基于近邻传播算法的高光谱波段选择

Hyperspectral Band Selection Based on Affinity Propagation

由于高光谱图像具有波段之间相关性高,信息冗余性强等特点,高光谱图像降维是高光谱图像预处理中的重要一步。波段选择在降维的同时能够保留原始数据的物理意义,在很多方面有着应用。近邻传播算法(Affinity PropagationClustering,AP)是Fray等在2007年提出的一种聚类方法。它将全部数据点看作潜在聚类中心,根据数据点之间的相关性进行聚类。论文提出一种基于AP聚类的波段选择方法,将小波变换引入聚类算法中相似度和偏好值的计算。将降维结果输入最小距离分类器进行分类,计算分类准确性,并通过数据集Indiana Pines验证,对比实验结果验证了论文提出方法的有效性。

Because hyperspectral images have the characteristics of high correlation between bands and strong information re-dundancy,the reduction in dimension of hyperspectral images is an important step in the preprocessing of hyperspectral images.Band selection can preserve the physical meaning of the original data while reducing dimension,and has application in many as-pects.Affinity Propagation Clustering(AP)is a clustering method proposed by Fray et al.in 2007.It regards all data points as poten-tial clustering centers and clusters the correlations between data points.This paper proposes a band selection method based on AP clustering,which introduces the wavelet transform into the computation of similarity and preference values in the clustering algo-rithm.The dimensionality reduction results are input into the minimum distance classifier for classification,and the classification ac-curacy is calculated.The dataset is validated by the Indiana Pines dataset and the experimental results are compared to verify the ef-fectiveness of the proposed method.