一种顾及形状特征的遥感图像道路提取方法  

Remote Sensing Image Road Extraction Method Considering Shape Feature

作　　者：裔阳 周绍光[1] 刘文静[1] 赵鹏飞[1] 

机构地区：[1]河海大学地球科学与工程学院,江苏南京211100

出　　处：《地理空间信息》2017年第4期47-50,共4页Geospatial Information

基　　金：国家自然科学基金资助项目(41271420/D010702)

摘　　要：依照单类分类和主动学习的基本原理,利用光谱信息和道路几何信息共同实现道路提取。首先人工标记部分道路与非道路样本,用SVDD训练筛选出初始道路与非道路样本点代入SVM得到初始分类图,路径开运算处理之后进行直线匹配获取二值道路图。接下来是一个主动学习过程,根据样本点离超平面的距离及与二值道路图的匹配结果选取最终样本。将留下的样本点代入SVM中并赋以一定权值迭代训练,迭代固定次数之后选取正确率最高的SVM模型。最后利用路径开运算处理获取初始道路,经过形态学后处理得到最终的道路。实验结果表明,该方法可以有效地从高光谱遥感影像中提取道路。According to the basic principle of one class classification and active learning,this paper used spectral information and road geometry information to realize road extraction.Firstly,the paper marked some road and non-road samples manually,through training SVDD to select initial road and non-road samples,and brought them into SVM to get initial classification map.After path opening calculation processing,the paper continued to linear match to get binary road map.Next was an active learning process.According to the distance to the hyperplane and the match result with the binary road map to select final samples,the paper gave a certain weight to conduct SVM iterative training,and selected the model which had the highest accuracy.Finally,the paper used path opening calculation to achieve initial road and got final road by morphological processing.The experimental result shows that this method can effectively extract the road from hyperspectral remote sensing images.

关 键 词：支持向量数据描述 支持向量机 道路提取 路径开运算 主动学习 

分 类 号：P237[天文地球—摄影测量与遥感]