Abstract: The road network extracted by the traditional data model is not accurate enough, so a new data model is defined. The traditional birth-death process of searching is blindness and a new proposal nuclear is proposed based on edge detection, and it is greatly accelerating the convergence speed. For traditional proposal will destroy connectivity of the network, it designs a new proposal nuclear based on birth and death of the neighborhood and a new parameters proposal nuclear. Experimental results show that the improved algorithm greatly improves the convergence rate, and the extracted road networks are more accurate and continuous.

Key words: road extraction, marked point process, reversible jump Markov chain Monte Carlo method, proposal nuclear

摘要： 研究了标值点过程的道路提取算法，针对传统数据模型提取道路不够准确的缺点，改进了数据模型。提出了基于边缘检测的生灭转移核，避免了传统的生灭过程搜索的盲目性，大大加快了算法的收敛速度。针对传统转移核容易破坏线段的连接性的缺点，定义了多种新型的RJMCMC转移核，重新设计了基于邻域的生灭转移核及线段参数转移核。仿真结果表明，改进算法大大提高了收敛速度，并且提取的道路网络更准确，更连续。

关键词: 道路提取, 标值点过程, 可逆跳跃马尔可夫链蒙特卡罗方法（RJMCMC）, 转移核