基于超图的多波段联合稀疏表示遥感图像时空融合

对于给定的卫星传感器,其在空间分辨率与时间分辨率之间总是存在权衡,现有的基于稀疏表示的融合算法大多在单波段上进行,忽略了多波段间的相关性,并且稀疏表示算法应用于分辨率差距过大的场景下,融合效果较差。针对以上问题,文章提出一种基于超图的多波段联合稀疏表示算法。该算法将超图约束引入稀疏表示模型,保持多波段之间的关联性,并通过细节增强模型对初步预测结果进行细节增强,从而提高融合精度。实验结果表明:在 BOREAS 数据集上,相比于次优算法,文章所提算法的结构相似度(structural similarity, SSIM)提高了 3.0%,光谱角匹配(spectral angle match, SAM)提高了 5.4%;在 CIA 数据集上,相比于次优算法,文章所提算法的 SSIM 提高了 2.0%,SAM 提高了 4.0%;在 LGC 数据集上,相比于次优算法,文章所提算法的 SSIM 提高了 2.4%,SAM 提高了 4.5%。表明文中所提出的算法在光谱的保持和空间信息的预测上都表现出一定的优势。

There is always a trade-off between spatial resolution and temporal resolution for a given satellite sensor. Most of the existing sparse representation based fusion algorithms are carried out on a single band, ignoring the correlation between multiple bands, and the sparse representation algorithm has poor fusion effect when applied to the scene with a large resolution gap. To solve the above problems, this paper proposes a multi-band joint sparse representation algorithm based on hypergraph. The hypergraph constraint is introduced into the sparse representation model to maintain the correlation between multi-bands, and the detail enhancement model is used to enhance the details of the preliminary prediction results, so as to improve the fusion accuracy. The experimental results show that the structural similarity (SSIM) of the proposed algorithm is improved by 3.0% and the spectral angle match (SAM) is improved by 5.4% on BOREAS dataset compared with the suboptimal algorithm. On the CIA dataset, compared with the suboptimal algorithm, the SSIM of the proposed algorithm is improved by 2.0% and the SAM is improved by 4.0%. On the LGC dataset, compared with the suboptimal algorithm, the SSIM of the proposed algorithm is increased by 2.4% and the SAM is increased by 4.5%. The algorithm in this paper shows advantages in spectrum preservation and spatial information prediction.