Compressed hyperspectral imaging is a powerful technique for satellite-borne and airborne sensors that can effectively shift the complex computational burden from the resource-constrained encoding side to a presumably more capable base-station decoder. Reconstruction algorithms play a pivotal role in compressive imaging systems. Traditional model-based reconstruction approaches are computationally burdensome and achieve limited success. Deep learning-based approaches, while improving in reconstruction accuracy and speed, depend heavily on data, which is a major challenge for satellite-borne hyperspectral compressed imaging. In this article, we combine the respective advantages of model-based and learning-based approaches in a distributed compressed hyperspectral sensing framework, employing linear mixed model assumptions and spectral library learning to simultaneously improve the reconstruction speed and accuracy without using a large amount of additional hyperspectral data. First, the relationship between the CS band and the key band is learned from the spectral library to ensure that the key band endmembers can be accurately predicted. Then, the joint horizontal and vertical difference operators are proposed to enhance the estimation of the initial values of abundance. Finally, the CS band endmembers and residuals are updated in the reconstruction module to deal with the endmember and abundance mismatch. Extensive experimental results on five real hyperspectral datasets demonstrate that the proposed spectral library learning, abundance initialization, and reconstruction strategy can effectively improve the compressed sensing reconstruction accuracy, outperforming the existing state-of-the-art methods.

中文翻译：

---

分布式压缩高光谱传感成像结合光谱分解和学习

压缩高光谱成像是一种强大的星载和机载传感器技术，可以有效地将复杂的计算负担从资源受限的编码端转移到可能更强大的基站解码器上。重建算法在压缩成像系统中起着举足轻重的作用。传统的基于模型的重建方法计算量大，成功率有限。基于深度学习的方法在提高重建精度和速度的同时，严重依赖数据，这是星载高光谱压缩成像的主要挑战。在本文中，我们在分布式压缩高光谱传感框架中结合了基于模型和基于学习的方法的各自优势，采用线性混合模型假设和光谱库学习来同时提高重建速度和准确性，而无需使用大量额外的高光谱数据。首先，从谱库中学习CS波段和关键波段之间的关系，以确保关键波段端元可以被准确预测。然后，提出了联合水平和垂直差分算子来增强丰度初始值的估计。最后，在重建模块中更新 CS 带端元和残差以处理端元和丰度不匹配。五个真实高光谱数据集的广泛实验结果表明，所提出的光谱库学习、丰度初始化、