欢迎来到天天文库
浏览记录
ID:36780995
大小:1.50 MB
页数:8页
时间:2019-05-15
《小波域马铃薯典型虫害图像特征选择与识别》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、2017年9月农业机械学报第48卷第9期doi:10.6041/j.issn.10001298.2017.09.003小波域马铃薯典型虫害图像特征选择与识别肖志云刘洪(内蒙古工业大学电力学院,呼和浩特010080)摘要:为准确、快速地识别马铃薯典型虫害,提出了一种基于小波域的马铃薯典型虫害特征提取与识别方法。该方法以自然环境下的马铃薯虫害分割图像为对象,提取小波域高斯空间模型的高频协方差阵特征值与低频低阶矩(HELM)的12个不变纹理特征、空间域Hu不变矩的4个形状特征,进行支持向量机(SVM)的虫害分类识别。通过对8类典型虫害的识别,试验结果表明:在SVM识别
2、方法下,本文HELM特征提取方法,相比传统纹理特征提取方法,在特征计算量不增加的同时,平均识别率至少提高了17个百分点;在HELM特征与Hu矩特征下,本文SVM的运行时间为0481s,比人工神经网络快了近2s,平均识别率为975%,比人工神经网络、贝叶斯分类器识别率提高了至少6个百分点,有明显的识别优势。关键词:马铃薯虫害;小波域;高斯空间模型;特征选择;图像识别;支持向量机中图分类号:TP39141文献标识码:A文章编号:10001298(2017)09002408FeaturesSelectionandRecognitionofPotatoTypi
3、calInsectPestImagesinWaveletDomainXIAOZhiyunLIUHong(CollegeofElectricPower,InnerMongoliaUniversityofTechnology,Huhhot010080,China)Abstract:Inordertorecognizepotatotypicalinsectpestsaccuratelyandquickly,anewfeatureextractionandrecognitionmethodbasedonwaveletandspacedomainwasproposed.Th
4、eprocessingobjectinthemethodwasthesegmentedimageofinsectpestsseparatedfromcomplexbackgroundbythetwodimensionalOtsumethodandmorphologicalmethod.Aimingattheprocessingobject,totally12invarianttexturefeaturesofhighfrequencycovariancematrixeigenvaluesandlowfrequencylowerordermoments(HELM)
5、wereextractedfromthehighfrequencyimagesinthehorizontal,verticalanddiagonaldirections,formingaGaussianspacemodel,andfromlowfrequencyimagedecomposedbysym8waveletfunction.Meanwhile,4Humomentswithinvariantshapefeatureswereextractedfromthebinaryimageoftheprocessingobject.Asthus,16pestfeatu
6、reswereputintosupportvectormachine(SVM),andtheresultsofinsectpestclassificationcouldbeobtained.ForSVMclassifier,theOnevsOnevotingstrategywasadopted,andtheparameters,includingradialbasiskernelfunctionparameter,errorcostcoefficientandrelaxationcoefficientweresetto00125,60and0001,res
7、pectively.Bytheclassificationof8kindsofpests,ontheonehand,usingthesameSVMmethod,thetestresultsshowedtheeffectivenessofproposedHELMfeatureextraction.Texturefeaturesinwaveletdomainweretraditionallyrelatedtosinglescalelowfrequencylowerordermoments(SLM),includingthemean,varianceandthethir
8、dorde
此文档下载收益归作者所有