[1]吴新浪,叶 军.全变差 Cauchy 非负张量分解高光谱解混算法[J].计算机技术与发展,2022,32(12):21-28.[doi:10. 3969 / j. issn. 1673-629X. 2022. 12. 004]
　WU Xin-lang,YE Jun.Hyperspectral Unmixing Algorithm Based on Total Variation Cauchy Nonnegative Tensor Factorization[J].,2022,32(12):21-28.[doi:10. 3969 / j. issn. 1673-629X. 2022. 12. 004]
点击复制

全变差 Cauchy 非负张量分解高光谱解混算法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年12期
页码:
21-28
栏目:
嵌入式计算
出版日期:
2022-12-10

文章信息/Info

Title:
Hyperspectral Unmixing Algorithm Based on Total Variation Cauchy Nonnegative Tensor Factorization
文章编号:
1673-629X(2022)12-0021-08
作者:
吴新浪; 叶 军
南京邮电大学 理学院,江苏 南京 210023
Author(s):
WU Xin-lang; YE Jun
School of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
关键词:
高光谱图像解混; 非负张量分解; 全变差; Cauchy 损失; 交替方向乘子法
Keywords:
hyperspectral images unmixing; nonnegative tensor decomposition; total variation; Cauchy loss; alternative direction methodof multipliers
分类号:
TP751
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 12. 004
摘要:
高光谱图像解混是高光谱图像应用的一项重要任务,包括了对端元的提取和丰度的估计。 基于非负张量分解的光谱解混方法能很好地保留高光谱图像的空间结构信息,但是却没有很好地利用丰度张量的分段光滑性约束,同时噪声的存在也严重影响了高光谱图像的解混性能。 针对上述问题, 全变差 Cauchy 非负张量分解 ( Total Variation CauchyNonnegative Tensor Factorization,TV-CNTF) 方法被提出用于高光谱图像解混。 该方法利用 Cauchy 损失来代替传统的最小二乘损失,通过减小噪声点在解混模型中的权重,来降低噪声对解混结果的影响,同时在模型中加入了全变差算子,保证了丰度张量的分段平滑结构。 此外,采用交替方向乘子法求解所提出的 TV-CNTF。 经过模拟和真实数据实验,同现有的其他方法相比,TV-CNTF 方法的解混效果和鲁棒性都更好。
Abstract:
Hyperspectral image unmixing is an important task in hyperspectral image application, which includes the extraction ofendmembers and the estimation of abundance. The spectral unmixing method based on nonnegative tensor decomposition can well retainthe spatial structure information of hyperspectral image,　　but does not make good use of the piecewise smoothness constraint of abundancetensor. At the same time,noise also seriously affects the performance of hyperspectral image unmixing. Therefore,a method named TotalVariation Cauchy Nonnegative Tensor Factorization ( TV-CNTF) is proposed for hyperspectral 　image unmixing. Cauchy loss was usedto replace the traditional least square loss,and the influence of noise on the results was reduced by reducing the weight 　　of noise points tothe model. At the same time,total variation operator was added into the model to ensure the piecewise smooth structure of abundancetensor. Besides,the alternate direction method of multipliers is adopted to solving the proposed TV-CNTF. Through simulation and realdata experiments and compared 　with other existing methods,the proposed TV-CNTF method has better unmixing effect and robustness.
更新日期/Last Update: 2022-12-10