資源描述:
《基于bp神經(jīng)網(wǎng)絡(luò)的高光譜遙感數(shù)據(jù)分類研究—論文》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在工程資料-天天文庫。
1、河南農(nóng)業(yè)大學(xué)本科生畢業(yè)論文(設(shè)計)目基于BP神經(jīng)網(wǎng)絡(luò)的高光譜遙感數(shù)據(jù)分類研究院信息與管理科學(xué)學(xué)院業(yè)計算機(jī)科學(xué)與技術(shù)學(xué)生姓名指導(dǎo)教師摘要:11引言22高光譜遙感32.1高光譜遙感的介紹32.2高光譜遙感國外發(fā)展32.3高光譜遙感國內(nèi)發(fā)展32.4高光譜遙感數(shù)據(jù)的特征52.5高光譜遙感數(shù)據(jù)的常用處理方法53人工神經(jīng)網(wǎng)絡(luò)53.1人工神經(jīng)網(wǎng)絡(luò)的概念53.2人工神經(jīng)網(wǎng)絡(luò)的特點63.3人工神經(jīng)元的模型63.4人工神經(jīng)網(wǎng)絡(luò)模型83.5BP網(wǎng)絡(luò)93.5.1BP網(wǎng)絡(luò)概念93.5.2BP算法的原理93.5.3BP算法的執(zhí)行步驟103.5.4BP神經(jīng)網(wǎng)
2、絡(luò)的局限性124BP神經(jīng)網(wǎng)絡(luò)在小麥口粉病高光譜遙感數(shù)據(jù)分類研究134.1試驗設(shè)i十134.2光譜數(shù)據(jù)采集134.3數(shù)據(jù)預(yù)處理134.4BP網(wǎng)絡(luò)模型建立及實現(xiàn)過程154.4.1BP網(wǎng)絡(luò)的分析流程154.4.2BP網(wǎng)絡(luò)層數(shù)的確定及各層神經(jīng)元數(shù)目的確定162.4.3BP網(wǎng)絡(luò)參數(shù)的選擇162.4.4讀入數(shù)據(jù)、劃分?jǐn)?shù)據(jù)及歸一化處理172.4.5BP網(wǎng)絡(luò)的模型建立、訓(xùn)練及預(yù)測174.4實驗總結(jié)195結(jié)論與討論19參考文獻(xiàn):20致謝21附件:22基于BP神經(jīng)網(wǎng)絡(luò)的高光譜遙感數(shù)據(jù)分類研究朱青信息與管理科學(xué)學(xué)院計算機(jī)科學(xué)與技術(shù)專業(yè)摘要:本文探討了
3、多層誤差反向傳播(BP)神經(jīng)網(wǎng)絡(luò)分類算法應(yīng)用于高光譜遙感數(shù)據(jù)的分類研究。首先介紹了高光譜遙感,高光譜遙感國內(nèi)外發(fā)展現(xiàn)狀,高光譜遙感數(shù)據(jù)的特點,以及高光譜遙感數(shù)據(jù)常用的處理方法。簡要介紹了神經(jīng)元的模型,神經(jīng)網(wǎng)絡(luò)模型,BP神經(jīng)網(wǎng)絡(luò)的算法以及執(zhí)行過程,BP網(wǎng)絡(luò)自身的局限性。將BP神經(jīng)網(wǎng)絡(luò)與高光譜遙感結(jié)合,用采集到的白粉病小麥冠層高光譜數(shù)據(jù)進(jìn)行實驗。利用ASD手持式高光譜儀測定了患有白粉病的小麥冠層光譜反射率,并對采集到的高光譜數(shù)據(jù)用特征選擇的方法,選出20個特征波長對應(yīng)的反射率作為神經(jīng)網(wǎng)絡(luò)的輸入,病情指數(shù)0、1、2、3、4作為神經(jīng)網(wǎng)絡(luò)
4、的輸出。借助Matlab神經(jīng)網(wǎng)絡(luò)模塊,選用經(jīng)典三層的BP神經(jīng)網(wǎng)絡(luò),將150組光譜數(shù)據(jù)隨機(jī)劃分120組訓(xùn)練數(shù)據(jù)和30組測試數(shù)據(jù),經(jīng)過訓(xùn)練,預(yù)測正確率為83.33%。這表明BP神經(jīng)網(wǎng)絡(luò)對高光譜遙感數(shù)據(jù)進(jìn)行識別分類,是一種很好的應(yīng)用。關(guān)鍵詞:高光譜遙感;人工神經(jīng)網(wǎng)絡(luò);BP神經(jīng)網(wǎng)絡(luò)HyperspectralremotesensingdataclassificationbasedonBPneuralnetworkAbstract:Thispaperdiscussesthemulti-layerbackpropagation(BP)neur
5、alnetworkclassificationalgorithmisappliedtotheclassificationofhyperspectralremotesensingdata.Firstintroducedhyperspectralremotesensing,hyperspectralremotesensingdomesticandforeigndevelopmentstatus,thecharacteristicsofhyperspectralremotesensingdataandthecommonlyusedap
6、proachofhyperspectralremotesensing.Brieflyintroducedthemodeloftheneuron,neuralnetworkmodel,BPneuralnetworkalgorithmandimplementationprocess,BPnetworklimitations.BPneuralnetworkcombinedwithhyperspectralremotesensing,experimentwithpowderymildewofwheatcanopyhyperspectra
7、ldata.ASDhandheldspectrometermeasuredcanopyspectralreflectancewithpowderymildew.Andhyperspectraldatacollectedwithafeatureselectionmethodselected20characteristicwavelengthscorrespondingtothereflectivityasaneuralnetworkinput,thediseaseindex0,1,2,3,4astheoutputoftheneur
8、alnetwork.WithMatlabneuralnetworkmodule,aclassicthree-tierBPneuralnetwork,150spectraldatawererandomlydividedinto120trainingdataand3