高光谱成像结合BiTCN-SA的马铃薯晚疫病早期识别

  • 打印
  • 收藏
收藏成功


打开文本图片集

Abstract:Earlyidentificationofpotatolateblightiscrucialforcontrollingitsdevelopment.Tofullutilizetheinter-band characteristicinformationofhyperspectraldataandimprovetheaccuracyofmodels intheearlyidentificationofpotatolate blight,this study proposesa potato late blightearlyidentification model (BiTCN-SA)basedona Bidirectional Temporal Convolutional Network (BiTCN)fused witha Self-Atention (SA)mechanism.The BiTCNcaptures inter-bandcorrelation features through forwardand backward convolution branches,and fully exploits the asociations betweenprecedingand subsequent bands.The self-atention mechanism dynamically assigns importance weights to different bands,enhancing the contributionofkeybands tomodelclasification.TheBiTCN-SAmodel integratesself-atentionwith BiTCNtoachievea combinationoflocalconvolutionalfeaturesandglobalatention weights inboth directions,realizingdualfeatureextraction and improving the model's identificationaccuracy.This studycollectshyperspectral potato leaf data from three stages (healthy,asymptomatic,andearlysymptomatic),conductsmodelingandanalysis.Itvrifiesthesuperiorityofteroosed model bycomparing machine learning methods such as SVM and RF,and deep learning models including CNN,LSTM, TCN,and BiTCN.Theresults show thatthe BiTCN-SA model converges fasterthan standalone TCNand BiTCN models, withsignificantlyimprovedaccuracy.Itdemonstratesstronger featureextractioncapabilitythanothermachinelearningand deep learningmethods,achieving an overall accuracyof 98% and an identification rate of 96% forasymptomatic diseased leaves.Thismethodfullyutilizesdeepinter-band informationfromhyperspectraldataanditsidentificationrateshows substantial improvement over other machine learning and dep learning methods,providing technicalsupport forearly warning and control of potato late blight.

Keywords:Potatolate blight;hyperspectralimaging;earlyidentification;bidirectionaltemporalconvolutionalnetwork;selfattentionmechanism; feature extraction

马铃薯是全球第四大重要的粮食作物,我国是世界上马铃薯种植面积最大的国家,而由致病疫霉引起的马铃薯晚疫病是一种毁灭性病害[],其爆发性强、蔓延速度快[2],严重威胁我国粮食安全。(剩余11994字)

monitor
客服机器人