Abstract
Hyperspectral images with high spatial resolution play an important role in material classification, change detection, and others. However, owing to the limitation of imaging sensors, it is difficult to directly acquire images with both high spatial resolution and high spectral resolution. Therefore, the fusion of remotely sensed images is an effective way to obtain high-resolution desired data, which is usually an ill-posed inverse problem and susceptible to noise corruption. To address these issues, a low-rank model based on tensor decomposition is proposed to fuse hyperspectral and multispectral images by incorporating graph regularization, in which the logarithmic low-rank function is utilized to suppress the small components for denoising. Furthermore, this article takes advantage of the local spatial similarity of remotely sensed images to enhance the reconstruction performance by constructing spatial graphs, and also promotes signature smoothing between adjacent endmember spectra using the neighborhood-based spectral graph regularization. Finally, a set of efficient solvers is carefully designed via alternating optimization for closed-from solutions and computational reduction, in which vector-matrix operators are adapted to solve the 3-D core tensor. Experimental tests on several real datasets illustrate that the proposed fusion method yields better reconstruction performance than the current state-of-the-art methods, and can significantly suppress noise at the same time.
Highlights
W ITH the rapid development of imaging sensors in remote sensing, hyperspectral images (HSIs) are widely used in many fields, such as objective classification and target detection [1] [2]
This paper proposes a low-rank tensor decomposition approach, termed as graph-regularized low-rank tensor fusion algorithm (G-LRTF), by incorporating spatial-spectral graph regularization for hyperspectral image super-resolution
To capture the correlation of different dimensions in hyperspectral data, we focus on the local manifold information in hyperspectral and multispectral data to build the spatialspectral graph Laplacian matrices, and reconstruct the images based on coupled Tucker decomposition
Summary
W ITH the rapid development of imaging sensors in remote sensing, hyperspectral images (HSIs) are widely used in many fields, such as objective classification and target detection [1] [2]. Zare et al [13] brought forward a fusion model based on smoothed graph signals, and took the clustering method to redefine the spatial similarity of HSIs. In general, matrix factorization-based methods inevitably have to unfold the original three-dimensional data into matrices, so that the structure information of the observed HSI is lost, resulting in a certain degree of spatial and/or spectral distortion of the reconstructed image. Deep learning-based approaches can address non-convex models in remote sensing [19] These methods generally require the establishment of training samples (i.e., image elements or features), and involve the adjustment of a large number of hyperparameters. To utilize the advantages of tensor models, this paper incorporates low-rank and graph regularizers into Tucker decomposition to redefine the fusion of HSIs and MSIs as an optimization problem. The matrices A(n) ∈ RIn ×I1 I2,...,In−1 In+1,...,IN and A ∈ RI1 I2,...,In ×In+1,...,IN denote the mode-n unfolding and the mode-n canonical unfolding of the tensor A, respectively
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
