基于特征优化的高光谱遥感影像降维算法  

Dimension Reduction Algorithm for Hyperspectral Remote Sensing Image Based on Feature Optimization

作　　者：王小宁 宋伟东[1] WANG Xiaoning;SONG Weidong(School of Geomatics,Liaoning Technology University,Fuxin,Liaoning 123000,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000

出　　处：《遥感信息》2020年第6期122-128,共7页Remote Sensing Information

摘　　要：高光谱影像数据量大、波段间相关性强、信息冗余度高等特点为地物高效识别与分类带来挑战。鉴于降低维度在有效利用高光谱数据方面的重要性,文章提出高光谱影像特征优化降维算法。相关系数矩阵用以确定初始子空间,以此作为先验确定聚类个数及初始聚类中心。依据相似性度量准则,应用K-means算法进行波段聚类,取不同准则下聚类结果交集,实现子空间的自动划分,并利用PCA变换提取第一主成分作为子空间降维结果。对于未被子空间覆盖的剩余波段,采用BSMM算法进行降维处理。叠加2次降维结果,实现最终降维。通过对华盛顿哥伦比亚特区和帕维亚大学2幅影像降维结果的定性定量评价,验证本文算法的可行性与有效性。实验表明,该算法能够在更好实现影像降维的同时极大限度地保留原始影像信息,为后续高光谱影像快速解译提供可能。Hyperspectral images with large amount of data,strong correlation between bands,and high information redundancy pose challenges for the efficient identification and classification of ground objects.It is especially important to reduce the dimension on the basis of effective use of hyperspectral data.In this paper,a dimension reduction algorithm for hyperspectral image based on feature optimization is proposed.The correlation coefficient matrix is used to determine the initial subspace,which is used as a priori to determine the number of clusters and the initial cluster center.According to the similarity measurement criteria,K-means algorithm is applied for band clustering.The intersection of clustering results under different criteria is taken to realize the automatic subspace partition and PCA transform is applied on each subspace to extract the first principal component as the result of subspace dimension reduction.For the remaining bands not covered by the subspace,BSMM algorithm is used to reduce the dimension.The final dimension reduction is achieved by superimposing two dimension reduction results.The feasibility and effectiveness of the proposed algorithm are verified by the qualitative and quantitative evaluation of the two dimension reduction results of Washington D.C.and Pavia University.Experimental results show that the algorithm can better achieve image dimension reduction while retaining original image information to a great extent,which provides a possibility for fast interpretation of subsequent hyperspectral images.

关 键 词：高光谱影像 特征优化 PCA变换 波段选择 降维 子空间 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]