基金项目:国家重点研发计划项目(2018YFC1903101)收稿日期:2021-04-19修回日期:2021-04-22第40卷第2期计算机仿真2023年2月文章编号:1006-9348(2023)02-0214-04基于深度学习的马铃薯病害智能识别陈从平1,钮嘉炜1,丁坤1,姜金涛2(1.常州大学机械与轨道交通学院,江苏常州213164;2.内蒙古智诚物联股份有限公司,内蒙古乌兰察布012000)摘要:针对现有通过叶片病斑识别马铃薯病害的方法中,只对简单背景下的单片叶片进行病害识别处理,难以应用于真实复杂环境的问题,提出了一种基于深度学习的现场环境下马铃薯病害智能识别的方法。首先采用Deeplabv3+语义分割网络在生长背景中分割出马铃薯叶片,然后使用自适应对比度增强和颜色空间转换的方法提取马铃薯病斑,最后结合病斑的纹理特征和VGG16网络提取的特征,通过搭建一维卷积神经网络识别出病害。实验结果表明,上述方法能准确有效地识别复杂背景下的马铃薯病害。关键词:病害识别;语义分割;特征融合中图分类号:TP183文献标识码:BIntelligentIdentificationofPotatoDiseasesBasedonDeepLearningCHENCong-ping1,NIUJia-wei1,DINGKun1,JIANGJin-tao2(1.SchoolofMechanicalEngineeringandRailTransit,ChangzhouUniversity,ChangzhouJiangsu213164,China;2.InnerMongoliaZhichengIOTCo.,Ltd.,UlanqabInnerMongolia012000,China)ABSTRACT:Inviewoftheexistingmethodsofidentifyingpotatodiseasesbyleafspot,itisdifficulttoapplytoarealandcomplexenvironment,andonlyasingleleafundersimplebackgroundisidentifiedandprocessed.Amethodofin-telligentidentificationofpotatodiseasesinafieldenvironmentbasedondeeplearningisproposed.First,theDeeplabv3+semanticsegmentationnetworkwasusedtosegmentthepotatoleavesinthegrowingbackground,andthenthepo-tatolesionswereextractedusingthemethodofadaptivecontrastenhancementandcolorspaceconversion.Finally,combiningthetexturefeaturesofthelesionsandthefeaturesextractedbytheVGG16network,thediseasewasidenti-fiedbybuildingaone-dimensionalconvolutionalneuralnetwork.Experimentalresultsshowthatthismethodcanac-curatelyandeffectivelyidentifypotatodiseasesinacomplexbackground.KEYWORDS:Diseaserecognition;Semanticsegmentation;Featurefusion1引言马铃薯生长过程中出现的各种病害严重影响了...