为了提高复杂场景弱小目标高光谱融合图像的质量, 提出了基于相似性分类的主成分融合方法。光谱数据像素向量的相似性测度分类产生类矩阵, 通过由类矩阵主成分变换的降维投影矩阵来投影变换原有光谱数据, 获得降维数据矩阵。对比了传统PCA与基于欧式距离分类的PCA(ED_PCA)、基于光谱角分类的PCA(SA_PCA)、基于光谱信息散度分类的PCA(SID_PCA)和基于正交投影散度分类的PCA(OPD_PCA)四种改进方法的融合性能。实验结果表明: SA_PCA和SID_PCA方法兼具了ED_PCA和OPD_PCA的优点, 对比度提升较好, 阈值参数不敏感, 运行时间较短。

A principal component fusion method based on similarity classification is proposed, to enhance the fusion image quality of the complex scene hyperspectral images with weak and small targets. The class pixel matrix is produced by the spectral data’s pixel vectors classified in similarity measures, the dimension reduction projection matrix is calculated by principal component transforming the class pixel matrix, and dimension reduction data is obtained by the dimension reduction projection matrix projecting the original spectral data. The fusion performance is compared among the traditional PCA (Principal Component Analysis) and the four improving methods, ED_PCA (Euclidean Distance based PCA), SA_PCA (Spectral Angle based PCA), SID_PCA (Spectral Information Divergence based PCA) and OPD_PCA (Orthogonal Projection Divergence based PCA). Experiment results show that SA_PCA and SID_PCA have both advantages of ED_PCA and OPD_PCA, and have high target contrast ratio, robust threshold value and shorting runtime.