臺灣博碩士論文加值系統

害蟲發生的頻率一直是農業時間和勞動的任務。本文試圖通過深度學習與農業的結合來解決上述問題。我們提出了一種全部和部分特徵轉移學習方案來進行有害生物檢測，分類和計數，以提供有害生物發生頻率的結果。在部分特徵轉移學習中，部分特徵轉移學習的細粒度特徵圖用於加強整個特徵轉移學習。最後，對整個特徵轉移學習使用權重方案加強了不同的細粒度特徵映射，並將整個特徵網絡的跨層與多尺度特徵映射相結合。整體特徵轉移學習方法通過使用跨層機制創建快捷拓撲來減少梯度消失問題來增強特徵。實驗結果表明，整體和部分特徵轉移學習機制的檢測和分類可以得到顯著改善，該方法可以達到90.2％。

The frequency of pest occurrence has always been a task of agricultural time and labor. This paper attempts to solve the above problems through the combination of deep learning and agriculture. We propose an entire-and-partial feature transfer learning scheme to perform pest detection, classification and counting, to offer the result of pest occurrence frequency.
In the partial-feature transfer learning, the fine-grained feature map of the partial-feature transfer learning is used to strengthened the entire-feature transfer learning.
Finally, different fine-grained feature map are strengthened to the entire-feature transfer learning use weight scheme and the cross-layer of the entire-feature network is combined with multi-scale feature map. The entire-feature transfer learning approach enhances the feature by creating a shortcut topology using cross layer mechanism to reduce the gradient disappearance problem.
The experimental results shows that the detection and classification of the entire-and partial feature transfer learning mechanism can be significantly improved, and the method can reach 90.2%.

1 Introduction1
2 RelatedWorks5
2.1 RelatedWorks.................................... 5
2.2 Motivation...................................... 7
3 Preliminaries8
3.1 Systemmodel.................................... 8
3.2 ProblemFormulation................................ 9
3.3 BasicIdea...................................... 11
4 AnEntire-and-PartialFeatureTransferLearningApproachforPestOccur-
rence Frequency13
4.1 Entire-and-Partialfeaturelearningphase..................... 15
4.2 Partial-featuretransferringphase......................... 17
4.3 Entire-featuretransferringphase.......................... 20
4.4 Pestdetectionandclassicationphase...................... 22
5 Simulationresult24
5.1 Averageprecision(AP)............................... 26
5.2 Recall........................................ 27
5.3 Trainingloss..................................... 30
5.4 Classaccuracy.................................... 30
5.5 CDF......................................... 32
6 Conclusions34
6.1 Acknowledgments.................................. 34

