高光谱图像的高维多目标压缩感知技术研究

Research on Compressed Sensing of Hyperspectral Image with Many-objective Optimization

高光谱图像(Hyperspectral image, HSI)相对于普通图像而言,不仅具有空间信息,还有光谱维信息.由于HSI占用的存储空间较大,因此在数据传输中对传输带宽有很高的要求.采用压缩感知理论可以在很大程度上缓解高光谱图像的传输和存储压力.本文提出了一种基于NSGA-Ⅲ优化的OMP算法(MO-OMP)对高光谱图像进行重构.通过考虑高光谱图像的空间相似性和光谱间相似性,对高光谱图像压缩感知的重构过程进行高维多目标建模,在求解模型方面,将OMP算法中的非零元素原子指标集作为种群,并采用NSGA-Ⅲ算法对模型进行求解,提高了模型的求解精度.本文在三组公共数据集上对模型进行测试,实验结果表明MO-OMP算法在高光谱图像压缩感知的问题下有良好的效果,相比于传统的高光谱图像压缩感知模型,高维多目标压缩感知模型在高光谱图像的重建问题上更具鲁棒性.

Hyperspectral image(Hyperspectral image, HSI)has not only spatial information but also spectral dimension information compared to ordinary images.It has high requirements for transmission bandwidth in data transmission because of storage.The use of compressed sensing theory can ease the transmission and storage pressure of HSI to a large extent.This paper proposes an OMP algorithm(MO-OMP)based on fast non-dominated sorting Ⅲ optimization to reconstruct HSI.By considering the spatial and the similarity of hyperspectral images, the reconstruction process of hyperspectral image compressed sensing is modeled with Many-objectives and the NSGA-Ⅲ algorithm is used to solve the model, which improves the accuracy of the model.In this paper, the algorithm is tested on three sets of public data sets.The experimental results show that the MO-OMP algorithm has a good effect on the problem of hyperspectral image compressed sensing.The Many-objective compressed sensing model, compared with the traditional hyperspectral image compressed sensing model, has more robust in the reconstruction of hyperspectral images.