Sparse representation-based color visualization method for hyperspectral imaging

基于高光谱图像稀疏表示的彩色可视化模型(英文)

In this paper, we designed a color visualization model for sparse representation of the whole hyperspectral image, in which, not only the spectral information in the sparse representation but also the spatial information of the whole image is retained. After the sparse representation, the color labels of the effective elements of the sparse coding dictionary are selected according to the sparse coefficient and then the mixed images are displayed. The generated images maintain spectral distance preservation and have good separability. For local ground objects, the proposed single-pixel mixed array and improved oriented sliver textures methods are integrated to display the specific composition of each pixel. This avoids the confusion of the color presentation in the mixed-pixel color display and can also be used to reconstruct the original hyperspectral data. Finally, the model effectiveness was proved using real data. This method is promising and can find use in many fields, such as energy exploration, environmental monitoring, disaster warning, and so on.

提出一种对整幅高光谱图像的稀疏表示结果进行直接显示的方法,图中不仅包含了稀疏表示中保留的光谱信息,还可显示整幅图像的空间信息。稀疏表示后,将字典中的各有效原子根据光谱特性选择颜色标签,之后根据稀疏系数进行混合颜色显示,此时的图像能够同时满足可分性及距离保持特性。针对局部地物时,提出的单像素混合阵列表示法及改进的裂片纹理技术能够直观且完整的显示出每个像元的具体组成情况,还能够根据所生成图像中的信息对原始HSI进行重建,进而提高数据的利用率。该模型不仅能够良好地显示地物的空间特性,同时能够显示稀疏系数的组成,同时单像素混合阵列表示法及裂片纹理技术弥补了混合像素彩色显示中颜色表达混乱的弊端。对真实地物数据进行实验,结果证明该模型产生的彩色图像具有良好的视觉效果及可分性,满足距离保持特性。