浙江农业学报 ›› 2018, Vol. 30 ›› Issue (6): 1073-1081.DOI: 10.3969/j.issn.1004-1524.2018.06.26

• 农业经济与发展 • 上一篇    下一篇

基于参数指数非线性残差神经网络的脐橙病变叶片识别

杨国亮, 许楠*, 康乐乐, 龚曼, 洪志阳   

  1. 江西理工大学 电气工程与自动化学院,江西 赣州 341000
  • 收稿日期:2017-08-17 出版日期:2018-06-20 发布日期:2018-07-02
  • 通讯作者: 许楠,E-mail: xunan003@126.com
  • 作者简介:杨国亮(1973—),男,江西丰城人,博士,教授,主要研究方向为图像处理与模式识别。E-mail: ygliang30@126.com
  • 基金资助:
    国家自然科学基金(51365017)

Identification of navel orange lesions leaves based on parametric exponential non-linear residual neural network

YANG Guoliang, XU Nan*, KANG Lele, GONG Man, HONG Zhiyang   

  1. School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou 341000, China
  • Received:2017-08-17 Online:2018-06-20 Published:2018-07-02

摘要: 提出一种参数指数非线性(PENL)函数来改进残差网络,利用深度学习的新方法识别脐橙叶面病变,减少了整流线性损失,提升了训练效果。以脐橙叶面图像为样本,进行CNN训练,以区分出病变、缺素、正常及非此类物种4种类型,实现了对于脐橙疾病检测迅速且方便应用的分类模型,相比于传统植物病变识别方法具有极大的优势,最终识别准确率达到了97.18%~98.86%。

关键词: 神经网络, 脐橙, 病变识别, 深度学习, 残差

Abstract: In order to better identify navel orange leaf lesions. The paper proposed a parametric exponential nonlinear function asactivation function to improve the residual network and provided a new method of deep learning to identify lesions of navel orange leaves, which reduced the linear loss of rectification and improves the training effect. In this paper, CNN was used as a training tool, and navel orange leaf images were selected as training samples to discriminate the four types of diseased, deficient, normal and non-species. This new identification method had the great advantage of being convenient to use and more accurate for the identification of traditional plant diseases, and finally reached the accuracy rate of 97.18%-98.86%.

Key words: neural networks, navel orange, disease identification, deep learning, residual

中图分类号: