基于ADE的ReliefF特征选择在高光谱图像分类中的应用

针对高光谱图像谱段数目较多、近邻谱段相关性过高而导致分类困难的问题,提出了一种自适应差分进化特征选择的高光谱图像分类算法.首先初始化种群向量集,利用自适应差分进化算法搜索特征的自适应性生成特征子集;然后,通过使用ReliefF技术根据特征排序去除重复特征,从而为所有的特征构建一个特征列表;最后,借助于模糊k-近邻分类器计算每个向量的分类精度,利用包裹模型评估特征子集.在印第安纳数据集和KSC数据集上的实验结果验证了算法的有效性及可靠性,实验结果表明,相比其他几种特征选择算法,该算法取得了更高的总分类精度和更好的Kappa系数.

Application of Selecting Features Using ReliefF Based on ADE in Hyperspectral Image Classification

For the issue that the number of frequencies is large and nearest neighbor frequencies has high correlation in hyperspectral images which causes its classification is difficult, a hyperspectral images classification algorithm with adaptive differential evolution selecting features is proposed. Firstly, population vector set is initialized and feature subset is generated by using adaptive differential evolution searching adaptivity of features. Then, ReliefF technique is used to wipe repetitive features out so as to constructing a feature tables. Finally, fuzzy k-nearest neighbor classifier is used to calculate of classification accuracy each vector, and wrapper model is used to estimate feature subsets. The effectiveness of proposed algorithm has been verified by experiments on Indiana data sets and KSC data sets. Experimental results show that proposed algorithm has higher overall accuracy and better Kappa coefficient than several other feature selecting algorithms.