基于多分类器的C5.0决策树植被分类方法

发布时间：2018-01-12 11:40

本文关键词：基于多分类器的C5.0决策树植被分类方法　出处：《图学学报》2017年05期 　论文类型：期刊论文

更多相关文章： 高光谱图像 遥感 SVM SAM MLC C.

【摘要】：针对光谱角制图(SAM)和最大似然(MLC)分类器对AVIRIS高光谱遥感图像进行植被分类精度均不高的问题,提出了一种基于多分类器的C5.0决策树植被分类方法。首先,利用支持向量机(SVM),进行核函数以及核函数参数选择,提取出AVIRIS高光谱图像中的植被信息。其次,利用C5.0算法将光谱角制图和最大似然分类器组合,作为决策树的特征属性,学习样本训练并生成分类规则;根据C5.0算法计算植被样本中对应分类器的信息增益率,选择信息增益率最大的属性去分类样本;当叶样本的分类结果满足停止生长的阈值,输出样本分类的结果,否则,回到开始,递归调用以上方法继续分类叶样本,直到所有子集仅包含一个植被类别的样本完成决策。实验结果表明,与光谱角制图和最大似然分类器相比,本文提出的方法整体精度分别提高了6.04%、2.92%,不仅证实了多分类器组合的可行性和有效性,而且更加适用于AVIRIS高光谱图像中的植被调查。
[Abstract]:The accuracy of vegetation classification for AVIRIS hyperspectral remote sensing images by spectral angle mapping (sam) and maximum likelihood (MLC) classifier is not high. A multi-classifier based C5.0 decision tree vegetation classification method is proposed. Firstly, support vector machine (SVM) is used to select kernel function and kernel function parameters. The vegetation information in AVIRIS hyperspectral image is extracted. Secondly, the spectral angle mapping and the maximum likelihood classifier are combined as the feature attributes of the decision tree using C5.0 algorithm. Learning sample training and generating classification rules; According to C5.0 algorithm, the information gain rate of the corresponding classifier in vegetation samples is calculated, and the attributes with the largest information gain rate are selected to classify the samples. When the classification result of the leaf sample meets the threshold of stop growing, the result of the sample classification is output, otherwise, back to the beginning, recursively call the above method to continue classifying the leaf sample. The experimental results show that compared with the spectral angle mapping and the maximum likelihood classifier, the overall accuracy of the proposed method is 6.04% higher than that of the spectral angle mapping and the maximum likelihood classifier. 2.92, which not only proves the feasibility and validity of the combination of multiple classifiers, but also is more suitable for vegetation investigation in AVIRIS hyperspectral images.
【作者单位】：中北大学信息与通信工程学院;英国雷丁大学系统工程学院;
【基金】：国家自然科学基金项目(61672472) 中北大学科学研究基金项目(XJJ2016024);中北大学电子测试技术重点实验室开放基金项目(ZDSYSJ2015005)
【分类号】：TP751
【正文快照】： 高光谱影像在保留较高空间分辨率同时,光谱分辨率有极大地提高,这使得无论在描述同类地物的细节方面,还是识别不同类别地物的能力等方面都有大幅提高。植被是地理环境重要的组成部分,陆地生态系统植被生产力一直是全球变化领域内的研究热点[1]。利用高光谱遥感技术对植被信息

【相似文献】