文献类型：期刊文献

中文题名：多分类器组合森林类型精细分类

英文题名：Multiple Classifiers Combination Method for Precise Classification of Forest Type

作者：王怀警[1] 谭炳香[1] 王晓慧[1] 房秀凤[1] 李世明[1]

第一作者：王怀警

机构：[1]中国林业科学研究院资源信息研究所

年份：2019

卷号：34

期号：2

起止页码：104-112

中文期刊名：遥感信息

外文期刊名：Remote Sensing Information

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD2019_2020】；

基金：浙江省省院合作林业科技项目(2017SY04);高分辨率对地观测系统重大专项(30-Y20A37-9003-15/17-3);国家自然科学基金(31370635)

语种：中文

中文关键词：Hyperion;支持向量机;随机森林;自适应权值;分层分类;森林类型分类;高光谱

外文关键词：Hyperion;support vector machine;random forest;adaptive weight method;hierarchical classification;forest type classification;hyperspectral

分类号：TP751

摘要：针对高光谱遥感数据树种识别精度不高,现有多分类器组合策略难以避免人为因素干扰的问题,利用自适应权值模型组合2种机器学习算法,有效改善森林类型精细识别精度。研究综合利用影像的光谱和纹理特征、地形特征及森林类型外业调查样本数据,采用分层分类的策略,分别利用支撑向量机(support vector machine,SVM)和随机森林算法(random forest classifier,RFC)对森林类型进行精细识别;为进一步提高森林类型识别精度,采用自适应权值组合模型综合2种分类器,并采用分层随机抽样的独立检验样本进行精度验证。结果表明,自适应权值组合模型可综合不同分类器的优势,避免人为因素干扰且提高识别精度和稳定性,对高分五号(GF-5)星载高光谱遥感数据应用具有借鉴意义和参考价值。
In view of the low accuracy of tree species recognition in hyperspectral remote sensing data,and the multi-classifier combination strategy is difficult to avoid the interference of human factors,the adaptive weight model is used to combine two machine learning algorithms to effectively improve the precision of forest type fine recognition.Based on the spectral and texture features of images,topographic features and forest type field survey sample data,a hierarchical classification strategy was used to identify forest types using support vector machine (SVM) and random forest algorithm (RFC) respectively;two classifiers were synthesized by adaptive weight combination model to further improve the accuracy of forest type recognition,and the accuracy was verified by independent test samples of stratified random sampling.The results show that the adaptive weight combination model can combine the advantages of different classifiers,avoid human factors interference and improve the recognition accuracy and stability.It has reference significance and reference value for the application of GF-5 satellite-borne hyperspectral remote sensing data.