基于迁移学习对棉花受海水胁迫情况判断的模型研究  

Study on the Model for Judging the Situation of Cotton under Seawater Stress Based on Transfer Learning

作　　者：施洋 高进[1] 陈建平[1] 杨华[1] 陆镇威 王永慧[1] 施庆华[1] 孙艳茹[1] SHI Yang;GAO Jin;CHEN Jian-ping(Jiangsu Coastal Area Institute of Agricultural Sciences,Yancheng,Jiangsu 224002)

机构地区：[1]江苏沿海地区农业科学研究所,江苏盐城224002

出　　处：《安徽农业科学》2022年第20期226-229,239,共5页Journal of Anhui Agricultural Sciences

基　　金：国家重点研发计划课题(2016YFD0101421);江苏省自然科学基金项目(BK20181210);江苏省重点研发计划项目(BE2019413);农业部沿海盐碱地科学观测实验站开放课题(YHS201709)。

摘　　要：为探讨人工智能技术热点之一的迁移学习技术对棉花受海水胁迫程度情况判断进行端到端识别的可行性,以浓度为0(蒸馏水)、25%、50%和100%的海水分别对30个棉花种质资源进行苗期胁迫20 d,将迁移学习应用于VGG16卷积神经网络,对不同浓度海水胁迫下棉花的顶视图和侧视图进行分类研究。结果表明,网络对棉花侧视图的测试准确率为80.00%,对顶视图的测试准确率为77.14%。基于VGG16和迁移学习可构建识别棉花受海水胁迫情况的模型,侧视图更有利于网络识别。2种视图下,网络对0和100%浓度海水处理的识别能力更强。The aim was to explore the feasibility of end-to-end identification of cotton's degree under seawater stress by the transfer learning technology,which is one of the popular artificial intelligence technologies.30 cotton varieties were stressed for 20 d at seedling stage with seawater at the concentrations of 0(distilled water),25%,50%and 100%.The migration learning was applied to VGG16 convolutional neural network to classify the top and side views of cotton plants under different concentrations of seawater stress.Results showed that the test accuracy of the network to the side views verification set of cotton plant was 80.00%,and the test accuracy of the top views was 77.14%.The model for identifying cotton under seawater stress could be constructed based on VGG16 and migration learning,and the side views were more conducive to network recognition.Under the two views,the network has stronger recognition ability for 0 and 100%concentration seawater treatment.

关 键 词：棉花 海水胁迫 VGG16 迁移学习 模型研究 

分 类 号：S126[农业科学—农业基础科学]