摘要
为提高滨海区域湿地信息提取的精度,以曹妃甸地区为例,通过融合主成分分析、形状、纹理、几何、水体指数、植被指数等52个特征变量,采用Relief-F算法模型优选出20个特征变量,对比分析C5.0、CART、QUEST决策树算法在滨海区域的分类精度.研究结果表明:特征优选下QUEST决策树方法的分类精度最高,总体分类精度为86.9%,Kappa系数为0.84.基于特征优选的决策树分类精度均高于未特征优选下的决策树与未特征优化的QUEST决策树相比,在草本沼泽和泥沙质滩涂两种土地类型上,分类精度有明显提高.
In order to improve the accuracy of wetland information extraction in coastal areas,taking Caofeidian area as an example,by fusing 52 feature variables such as principal component analysis,shape,texture,geometry,water index,and vegetation index,20 features were selected using the Relief-F algorithm model.Variables,comparative analysis of the classification accuracy of C5.0,CART,QUEST decision tree algorithm in the coastal area.The results of study show that based on feature optimization,the accuracy of QUEST decision tree method is the highest,the overall classification accuracy is 86.9%,and Kappa coefficient is 0.84.The classification accuracy of decision tree based on feature optimization is higher than that of decision tree without feature optimization.Compared with the QUEST decision tree without feature optimization,the classification accuracy of herbaceous swamp and muddy beach has been improved significantly.
作者
郝玉峰
满卫东
汪金花
刘明月
张阔
HAO Yufeng;MAN Weidong;WANG Jinhua;LIU Mingyue;ZHANG Kuo(College of Mining Engineering,North China University of Science and Technology,Tangshan 063210,China;Hebei Key Laboratory of Mining Development and Security Technology,Tangshan 063210,China;Hebei Industrial Technology Institute of Ecological Remediation,Tangshan 063210,China)
出处
《辽宁工程技术大学学报（自然科学版）》
CAS
北大核心
2021年第3期225-233,共9页
Journal of Liaoning Technical University (Natural Science)
基金
河北省自然科学基金(D2019209317)
河北省自然科学基金(D2019209322)
唐山市科技计划重点研发项目(19150231E)
关键词
湿地
多尺度分割
信息提取
特征优选
决策树
wetland
multi-scale segmentation
information extraction
feature optimization
decision tree
