摘要
针对基于离散度的特征选择算法存在无法自动确定特征阈值且无法创造新特征等问题,提出了一种基于离散度及支持向量机的遥感影像特征提取算法。该算法首先对原始特征库进行去相关,然后利用优化的离散度指标进行特征优选,进而利用线性不可分支持向量机模型将特征选择结果提取出具有更好判别特性的新特征,并将决策函数作为特征阈值。利用陕西陇县光伏场地无人机数据进行对比实验,分析结果表明,利用新算法进行分类的总体精度为93.5%,Kappa系数为0.9,相比基于离散度的特征提取算法分别提高7%和0.1,而且在各地物的生产者精度和用户精度方面均有一定提升,是一种更优的分类规则集构建方法。
Aiming at the shortcomings of feature selection algorithm based on discretization,such as not being able to determine the feature threshold automatically and not being able to create new features,a remote sensing image feature extraction algorithm based on discretization and support vector machine is proposed.The algorithm first de-correlates the original feature library,then uses the optimized discretization index for feature selection,next extracts new features with better discriminative properties using the linear indivisible support vector machine model with the feature selection results,and uses the decision function as the feature threshold.The results of comparative experiments using UAV data from photovoltaic sites in Long county,Shaanxi show that the overall accuracy of classification using the new algorithm is 93.5%,and the Kappa coefficient is 0.9,which is 7%and 0.1 higher than that of the discretization-based feature extraction algorithm,and there is a certain degree of improvement in the accuracy of the producer and the accuracy of the user for each place,which is a better method of constructing the classification rule set.
作者
曹红新
王宇豪
秦增忍
王帆
朱镇
CAO Hongxin;WANG Yuhao;QIN Zengren;WANG Fan;ZHU Zhen(China Energy Engineering Group Shaanxi Electric Power Design Institute Co.Ltd.,Xi’an 710054,China)
出处
《遥感信息》
CSCD
北大核心
2024年第4期80-86,共7页
Remote Sensing Information
基金
中国能源建设集团陕西省电力设计院有限公司重大科技项目(61-K2022-10)。
关键词
离散度
支持向量机
特征提取
特征选择
面向对象
光伏用地
scatter degree
SVM
feature extraction
feature selection
object-oriented
photovoltaic site
