現(xiàn)通過紅色、紅邊、近紅外波段計算得到的五種植被指數(shù)與LAI具有較高的相關(guān)性,包括比值植被指數(shù)(ratio vegetation index,RVI)、歸一化差分植被指數(shù)(normalized differencevegetationindex,NDVI)、優(yōu)化土壤調(diào)節(jié)植被指數(shù)(optimizedsoil adjusted vegetation index,OSAVI)、歸一化差分紅邊植被指數(shù)(normalized difference red edgevegetationindex,NDRE)、紅邊葉綠素植被指數(shù)(red edge chlorophyll index, CIrededge )。 Zhang等[14]通過偏最小二乘回歸(partial least squaresregression,PLSR)、支持向量回歸(support vectorregression,SVR)、極端梯度增強(qiáng)(extreme gradientboosting,XGBoost),結(jié)合競爭自適應(yīng)重加權(quán)采樣( competitiveadapativereweightedsampling,CARS)與連續(xù)投影算法(successive projections al-gorithm,SPA)從無人機(jī)高光譜成像(hyperspectralimaging,HSI)中提取特征,構(gòu)建三種小麥(Tritic-umaestivum)LAI預(yù)測模型,其中XGBoost模型為最佳模型,可用于估算小麥生長信息。此外,結(jié)合輻射傳輸(radiativetransfer)模型和隨機(jī)森林回歸(randomforestregression,RFR)模型,根據(jù)在野外條件下獲得的無人機(jī)多光譜圖像計算冠層反射率,可準(zhǔn)確地預(yù)測小麥LAI的時空變化,幫助鑒定不同生長時期、不同處理方式、不同小麥品種等帶來的LAI差異,進(jìn)而可指導(dǎo)優(yōu)化耕作方案(種植密度、灌溉、施肥等)[15] C

圖1機(jī)器學(xué)習(xí)在植物形態(tài)結(jié)構(gòu)研究中的相關(guān)處理流程

然而,傳統(tǒng)機(jī)器學(xué)習(xí)方法在特征提取方面通常較為復(fù)雜和耗時,且難以保證特征提取質(zhì)量,適用性較為有限。近年來,深度學(xué)習(xí)(deeplearning,DL)在預(yù)測LAI方面表現(xiàn)出更突出的優(yōu)勢。Ilniyaz等[16]基于深度殘差神經(jīng)網(wǎng)絡(luò)(deepresidu-alnetwork,ResNet)預(yù)測葡萄(Vitisvinifera)的LAI,該模型直接采用裁剪圖像作為輸入,無需手動提取特征。Li等[17]提出一種作物雙學(xué)習(xí)生成對抗網(wǎng)絡(luò)模型(命名為CROP-DualGAN),用于小樣本數(shù)據(jù)的增強(qiáng),可預(yù)測谷物、玉米、油菜(Brassicanapus)等的LAI。Wang等[18]基于卷積神經(jīng)網(wǎng)絡(luò)(convolutionalneuralnetwork,CNN)和生成對抗網(wǎng)絡(luò)(generative adversarial network,GAN)制作了新型無線LAI傳感器(名為LAINET),并用其采集和分割樹冠圖像,提取植被間隙分?jǐn)?shù)來估算LAI值,結(jié)果表明語義分割(即CNN部分)的平均精度可達(dá) 97.80% ,通過GAN進(jìn)行圖像超分辨率重建,可使間隙分?jǐn)?shù)測量精度提高 5.50% 以上。

1.2 果實特征

植物果實特征包括果實的重量、顏色、大小等形態(tài)特征。通過ML方法識別與高品質(zhì)果實特征相關(guān)的基因,有助于培育優(yōu)質(zhì)果蔬品種。例如,Kang等[19]利用全基因組關(guān)聯(lián)研究(genome-wideassociationstudies,GWAS)鑒定了與日本栗(Cas-taneacrenata)重量相關(guān)的SNPs,并結(jié)合SVM、決策樹(decision tree,DT)、K-最鄰近分類(K-nea-restneighbors,KNN)、PLS、RF等方法預(yù)測日本栗的重量,這些SNPs可作為分子標(biāo)記輔助選育栗品種。Liu等[20]開發(fā)了一種基于極限學(xué)習(xí)機(jī)(ex-tremelearningmachine,ELM)集成貝葉斯方法的新模型(命名為BELM),用于預(yù)測干燥過程中杏鮑菇(Pleurotuseryngi)的顏色變化,結(jié)果表明,干燥溫度和空氣流速對杏鮑菇顏色有顯著影響,BLEM在預(yù)測過程中表現(xiàn)出魯棒性,模型相對誤差低于 8.50% 。Tong等[21]利用基因組選擇(ge-nomicselection,GS)方法預(yù)測兩個茄科作物番茄(Solanum lycopersicum)和辣椒(Capsicum annu-um)的形態(tài)計量學(xué)特征和比色特征,通過基于脊回歸最佳線性無偏預(yù)測(best linearunbiased pre-diction,BLUP)的GS模型預(yù)測茄科果實的形態(tài)和顏色等相關(guān)特征。

果實圖像可直觀地展示果實的形態(tài)特征Minamikawa等[2]通過多元線性回歸(multiplelinearregression,MLR)RF、貝葉斯網(wǎng)絡(luò)分析多個柑橘屬(Citrus)品種果實的形態(tài)特征與易剝性(果實硬度)之間的關(guān)聯(lián)性,發(fā)現(xiàn)果實核心的降解面積與去皮和硬度相關(guān),種子面積僅與硬度相關(guān),為柑橘的品質(zhì)評價提供了一種新方法。Cetin等[23]利用高光譜相機(jī)采集了100個蘋果(Malusdomestica)果實在 386~1028nm 范圍內(nèi)的反射率數(shù)據(jù),通過ANN、PLSR、MLR、DT、KNN等方法對不同收獲期的蘋果硬度進(jìn)行預(yù)測,結(jié)果顯示高光譜成像結(jié)合ANN和DT方法對硬度預(yù)測最有效。不過,基于果實二維圖像的識別分類結(jié)果易受到光照變化、葉枝遮擋、果實重疊等因素的影響,而基于果實三維圖像的識別方法可有效克服上述缺點。近年來,CNN在果實三維圖像分割方面取得了不少進(jìn)展。例如,三維卷積神經(jīng)網(wǎng)絡(luò)可對X射線CT掃描的甜菜(Betavulgaris)果實和種子圖像進(jìn)行語義分割[24],還可對立體相機(jī)和飛行時間相機(jī)拍攝的草莓 (Fragaria×ananassa) 圖像進(jìn)行形狀分割[25]。Nyalala等[26]基于計算機(jī)視覺與ML算法開發(fā)了預(yù)測番茄質(zhì)量與體積的方法,通過提取番茄果實的二維和三維圖像特征,建立了基于徑向基函數(shù)核(radialbasisfunctionkernel,RBF核)的SVM回歸模型(RBF-SVM),該模型對番茄質(zhì)量和體積的估計精度分別為 97.06% 和 96.94% 。Chen 等[27]基于視覺顯著性和CNN提出了一種識別柑橘果實三種成熟度等級的方法:第一階段使用YOLOv5對圖像中的柑橘果實進(jìn)行目標(biāo)檢測,第二階段將RGB圖像和基于改進(jìn)最大對稱環(huán)繞顯著性檢測算法(maximum symmetric surroundsaliencydetection)生成的顏色對比度熱力圖相結(jié)合,使用4通道ResNet34網(wǎng)絡(luò)判斷果實的成熟度等級。

1.3 根系結(jié)構(gòu)

