基于标值点过程的遥感影像道路网络提取

Road Network Extraction from Remote Sensing Image Based on a Marked Point Process

针对高分辨率遥感影像道路网络的特点,采用基于贝叶斯理论的全自动方法从遥感影像中提取道路。根据道路的局部和全局特征,使用标值点过程对道路建模,采用结合可逆跳跃马尔可夫链蒙特卡罗算法的模拟退火算法优化求得全局最优解。提出新的预处理方法得到道路的位置和方向信息,提出基于预处理的生灭转移核以降低算法的搜索空间,提出基于连接的移动转移核以加快算法的收敛速度。实验结果表明,该方法可以快速、有效地从不同的遥感影像(光学、SAR)提取道路网络。

A method based on the Bayesian theory is presented to extract road networks in remote sensing images. A model based on a marked point process is designed to exploit as fully as possible the properties of the network, and the optimization is done via simulated annealing using a Reversible Jump Markov Chain Monte Carlo algorithm. A new preprocessing method is proposed to extract the location and orientation information. A birth - and - death proposal kernel based on preprocessing is proposed to reduce the searching space. A move proposal kernel based on connection is proposed to accelerate the convergence of the algorithm. The experimental results show that this method can extract road networks from different kinds of remote sensing images fast and efficiently.