基于粗糙集和BP神经网络的棉花病害识别

为了提高棉花病害的识别率,提出了一种在自然环境条件下基于粗糙集和BP神经网络的棉花病害识别方法。该方法以轮纹病、角斑病、褐斑病和盲椿象为研究对象,将病害棉花图像从RGB颜色空间转换到HSI和L＊a＊b＊颜色空间,应用Otsu算法对H分量、a＊分量和b＊分量进行阈值分割,通过H＋a＊＋b＊分量与原始图像的交集提取棉花病斑区域,利用颜色矩和灰度共生矩阵分别提取病斑的颜色和纹理特征,并结合粗糙集理论和BP神经网络,实现特征向量的优选,和棉花病害的识别。通过比较试验发现,粗糙集理论能有效减少特征维数,使提取的全部特征向量16个减少到5个,使BP神经网络的训练时间缩短到原来的1/4,且棉花病害平均识别正...

Full description

Saved in:
Bibliographic Details
Published in农业工程学报 Vol. 28; no. 7; pp. 161 - 167
Main Author 张建华 祁力钧 冀荣华 王虎 黄士凯 王沛
Format Journal Article
LanguageChinese
Published 现代农业装备优化设计北京市重点实验室%中国农业大学信息与电气工程学院,北京100083 2012
现代农业装备优化设计北京市重点实验室
中国农业大学工学院,北京100083
Subjects
BP神经网络
棉花
病害
粗糙集
识别
病害
识别
BP神经网络
棉花
粗糙集
Online AccessGet full text
ISSN1002-6819
DOI10.3969/j.issn.1002-6819.2012.07.027

Cover

More Information
Summary:为了提高棉花病害的识别率,提出了一种在自然环境条件下基于粗糙集和BP神经网络的棉花病害识别方法。该方法以轮纹病、角斑病、褐斑病和盲椿象为研究对象,将病害棉花图像从RGB颜色空间转换到HSI和L＊a＊b＊颜色空间,应用Otsu算法对H分量、a＊分量和b＊分量进行阈值分割,通过H＋a＊＋b＊分量与原始图像的交集提取棉花病斑区域,利用颜色矩和灰度共生矩阵分别提取病斑的颜色和纹理特征,并结合粗糙集理论和BP神经网络,实现特征向量的优选,和棉花病害的识别。通过比较试验发现,粗糙集理论能有效减少特征维数,使提取的全部特征向量16个减少到5个,使BP神经网络的训练时间缩短到原来的1/4,且棉花病害平均识别正确率达到92.72%。研究结果表明,该方法准确识别了4种棉花病害,为棉花病害的防治提供了有效的技术支持。
Bibliography:Zhang Jianhua ,Qi Lijun ,※,Ji Ronghua .,Wang Hu ,Huang Shikai ,Wang Pei ,（1.College of Engineering,China Agricultural University,Beijing 100083,China;2.College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China 3.Key Laboratory of Optimal Design of Modern Agricultural Equipment in Beijing,Beijing 100083,China）
11-2047/S
cotton; diseases; identification; rough set; BP neural network
In order to improve the recognition rate of cotton diseases,an identification method of cotton diseases based on rough sets and BP neural network under natural environmental conditions was presented.In this method,Otsu method was used to get the threshold of H,a＊and b ＊components from four cotton diseases colored images in the HSI and L ＊a ＊b ＊ color spaces,and diseased regions of cotton were extract by intersection with H＋a ＊ ＋b ＊ component and original image.Color moments and GLCM were used to extract texture features and color features from diseased regions.Features were then used as inputs
ISSN:1002-6819
DOI:10.3969/j.issn.1002-6819.2012.07.027