中国农业科学2010,43(18):3882-3891ScientiaAgriculturaSinicadoi:10.3864/j.issn.0578-1752.2010.18.025基于计算机视觉的花生品质分级检测研究韩仲志，赵友刚（青岛农业大学理学与信息学院，山东青岛266109）摘要：【目的】建立能够对花生进行品质分级的计算机视觉无损检测方法。【方法】同步拍摄和扫描11类品质，每类品质100颗和100宗，每宗100颗不同等级的花生籽粒的正反面图像；参照国家标准量化花生品质籽粒的11个限制性检测项目，设计花生规格和品质等级的判别方法；测量每个籽粒的形态、纹理、颜色共3大类54个外观特征，采用主分量分析（PCA）进行特征优化，构建并比较BP神经网络（ANN）和支持向量机（SVM）品质检测模型；分别应用Matlab和Spss工具软件实现检测过程和对结果进行统计分析。【结果】前16个主分量的SVM模型，能够鉴别95%以上的不完善粒、霉变、杂质、异品种等不同品质的籽粒，与人工检测结果吻合度达到了93%，对100宗待检样品进行检测，规格和等级检测完全正确率达到了92%。【结论】研究结果为花生的品质分级检测提供了比较系统全面的量化标准和检测方法，该方法可推广应用于花生品质鉴别、分级筛选加工和商品分级定价等领域。关键词：花生仁；计算机视觉；品质分级；无损检测QualityGradeDetectioninPeanutUsingComputerVisionHANZhong-zhi,ZHAOYou-gang(DepartmentofScienceandInformation,QingdaoAgriculturalUniversity,Qingdao266109,Shandong)Abstract:【Objective】Theobjectiveofthisstudyistoestablishakindofqualitynondestructivetestingmethod,whichcanbeusedforgradingpeanutquality,basedoncomputervision.【Method】Digitalcolorimageofpeanutsweretakenandscannedfrom2sidefaceseachof100kernelseachof11differentkindsofqualityandof100kernelseachof100groups.ReferringtonationalstandardsofChina,11restrictiveitemsofpeanutskernelsofdifferentkindsofqualityhavebeendevisedandquantized.Alsothedistinguishablemethodsofsizeandgradesofpeanutshavebeendevised.Fifty-fourappearancecharactersbelongingto3categoriesofshape,colorandtexturehadbeenmeasured.AndthenthecharacterswereoptimizedbasedonPCA.ANNandSVMqualitytestingmodelswerebuiltandcompared.UsingMATLABandSPSS,theresultswereanalyzed.【Result】TheSVMmodelbasedonthefirst16PCscoulddetectat95%accuracydifferentqualitiesofunsound/mildew/impurity/differentpeanutvarieties.Alsotheseresultsfittedat93%accuracyclosetothatoftestedbymanual.Bytesting100groupsofpeanuts,thecorrectrateofsizeandgradewas92%.【Conclusion】Theresultofthisstudyhasprovidedanewmethodwhichcanbeusedinpeanutqualitytestingandgradetesting,andthismethodisgoodandstable.Thismethodcanbegeneralizedandusedinpeanutstestingofdifferentqualities,gradescreening,processing,andcom