应用RF和SVM的城镇土地利用面向对象分类  

Object-Based for Land Corver Classification of Urben Areas by Methods of RF and SVM

作　　者：涂梨平 饶俊 舒斯红 刘利敏 TU Liping;RAO Jun;SHU Sihong;LIU Limin(Jiangxi Nuclear Surveying and Mapping Institute,330038,Nanehang,PRC;Colledge of Computer,Hubei University of Education,430205,Wuhan,PRC)

机构地区：[1]江西核工业测绘院,南昌330038 [2]湖北第二师范学院计算机学院,武汉430205

出　　处：《江西科学》2018年第5期771-776,共6页Jiangxi Science

基　　金：湖北省教育厅科学研究计划资助项目(编号:Q20173006)

摘　　要：中尺度城镇土地资源空间信息提取是资源环境监测的重要内容。以鄂州、黄冈区域城镇为案例,基于Landsat 8数据,使用面向对象方法提取地类光谱、纹理、几何和地形特征,并应用RF和SVM算法实施城镇土地利用分类。结果表明,合理尺度分割能够增强用地类型可识别性,提升解译效率; RF和SVM算法很好地模拟了地类对象属性特征地物类别间的模式规则,RF分类模型总体精度达89. 18%,Kappa系数为86. 33%,SVM模型总体精度为88. 03%,Kappa系数为81. 60%,整体而言RF分类结果优于SVM。该方案兼具可操作性、准确性,对大中尺度的土地资源信息提取适用性良好。The spatial information extraction of mesoscale urban land resources is an important part of resource environment monitoring. Taking the towns of Ezhou and Huanggang as examples, based on Landsat 8 data, object-oriented methods were used to extract the spectral, textron, geometric and topographic features of the landforms, and RF and SVM algorithms were applied to implement urban land use classification. The results show that the reasonable scale segmentation can enhance the recognizability of land type and improve the interpretation efficiency. The RF and SVM algorithms simulate the pattern rules between the feature categories of the terrestrial object attributes. The overall accuracy of the RF classification model is 89.18%. The Kappa coefficient is 86.33%, the overall accuracy of the SVM model is 88.03% ,and the Kappa coefficient is 81.60%. Overall,the RF classification result is better than SVM. The program has both operability and accuracy, and has good applicability to large and medium-scale land resource information extraction.

关 键 词：面向对象 RF SVM 城镇 

分 类 号：F301.2[经济管理—产业经济] TP75[自动化与计算机技术—检测技术与自动化装置]