基于适应度函数和染色体信息量排序的高光谱影像特征选择方法

Hyperspectral image feature selection method based on fitness function and chromosome information quantity ranking

针对高光谱遥感影像数据中存在较多冗余信息的问题,以染色体的信息量排序为基础,构建联合条件互信息和多元互信息的适应度函数,提高所选特征可以提供的信息量,将适应度函数作为差分进化算法的评价标准,通过最大化适应度函数获得最优特征子集,提出一种新型光谱特征选择算法,使用每条染色体中所选特征的信息量来计算相关性。实验结果表明:在16类地物中该算法在9类上分类准确度最高,说明将基于信息量的相关性的估算作为适应度函数与群体智能优化算法相结合能更好地应用于高光谱遥感影像的光谱特征选择。

To solve the problem of redundant information in hyperspectral remote sensing image data,a fitness function of joint conditional mutual information and multivariate mutual information is constructed based on the information ranking of chromosomes to improve the amount of information provided by the selected features.The fitness function is used as the evaluation standard of differential evolution algorithm,and the optimal feature subset is obtained by maximizing the fitness function.A new spectral feature selection algorithm is proposed.The correlation is calculated using the amount of information of the selected feature in each chromosome.The experimental results show that the algorithm achieves the maximum classification accuracy on 9 out of 16 categories of ground objects,indicating that the estimation based on the correlation of information content as the fitness function combined with the swarm intelligence optimization algorithm can be better applied to the spectral feature selection of hyperspectral remote sensing images.