根系結(jié)構(gòu)(RSA)是描述植物根系形態(tài)和空間分布的指標(biāo),能反映植物對土壤環(huán)境的適應(yīng)能力。回歸或分類模型(多項式回歸、SVM、ANN等)可基于光譜和熱成像數(shù)據(jù)無創(chuàng)地測量小型植物的根系深度[28]。Wedger等[29]利用RF模型研究雜草稻(一種具有雜草特性的變種水稻)RSA的演化及遺傳機(jī)制,通過分析98個根系特征,發(fā)現(xiàn)雜草稻與水稻在根系表型上獨立演化。Awika等[30]基于ranger超參數(shù)網(wǎng)格搜索增強(qiáng)的RF方法研究不同土壤氮濃度下菠菜(Spinaciaoleracea)RSA特征與莖重量(shootweight)的關(guān)系,結(jié)果表明,在高氮條件下,根尖數(shù)和根長與莖重量相關(guān)性最強(qiáng);在低氮條件下,分枝點數(shù)量和根平均直徑與莖重量相關(guān)性最強(qiáng)。Changdar等[31使用RF模型分析小麥根系在不同土壤深度的分布及其與深層土壤氮吸收和干旱恢復(fù)潛力之間的關(guān)系,使用微根管(minirhi-zotron)技術(shù)采集了 80~250cm 深度范圍的根圖像,通過CNN提取根在土壤中的分布密度,結(jié)果表明,小麥根的分布密度與同位素示蹤劑 15N 和干旱脅迫指標(biāo) 13C 的吸收存在明顯的相關(guān)性,其中 150~170cm 深度的土層 15N 和 13C 同位素含量最多。

近年來,ML成為苜蓿(Medicagosativa)根系選擇的一種新方法,通過選擇分枝更多、根系更發(fā)達(dá)的首蓿,可以增加苜蓿生長過程中有機(jī)物質(zhì)的積累[32]。Bucciarelli等[33]通過14日齡紫花苜蓿幼苗的表型分析確定可測量的根系性狀,利用平板掃描儀進(jìn)行根系成像,然后使用RF和梯度提升機(jī)(gradientboostingmachine,GBM)將其分為分枝根型和直根型苜蓿,可以快速篩選出目標(biāo)表型的苜蓿幼苗。 Xu 等[34]利用RF和神經(jīng)網(wǎng)絡(luò)對617株成熟首蓿的RSA圖像進(jìn)行客觀分類,可成功將根類型分為分枝型、主根型、中間主根分枝型,準(zhǔn)確率高達(dá) 97.00% 。

1.4 其他形態(tài)結(jié)構(gòu)特征

機(jī)器學(xué)習(xí)方法也應(yīng)用于植物株高、花色、冠層覆蓋等形態(tài)結(jié)構(gòu)特征的研究。例如,RF能夠揭示玉米株高相關(guān)的遺傳變異[35],還能識別桔梗(Platycodongrandiflorus)花色相關(guān)的關(guān)鍵SNPs[36],為玉米和桔梗的育種提供了幫助。Bue-no 等[37]使用 ANN、SVM、RF 等算法對巴西塞拉多草原的200棵喬木的樹冠直徑進(jìn)行評估,ANN表現(xiàn)最佳。Apriyanti等[38]構(gòu)建了包含蘭花(Cym-bidium)種類、文本描述、數(shù)字圖像在內(nèi)的數(shù)據(jù)集,使用五種神經(jīng)網(wǎng)絡(luò)架構(gòu)(VGG16、Inception、Res-net50、Xception、Nasnet)確定遷移學(xué)習(xí)的最佳方案,無需進(jìn)行圖像分割即可高效檢測出花唇瓣的顏色。研究人員發(fā)現(xiàn)隨機(jī)森林回歸(RFR)在分析衛(wèi)星數(shù)據(jù)時表現(xiàn)出色。Xie等[39使用RFR和SVR方法,利用加拿大農(nóng)業(yè)地區(qū)多年的星載極化合成孔徑雷達(dá)(polarimetric synthetic aperture ra-dar)數(shù)據(jù)反演玉米的高度。Iizuka等[40]用RFR和SVR方法,通過無人機(jī)和衛(wèi)星獲取冠層高度、冠層大小、冠層覆蓋等數(shù)據(jù),并將其作為輸人變量訓(xùn)練模型以預(yù)測日本針葉林的森林結(jié)構(gòu)參數(shù),結(jié)果表明RFR精度更高。

2 機(jī)器學(xué)習(xí)在脅迫環(huán)境下植物抗逆性研究中的應(yīng)用

2.1 機(jī)器學(xué)習(xí)在植物生物脅迫研究中的應(yīng)用

生物脅迫是指植物生長和發(fā)育過程中各種病原體和害蟲侵襲導(dǎo)致的生存壓力。研究植物生物脅迫的抗性機(jī)制,有助于揭示植物在病蟲害脅迫下信號通路的改變和基因網(wǎng)絡(luò)的應(yīng)答等分子機(jī)理,這些分子層面的調(diào)控網(wǎng)絡(luò)直接影響到植物的表型特征。機(jī)器學(xué)習(xí)方法可以通過構(gòu)建植物的基因網(wǎng)絡(luò)和表型圖譜,揭示植物的抗病機(jī)制和遺傳變異。Cao 等[41]利用大規(guī)模RNA 測序數(shù)據(jù),從馬鈴薯塊莖中鑒定出2857個長鏈非編碼RNA(long non-codingRNAs,lncRNAs)和 33 150 個信使RNA(messengerRNAs,mRNAs),并采用RF模型構(gòu)建lncRNAs和mRNAs的交互網(wǎng)絡(luò),預(yù)測與抗病相關(guān)的lncRNAs和靶基因。Chi等[42]從兩方面分析大豆疫霉病菌(Phytophthorasojae)侵染大豆(Glycinemax)根莖后急劇表達(dá)的768個小RNA(smallRNA,sRNA)序列:大豆sRNA靶向節(jié)點的功能富集分析表明,存在多個GO(Gene Ontology)生物過程(biological process,BP)和 KEGG(KyotoEncyclopedia of Genes and Genomes)信號通路與大豆生長防御和反向抗性相關(guān);通過對比 RF、SVM和XGBoost三種ML方法構(gòu)建的大豆抗病sRNA預(yù)測模型,發(fā)現(xiàn)XGBoost預(yù)測效果最好(準(zhǔn)確率為 86.98% )。Bhattarai等[43]繪制了菠菜葉片上霜霉病(downymildew)抗性基因座RPF3的精細(xì)圖,使用嶺回歸BLUP、貝葉斯模型Bayes B、貝葉斯LASSO、貝葉斯嶺回歸、SVM、基因組預(yù)測(genomicprediction,GP)等方法預(yù)測抗性相關(guān)的SNPs。

蟲害是另一種常見的植物生物脅迫。CanellaVieira等[44]利用基于RF和SVM的大豆根部GWAS方法預(yù)測大豆基因組中與美國南方根結(jié)線蟲(Meloidogyneincognita)抗性相關(guān)的染色體新區(qū)域,發(fā)現(xiàn)除了已知的位于10號染色體上的主效數(shù)量性狀位點(quantitativetraitlocus,QTL)之外,還有一些位于10 號和11號染色體的次效SNPs。Kortbeek等[45]使用RF算法從野生番茄葉片的酰基糖(acylsugars)和揮發(fā)物(volatiles)中確定與植物寄生昆蟲煙粉虱(Bemisiatabaci)和西花薊馬(Frankliniellaoccidentalis)抗性相關(guān)的代謝物,結(jié)果顯示,兩種特定的酰基糖與煙粉虱抗性相偶聯(lián),86 種揮發(fā)物中倍半萜烯 ∝ 蛇麻烯(sesquiterpeneα -humulene)與煙粉虱易感有關(guān)。此外,我們還總結(jié)了近期機(jī)器學(xué)習(xí)在植物病害檢測方面的研究成果,見表1。

表1機(jī)器學(xué)習(xí)在植物病害檢測方面的應(yīng)用研究成果
注;LS-SVM(leastsquaressupport vector machines),,最小二乘支持向量機(jī);MLP(multi-ayerper-ceptron),多層感知器;LR-L2(regularizedlogisticregresion),正則邏輯回歸;GBD(gradientboostigdecisiontree),梯度增強(qiáng)決策樹;GMM(gaussian mixture model),高斯混合模型;K-Means(K-meansclustering),K均值聚類;LDA(lineardiscriminant analysis),線性判別分析。

2.2 機(jī)器學(xué)習(xí)在植物非生物脅迫研究中的應(yīng)用

非生物脅迫是指不利于植物生長發(fā)育(嚴(yán)重時甚至導(dǎo)致植物死亡)的環(huán)境條件,包括低溫、高溫、干旱、高鹽、水淹、過量光等。ML方法結(jié)合高光譜圖像數(shù)據(jù),能夠有效鑒定植物對非生物脅迫的響應(yīng)。Chen等[55]利用高光譜相機(jī)獲取茶樹(Camelliasinensis)在模擬干旱處理下的樹葉光譜數(shù)據(jù),使用SVM、RF、PLS算法提取多項生理指標(biāo)(丙二醛含量、電解質(zhì)泄漏等),并結(jié)合CARS、無信息變量消除(uninformativevariable elimination,UVE)等方法,用于監(jiān)測干旱脅迫下茶樹幼苗的應(yīng)答程度。此外,Chen等[56]還發(fā)現(xiàn)丙二醛(malond-ialdehyde)、可溶性糖(solublesugar)、總多酚(totalpolyphenols)是評估茶樹抗旱性的關(guān)鍵指標(biāo),且不同茶樹品種的抗旱性有所差異,最終篩選出最佳預(yù)測模型 MSC-2D-UVE-SVM 。Lu 等[57]結(jié)合SVM算法和無人機(jī)高光譜成像篩選火炬松(Pinustaeda)的耐凍性幼苗,根據(jù)表型將幼苗分為健康型和冷凍脅迫型,對冷凍脅迫41天幼苗的分類準(zhǔn)確率高達(dá) 96.00% 。Das 等[58]收集56個耐鹽敏感水稻的基因型、葉片光譜特征、養(yǎng)分濃度數(shù)據(jù)(鉀、鈉、鈣等),利用PLSR確定了鹽脅迫下與葉片營養(yǎng)狀態(tài)密切相關(guān)的光譜特征。

在非生物脅迫下,植物通過調(diào)整特定基因的轉(zhuǎn)錄水平可以激活或抑制其表達(dá),從而產(chǎn)生適應(yīng)性蛋白質(zhì)或其他分子[59]。Sprenger 等[60]利用 RF從31個馬鈴薯品種的葉片樣本中篩選出與淀粉產(chǎn)量相關(guān)的特征代謝物和轉(zhuǎn)錄物,發(fā)現(xiàn)僅需20個特征就能將馬鈴薯抗旱性的預(yù)測誤差降低到4.30% ,且這些特征在16個獨立的農(nóng)業(yè)試驗中表現(xiàn)出良好的重復(fù)性和穩(wěn)定性。Smet等[61]利用RF預(yù)測高溫或干旱脅迫下水稻的基因表達(dá),并評估了啟動子和基因組序列特征在訓(xùn)練ML模型中的重要性。

微小RNA(microRNA,miRNA)是一類能夠在轉(zhuǎn)錄后水平調(diào)控基因表達(dá)的小分子RNA,在植物對各種非生物脅迫的應(yīng)激反應(yīng)中起作用。Asef-pour Vakilian[62]利用信息論(information theory)原理進(jìn)行特征選擇,從擬南芥(Arabidopsisthali-ana)的miRNA表達(dá)數(shù)據(jù)中篩選出在干旱、鹽分、低溫、高溫脅迫下響應(yīng)貢獻(xiàn)度最高的miRNA(miRNA-169、miRNA-159、miRNA-396、miRNA-393),然后以這些miRNA的表達(dá)豐度為輸入特征,通過DT、SVM、樸素貝葉斯(NaiveBayes,NB)等方法構(gòu)建模型,預(yù)測植物對不同脅迫條件的反應(yīng)程度,結(jié)果表明SVM的預(yù)測準(zhǔn)確度最高,決定系數(shù)(coefficient of determination, R2 )達(dá)到了0.96。Pradhan 等[63]對比 SVM、RF、XGBoost 等 ML方法,選擇SVM模型來預(yù)測與寒冷、干旱、高溫和鹽脅迫相關(guān)的miRNA,預(yù)測精度優(yōu)于其他深度學(xué)習(xí)模型,同時建立了在線預(yù)測網(wǎng)站“ASmiR”(ht-tps://iasri-sg.icar.gov.in/asmir/)。

3 機(jī)器學(xué)習(xí)在植物生化組分研究中的應(yīng)用

近年來,研究人員通過高效液相色譜(highperformance liquid chromatography,HPLC)、高光譜、近紅外光譜(near infrared spectroscopy,NIR)等多源數(shù)據(jù),結(jié)合機(jī)器學(xué)習(xí)方法分析植物生化組分(如葉綠素、多酚、脂肪酸等),取得顯著進(jìn)展[64],為評估植物的生長狀況、抗逆能力、營養(yǎng)價值提供了新的技術(shù)手段,并為農(nóng)業(yè)生產(chǎn)和食品加工提供了科學(xué)依據(jù)。我們總結(jié)了近期相關(guān)研究進(jìn)展,見表2。

另外,ML也為植物發(fā)育過程中土壤和葉片不同營養(yǎng)元素(氮磷鉀等)的含量估計提供了新途徑。Fu等[71]對堆疊稀疏自編碼器(stacked sparseautoen-coder,SSAE)進(jìn)行改進(jìn)(命名為MSSAE),將深度光譜特征作為輸人,采用SVR和最小二乘支持向量回歸(least squares support vector regression,LS-SVR)的方法預(yù)測油菜葉片中鋅的含量,結(jié)果顯示MSSAE-LSSVR模型的預(yù)測效果最佳,預(yù)測集的R2 和均方根誤差(root mean square error,RMSE)分別為0.9566和 1.0240mg/kg 。Wang等[80]利用標(biāo)準(zhǔn)正態(tài)變量(standardnormalvariate)對茶樹葉片的高光譜數(shù)據(jù)進(jìn)行預(yù)處理,采用偏最小二乘判別分析(partial least squares discriminant analy-sis,PLS-DA)和最小二乘支持向量機(jī)(leastsquares support vector machines,LS-SVM) 對不同氮狀態(tài)的葉片進(jìn)行分類,LS-SVM模型分類準(zhǔn)確率達(dá)到了 92.00% 。Siedliska等[8i]使用4種不同劑量的磷對芹菜(Apiumgraveolens)甜菜、草莓施肥,采用高光譜成像在植株發(fā)育的不同階段測量磷含量,并通過BPNN、RF、NB、SVM方法對不同施磷量的植物進(jìn)行分類。Zhang等[8使用RF算法、光譜指數(shù)(spectralindices)、連續(xù)小波特征(continuouswavelettransform)相結(jié)合的方法,對水稻葉片磷濃度進(jìn)行遠(yuǎn)程估算,為不同供磷水平土壤的水稻葉磷濃度光譜監(jiān)測提供了有益參考。

表2機(jī)器學(xué)習(xí)在植物生化組分檢測中的應(yīng)用研究成果
注:ERT(extremerandom tree),極端隨機(jī)樹;KRR(kermelridge regression),核嶺回歸;LS-SVR(leastsquaressupport vectorgresson)最小二乘支持向量回歸;PLS-DA(partial least squares discriminant analysis),偏最小二乘判別分析。

