文献类型：期刊文献

中文题名：基于分割评价函数的多尺度分割参数的选择

英文题名：Parameters of Multi-Segmentation based on Segmentation Evaluation Function

作者：施佩荣[1] 陈永富[1] 刘华[1] 吴云华[2]

第一作者：施佩荣

机构：[1]中国林业科学研究院资源信息研究所;[2]西藏自治区林业调查规划研究院

年份：2018

卷号：33

期号：4

起止页码：628-637

中文期刊名：遥感技术与应用

外文期刊名：Remote Sensing Technology and Application

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD2017_2018】；

基金：国家科技基础性工作专项"中国森林植被调查"(2013FY111600)

语种：中文

中文关键词：分割评价函数;分割尺度;形状指数;紧致度;Landsat;8;OLI;GF-1

外文关键词：Segmentation evaluation function;Scale;Shape;Compactness;Landsat 80LI;GF-1

分类号：TP75

摘要：利用遥感影像进行林地分类时,面向对象分割参数的设置对影像分割结果至关重要,进而影响遥感影像的分类结果。分割评价函数是检验分割效果的重要标准,利用分割尺度、形状指数和紧致度3个参数在不同水平下组合得到的分割评价函数值来评价分割效果的优劣。在已有分割评价函数的基础上进行改进,并对加入面积权重因子后是否对参数选择产生影响进行研究。基于统计分析的方法,利用方差分析和相关性分析法研究4种分割评价函数与分割参数的关系。从全国范围内随机选取Landsat 8OLI和高分一号(GF-1)影像数据各10景作为实验样本,探讨3个参数对分割结果的影响,实验结果表明:(1)分割尺度对分割的影响最大,形状指数次之,紧致度最小;(2)形状指数取值偏小,紧致度取值偏大,对分割效果好;(3)加入面积权重因子提高分割评价函数的稳定性;(4)改进的方法与现有的方法相关性显著,因此适宜作为评价的标准;(5)影像分辨率不同不会对参数的选择产生显著影响。
The segmentation parameters is key to the segmentation result in the object-oriented classification.Further,it would effect the result of the classification.Segmentation evaluation function is a standard which is significant to the quality of segmentation.Scale,shape and compactness could evaluate the quality of the segmentation by combining from the different levels of the three parameters.We improved the methods on the segmentation evaluation function,and digged into the affectation of the weight of area.The methods of variance analysis and correlation analysis were used to analyze the effect of the four segmentation evaluation functions with scale,shape and compactness.There were 10 pieces of images of Landsat 8 OLI and GF-1 as samples of the experience,which were selected from the county.It turned out that：First,the segmentation scale is the most important parameter to the result and the shape is heavier than the compactness.Second,the high quality of the segmentation ask for small shape and big compactness.Third,the area could improve the stability of the segmentation evaluation function.Forth,the proposed method correspond to the existed method and it could evaluate the segmentation.Fifth,the different resolution had the same effect on the selection of segmentation parameters.