摘要
针对准确识别小麦常见病害的需要,提出了一种基于卷积神经网络的小麦病害识别方法。该方法首先以小麦病害图片资料为基础,利用中值滤波法、直方图阈值法等对图像进行去背景、去噪、病斑分割等预处理形成样本库,然后利用卷积神经网络构建一个具有五层结构的深度学习模型进行样本学习,并利用随机梯度下降法进行学习过程控制,最后以获取的特征集对小麦图片进行病害识别,并形成一个在线识别系统。在泰安市4样点的试验结果表明,利用该方法可以有效实现对小麦常见病害——纹枯病、条锈病、叶锈病、秆锈病、赤霉病和白粉病的识别,综合识别率可达99%以上,可以应用于实际生产管理。
Based on the needs of identifying wheat's common diseases accurately,a wheat disease recognition method based on convolutional neural network was put forward in this study. This method was based on the wheat disease image data,firstly it formed a sample database after a series of pre-treatments including removing background,denoising,segmentation of lesion with median filter and histogram thresholding method;then it built a five-layer-structure model of deep learning sample for learning using convolutional neural network,and using stochastic gradient descent method to control learning process; finally,it identified the wheat's pictures with obtained feature set,and formed an online identification system. The experimental results in Taian showed that this method could effectively identify the common diseases of wheat,and the comprehensive recognition rate reached more than 99%,which could be applied to the actual production management.
出处
《山东农业科学》
2018年第3期137-141,共5页
Shandong Agricultural Sciences
基金
山东省自然科学基金(ZR2017MD018)
中国气象局旱区特色农业气象灾害监测预警与风险管理重点实验室开放研究项目(CAMF-201701)
关键词
小麦病害
卷积神经网络
在线识别
病害识别
Wheat disease
Convolutional Neural Network
Online identification
Disease identification
作者简介
张航（1994-）,硕士,研究方向：深度学习图像提取.E-mail：sdauzh＠163.com;通讯作者：张承明（1972-）,博士,教授,研究方向：深度学习图像提取.E-mail：chming＠sdau.edu.cn
