基于无人机多光谱和改进BPNN的烟草病毒病检测

打开文本图片集

中图分类号：S572；S126 文献标识码：A文章编号：1007-5119（2025）03-0098-11

Tobacco Virus Disease Detection Based on UAV Multispectral and Improved BPNN

ZHANG Yangliang1²， JIANG Xueyan1， LI Min²， JIANG Houlong³， JIANG Lianqiang4， GUO Leifeng²*，WANG Xinweil

（1.KeyLaboratoryofTobaco PestMonitoring &Integrated Management/Istitute ofTobaccoResearchofCAAS，Qingdao66101，

China;2.AgricuturalInfomationItuteofCeseAcademyofAgriculturalSienes，BeiingO1Chia;3.obaccoLeaf

Company，ChongingBrachf，hongingo，Cina;4.ingshanBrachoficuanobaccoopaniaa 615000， Sichuan， China）

Abstract：Thisstudyaims toidentifytobacovirus diseasebyintegrating UAV-based multispectralremotesensing technologywith animprovedBPneuralnetwork （BPNN）.Multispectral imagesof healthyanddiseasedtobacoplants withvaryingdegreesof virus infectionwerecapturedusingtheDJIP4MUAV.Atotalof19 vegetation indiceswerecalculated toconstructfeaturesets for correlationanalysis.K-nearest neighbors （KNN），random forest （RF），supportvectormachine （SVM），traditional BPNN，and improved BPNN were used toperformcomparative testsonbinaryandtermaryclasificationsamples.Theimproved BPNN，with optimizationsinetworktructure，imbalanceatahandling，actiationfunctioneplacing，andotimzerancementchied 89% accuracyand anF1 scoreof O.88 forbinaryclassification， and 79% accuracywith anF1 scoreof O.76 for ternaryclassification-both outperforming traditionalalgorithms.TheseresultsindicatethatUAVmultispectraldatacombinedwithanimprovedBPNNholds applicationpotential forthedetectionof tobaccovirusdisease，providing technical supportforearlywamingand preventionof agricultural diseases.

Keywords： tobacco; virus disease; UAV; multispectral; improved BPNN

烟草病毒病是烟草生产过程中最常见且最具破坏性的病害，常年造成巨大经济损失[1]。（剩余17473字）