最小体积约束的高光谱图像分辨率增强算法
|
摘要
针对现有的高光谱多光谱图像融合算法解空间较大、未考虑高光谱数据的物理意义以及存在局部最优的问题,提出了一种基于单形体最小体积约束的耦合非负矩阵分解的高光谱与多光谱图像融合算法(MVC-CNMF)。该算法在混合像元解混的过程中,考虑图像的物理意义,加入了端元单形体最小体积约束。由仿真结果可以看出,该算法能有效地克服现有融合算法中的缺陷,实现了高光谱与多光谱图像的端元与丰度的精确匹配,获得高空间分辨率的融合图像,尤其适用于端元数目较多的高光谱图像。
The current hyperspectral and multi-spectral image fusion algorithms have such defects as having large solution space not considering the physical meaning of hyperspectral data and being prone to local optimal solutions. To solve these problems a hyperspectral and multi-spectral image fusion algorithm is proposed based on Minimum Volume Constraint and Coupled Non-negative Matrix Factorization( MVCCNMF). In the process of separating the mixed pixels the algorithm takes the physical meaning of the image into consideration and adds the minimum volume constraint of the endmember single body. Simulation results show that the proposed algorithm can effectively overcome the defects in the existing fusion algorithmsaccurately match the endmember with the abundance of hyperspectral and multi-spectral images and obtain high-spatial-resolution fused images. This algorithm is especially suitable for the hyperspectral images with a large number of endmembers.
The current hyperspectral and multi-spectral image fusion algorithms have such defects as having large solution space not considering the physical meaning of hyperspectral data and being prone to local optimal solutions. To solve these problems a hyperspectral and multi-spectral image fusion algorithm is proposed based on Minimum Volume Constraint and Coupled Non-negative Matrix Factorization( MVCCNMF). In the process of separating the mixed pixels the algorithm takes the physical meaning of the image into consideration and adds the minimum volume constraint of the endmember single body. Simulation results show that the proposed algorithm can effectively overcome the defects in the existing fusion algorithmsaccurately match the endmember with the abundance of hyperspectral and multi-spectral images and obtain high-spatial-resolution fused images. This algorithm is especially suitable for the hyperspectral images with a large number of endmembers.
引文
[1]浦瑞良,宫鹏.高光谱遥感及其应用[M].北京:高等教育出版社,2000.
[2]王楠,张良培,杜博.最小光谱相关约束NMF的高光谱遥感图像混合像元分解[J].武汉大学学报:信息科学版,2014,39(1):22-25.
[3] YOKOYA NYAIRI TIWASAKI A. Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion[J]. IEEE Transactions on Geoscience and Remote Sensing201250(2):528-537.
[4] ZHANGY F. Spatial resolution enhancement of hyperspectral image based on the combination of spectral mixing model and observation model[C]//Proceedings of SPIEthe International Society for Optical Engineering2014:201-204.
[5] LIU DBOUFOUNOS P T. Dictionary learning based pansharpening[C]//IEEE International Conference on AcousticsSpeech and Signal Processing(ICASSP)2012:2397-2400.
[6] LEE D DSEUNG H S. Learning the parts of objects by non-negative matrix factorization[J]. Nature1999401(6755):788-791.
[7]魏一苇,黄世奇,王艺婷,等.基于体积和稀疏约束的高光谱混合像元分解算法[J].红外与激光工程,2014,43(4):1247-1254.
[8] HOYER P O. Non-negative matrix factorization with sparseness constraints[J]. Journal of Machine Learning Research20045(9):1457-1469.
[9]郭连坤.基于多核Boosting多特征组合高光谱分类技术研究[D].西安:西安科技大学,2015.
[10] WEI QBIOUCAS-DIA J MDOBIGEON Net al. Fusion of multispectral and hyperspectral images based on sparse representation[C]//The 22nd European Signal Processing Conference(EUSIPCO)2014:1-5.
[2]王楠,张良培,杜博.最小光谱相关约束NMF的高光谱遥感图像混合像元分解[J].武汉大学学报:信息科学版,2014,39(1):22-25.
[3] YOKOYA NYAIRI TIWASAKI A. Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion[J]. IEEE Transactions on Geoscience and Remote Sensing201250(2):528-537.
[4] ZHANGY F. Spatial resolution enhancement of hyperspectral image based on the combination of spectral mixing model and observation model[C]//Proceedings of SPIEthe International Society for Optical Engineering2014:201-204.
[5] LIU DBOUFOUNOS P T. Dictionary learning based pansharpening[C]//IEEE International Conference on AcousticsSpeech and Signal Processing(ICASSP)2012:2397-2400.
[6] LEE D DSEUNG H S. Learning the parts of objects by non-negative matrix factorization[J]. Nature1999401(6755):788-791.
[7]魏一苇,黄世奇,王艺婷,等.基于体积和稀疏约束的高光谱混合像元分解算法[J].红外与激光工程,2014,43(4):1247-1254.
[8] HOYER P O. Non-negative matrix factorization with sparseness constraints[J]. Journal of Machine Learning Research20045(9):1457-1469.
[9]郭连坤.基于多核Boosting多特征组合高光谱分类技术研究[D].西安:西安科技大学,2015.
[10] WEI QBIOUCAS-DIA J MDOBIGEON Net al. Fusion of multispectral and hyperspectral images based on sparse representation[C]//The 22nd European Signal Processing Conference(EUSIPCO)2014:1-5.