资源描述：
《基于人工鱼群算法储粮害虫特征选择》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、do:i10.3969/.jissn.1671-7775.2010.05.002基于人工鱼群算法的储粮害虫特征选择1,211张红涛,毛罕平,张晓东(1.江苏大学现代农业装备与技术省部共建教育部/江苏省重点实验室,江苏镇江212013;2.华北水利水电学院电力学院,河南郑州450011)摘要:储粮害虫特征选择是粮虫图像识别中的一个核心问题.提出基于人工鱼群算法的特征选择,并给出了基于二进制编码寻优的实现方法.以交叉验证训练模型的识别率作为特征子集的性能评价准则,将人工鱼群算法应用于粮虫的特征选择.该算法从粮虫的17维形态学特征中自动选择出面积、周长等7个特征所组成的最优特征子集,采用参数优化
2、之后的SVM分类器对90个粮虫样本进行分类,识别率达到955%以上,并与PCA法、GA法和原始特征法进行对比,结果表明人工鱼群算法降低了特征空间的维数,提高了分类器的识别率,证实了基于人工鱼群算法的粮虫特征选择是可行的.关键词:储粮害虫;人工鱼群算法;特征选择;支持向量机;图像识别中图分类号:S24;TP39141文献标志码:A文章编号:1671-7775(2010)05-0502-04Featureselectionofstoredgraininsectsbasedonartificialfishswarmalgorithm1,211ZhangHongtao,MaoHanp
3、ing,ZhangXiaodong(1.KeyLaboratoryofModernAgriculturalEquipmentandTechnology,MinistryofEducation&JiangsuProvince,JiangsuUniversity,Zhenjiang,Jiangsu212013,China;2.InstituteofElectricPower,NorthChinaInstituteofWaterConservancyandHydroelectricPower,Zhengzhou,Henan450011,China)Abstract:Thefeaturesel
4、ectionisthecoreissueintheimagerecognitionofthestoredgraininsects.Thefeatureselectionwasfirstlyproposedbasedontheartificialfishswarmalgorithm,andtheoptimizationandrealizationweregivenbasedonbinarycode.TherecognitionaccuracyoftheVfoldcrossvalidationtrainingmodelwastakenastheevaluationprincipleoft
5、hefeaturesubse.tTheartificialfishswarmalgorithmwasappliedtothefeatureselectionofthestoredgraininsects.Thealgorithmselectedsevenfeaturescomposingtheoptimalfeaturespacefrom17morphologicalfeatures,suchasareaandperimeter.Theninetyimagesamplesofthestoredgraininsectsingraindepotwereautomaticallyrec
6、ognizedbythesupportvectormachineclassifiersothattheparameterswereoptimized,andthecorrectidentificationratiowasover955%.Theartificialfishswarmalgorithmwascomparedwiththeprincipalcomponentanalysis,thegeneticalgorithmandtheoriginalfeature.Theexperimentalresultsshowthattheartificialfishswarmalgorith
7、mgreatlyreducesthefeaturedimensionsandimprovestherecognitionaccuracy.Itispracticalandfeasible.Keywords:storedgraininsects;artificialfishswarmalgorithm;featureselection;supportvectormachine;imagerecognition收稿日期