利用ML檢測植物的次生代謝物,可幫助更深入地研究植物的適應(yīng)性、抗性等機(jī)制。Li等[73]使用SPA、CARS、堆疊自動編碼器(stackedau-toencoder,SAE)對桑樹(Morusalba)果實在3個成熟期的可見近紅外高光譜圖像數(shù)據(jù)進(jìn)行降維處理,利用LS-SVM和ELM算法建立檢測桑椹花青素含量的模型,并通過遺傳算法(geneticalgo-rithm,GA)優(yōu)化模型的主要參數(shù),結(jié)果表明,基于SAE-GA-ELM的模型取得最佳性能,可有效檢測桑椹中花青素含量的分布。Shen等[74]利用高光譜成像技術(shù)對鎘脅迫下水稻葉片重金屬脅迫指示物游離脯氨酸(freeproline,F(xiàn)P)進(jìn)行高通量篩選,并使用PLS、LS-SVM、ELM 模型檢測FP含量,結(jié)果表明,光譜差異隨著鎘脅迫時間的延長而增加,F(xiàn)P含量隨著鎘脅迫濃度的增加而增加。Wu 等[79]使用傅里葉變換紅外光譜(fourier trans-forminfrared spectroscopy,F(xiàn)T-IR)和高效液相色譜對滇龍膽(Gentianarigescens)的活性成分進(jìn)行鑒定,結(jié)果表明PLS-DA能有效鑒別出龍膽不同的來源產(chǎn)地,并通過SVR和七重交叉驗證的網(wǎng)格搜索算法預(yù)測了活性成分龍膽苦苷的含量。Mao等[83]采用高光譜成像技術(shù)獲取茶葉在干燥和發(fā)酵過程中的光譜數(shù)據(jù),采用SPA、CARS、UVE來選擇光譜的特征波段,結(jié)合SVM、RF和PLS方法建模監(jiān)測茶葉中風(fēng)味物質(zhì)茶多酚、FP的含量,可用于判斷加工過程中紅茶的萎凋和發(fā)酵程度。

4機(jī)器學(xué)習(xí)在作物改良及產(chǎn)量預(yù)測方面的應(yīng)用

4.1 機(jī)器學(xué)習(xí)在作物改良方面的應(yīng)用

機(jī)器學(xué)習(xí)方法結(jié)合光譜、基因組等多源數(shù)據(jù)在輔助作物改良方面展現(xiàn)出巨大的優(yōu)勢。Feng等[84]研究了冬、春、半冬3種生態(tài)型油菜品種的基因組足跡和葉片形態(tài)結(jié)構(gòu)的變化,通過高通量表型分析獲得171種油菜的性狀差異,利用RF模型鑒定出19種表型性狀對3種生態(tài)型的分化有貢獻(xiàn),可作為預(yù)測生態(tài)型的生物標(biāo)志物,QTL信息分析發(fā)現(xiàn)3種生態(tài)型油菜分別有213、237、184個QTLs與生態(tài)型分化信號重疊。Cui等[85]利用基因組BLUP對1495個水稻雜交種的雜交性能進(jìn)行預(yù)測,通過10倍交叉驗證,對10個農(nóng)藝性狀的預(yù)測能力在 0.35~0.92 之間,基于3K水稻基因組計劃和10個育種性狀指數(shù),從3000個水稻雜交種中預(yù)測出200個表現(xiàn)最優(yōu)和最差的雜交種

大豆GWAS和GS研究方面,基于機(jī)器學(xué)習(xí)的方法同樣顯示出輔助農(nóng)藝性狀改良的潛力。2021年,Yoosefzadeh-Najafabadi等[86]使用基于高光譜寬關(guān)聯(lián)研究(hyperspectral wide associationstudy)和SVR的GWAS方法,確定了與大豆產(chǎn)量相關(guān)的高光譜反射帶和QTLs。2022年,該課題組又采用基于單倍型的精準(zhǔn)GS方法,使用裝袋集成(ensemble bagging)策略整合了徑向基函數(shù)回歸、SVR、RF三種ML方法,并聯(lián)合優(yōu)化的遺傳信息來增強(qiáng)模型,提高了預(yù)測大豆產(chǎn)量及其組成性狀(單株節(jié)數(shù)、非生殖節(jié)數(shù)、生殖節(jié)數(shù)、莢果數(shù))的準(zhǔn)確性,并在大豆19號染色體上發(fā)現(xiàn)了一個對產(chǎn)量及其組成具有顯著影響的單倍型塊(haplo-type block),可用于篩選高產(chǎn)大豆[87]。同年,他們評估了SVR和RF兩種算法在大豆GWAS中的應(yīng)用,對比混合線性模型(mixedlinearmodels)

及固定和隨機(jī)模型交替概率統(tǒng)一(fixedand ran-dom model circulating probability unification)兩種傳統(tǒng)方法,發(fā)現(xiàn)使用SVR耦合GWAS的方法可預(yù)測與QTL共定位的“標(biāo)記-性狀關(guān)聯(lián)”(marker-traitassociations),功能注釋和報道文獻(xiàn)支持預(yù)測結(jié)果[88]。此外,他們還使用基于SVR的GWAS檢測大豆耐久存儲(蛋白質(zhì)、油脂和百粒重等性狀)相關(guān)的QTLs,結(jié)果表明大豆籽粒蛋白與油脂濃度呈顯著負(fù)相關(guān),遺傳力值分別為0.69和0.67[89] O

4.2機(jī)器學(xué)習(xí)在作物產(chǎn)量預(yù)測方面的應(yīng)用

作物產(chǎn)量預(yù)測是農(nóng)業(yè)管理和決策的重要依據(jù),也是機(jī)器學(xué)習(xí)方法在農(nóng)業(yè)領(lǐng)域應(yīng)用的重要方面。目前結(jié)合遙感、土壤、氣候、歷史產(chǎn)量等多源數(shù)據(jù),利用ML進(jìn)行作物產(chǎn)量預(yù)測的研究已取得了一定的成就,我們總結(jié)了近期相關(guān)研究成果,見表3。

隨著遙感技術(shù)的發(fā)展,RF、SVM、長短期記憶網(wǎng)絡(luò)(long short-term memory,LSTM)等 ML方法被廣泛用于分析遙感數(shù)據(jù)以預(yù)測作物產(chǎn)量。然而,這些監(jiān)督學(xué)習(xí)模型的訓(xùn)練往往受限于高質(zhì)量標(biāo)簽數(shù)據(jù)的缺乏,尤其是在特定區(qū)域的詳細(xì)地面真實數(shù)據(jù)方面。此外,由于域轉(zhuǎn)移問題,模型在訓(xùn)練時表現(xiàn)很好但應(yīng)用到新區(qū)域時性能可能下降[98]。半監(jiān)督學(xué)習(xí)(semi-supervised learning)和無監(jiān)督學(xué)習(xí)(unsupervised leaning)方法在作物產(chǎn)量預(yù)測中的應(yīng)用有助于解決此類不足。Khaki等[99]基于半監(jiān)督深度學(xué)習(xí)方法和高通量圖像開發(fā)了DeepCorn軟件,可利用改進(jìn)的VGG-16網(wǎng)絡(luò)預(yù)測玉米籽粒數(shù)和估計產(chǎn)量,并通過融合不同尺度的特征圖增強(qiáng)了對圖像尺寸變化的魯棒性,其預(yù)測結(jié)果的平均絕對誤差(mean absolute error,MAE)和RMSE分別為41.36和60.27。2022年,Ma等[10]提出了貝葉斯域?qū)股窠?jīng)網(wǎng)絡(luò)(Bayes-ian domain adversarial neural network,BDANN)的無監(jiān)督域適應(yīng)方法來預(yù)測玉米產(chǎn)量,BDANN通過對抗性學(xué)習(xí)和貝葉斯推理減少域轉(zhuǎn)移,從源域和目標(biāo)域中提取特征,準(zhǔn)確預(yù)測了美國玉米種植帶的玉米產(chǎn)量。2023年,Ma等[101]進(jìn)一步提出了一種多源最大預(yù)測差異(multisource maximum pre-dictordiscrepancy,MMPD)的神經(jīng)網(wǎng)絡(luò),用于無監(jiān)督域適應(yīng)的縣級玉米產(chǎn)量預(yù)測,在美國和阿根廷玉米種植帶的實驗結(jié)果表明,MMPD模型表現(xiàn)優(yōu)于其他深度學(xué)習(xí)和無監(jiān)督域適應(yīng)方法。

