利用稀疏自表达实现高光谱影像波段选择

发布时间：2018-07-17 22:13

【摘要】：提出一种稀疏自表达方法来研究高光谱影像分类中的波段选择问题。该方法利用字典矩阵等于测量矩阵的条件来改进多观测向量的稀疏表达模型,将波段子集看作高光谱影像波段集合中的代表子集。稀疏自表达方法将波段选择转换为寻求多观测向量中稀疏系数矩阵的非零行向量问题,通过引入混合范数来限定非零元素行向量的个数,利用快速交替方向乘子方法求解稀疏系数矩阵,并聚类非零行向量,实现波段的有效选择。基于两个公开高光谱影像数据集并对比其他4种波段选取方法来验稀疏自表达方法。实验结果证明,稀疏自表达方法能够在计算效率明显优于基于波段相关性的线性限制最小方差方法的同时,取得与该方法和非负稀疏矩阵分解方法相匹甚至略高的总体分类精度。
[Abstract]:A sparse self-representation method is proposed to study the band selection in hyperspectral image classification. The sparse representation model of multi-observation vectors is improved by using the condition that the dictionary matrix is equal to the measurement matrix, and the band subset is regarded as the representative subset in the band set of hyperspectral images. The sparse self-representation method transforms the band selection into the nonzero row vector problem which seeks the sparse coefficient matrix in the multi-observation vector, and defines the number of the non-zero element row vectors by introducing the mixed norm. The sparse coefficient matrix is solved by using the fast alternating direction multiplier method and the nonzero row vector is clustered to realize the effective band selection. Based on two open hyperspectral image data sets and comparing with the other four band selection methods, a sparse self-representation method is tested. The experimental results show that the sparse self-representation method is more efficient than the linear constrained minimum variance method based on band correlation. Compared with this method and the non-negative sparse matrix decomposition method, the overall classification accuracy is even higher.
【作者单位】：航空遥感技术国家测绘地理信息局重点实验室;宁波大学建筑工程与环境学院;武汉大学测绘遥感信息工程国家重点实验室;上海师范大学城市发展研究院;
【基金】：国家自然科学基金(41671342,41401389) 航空遥感技术国家测绘地理信息局重点实验室经费资助课题(2015B12) 中国博士后科学基金(2016T90732,2015M570668) 浙江省科技厅公益项目(2016C33021)~~
【分类号】：P237

【相似文献】