摘要
针对面向对象分类方法中多尺度分割实验参数选取问题,提出一种双变量随机组合多尺度分割方法,对形状指数选取进行探讨。首先在eCognition软件中利用试错法寻找最佳分割尺度可能存在的区间,然后采用区间内分割尺度和形状指数随机组合方式进行多尺度分割和控制训练样本分类实验,最后利用混淆矩阵对分类结果进行评价。实验结果表明,多尺度分割过程中,当形状指数选择0.3时,结合较为合适的分割尺度,图像的分割结果能够达到更加理想的效果。该方法可为高分辨率遥感影像多尺度分割参数选取提供参考,具有一定的普适性。
Aiming at the selection of experimental parameters for multi-scale segmentation in object-oriented classification methods,a bivariate random combination multi-scale segmentation method was proposed to discuss the selection of shape index.First of all,the method of trial and error in the eCognition software was used to find the best possible interval segmentation scale.Then the multi-scale segmentation and control training samples classification experiments were carried out by random combination of segmentation scale and shape index in the interval.Finally,the confusion matrix is used to evaluate the classification results.The experimental results show that in the process of multi-scale segmentation,when the shape index is 0.3 and the appropriate segmentation scale is selected,the image segmentation results can achieve a more ideal effect.The method in this paper can provide reference for multi-scale segmentation parameters selection of high-resolution remote sensing images,which has certain universality.
作者
罗文强
何浩
LUO Wenqiang;HE Hao(School of Marine Technology and Geomatics,Jiangsu Ocean University,Lianyungang 222005,China;Heilongjiang Fertile Soil Huayuan Technology Development Co.Ltd.,Harbin 150000,China)
出处
《江苏海洋大学学报（自然科学版）》
CAS
2020年第3期72-77,共6页
Journal of Jiangsu Ocean University:Natural Science Edition
关键词
高分辨率遥感影像
多尺度分割
分割尺度
形状指数
high resolution remote sensing image
multi-scale segmentation
segmentation scale
shape index
