資源描述:
《基于BP神經(jīng)網(wǎng)絡(luò)的草莓圖像分割》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在教育資源-天天文庫(kù)。
1、中國(guó)農(nóng)業(yè)大學(xué)學(xué)報(bào)2004,9(4):65~68JournalofChinaAgriculturalUniversity草莓采摘機(jī)器入的研究:工.基于BP神經(jīng)網(wǎng)絡(luò)的草莓圖像分割張鐵中周天娟(中國(guó)農(nóng)業(yè)大學(xué)工學(xué)院,北京100083)摘要草莓成熟度和空間位置的識(shí)別是草莓采摘機(jī)器人研究的重要環(huán)節(jié),解決此問題必須首先對(duì)采集的草莓圖像進(jìn)行分割。采用三層BP神經(jīng)網(wǎng)絡(luò),通過分析選取3×3鄰域像素的H通道值作為草莓圖像的特征;選取HSV模型中與亮度無關(guān)的通道以排除圖像的明暗對(duì)分割效果的影響;采用單通道以縮短圖像處理時(shí)間。選取20幅圖像作為訓(xùn)練樣本,CA.h~5_借助Photoshop軟件分割后的圖
2、像作為教師信號(hào),采用BP算法對(duì)神經(jīng)網(wǎng)絡(luò)的權(quán)值進(jìn)行訓(xùn)練。經(jīng)過100次循環(huán)后(誤差為0.001),獲得了有效的網(wǎng)絡(luò)權(quán)值。試驗(yàn)結(jié)果表明。利用BP神經(jīng)網(wǎng)絡(luò)能較好地實(shí)現(xiàn)成熟草莓果實(shí)與背景的分離,經(jīng)過提取大區(qū)域和腐蝕、膨脹等算法的進(jìn)一步處理后。效果更好;而且,只要改變勘l練時(shí)的教師信號(hào),即可實(shí)現(xiàn)對(duì)草莓果梗、萼片等圖像的分割。關(guān)鍵詞BP神經(jīng)網(wǎng)絡(luò);草莓;圖像分割;特征中圖分類號(hào)TP274文章編號(hào)1007—4333(2004)04—0065—04文獻(xiàn)標(biāo)識(shí)碼AStrawberryharvestingrobot:I.SegmentationofstrawberryimagebyBPneuralnet
3、workZhangTiezhong,ZhouTianjuan(ColegeofEngineering,ChinaAgriculturalUniversity,Beijing100083。China)AbstractForthestrawberryharvestingrobot,thedegreeofmaturationandrecognitionofthepositioninspacearetheimportantsteps.Firstofall,segmentationofstrawberryimageisrequired.ThethreelayersBPneuralnetw
4、orkmethodforsegmentingthetruecolorofstrawberryimagewasstudied.TheHchannelvaluesof3×3neighborhoodpixelswereobtainedasfeaturesthroughanalysis.IntheHSVmodel,channelHwaschosentogetridoftheeffectoflightbecauseithadnorelationwithlight.Singlechannelwasselectedtoshortenprocessingtime.Inthenetwork.20
5、strawberryim—agesweretakenastrainingsamples.Andresultsofmanualsegmentationimagesweretakenasteachersignals.TheBPalgorithmwasusedtotraintheparameterofthenetwork.Theeffectiveparameterwasachievedafter100times’training.Theerrorwas0.001.Theresultofthisexperimentshowedthatexpectedsegmentationresult
6、couldbeachievedbyusingBPnetworkmethod.Aftertakingfurtherprocessingsuchaspickingupbiggerareas,erosionanddi—lation,etc.,theresultcouldbebetter.Moreover,iftheteachersignalswerechanged,theimagesegmentationofpe—duncleorsepalcouldberealizedeasily.KeywordsBPneuralnetwork;strawberry;imagesegmentatio
7、n;feature我國(guó)草莓栽培面積1985年約為3300hm,近(約30d),草莓日成熟量可達(dá)150~200kg·年來迅速擴(kuò)大,到2000年已達(dá)到66700hm,北起667m_。,2個(gè)勞動(dòng)力需工作約10h·d-1才能采摘黑龍江,南至廣東,東起山東,西到新疆均有種完。草莓采摘工作的復(fù)雜性和勞動(dòng)強(qiáng)度已成為制約植。草莓成熟后必須及時(shí)采摘,尤其是盛果期草莓種植業(yè)發(fā)展的重要因素。然而,由于草莓形狀收稿日期:2004—04—23基金項(xiàng)目:國(guó)家高技術(shù)研究發(fā)展計(jì)劃資助項(xiàng)目(2001AA422300)作者簡(jiǎn)介