Scholarlink

<p>Due to the limitations of satellite optical sensors, remote sensing images exhibit significant imbalances in spatial and spectral resolution. The hyperspectral image (HSI) has high spectral resolution but low spatial resolutions, while the multispectral image (MSI) does the opposite. Naturally, a popular way to get both high spectral and high spatial resolution image, known as super-resolution image (SRI), is to fuse the HSI and MSI. In order to improve the image quality of SRI, it is necessary to explore its potential low rank, sparsity, and piecewise smoothness characteristics. Therefore, in this paper, we propose an unidirectional total variation (TV) regularized tensor CANDECOMP/PARAFAC (CP) decomposition model to solve the hyperspectral super-resolution problem. In our method, we make full use of the simplicity and unique decomposition advantages of tensor CP decomposition, and the regularization term effectively characterizes the sparsity and piecewise smoothness features of SRI using the <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\ell _{1}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>ℓ</mi> <mn>1</mn> </msub> </math></EquationSource> </InlineEquation> norm in TV. An efficient proximal alternating optimization algorithm is employed to solve our proposed non-convex optimization model. The global convergence of the iteration sequence to a critical point of the model is proved. Experiments on five semi-real datasets reflect the effectiveness and efficiency of our proposed method.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hyperspectral Super-Resolution Via Tensor CP Decomposition with Total Variation Regularization

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Jingya Chang,
  • Xiaofei Cui,
  • Deren Han,
  • Zhenyou Wang

摘要

Due to the limitations of satellite optical sensors, remote sensing images exhibit significant imbalances in spatial and spectral resolution. The hyperspectral image (HSI) has high spectral resolution but low spatial resolutions, while the multispectral image (MSI) does the opposite. Naturally, a popular way to get both high spectral and high spatial resolution image, known as super-resolution image (SRI), is to fuse the HSI and MSI. In order to improve the image quality of SRI, it is necessary to explore its potential low rank, sparsity, and piecewise smoothness characteristics. Therefore, in this paper, we propose an unidirectional total variation (TV) regularized tensor CANDECOMP/PARAFAC (CP) decomposition model to solve the hyperspectral super-resolution problem. In our method, we make full use of the simplicity and unique decomposition advantages of tensor CP decomposition, and the regularization term effectively characterizes the sparsity and piecewise smoothness features of SRI using the \(\ell _{1}\) ℓ 1 norm in TV. An efficient proximal alternating optimization algorithm is employed to solve our proposed non-convex optimization model. The global convergence of the iteration sequence to a critical point of the model is proved. Experiments on five semi-real datasets reflect the effectiveness and efficiency of our proposed method.