表3機(jī)器學(xué)習(xí)在作物產(chǎn)量預(yù)測方面的應(yīng)用研究成果
注:OLS(ordinaryleast squares),普通最小二乘法;LSTM(longshort-term memory),長短期記憶網(wǎng)絡(luò);GPR(Gaussianprocessregression),高斯過程回歸;R(logisticregresson),邏輯回歸;ASSO(leastabsoluteshrinkageandselectionoperator),最小絕對收縮和擇算子;LightG-BM(light gradientbostingmachine),輕量級梯度提升;DN(deepneuralnetwork),深度神經(jīng)網(wǎng)絡(luò);AdaBost(adaptivebosting),自適應(yīng)增強(qiáng);GRU(gated recurrent unit),門控循環(huán)單元。

5 總結(jié)與展望

隨著高通量測量技術(shù)的成熟和無人機(jī)等表型采集技術(shù)的發(fā)展,積累的植物表型數(shù)據(jù)日益豐富,如何有效地從這些數(shù)據(jù)中挖掘出有用信息,對于構(gòu)建全面、準(zhǔn)確的植物表型數(shù)據(jù)庫和知識圖譜,以及揭示植物表型的遺傳和分子機(jī)制有著重要意義。盡管機(jī)器學(xué)習(xí)在植物表型研究領(lǐng)域已取得一定成果,但仍面臨著許多挑戰(zhàn),例如,如何更好地集成多源異構(gòu)數(shù)據(jù)[102],如何更好地處理高維度和稀疏的表型數(shù)據(jù)[103],如何在非結(jié)構(gòu)化(如目標(biāo)植物被其他物體遮擋,以及因天氣影響而產(chǎn)生的動態(tài)變化等)的數(shù)據(jù)中提取有價值的信息[103],如何解決模型的可解釋性[104]等問題。

為了克服這些不足,基于機(jī)器學(xué)習(xí)的植物表型研究需在以下3個方面取得突破: ① 利用半監(jiān)督學(xué)習(xí)和無監(jiān)督學(xué)習(xí)等技術(shù),從未標(biāo)注數(shù)據(jù)中挖掘復(fù)雜表型特征,提高模型泛化能力和魯棒性[105],在此基礎(chǔ)上,以遷移學(xué)習(xí)作為補(bǔ)充,將一個領(lǐng)域的知識遷移到另一個相關(guān)領(lǐng)域,通過預(yù)訓(xùn)練模型來提升對未標(biāo)注數(shù)據(jù)的理解,從而充分利用現(xiàn)有的數(shù)據(jù)和知識[106]; ② 發(fā)展多模態(tài)學(xué)習(xí)算法和模型,整合不同類型(基因組、表型、環(huán)境、圖像、聲音等)的數(shù)據(jù),提高表型數(shù)據(jù)的綜合性和描述精度[107]; ③ 結(jié)合機(jī)器學(xué)習(xí)和其他數(shù)學(xué)方法(遺傳算法、梯度下降、模擬退火等),開發(fā)出更有效的特征選擇方法和降維方法,優(yōu)化評估模型,提高植物表型的預(yù)測精度和模型可解釋性[104] 。

近年來,應(yīng)用機(jī)器學(xué)習(xí)方法結(jié)合多源圖像數(shù)據(jù)進(jìn)行研究已成為植物表型研究的主流[108]。模型的有效性很大程度上依賴于訓(xùn)練數(shù)據(jù)的多樣性,如果數(shù)據(jù)不夠全面,可能會導(dǎo)致模型在實際環(huán)境應(yīng)用表現(xiàn)不佳。因此,未來應(yīng)在數(shù)據(jù)共享、數(shù)據(jù)集成、算法改進(jìn)等方面加強(qiáng)研究,以增強(qiáng)模型的泛化能力[109]。此外,一些新的神經(jīng)網(wǎng)絡(luò)算法如圖卷積神經(jīng)網(wǎng)絡(luò)(graph convolutional neural network,GCNN)、圖注意力網(wǎng)絡(luò)(graph attention network,GAT)等方法的提出,也有助于預(yù)測更為復(fù)雜的關(guān)聯(lián),但目前這些新興網(wǎng)絡(luò)在植物表型研究方面應(yīng)用的相關(guān)報道還很少。今后可結(jié)合更多來源的生物數(shù)據(jù)(如組學(xué)數(shù)據(jù)、宏觀生態(tài)數(shù)據(jù)等)將更優(yōu)化的機(jī)器學(xué)習(xí)算法應(yīng)用于植物表型研究,為揭示植物生長發(fā)育、產(chǎn)量和品質(zhì)形成等復(fù)雜性狀的分子機(jī)理提供新視角。我國已于2024年2月27日在武漢啟動國家作物表型組學(xué)研究重大科技基礎(chǔ)設(shè)施(“神農(nóng)設(shè)施”)項目,該設(shè)施將采用尖端技術(shù)進(jìn)行植物表型篩選研究,預(yù)計每年處理50萬至100萬株植物的表型數(shù)據(jù),為推動我國“常規(guī)育種 + 生物技術(shù) + 信息化”育種4.0時代的到來提速助力[110]。總的來說,機(jī)器學(xué)習(xí)在植物表型研究中的應(yīng)用前景十分廣闊,有望為實現(xiàn)作物“數(shù)字育種”[1]提供強(qiáng)大的預(yù)測工具,為推動植物生長發(fā)育機(jī)理研究和作物育種添磚加瓦。

參考文獻(xiàn):

[1] SunD W,Robbins K,Morales N,et al.Advancesin optical phenotyping of cereal crops[J]. Trends in Plant Science, 2022,27(2):191-208.

[2] Chebrolu N,MagistriF,Labe T,etal.Registration of spatiotemporal point clouds ofplants for phenotyping[J].PLoS ONE,2021,16(2):e0247243.

[3] TardieuF,Cabrera-BosquetL,PridmoreT,etal.Plantphenomics,from sensors to knowledge[J].Current Biology,2017,27 (15):R770-R783.

[4] Pieruschka R,Schurr U. Plant phenotyping: past, present,and future[J].PlantPhenomics,2019,2019:7507131.

[5] Esposito S,CarputoD,Cardi T,et al.Applications and trends of machine learning in genomics and phenomics for next-generationbreeding[J].Plants(Basel),2019,9(1):34.

[6] SinghA,GanapathysubramanianB,Singh AK,et al.Machine abolic biomarkers for phenotype prediction using an untargeted approach[J]. Plant Biotechnology Journal,2010,8(8) :900- 911.

[8]Yang X H,HuangJF,Wu YP,et al. Estimating biophysical parameters of rice with remote sensing data using support vector machines[J]. Science China(Life Sciences),2011,54(3): 272-281.

[9]Hall R D,D'Auria JC,Silva Ferreira A C,et al. High-throughput plant phenotyping: a role for metabolomics?[J]. Trends in Plant Science,2022,27(6) :549-563.

[10]QiH X,Zhu B Y,Wu Z Y,et al. Estimation of peanut leaf area index from unmanned aerial vehicle multispectral images[J]. Sensors(Basel),2020,20(23) :6732.

[11]Du L P,Yang H,Song X,et al. Estimating leaf area index of maize using UAV-based digital imagery and machine learning methods[J]. Scientific Reports,2022,12(1) :15937.

[12]Shu M Y,F(xiàn)ei SP,Zhang B Y,et al. Application of UAV multisensor data and ensemble approach for high-throughput estimation of maize phenotyping traits[J].Plant Phenomics,2022, 2022 :9802585.

[13] Sun X K,Yang Z Y,Su P Y,et al. Non-destructive monitoring of maize LAI by fusing UAV spectral and textural features[J]. Frontiers in Plant Science,2023,14:1158837.

[14] Zhang JJ,Cheng T,Guo W,et al.Leaf area index estimation model for UAV image hyperspectral data based on wavelength variable selection and machine learning methods[J].Plant Methods,2021,17(1) :49.

[15] Chen Q M,Zheng B Y,Chenu K,et al. Unsupervised plot-scale LAI phenotyping via UAV-based imaging,modelling,and machine learning[J]. Plant Phenomics,2022,2022:9768253.

