L_1稀疏正则化的高光谱混合像元分解算法比较  

Hyperspectral unmixing algorithm based on L_1 regularization

作　　者：邓承志[1] 张绍泉 汪胜前[1] 田伟[1] 朱华生[1] 胡赛凤[1] 

机构地区：[1]南昌工程学院信息工程学院,江西南昌330099

出　　处：《红外与激光工程》2015年第3期1092-1097,共6页Infrared and Laser Engineering

基　　金：国家自然科学基金(61162022;61362036);江西省自然科学基金(20132BAB201021);江西省科技落地计划(KJLD12098);江西省教育厅科技项目(GJJ12632)

摘　　要：基于稀疏性的高光谱解混是近年来高光谱混合像元分解的研究热点。主要研究了L1正则化的高光谱混合像元分解算法。首先分析了L1正则化的三种解混模型,即无约束、非负约束和全约束模型;然后给出了三种模型对应的数值求解算法;最后,采用模拟的和真实的高光谱数据进行实验,比较了三种高光谱混合像元分解算法的效果。实验结果表明:三种模型均具有很好的高光谱混合像元分解精度(SRE),其中全约束模型最好,非负约束模型次之,无约束模型最差;全约束模型在信噪比低和端元数多的情况下,仍然获得较高的SRE。Hyperspectral unmixing based on sparsity is a research hotspot in recent years. This paper studies the hyperspectral unmixing algorithms based on L1 regularization. First we analyzed three unmixing models, including unconstrained model, non-negative constraint model and full-constrained model. And then the corresponding algorithms are presented. In the end, both simulated and real hyperspectral data sets are used to compare and evaluate the proposed three hyperspectral unmixing algorithms. Experimental results demonstrate that three models all have good high-precision. The full constrained model achieves the best unmixing precision（SRE）. The non-negative constrained model is better. And the unconstrained model is worst. In particular, the fully constrained model achieves the higher SRE under the low signal to noise ratio and a large amount of endmembers situation.

关 键 词：高光谱 混合像元分解 稀疏性 增广拉格朗日 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]