摘要
针对高分遥感影像分类过程中面临的特征维数高、数据冗杂度严重问题,从机器学习的角度提出了混合粒子群优化遗传算法的特征优化方法。此方法发挥2种机器学习算法优势,以Relief F算法进行初步特征筛选,再利用新二进制粒子群优化遗传算法确定优化特征集用于随机森林分类器进行城市用地信息的提取。通过与全特征、Relief F算法、GABPSO算法3种特征提取方法进行比较,验证此方法的优越性。结果表明,基于Relief F和GANBPSO算法的混合特征选择方法能够在提取较少特征变量的情况下获得较高的精度,总精度和Kappa系数分别为91.17%和0.874,与传统方法相比具有更好的分类效果。
Aiming at the problems of high feature dimension and data complexity in the process of high resolution remote sensing image classification,a hybrid particle swarm optimization genetic algorithm is proposed from the perspective of machine learning.This method takes advantage of two machine learning algorithms,and uses the Relief F algorithm to perform preliminary feature screening.Then,it uses the new binary particle swarm optimization genetic algorithm to determine the optimal feature set for random forest classifier to extract urban land.The advantages of this method are verified by comparison with the full feature,Relief F algorithm and GABPSO algorithm.The results show that the hybrid feature selection method based on Relief F and GANBPSO can obtain higher accuracy with fewer feature variables extracted.The total accuracy and Kappa coefficient are 91.17%and 0.874 respectively,which has better classification effect than the traditional method.
作者
唐晓娜
张和生
TANG Xiaona;ZHANG Hesheng(Taiyuan University of Technology,Taiyuan 030024,China)
出处
《遥感信息》
CSCD
北大核心
2019年第6期113-118,共6页
Remote Sensing Information
关键词
高分遥感影像
随机森林
RELIEF
F算法
粒子群优化遗传算法
特征选择
high resolution remote sensing image
random forest
Relief F algorithm
particle swarm optimization genetic algorithm
feature selection
