摘要
稀疏解混能够有效地规避高光谱场景中缺少纯像元和估计端元数目的两个瓶颈问题,因而成为目前广泛研究的光谱解混技术。针对协同稀疏解混模型在边界上容易出现错误识别的问题,结合字典削减策略和低秩表示,提出一种协同稀疏低秩的解混模型。该方法同时施加稀疏和低秩约束在丰度矩阵上,并对协同稀疏模型的?2,1混合范数采用加权策略,使得行稀疏性得到了增强,同时也使用非凸logdet(·)作为秩的光滑替代函数。由于提出方法充分利用了高光谱数据的空间信息和光谱信息,因此获得了比协同稀疏回归模型更准确的解混结果。最后利用著名的交替方向乘子方法(ADMM)对提出的非凸模型进行有效求解,实验结果验证了提出算法的有效性。
Sparsity unmixing has become a widely-studied spectral unmixing technique as it can effectively avoid the bottleneck associated with the absence of the number of the pure pixels and the estimated end members in hyperspectral scenes.Since the unmixing model with collaborative sparsity is prone to making wrong recognition on the boundary,an unmixing model with collaborative sparsity and low-rank is proposed by combining dictionary pruning strategy and low-rank representation. In this method,sparsity and low-rank constraints are imposed on the abundance matrix simultaneously,and the weighted strategy is adopted to deal with the mixed norm ?2,1 of the collaborative sparsity model,which further enhances the low sparsity of the abundance matrix. At the same time,a nonconvex logdet( ·)is used as a smooth surrogate function of the rank. Since the spatial information and spectral information of hyperspectral data is fully utilized in the proposed method,more accurate results of unmixing are obtained than that of the cooperative sparse regression model. Finally,the famous alternating direction multiplier method(ADMM) is used to effectively solve the proposed non-convex model. The results of experiments demonstrate the effectiveness of the proposed algorithm.
作者
韩红伟
冯向东
郭科
HAN Hongwei;FENG Xiangdong;GUO Ke(The Engineering&Technical College of Chengdu University of Technology,Leshan 614000,China;Geomathematics Key Laboratory of Sichuan Province,Chengdu University of Technology,Chengdu 610059,China)
出处
《现代电子技术》
2022年第5期67-73,共7页
Modern Electronics Technique
基金
国家重点研发计划课题(2017YFC0601505)
数学地质四川省重点实验室开放基金(scsxdz2019yb01)。
关键词
协同稀疏回归
稀疏表示
高光谱解混
低秩表示
高光谱图像
解混模型
实验分析
cooperative sparse regression
sparse representation
hyperspectral unmixing
low-rank representation
hyperspectral image
unmixing model
experimental analysis
