中国水稻科学

中国水稻科学 ›› 2012, Vol. 26 ›› Issue (5): 619-623.DOI: 10.3969/j.issn.10017216.2012.05.016

• 研究简报 • 上一篇    下一篇

基于模板匹配的多目标水稻灯诱害虫识别方法的研究

吕军1 ,姚青1,刘庆杰1,薛杰1,陈宏明3,杨保军2,唐健2，*   

  1. 1浙江理工大学 信息电子学院， 浙江 杭州 310018；2 中国水稻研究所， 浙江 杭州 310006；3象山县植保植检站， 浙江 象山  315700；
  • 收稿日期:2011-11-16 修回日期:2012-05-16 出版日期:2012-09-10 发布日期:2012-09-10
  • 通讯作者: 唐健2，*
  • 基金资助:

    国家自然科学基金资助项目（31071678）； 浙江省重大科技项目（2010C12026）； 中央级公益性科研院所基本科研业务费专项资金资助项目（2009RG0042）； 宁波市科技项目（2010C10044）； 浙江理工大学研究生创新基金资助项目（YCXS11022）。

Identification of Multiobjective Rice Lighttrap Pests Based on Template Matching

LV  Jun1 , YAO  Qing1, LIU  Qingjie 1, XUE  Jie 1, CHEN  Hongming 3, YANG  Baojun 2, TANG Jian 2,*   

  1. 1 College of Informatics and Electronics, Zhejiang SciTech University, Hangzhou 310018, China; 2 China National Rice Research Institute, Hangzhou 310006, China; 3 Xiangshan County Plant Protection and Quarantine Station, Xiangshan 315700, China;
  • Received:2011-11-16 Revised:2012-05-16 Online:2012-09-10 Published:2012-09-10
  • Contact: TANG Jian2,*

PDF

摘要： 水稻灯诱害虫的识别与计数在水稻田间害虫监测中是非常重要的。由于水稻害虫被黑光灯诱集后姿态各异，存在虫体残缺现象，增加了图像自动识别的难度。在获取水稻灯诱害虫非粘连图像基础上，利用模板匹配和K折交叉验证方法进行多目标水稻灯诱害虫的识别。首先，提取每个水稻害虫图像中包括颜色、形态和纹理共156个特征参数；然后，利用主成分分析法进行数据降维，选取前6个主成分作为害虫特征参数；最后，根据每种灯诱害虫的姿态确定模板数，通过模糊C均值获得聚类中心作为模板参数，分别利用单模板和多模板匹配方法进行水稻害虫的识别。结果表明，针对姿态各异且有虫体残缺的多目标水稻灯诱害虫，多模板和单模板匹配法的识别率分别为83.1%和59.9%。

关键词: 水稻灯诱害虫, 害虫识别, 图像处理, 特征提取, 模板匹配

Abstract: Identification and count of rice lighttrap pests are very important in monitoring rice pests. The pests trapped by black light lamps show different postures and incomplete bodies, which increase the difficulty of image automatic identification.  Template matching and Kfold cross validation methods were used to identify multiobjective rice lighttrap pests based on nontouching pest images. Firstly, one hundred and fiftysix features including color, shape and texture features were extracted from each pest image. Secondly, the principal component analysis was employed for reducing data dimensionality, and first six principal components were selected as pests’ features. Then, the template number was determined according to the gesture of each pest species  and the template parameters were obtained from the cluster centers by fuzzy Cmean clustering method. Finally, single and multitemplate matching methods were used to identify rice pests. The results showed that the accurate rate of multitemplate matching and single template matching were 83.1% and 59.9%, respectively for rice lighttrap pests with multiple postures and some incomplete bodies.

Key words: rice lighttrap pest, pest identification, image processing, feature extraction, template matching

中图分类号: