融合光谱子空间和模型导向的高光谱图像超分辨率研究  

Research on Hyperspectral Image Super-resolution Reconstruction Integrating Spectral Subspace and Model Guided

作　　者：刘丛[1] 梅海闽 LIU Cong;MEI Haimin(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《小型微型计算机系统》2025年第2期373-380,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61703278)资助。

摘　　要：针对现有的基于深度学习的高光谱图像超分辨率重建方法无法通用于不同波段的高光谱图像以及缺乏可解释性等问题.提出一种融合光谱子空间映射和模型引导的高光谱图像超分辨率算法.首先,使用光谱子空间分解将原始图像映射到低维空间中,既可以增加光谱间的相关性又可以去除不同波段高光谱图像对网络的限制.其次,使用小波变换将稀疏矩阵分解为高频特征和低频特征,挖掘图像中的纹理和结构等高频信息.再者,以超分辨率重建模型为指导,将ADMM分解后的子模型优化展开为深度网络的形式,增加了深度网络设计的可解释性.最终,使用逆小波变换后将重建的系数矩阵映射到原始的全谱空间中.实验表明,提出的方法在定量指标和主观视觉方面均表现优异.Existing deep learning based hyperspectral image super-resolution reconstruction methods usually cannot common be applied in various hyperspectral images with different spectral bands and also lack of interpretability.This paper introduces a spectral subspace mapping and model-guided super-resolution approach for hyperspectral images.Initially,the original hyperspectral image is mapped into a low-dimensional space by using the spectral subspace decomposition,which can not only increase the spectral correlation but also remove the limitation of different spectral bands in different hyperspectral images.Subsequently,the sparse matrix is decomposed into the low-frequency and high-frequency by using the wavelet transform to capture the high-frequency information such as the texture and structure information in the image.Furthermore,we apply the super-resolution reconstruction model as the guidance and unfold the sub-model optimization processes obtained by ADMM to deep networks,which can enhance the interpretability of the designed network.Ultimately,the reconstructed coefficient matrix are mapped back into the original full spectrum space by using the inverse wavelet transform.Experimental results demonstrate the effectiveness of the proposed method in both quantitative metrics and subjective assessment of visual quality.

关 键 词：高光谱图像 超分辨率 模型引导 光谱子空间 小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]