一种基于分布差异的分形网络演化分割方法  

A fractal net evolution multi-scale segment method based on distribution difference

作　　者：吕剑峰[1,2] 李海涛[2] 韩颜顺[2] 顾海燕 余凡[2] 

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]中国测绘科学研究院,北京100039

出　　处：《测绘科学》2015年第7期45-48,72,共5页Science of Surveying and Mapping

基　　金：国家科技支撑计划项目(2012BAH28B03)

摘　　要：针对分形网络演化多尺度分割方法对高分辨率遥感影像存在的欠分割问题,文章提出以KL散度为核心的区域合并标准,研究出一种基于KL散度原理的改进分形网络演化多尺度分割方法。该方法相对于分形网络演化多尺度分割方法,在合并标准上更充分利用了遥感影像的光谱特征,能够很好地消除因光谱差异造成的欠分割,对同质性较强的地物效果尤为明显。通过WorldView-2高分辨率影像分割实验,比较欠分割率,结果证明该方法更适合于高分辨率遥感影像分割,能够为地物信息提取提供与自然更吻合的实体对象。Aiming at the problem of lack-of segmentation existed in fractal net evolution multi-scale segmentation methods for high resolution remote sensing image, this paper presented a region merging standard with KL divergence as the core, improved a new fractal net evolution multi-scale segmentation method based on the principle of KL divergence. Compared with the fractal net evolution multi-scale segmentation method, this method could efficiently eliminate the lack-of segmentation caused by the spectral differences with a better combination standard which makes full use of the spectral characteristics of remote sensing image, especially for the objects with strong homogeneity. Through the experiment of segmentation for WorldView-2 high resolution image, by comparing the lack-of segmentation rate, it proved that this method could provide a more consistent entity object with the nature for extraction of the object information, which would suit for high resolution remote sensing image segmentation.

关 键 词：分形网络演化 多尺度分割 KL散度 高斯分布 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]