[1] J. Cho,J.Choi,M.Qiao,C.Ji,andK.H.Y,"AutomaticIdenticationofWhite
ies,
Aphids andThripsinGreenhousebasedonImageAnalysis", International Journalof
Mathematics andComputersinSimulation, vol.1,no.1,pp.46{53.
[2] H. Gassoumi,N.Prasad,andJ.Ellington,"Neuralnetwork-basedApproachforInsect
Classication inCottonEcosystems",in ProceedingsofInternationalConferenceonIn-
telligentTechnologies, pp.1{7,Thailand,Jan.2007.
[3] K.VakilianandJ.Massah,"PerformanceEvaluationofaMachineVisionSystemfor
Insect PestsIdenticationofFieldCropsusingArticialNeuralNetworks", Archivesof
PhytopathologyandPlantProtection, vol.46,no.11,pp.1262{1269.
[4] J. Wang,C.Lin,L.Ji,andA.Liang,"ANewAutomaticIdenticationSystemofInsect
Images attheOrderLevel", Knowledge-basedSystems, vol.33,no.1,pp.102{110.
[5] S. Kang,S.Song,andS.Lee,"IdenticationofButter
ySpecieswithaSingleNeural
NetworkSystem", Journal ofAsia-pacicEntomology, vol.15,no.3,pp.431{435.
[6] C.WenandD.Guyer,"Image-basedOrchardInsectAutomatedIdenticationandClassi-
cation Method", Computers andElectronicsinAgriculture, vol.89,no.10,pp.110{115.
[7] Y.KayaandL.Kayci,"ApplicationofArticialNeuralNetworkforAutomaticSetection
of Butter
ySpeciesUsingColorandTexturefeatures", The VisualComputer, vol.30,
no. 1,pp.71{79.
35
[8] W. DingandG.Taylor,"AutomaticMothSetectionfromTrapImagesforPestManage-
ment", Computers andElectronicsinAgriculture, vol.123,pp.17{28.
[9] N. ArdandH.S.J.Hemming,"DetectionandClassicationofInsectsonStick-trapsin
a TomatoCropUsingFasterR-CNN",in ProceedingsoftheNetherlandsConferenceon
Computer Vision, pp.1{4,Eindhoven,Jan.2018.
[10] X. Chengjun,W.Rujing,andZ.Jie,"Multi-levelLearningFeaturesforAutomaticClas-
sication ofFieldCropPests", Computers andElectronicsinAgriculture, vol.152,no.7,
pp. 152{156.
[11] H. J.Muhammad,N.M.Humair,andY.K.Babar,"K-meansbasedAutomaticPests
Detection andClassicationforPesticidesSpraying", International JournalofAdvanced
Computer ScienceandApplications, vol.8,no.11,pp.174{176.
[12] K. G.Liakos,P.Busato,D.Moshou,S.Pearson,andD.Bochtis,"MachineLearning
in Agriculture:AReview", Computers andElectronicsinAgriculture, vol.18,no.8,
pp. 2674{2678.
[13] Y. Zhong,Q.L.JunyuanGao,andY.Zhou,"AVision-basedCountingandRecognition
System forFlyingInsectsinIntelligentAgriculture", Sensors, vol.18,no.5,pp.234{240.
[14] L. Deng,Y.Wang,andZ.H.R.Yu,"ResearchonInsectPestImageDetectionand
Recognition basedonBio-inspiredMethods", Biosystems Engineering, vol.169,no.6,
pp. 139{142.
[15] J. Redmon,S.Divvala,R.Girshick,andA.Farhadi,"YouOnlyLookOnce:Unied,
Real-time ObjectDetection",in ProceedingsofComputerVisionandPatternRecognition,
pp. 2{4,LasVegas,NV,USA,June.2016.
36
[16] P.Mohammad,S.Sherine,N.M.Sharon,andM.A.L,"ACluster-then-labelSemi-
supervisedLearningApproachforPathologyImageClassication", Scientic Reports,
vol.8,no.1,pp.2{4.
[17] D. Motta,A.lissonBandeiraSantos,I.Winkler,B.A.S.Machado,andD.A.D.I.Pereira,
"Application ofConvolutionalNeuralNetworksforClassicationofAdultMosquitoesin
the Field", Public LibraryofScience, vol.14,no.1,pp.1{18.
[18] S. LeeandW.Kim,"SentimentlabelingforExtendingInitialLabeledDatatoImprove
Semi-supervisedSentimentClassication", ElectronicCommerceResearchandApplica-
tions, vol.26,no.6,pp.35{43.
[19] M. Ebrahimi,M.Khoshtaghaza,S.Minaei,andB.Jamshidi,"Vision-basedPestDetec-
tion basedonSVMClassicationMethod", Computers andElectronicsinAgriculture,
vol.137,pp.52{58.
[20] K. L.Roldn-Serrato,J.Escalante-Estrada,andM.Rodrguez-Gonzlez,"AutomaticPest
Detection onBeanandPotatoCropsbyApplyingNeuralClassiers", Engineeringin
Agriculture, vol.11,no.4,pp.245{255.
[21] A. Nazri,N.Mazlan,andF.Muharam,"AutomatedBrownPlanthopperDetectionfrom
ImperfectStickyPadImagesUsingDeepConvolutionalNeuralNetwork", Public Library
of Science, vol.13,no.2,pp.1{13.
[22] J. Liu,D.Han,andY.Yang,"PestIdenticationbasedonMultipleClassierSystem",
in ProceedingsofChineseControlConference(CCC), pp.5{7,Wuhan,July.2018.
[23] P.C.D.Xia,J.Z.BWang,andC.Xie,"InsectDetectionandClassicationBasedonan
ImprovedConvolutionalNeuralNetwork", Sensors, vol.18,no.12,pp.4169{4181.
[24] T. B.Remboski,W.D.deSouza,M.S.deAguiar,andP.R.F.Jr,"Identicationof
FruitFlyinIntelligentTrapsUsingTechniquesofDigitalImageProcessingandMachine
37
Learning", in ProceedingsofACM33rdAnnualACMSymposiumonAppliedComputing,
pp. 260{267,Pau,France,April.2018.
[25] Z. Liu,G.Zhang,M.S.HaiyingYang,H.Dang,andX.Zhou,"ApplicationofObject
Detection AlgorithminIdenticationofRiceWeevilsandMaizeWeevils",in Proceedings
of InternationalConferenceonDeepLearningTechnologies, pp.76{80,Chongqing,June.
2018.
[26] Y. Shen,H.Zhou,J.Li,F.Jian,andD.S.Jayas,"DetectionofStored-grainInsectsUsing
Deep Learning", Computers andElectronicsinAgriculture, vol.145,no.4,pp.319{325.
[27] N. Murali,J.Schneider,J.Levine,andG.Taylor,"ClassicationandRe-identicationof
FruitFlyIndividualsAcrossDayswithConvolutionalNeuralNetworks",in Proceedings
of WinterConferenceonApplicationsofComputerVision(WACV), pp.7{11,Waikoloa
Village, HI,USA,USA,Jan.2019.
[28] M. Lin,Sheng,W.sheng,andL.Jie,"APestsImageClassicationMethodbasedon
ImprovedWolfPackAlgorithmtoOptimizeBayesianNetworkStructureLearning",
38