摘要
为提高村域尺度土地利用分类精度,本文基于高分辨率无人机影像,研究了融合多特征的两阶段分类方法。该方法首先利用基于平均J-M距离增量的“扩充特征子集法”获取最优纹理特征和可见光植被指数,并与原始影像融合;然后,根据地物的具体特征表现,基于规则结合最邻近法分两阶段进行提取。研究结果表明:1)纹理特征和可见光植被指数有助于提高影像分割质量,且基于平均J-M距离增量的“扩充特征子集法”选取的特征相较于同类其他特征更能体现地类间差异化程度;2)相较于全局最优分割尺度下的决策树、支持向量机及随机森林等分类,本文方法总体精度分别高出6.89%、2.66%、5.17%,Kappa系数分别高出11.86%、4.28%、9.04%,在村域土地利用分类方面表现出较强适用性。
In order to improve the accuracy of village-scale land use classification,a two-phase classification method for high resolution UAV imagery based on multi-features fusion was studied.This method firstly obtained the optimal textural feature and visible vegetation index using"extended feature subset method"based on average J-M distance increment,and further fused with spectral features of the original imagery;then rule-based method combined with nearest neighbor classifier was carried out in two stages to extract ground objects according its specific characteristics.The research results showed that:(1)texture features and visible vegetation index can effectively improve the quality of imagery segmentation,and the features selected by"extended feature subset method"based on average J-M distance increment can better reflect the degree of terrain differentiation than other similar features;(2)compared with decision tree,support vector machine and random forest classification with global optimal segmentation scale,the overall accuracy of the method is higher by 6.89%,2.66%and 5.17%,the Kappa coefficient is 11.86%,4.28%and 9.04%,accordingly,which has good applicability in village-scale land use classification.
作者
王宏胜
李永树
张天棋
张雷
王建成
赵乐
WANG Hongsheng;LI Yongshu;ZHANG Tianqi;ZHANG Lei;WANG Jiancheng;ZHAO Le(Survey Branch,POWERCHINA Guizhou Electric Power Engineering Co.,Ltd.,Guiyang 550081,China;Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University,Chengdu 611756,China;Southwest University of Science and Technology,School of Environment and Resource,Mianyang 621010,China)
出处
《测绘与空间地理信息》
2022年第3期44-49,共6页
Geomatics & Spatial Information Technology
基金
中国电建集团贵州电力设计研究院有限公司2020年度科技项目(GZEDKJ-2020-06)资助。
