一种新的多示例学习方法及其林木学分类应用

J4 ›› 2013, Vol. 35 ›› Issue (4): 150-156.

一种新的多示例学习方法及其林木学分类应用

窦立君1，张金凤2

(1.南京林业大学信息技术学院,江苏南京 210037；2.南京交通职业技术学院,江苏南京 211188)

收稿日期:2012-05-11 修回日期:2012-08-16 出版日期:2013-04-25 发布日期:2013-04-25
基金资助:
国家自然科学基金青年基金资助项目(40801020)；南京林业大学科技创新基金资助项目

A new multiinstance learning method
and its application on trees classification

DOU Lijun1,ZHANG Jinfeng2

(1.School of Information and Technology，Nanjing Forestry University,Nanjing 210037；
2.Nanjing Communications Institute of Technology，Nanjing 211188，China)

Received:2012-05-11 Revised:2012-08-16 Online:2013-04-25 Published:2013-04-25

摘要/Abstract

摘要：

通过设计一种全新的包与包之间的相似性度量方法，即混合型Hausdorff距离，改进了CitationKNN这一经典多示例算法；并通过针对林木自身特殊的成像特点，分析了林木类图像处理的难点，并利用基于小波域变换的处理技术，提出了专门的林木图像特征生成方法，使改进后的算法可以有效实现对林木种类的识别,进而成功将多示例学习引入了林木分类领域。实验证明：新算法不仅对林木分类领域问题的实现效果最佳，同时对公认数据集的测试也取得了良好的结果，与目前主流算法高度可比。

关键词: 多示例学习, CitationKNN, 混合型Hausdorff距离, 林木分类

Abstract:

A new method, named mixHausdorf distance, was designed to measure the similarity amongst packages. It can improve the performance of a classic multiinstance algorithm called CitationKNN. According to trees’ distinct imaging features, the paper analyzed the difficulty of handling trees graphs. Using wavelet transformation handling technique, a specialized method for generating the feature vector of tree images was proposed in order to make the improved algorithm identify the trees effectively. Our experiment proved that, compared with the main existing algorithms, the proposed algorithm has best effect for trees classification and good results when testing the recognized dataset.

Key words: multi-instances learning;CitationKNN;mix-Hausdorf distance;trees classification

窦立君1，张金凤2. 一种新的多示例学习方法及其林木学分类应用[J]. J4, 2013, 35(4): 150-156.

DOU Lijun1,ZHANG Jinfeng2. A new multiinstance learning method
and its application on trees classification [J]. J4, 2013, 35(4): 150-156.