作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 开发研究与工程应用 • 上一篇    下一篇

基于卷积网络的苹果病变图像识别方法

王细萍1a,黄婷1b,谭文学1b,吴华瑞2,孙闯1b   

  1. (1.湖南文理学院 a.经济与管理学院; b.计算机科学与技术学院,湖南 常德 415000;2.北京农业信息技术研究中心,北京 100097)
  • 收稿日期:2014-10-27 出版日期:2015-12-15 发布日期:2015-12-15
  • 作者简介:王细萍(1979-),女,讲师、硕士,主研方向:信息系统模型,智能优化;黄婷,学士;谭文学,教授、博士研究生;吴华瑞,研究员、博士;孙闯,学士。
  • 基金资助:
    国家自然科学基金资助项目(61271257);湖南省教育厅科学研究基金资助项目(16A151);湖南文理学院大学生研究性学习和创新性实验计划项目。

Apple Lesion Image Recognition Method Based on Convolutional Network

WANG Xiping  1a,HUANG Ting  1b,TAN Wenxue  1b,WU Huarui  2,SUN Chuang  1b   

  1. (1a.School of Economy and Management; 1b.School of Computer Science and Technology, Hunan University of Arts and Science,Changde 415000,China;2.Beijing Research Center for Information Technology in Agriculture,Beijing 100097,China)
  • Received:2014-10-27 Online:2015-12-15 Published:2015-12-15

摘要: 为更好地识别病变图像,提出基于卷积网络和时变冲量学习的苹果病变图像识别方法。引入卷积、采样算子,通过基于时变冲量的参数训练过程实现网络自我优化,自动提取果园物联网传感器采集的果体图像病变特征,并对病变类别予以识别。实验结果显示,与浅层学习方法 及深度学习方法相比,该方法识别性能优势明显,正确率为97.45%,收敛速度快,并能维持较好的后期稳定性,对于不同基准数据集有较好的泛化能力。

关键词: 病变图像, 卷积, 机器学习, 时变冲量, 预警

Abstract: In order to implement intelligent and real-time alerting to orchard disease-pest based on the image sensed by sensors,a method for pathologic image recognition-diagnosis based on convolution network and a variable impulse learning algorithm addressed at convolution operation are proposed.In addition,a novel scheme for updating parameter and optimizing network by variable impulse is designed.Experimental results on apple pathologic sensing image suggests that compared with shallow learning method and depth learning method,the proposed method presents an obvious edge of recognition performance,of which the accuracy rate stands at 97.45%,and it converges relatively fast,with a piecewise smooth,weak fluctuation in the convergence zone.Analysis of the data from the different benchmark data-sets also demonstrates the method presents a fair generalization.

Key words: lesion image, convolution, machine learning, time variable impulse, alert

中图分类号: