结合改进SWT和EMD的高分遥感影像道路提取

Road Extraction of High Resolution Remote Sensing Imagery Based on Improved SWT and EMD

针对利用传统笔画宽度变换(stroke width transform,SWT)提取遥感影像道路阈值无法自动确定以及精度不够高的问题,提出了一种结合改进SWT和融合多特征后的陆地移动距离(earth mover’s distance,EMD)进行高分遥感影像道路提取的方法。首先,结合连通分量分析,将笔画宽度变换改进成可自动确定提取阈值的方法,进行初步道路提取;其次,确定一个道路参考区域,利用连通区作为索引,分析道路参考区域与其他区域之间的特征相似度,通过提取纹理与光谱特征,计算加权融合后的特征EMD值,设定合理的阈值过滤大部分非道路区域,提取出其他道路区域;最后,进行形态学后处理得到最终提取结果。通过实验实现该方法并与现有道路提取方法进行对比。经过评估,证明该方法在提取不同场景的道路时均能达到理想效果且具有较高精度。

To address the problem of insufficient determination and accuracy of road thresholds for extracting remote sensing images using traditional stroke width transformation(SWT),this paper proposes a method to extract roads from high-resolution remote sensing images by combining improved SWT and EMD(earth mover’s distance)after fusion of multiple features. First,combined with the analysis of connected components,the stroke width transformation is improved into a method that can automatically determine its extraction road threshold to obtain the optimal result. Secondly,determine a road reference area,and use the connected area as an index to analyze the feature similarity between the road reference area and other areas. By extracting texture and spectral features and calculating the weighted and fused feature EMD value,set a reasonable threshold filter. For some non-road areas,other road areas are extracted. Finally,perform morphological post-processing to obtain the final extraction result. The method in this paper is realized through experiments and compared with the existing road extraction methods. After qualitative and quantitative evaluation,it is proved that the method can achieve the ideal effect and has high accuracy when extracting roads in different scenes.