一种多特征融合的高分辨率遥感图像道路提取算法

High Resolution Remote Sensing Image Road Extraction Algorithm Based on Multi-feature Fusion

利用高分辨率遥感图像提取道路网络对于规划导航等方面具有重要的意义。针对现有道路提取算法的深层次特征挖掘和特征间信息融合存在的不足,提出了一种多特征融合的高分辨率遥感图像道路提取算法。首先,通过图像分割的方式完成对象表达,并提取一种新的空间特征——二阶矩特征对复杂道路形状进行描述;然后利用一种多特征融合方式将光谱特征与空间特征融合,并通过基于机器学习的方法提取初始道路网络;最后利用二阶矩特征实现道路网络的精化。实验结果表明,该算法相比现有算法实验效果更优、鲁棒性更强。

Extracting the road network from high resolution remote sensing image is important for planning navigation and so on.The existing road extraction algorithms lack the feature fusion and deep-level feature mining.This paper proposes a multi-feature fusion road extraction algorithm for high resolution remote sensing imagery.Firstly,the object expression is completed by image segmentation,and a new spatial feature is extracted.The second moment feature can describe the complex road shape effectively.Then,a new multi-feature fusion algorithm is used to fuse the spectrum feature and space feature,and the method of machine learning is used to extract the initial road network.Finally,the second moment feature is used to refine the road network.The experiments show that the proposed algorithm has better accuracy and robust than the existing algorithms.