[16]Iniyaz O,Du QY,Shen HF,etal.Leaf area index estimation of pergola-trained vineyards in arid regions using classical and deep learning methods based on UAV-based RGB images[J]. Computers and Electronics in Agriculture,2023,207:107723.

[17]Li X L,Dong Y Y,Zhu Y N,et al. Enhanced leaf area index estimation with CROP-DualGAN network[J]. IEEE Transactions on Geoscience and Remote Sensing,2023,61:1-10.

[18] Wang H C,Wu Y,Ni Q,et al. A novel wirelessleaf area index sensor based on a combined U-Net deep learning model[J]. IEEE Sensors Jourmal,2022,22(16) :16573-16585.

[19]KangMJ,Shin AY,Shin Y,et al.Identification of transcriptome-wide, nut weight-associated SNPs in Castanea crenata [J]. Seientific Reports,2019,9(1) :13161.

[20]Liu Z L,Nan F,Zheng X,et al. Color prediction of mushroom slices during drying using Bayesian extreme learning machine [J].Drying Technology,2020,38(14) :1869-1881.

[21] Tong H,Nankar A N,Liu JT,et al. Genomic prediction of mor

[22]Minamikawa MF,Nonaka K,Hamada H,et al.Dissecting breeders’sense via explainable machine learning approach : application to fruit peelability and hardness in citrus[J]. Frontiers in Plant Science,2022,13:832749.

[23] Cetin N,Karaman K,Kavuncuoglu E,et al. Using hyperspectral imaging technology and machine learning algorithms for assessing internal quality parameters of apple fruits[J]. Chemometrics and Intelligent Laboratory Systems,2022,230:104650.

[24]Van De Looverbosch T,Vandenbussche B,Verboven P,et al. Nondestructive high-throughput sugar beet fruit analysis using X-ray CT and deep learning[J]. Computers and Electronics in Agriculture,2022,200:107228.

[25]Le Louedec J,Cielniak G. 3D shape sensing and deep learningbased segmentation of strawberries[J]. Computers and Electronics in Agriculture,2021,190:106374.

[26]Nyalala I,Okinda C,Nyalala L,et al. Tomato volume and mass estimation using computer vision and machine learning algorithms:cherrtomato modelJ]. JoumalofFoodEngineering, 2019,263:288-298.

[27] Chen S M,Xiong J T,Jiao JM,et al. Citrus fruits maturity detection in natural environments based on convolutional neural networks and visual saliency map[J]. Precision Agriculture, 2022,23(5) :1515-1531.

[28] Shi X,Hsieh S J,Romero R A F. FEA and machine learming techniques for hidden structure analysis[J]. Sensors (Basel), 2021,21(15) :5159.

[29]Wedger M J,Topp C N,Olsen K M. Convergent evolution of root system architecture in two independently evolved lineages of weedy rice[J]. The New Phytologist,2019,223(2) :1031- 1042.

[30]Awika H O,Mishra A K,Gill H,et al. Selection of nitrogen responsive root architectural traits in spinach using machine learning and genetic correlations[J]. Scientific Reports,2021, 11(1) :9536.

[31] Changdar S,Popovic O,Wacker T S,et al. Non-invasive phenotyping for water and nitrogen uptake by deep roots explored using machine learning[J]. Plant and Soil,2023,493(1) :603- 616.

[32] Gou J Q,Debnath S,Sun L,et al. From model to crop: functional characterization of SPL8 in M. truncatula led to genetic improvement of biomass yield and abiotic stress tolerance in alfalfa[J]. Plant Biotechnology Journal,2018,16(4) :951-962.

[33] Buciarelli B, Xu Z Y,Ao S,et al. Phenotyping seedlings for selection of root system architecture in alfalfa (Medicago sativa L.)[J].Plant Methods,2021,17:1-13.

[34] Xu Z Y,York L M,Seethepalli A,et al. Objective phenotyping of root system architecture using image augmentation and machine learning in alfalfa (Medicago sativa L.)[J]. Plant Phephenotyping,GWAS,and predictive models reveals the genetic architecture of plant height in maize[J]. Molecular Plant, 2023,16(2) :354-373.

[36]Yu G E,Shin Y,Subramaniyam S,et al. Machine learning, transcriptome,and genotyping chip analyses provide insights into SNP markers identifying flower color in Platycodon grandiflorus[J]. Scientific Reports,2021,11(1) :8019.

[37] Bueno G F,Costa E A,F(xiàn)inger C A G,et al. Machine learning: crown diameter predictive modeling for open-grown trees in the Cerrado biome,Brazil[J]. Forests,2022,13(8) :1295.

[38]Apriyanti D H,Spreeuwers L J,Lucas P J,et al. Automated color detection in orchids using color labels and deep learning[J]. PLoS ONE,2021,16(10) :e0259036.

[39] Xie Q H,Wang JF,Lopez-Sanchez JM,et al. Crop height estimation of corn from multi-year RADARSAT-2 polarimetric observables using machine learning[J]. Remote Sensing,2021, 13(3) :392.

[40]Iizuka K,Hayakawa Y S,Ogura T,et al. Integration of multisensor data to estimate plot-level stem volume using machine learning algorithms: case study of evergreen conifer planted forests in Japan[J].Remote Sensing,2020,12(10):1649.

[41]Cao WL,Gan L M,Wang C C,et al. Genome-wide identification and characterization of potato long non-coding RNAs associated with Phytophthora infestans resistance[J]. Frontiers in Plant Science,2021,12:619062.

[42] Chi J X,Song S Z,Zhang H,et al. Research on the mechanism of soybean resistance to Phytophthora infection using machine learning methods[J]. Frontiers in Genetics,2021,12:634635.

[43] Bhattarai G,Olaoye D,Mou B,et al. Mapping and selection of downy mildew resistance in spinach cv. whale by low coverage whole genome sequencing[J]. Frontiers in Plant Science, 2022,13;1012923.

[44] Canella Vieira C,Zhou J,Usovsky M,et al. Exploring machine learning algorithms to unveil genomic regions associated with resistance to southern root-knot nematode in soybeans[J]. Frontiers in Plant Science,2022,13:883280.

[45]Kortbeek R WJ,Galland M D,Muras A,et al. Natural variation in wild tomato trichomes; selecting metabolites that contribute to insect resistance using a random forest approach[J].BMC Plant Biology,2021,21:315.

[46]Feng Z H,Song L,Duan J Z,et al. Monitoring wheat powdery mildew based on hyperspectral, thermal infrared,and RGB image data fusion[J]. Sensors (Basel),2021,22(1) :31.

[47] Zhao JL,F(xiàn)ang Y,Chu G M,et al. Identification of leaf-scale wheat powdery mildew (Blumeria graminis f. sp. tritici) combining hyperspectral imaging andan SVMclasifierJ].Plants (Basel),2020,9(8) :936.

[48]Dhakal K,Sivaramakrishnan U,Zhang X,et al. Machine learn

[49]Koc A,Odilbekov F,Alamrani M,etal. Predicting yellow rust in wheat breeding trials by proximal phenotyping and machine learning[J]. Plant Methods,2022,18(1) :30.

[50] Liu S,Yu H Y,Sui Y Y,etal. Classification of soybean frogeye leaf spot disease using leaf hyperspectral reflectance[J]. PLoS ONE,2021,16(9):e0257008.

[51]Abdulridha J,Ampatzidis Y,Qureshi J,et al. Identification and classification of downy mildew severity stages in watermelon utilizing aerial and ground remote sensing and machine learning [J].Frontiers in Plant Science,2022,13:791018.

[52]Wang Z X,Niu Y,Vashisth T,et al.Nontargeted metabolomicsbased multiple machine learning modeling bosts early accurate detection for citrus Huanglongbing[J]. Horticulture Research, 2022,9:uhac145.

[53] Skelsey P.Forecasting risk of crop disease with anomaly detection algorithms[J].Phytopathology Φ(B) ,2021,111(2) : 321- 332.

