计算机与现代化

• 图像处理 • 上一篇    下一篇

 基于图像内容检索的作物病害识别数据库系统研究

  

  1. 德宏师范高等专科学校计算机科学系,云南德宏678400
  • 收稿日期:2014-12-24 出版日期:2015-04-27 发布日期:2015-04-29
  • 作者简介: 濮永仙(1975-),女,云南腾冲人,德宏师范高等专科学校计算机科学系副教授,硕士,研究方向:机器视觉诊断作物病害,智能农业。
  • 基金资助:
    云南省教育厅科研基金资助项目(2013Y571)

Research on Crop Disease Recognition Database System Based on Image Content Retrieval

  1. Computer Science Deptartment, Dehong Teachers College, Dehong 678400, China
  • Received:2014-12-24 Online:2015-04-27 Published:2015-04-29

摘要:

 针对作物病害诊断专家系统存在不足,提出基于图像检索的方式诊断病害。通过病害图像建立特征库、知识库、图像库等数据库,诊断时提取检索图像特征,求特征与特征库中特征的相似
度距离,按距离大小返回相似图像,依据返回结果获取病害的描述及防治措施。为提高检索的查准率和查全率,重点探索了病害图像数据库创建中特征库信息的获取和检索算法的设计。以烟草病害图像
为例,采用基于支持向量机与多特征选择检测彩色病斑边缘的方式分割病斑,提取病斑特征25个,利用双编码遗传算法和支持向量机对特征降维,以获取表征病害的有效特征17个及对应权重,对特征归
一化处理后建立数据库,并设计检索算法。实验结果表明,构建的图像数据库系统具有较高的查准率和查全率,其中融合病斑的多个特征检索的查准率比单一特征高。用这种方式诊断病害,除有较高的
病害识别率外,还有诊断结果的可视化,将其用于作物病害诊断专家系统中,将提高系统的鲁棒性,为实现病害的远程在线诊断提供了条件。

关键词: 基于内容的病害图像检索, 病斑分割, 特征提取, 图像数据库, 图像相似度

Abstract:

In view of the shortcomings for crop disease diagnosis expert system, the poper puts forward the way that disease diagnosis diseases based on image retrieval.
Try to establish image characteristics library database, knowledge base , image library and so on according to the plant disease images. Diagnosis, extract the characteristics
of the plant disease image, the characteristics and features of library matching query, through the query results for a description of the disease and prevention measures. In
order to improve the retrieval precision and recall rate, the poper mainly explores the characteristics of library information in the disease image database creation access and
retrieval algorithm design. Disease spot in tobacco plant disease image, for example, the color disease spot edge detection of based multifeature selection and on support vector
machine , 25 disease spot feature extracting, using double coding genetic algorithm and support vector machine (SVM) for feature dimension reduction, in order to get effective
characterization of disease features 17 and the corresponding weights, the characteristics of the normalized processing after establishing a database, and retrieval algorithm is
designed. Experiments show that the constructed image retrieval system has high precision and recall, the fusion of multiple features of disease spot retrieval precision is
higher than single feature. Diagnosis of disease in this way that includes higher disease recognition rate and diagnosis of visualization, the way is used for crop disease
diagnosis expert system that can improve the robustness of the system and achieve the remote online diagnosis of the disease.

Key words:  content-based disease image searching, disease spot segmentation, feature extraction, image database, image similarity