红外技术2026，Vol.48Issue(4)：468-475,8.
基于张量核范数框架表示和总变分的高光谱图像去噪
Hyperspectral Image Denoising Based on Tensor Nuclear Norm Framelet Representation and Total Variation
摘要
Abstract
During hyperspectral data acquisition,noise contamination inevitably degrades image quality and affects the accuracy of subsequent applications.To address this issue,this study proposes a hyperspectral image denoising model based on a tensor kernel norm framework combined with total variation regularization.First,the proposed model employs a tensor kernel norm framework tailored for highly correlated third-order tensors.In this framework,each tensor tube exhibits sparsity,and the sum of matrix ranks corresponding to the frontal slices of the transformed tensor is minimized,thereby fully capturing the low-rank characteristics of hyperspectral images.Second,a weighted spatial-spectral total variation term,expressed using the l2,1 norm,is incorporated to enhance sparsity while preserving local smoothness in the spatial-spectral domain.Finally,these two components are effectively integrated to jointly exploit the low-rank properties of hyperspectral images and the sparse smoothness of the spatial-spectral domain,thereby achieving removal of high-intensity Gaussian noise and strip noise.Both simulation and real-data experiments demonstrate that,compared with five classical denoising algorithms,the proposed model achieves superior denoising performance.The restored images exhibit improved clarity,better detail preservation,and well-maintained structural contours without excessive smoothing.关键词
高光谱图像/张量核范数框架表示/总变分/交替方向乘子法/图像去噪Key words
hyperspectral image/tensor nuclear norm framelet representation/total variation/alternating direction multiplier method/image denoising分类
信息技术与安全科学引用本文复制引用
徐光宪,王泽民,马飞,陶志勇..基于张量核范数框架表示和总变分的高光谱图像去噪[J].红外技术,2026,48(4):468-475,8.基金项目
辽宁工程技术大学鄂尔多斯研究院校地科技合作培育项目(YJY-XD-2024-B-010),辽宁省自然科学基金计划项目(2023-MS-314),辽宁省教育厅高校基本科研创新发展项目(LJ242410147006). （YJY-XD-2024-B-010）