[54]Reis Pereira M,dos Santos F N,Tavares F,et al. Enhancing host-pathogen phenotyping dynamics: early detection of tomato bacterial diseases using hyperspectral point measurement and predictive modeling[J].Frontiers in Plant Science,2023,14: 1242201.

[55]Chen S Z,Gao Y,F(xiàn)an K,et al. Prediction of drought-induced components and evaluation of drought damage of tea plants based on hyperspectral imaging[J].Frontiers in Plant Science, 2021,12:695102.

[56]Chen S Z,Shen J Z,F(xiàn)an K,et al. Hyperspectral machine-learning model for screening tea germplasm resources with drought tolerance[J]. Frontiers in Plant Science,2022,13:1048442.

[57]Lu Y Z,Payn K G,Pandey P,et al. Hyperspectral imaging with cost-sensitive learning for high-throughput screening of loblolly pine(Pinus taeda L.) seedlings for freeze tolerance[J]. Transactions of the ASABE,2021,64(6) :2045-2059.

[58]Das B,Manohara K,Mahajan G,et al. Spectroscopy based novel spectral indices,PCA-and PLSR-coupled machine learning models for salinity stress phenotyping of rice[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2020,229:117983.

[59]Kimotho R N,Baillo E H,Zhang Z. Transcription factors involved in abiotic stress responses in maize (Zea mays L.) and their roles in enhanced productivity in the post genomics era [J]. PeerJ,2019,7:e7211.

[60] Sprenger H,Erban A,Seddig S,et al. Metabolite and transcript markers for the prediction of potato drought tolerance[J]. Plant Biotechnology Jourmal,2018,16(4) :939-950.

[61]Smet D,Opdebeeck H,Vandepoele K. Predicting transcriptional responses to heat and drought stress from genomic features using a machine learning approach in rice[J]. Frontiers in Plant

[ Uz」 Aseipuui vakian Λ. imacnne ieang umpiover JuIKHowieuge about miRNA functions towards plant abiotic stresses[J]. Scientific Reports,2020,10(1) :3041.

[63]Pradhan U K,Meher P K,Naha S,et al. ASmiR:a machine learning framework for prediction of abiotic stress-specific miRNAs in plants[J].Functional amp; Integrative Genomics,2023,23 (2) :92.

[64] Zhang Y Y,Wang Y Z. Machine learning applications for multisource data of edible crops: a review of current trends and future prospects[J]. Food Chemistry: X,2023,19:100860.

[65]Burns M J,Renk JS,Eickholt D P,et al. Predicting moisture content during maize nixtamalization using machine learning with NIR spectroscopy[J]. Theoretical and Applied Genetics, 2021,134(11) :3743-3757.

[66] Zhang JJ,Zhang W,Xiong S P,et al. Comparison of new hyperspectral index and machine learning models for prediction of winter wheat leaf water content[J]. Plant Methods,2021,17 (1) :34.

[67]Cao YL,Jiang K L,Wu JX,et al. Inversion modeling of japonica rice canopy chlorophyll content with UAV hyperspectral remote sensing[J]. PLoS ONE,2020,15(9) :e0238530.

[68]Sun JH,Yang L,Yang X T,et al. Using spectral reflectance to estimate the leaf chlorophyll content of maize inoculated with arbuscular mycorrhizal fungi under water stress[J]. Frontiers in Plant Science,2021,12:646173.

[69] Wang W,Kong W W,Shen T T,et al. Application of laser-induced breakdown spectroscopy in detection of cadmium content in rice stems[J]. Frontiers in Plant Science,2020,11:599616.

[70]Luo S Y,Yuan X,Liang R Q,et al. Prediction and visualization of gene modulated ultralow cadmium accumulation in brown rice grains by hyperspectral imaging[J]. Spectrochim Acta Part A:Molecular and Biomolecular Spectroscopy,2023,297: 122720.

[71] Fu L H,Sun J,Wang S M,et al. Nondestructive evaluation of Zn content in rape leaves using MSSAE and hyperspectral imaging[J]. Spectrochim Acta Part A:Molecular and Biomolecular Spectroscopy,2022,281: 121641.

[72] Qin S Z,Ding Y R,Zhou Z X,et al. Study on the nitrogen content estimation model of coton leaves basedon “image-spectrum-fluorescence”data fusion[J].Frontiers in Plant Science, 2023,14 1117277.

[73]Li X L,Wei Z X,Peng F F,et al. Non-destructive prediction and visualization of anthocyanin content in mulberry fruits using hyperspectral imaging[J]. Frontiers in Plant Science,2023, 14:1137198.

[74]Shen T T,Zhang C,Liu F,et al. High-throughput screening of free proline content in rice leaf under cadmium stress using hyperspectral imaging with chemometrics[J]. Sensors (Basel), 2020,20(11) :3229. els[J」.Spectrochim Acta Part A:Molecular Biomolecular Spectroscopy,2020,237: 118403.

[76]Jin G,Wang Y J,Li M H,et al. Rapid and real-time detection of black tea fermentation quality by using an inexpensive data fusion system[J].Food Chemistry,2021,358:129815.

[77]Li C Y,Lv YB,Gao W,et al. A comprehensive strategy for prediction and qualityevaluation of standardized planting herbs based on plant metabolomics coupled with extreme learning machine : Astragali Radix as an example[J]. Phytochemical Analysis,2023,34(5) :606-616.

[78]Bertani FR,Businaro L,Gambacorta L,et al. Optical detection of aflatoxins B in grained almonds using fluorescence spectroscopy and machine learning algorithms[J]. Food Control,2020, 112:107073.

[79] Wu Z,Zhao Y L,Zhang J,et al. Quality assessment of Gentiana rigescens from diferent geogaphical origins using FT-IR spectroscopy combined with HPLC[J]. Molecules (Basel,Swizerland),2017,22(7) :1238.

[80]Wang Y J,Li TH,Jin G,et al. Qualitative and quantitative diagnosis of nitrogen nutrition of tea plants under field condition using hyperspectral imaging coupled with chemometrics[J]. Journal of the Science of Food and Agriculture,2020,100(1): 161-167.

[81] Siedliska A,Baranowski P,Pastuszka-Wozniak J,et al. Identification of plant leaf phosphorus content at different growth stages based on hyperspectral reflectance[J].BMC Plant Biology, 2021,21(1) :28.

[82] Zhang Y,Wang T,Li Z,et al. Based on machine learning algorithms for estimating leaf phosphorus concentration of rice using optimized spectral indices and continuous wavelet transform [J].Frontiers in Plant Science,2023,14:1185915.

[83]Mao YL,Li H,Wang Y,et al. Prediction of tea polyphenols, free amino acids and caffeine content in tea leaves during wilting and fermentation using hyperspectral imaging[J]. Foods, 2022,11( 16) :2537.

[84]Feng H,Guo C C,Li Z Y,et al. Machine learning assisted dynamic phenotypes and genomic variants help understand the ecotype divergence in rapeseed[J]. Frontiers in Plant Science, 2022,13:1028779.

[85] Cui YR,LiR D,Li G W,et al. Hybrid breeding of rice via genomic selection[J].Plant Biotechnology Journal,2020,18 (1) :57-67.

[86]Yoosefzadeh-Najafabadi M,Torabi S,Tulpan D,et al. Genomewide association studies of soybean yield-related hyperspectral reflectance bands using machine learning-mediated data integration methods[J].Frontiers in Plant Science,2021,12: 777028.

[87] Yoosefzadeh-Najafabadi M,Rajcan I,Eskandari M. Optimizing genomic selection in soybean:an important improvement in agricultural genomics[J]. Heliyon,2022,8(11) :e11873.

[88]Yoosefzadeh-Najafabadi M,Eskandari M,Torabi S,et al. Machine-learning-based genome-wide association studies for uncovering QTL underlying soybean yield and its components[J]. International Journal of Molecular Sciences,2022,23(10): 5538.

