结合ReliefF、GA和SVM的面向对象建筑物目标识别特征选择方法

提出结合ReliefF算法、遗传算法(Genetic algorithm,GA)和支持向量机(Support Vector Machine,SVM)的高分辨率遥感影像建筑物目标识别特征选择算法。首先使用ReliefF算法进行初步的特征筛选,然后将SVM参数和特征子集编码到GA染色体中,以SVM识别精度构建适应度函数,同时优化特征子集和SVM参数。实验结果表明,将文中算法应用于建筑物目标识别,能以较小的特征子集和较短的优化时间达到较高的识别精度。

Feature selection method for object-oriented building targets recognition based on ReliefF,GA and SVM

This paper proposes a feature selection algorithm for building targets recognition from high resolutionremotesensingimages ,whichcombinesReliefFalgorithm ,Geneticalgorithm(GA)andSupport Vector Machine(SVM ) .Firstly the algorithm uses ReliefF algorithm for preliminarily feature selection , then the parameters of SVM and feature subset are encoded to GA chromosome ,finally the fitness function is constructed with recognition precision ,white the feature subset and parameters of SVM are optimized simultaneously .The experiment demonstrates that the proposed algorithm can achieve higher recognition accuray with smaller feature subset and less optimizing time ,thus it has great practical value in recognizing building targets .