[89]Yoosefzadeh-Najafabadi M,Torabi S,Tulpan D,et al. Application of SVR-mediated GWAS for identification of durable genetic regions associated with soybean seed quality traits[J]. Plants (Basel),2023,12(14) :2659.

[90] Schwalbert R A,Amado T,Corassa G,et al. Satelite-based soybean yield forecast:integrating machine learning and weather data for improving crop yield prediction in southern Brazil[J]. Agricultural and Forest Meteorology,2020,284:107886.

[91]Han JC,Zhang Z,Cao J,et al. Prediction of winter wheat yield based on multi-source data and machine learning in China[ J]. Remote Sensing,2020,12(2) :236.

[92]Sun J,Di L P,Sun Z H,et al. County-level soybean yield prediction using deep CNN-LSTM model[J]. Sensors,2019,19 (20) :4363.

[93] Shahhosseini M,Hu G P,Huber I,et al. Coupling machine learning and crop modeling improves crop yield prediction in the US corn belt[J]. Scientific Reports,2021,11(1)) :1606.

[94] Wang Y M,Zhang Z,F(xiàn)eng L W,et al. Combining multi-source data and machine learning approaches to predict winter wheat yield in the conterminous United States[J].Remote Sensing, 2020,12(8) :1232.

[95]Guo Y H,F(xiàn)u Y S,Hao FH,et al. Integrated phenology and climate in rice yields prediction using machine learning methods [J].Ecological Indicators,2021,120:106935.

[96]Wang J,Wang P X,Tian HR,et al. A deep learning framework combining CNN and GRU for improving wheat yield estimates using time series remotely sensed multi-variables[J]. Computers and Electronics in Agriculture,2023,206:107705.

[97]Filippi P,Jones E J,Wimalathunge N S,et al. An approach to forecast grain crop yield using multi-layered,multi-farm data sets and machine learning[J]. Precision Agriculture,2019,20: 1015-1029.

[98] Alzubaidi L,Zhang J,Humaidi A J,et al. Review of deep learning:concepts, CNN architectures,challenges, applications, future directions[J]. Journal of Big Data,2O21,8(1):53.

[99]Khaki S,Pham H,Han Y,et al. DeepCorn:a semi-supervised deep learning method for high-throughput image-based corn kernel counting and yield estimation[J].Knowledge-Based Systems,2021,218:106874.

[100]Ma Y C,Zhang Z. A Bayesian domain adversarial neural network for corn yield prediction[J]. IEEE Geoscience and Remote Sensing Letters,2022,19:1-5.

[101]Ma Y C,Yang Z W,Zhang Z. Multi-source maximum predictor discrepancy for unsupervised domain adaptation on corn yield prediction[J]. IEEE Transactions on Geoscience and Remote Sensing,2023,61:1-15.

[102] Hesami M,Alizadeh M,Jones A M P,et al. Machine learning: its challenges and opportunities in plant system biology[J]. Applied Microbiology and Biotechnology,2022,106(9) :3507- 3530.

[103] van Dijk A D J,Kootstra G,Kruijer W,et al. Machine learning in plant science and plant breeding[J].iScience,2021,24 (1) :101890.

[104]Azodi C B,Tang JL,Shiu S H. Opening the black box;interpretable machine learning for geneticists[J]. Trends in Genetics,2020,36(6) :442-455.

[105]Yan J,Wang X F. Unsupervised and semi-supervised learning: the next frontier in machine learning for plant systems biology [J]. The Plant Journal,2022,111(6):1527-1538.

[106] Iman M,Arabnia H R,Rasheed K.A review of deep transfer learning and recent advancements[J]. Technologies,2023,11 (2) :40.

[107]Barua A,Ahmed M U,Begum S. A systematic literature review on multimodal machine learning:applications,challenges, gaps and future directions[J]. IEEE Access,2023,11:14804- 14831.

[108]Gill T,Gill S K,Saini D K,et al.A comprehensive review of high throughput phenotyping and machine learning for plant stress phenotyping[J]. Phenomics,2022,2(3) :156-183.

[109] Danilevicz M F,Gill M,Anderson R,et al.Plant genotype to phenotype prediction using machine learning[J]. Frontiers in Genetics,2022,13:822173.

[110]馮麗妃.國之重器“神農(nóng)設(shè)施\"在鄂啟動建設(shè)[N/OL].中國 科學(xué)報,2024-02-29(001).

[111]Jeon D,Kang Y,Le S,et al. Digitalizing breeding in plants:a new trend of next-generation breeding based on genomic prediction[J].Frontiers in Plant Science,2023,14:1092584.

猜你喜歡
表型機(jī)器光譜
基于多尺度卷積融合的高光譜圖像分類網(wǎng)絡(luò)
不同光譜變換形式對土壤有機(jī)質(zhì)偏最小二乘估算模型精度的影響
棉花種子油分含量QTL的定位及遺傳效應(yīng)分析
郁金香表型性狀的遺傳變異及多樣性分析
公益課堂 趣味科普
不同種源紫緞表型性狀多樣性分析與綜合評價
燈塔工廠的生產(chǎn)執(zhí)行研究
基于無人機(jī)獲取草地植物地上生物量的信息損失分析及修正
檢修記錄
青年文摘(2025年14期)2025-08-01 00:00:00
番茄植株生長過程中莖葉表型的無損分割與提取
主站蜘蛛池模板: 久久性妇女精品免费| 亚洲精品无码不卡在线播放| 日本免费福利视频| 欧美激情福利| 欧美另类图片视频无弹跳第一页| 久久久国产精品无码专区| 国产精品福利在线观看无码卡| 漂亮人妻被中出中文字幕久久| 99精品国产电影| 日韩大片免费观看视频播放| 精品日韩亚洲欧美高清a| 亚洲精品国产精品乱码不卞| AⅤ色综合久久天堂AV色综合 | 18禁高潮出水呻吟娇喘蜜芽| 黄片在线永久| 91日本在线观看亚洲精品| 久久精品午夜视频| 国产性生大片免费观看性欧美| 亚洲综合色区在线播放2019| 欧美成人精品在线| www.狠狠| 亚洲综合经典在线一区二区| 国产成人一区二区| 欧美在线伊人| 永久成人无码激情视频免费| 香蕉eeww99国产在线观看| 亚洲天堂视频在线免费观看| 免费无码在线观看| 久久国产高潮流白浆免费观看| 在线观看国产网址你懂的| 亚洲成人播放| 久久亚洲精少妇毛片午夜无码 | 日韩av在线直播| 97国产一区二区精品久久呦| 巨熟乳波霸若妻中文观看免费 | 国产成人精品第一区二区| 色吊丝av中文字幕| 欧美亚洲国产日韩电影在线| 久久综合色88| 国产农村妇女精品一二区| 激情乱人伦| 欧美一级在线看| 熟女视频91| 亚洲精品午夜天堂网页| 视频一区视频二区中文精品| 国产精品女人呻吟在线观看| 国产午夜福利片在线观看| 无码日韩视频| 三级欧美在线| 欧美怡红院视频一区二区三区| 国产特一级毛片| 国产一级裸网站| 网久久综合| 99偷拍视频精品一区二区| 中文字幕欧美日韩高清| 久久精品91麻豆| 日本精品中文字幕在线不卡 | 成人av专区精品无码国产| 国产在线观看91精品亚瑟| 国产精品太粉嫩高中在线观看| 久久这里只有精品2| 成人综合久久综合| 2021亚洲精品不卡a| 国产一区二区三区夜色| 国产高清不卡| 日韩无码真实干出血视频| 成人午夜福利视频| 亚洲精品制服丝袜二区| 国产亚洲男人的天堂在线观看| 日韩精品亚洲人旧成在线| 免费在线成人网| 激情乱人伦| 中文字幕无码av专区久久| 99热这里只有精品在线播放| 99久久国产综合精品2023| av在线人妻熟妇| 成人午夜天| 欧美啪啪一区| 亚洲妓女综合网995久久| 在线欧美一区| 日韩大片免费观看视频播放| 熟女成人国产精品视频|