Nonlocal Tensor Sparse Representation and Low-Rank Regularization for Hyperspectral Image Compressive Sensing Reconstruction

Xue, Jize; Zhao, Yongqiang; Liao, Wenzhi; Chan, Jonathan Cheung-Wai

doi:10.3390/rs11020193

Cited by 60 publications

(43 citation statements)

References 60 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our research process has four research directions: (1) the use of other bases, such as DWT and PCA, to improve the spectral or spatial representation capability; (2) optimization of the compression algorithm regarding the tradeoff between the compression performance and complexity; (3) constructing a more complex learning network consisting of the dictionary learning and tensor decomposing to implement an end-to-end compression scheme; (4) combining the distributed source coding (DSC) scheme to construct a DSC-CNN; (5) integrating compressive sensing [52] into the proposed scheme to construct a high performance compressor; and (6) optimization of our method for hardware design. These aforementioned issues will be investigated in future research.…”

Section: Resultsmentioning

confidence: 99%

“…(3) constructing a more complex learning network consisting of the dictionary learning and tensor decomposing to implement an end-to-end compression scheme; (4) combining the distributed source coding (DSC) scheme to construct a DSC-CNN; (5) integrating compressive sensing [52] into the proposed scheme to construct a high performance compressor; and (6) optimization of our method for hardware design. These aforementioned issues will be investigated in future research.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Multispectral Transforms Using Convolution Neural Networks for Remote Sensing Multispectral Image Compression

Liu

2019

Remote Sensing

View full text Add to dashboard Cite

A multispectral image is a three-order tensor since it is a three-dimensional matrix, i.e.one spectral dimension and two spatial position dimensions. Multispectral image compression canbe achieved by means of the advantages of tensor decomposition (TD), such as NonnegativeTucker Decomposition (NTD). Unfortunately, the TD suffers from high calculation complexity andcannot be used in the on-board low-complexity case (e.g., multispectral cameras) that the hardwareresources and power are limited. Here, we propose a low-complexity compression approach formultispectral images based on convolution neural networks (CNNs) with NTD. We construct anew spectral transform using CNNs, where the CNNs are able to transform the three-dimensionspectral tensor from large-scale to a small-scale version. The NTD resources only allocate thesmall-scale three-dimension tensor to improve calculation efficiency. We obtain the optimizedsmall-scale spectral tensor by the minimization of original and reconstructed three-dimensionspectral tensor in self-learning CNNs. Then, the NTD is applied to the optimized three-dimensionspectral tensor in the DCT domain to obtain the high compression performance. We experimentallyconfirmed the proposed method on multispectral images. Compared to the case that the newspectral tensor transform with CNNs is not applied to the original three-dimension spectral tensorat the same compression bit-rates, the reconstructed image quality could be improved. Comparedwith the full NTD-based method, the computation efficiency was obviously improved with only asmall sacrifices of PSNR without affecting the quality of images.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Multispectral Transforms Using Convolution Neural Networks for Remote Sensing Multispectral Image Compression

Liu

2019

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…HSpaLSpe is just an intermediate variable and is not necessary to be solved. PAN image contains abundant high-resolution spatial details and structures [30], used to train an over-completed spatial dictionary via sparse representation framework [31]. In Figure 2, virtual variable HSpaLSpe has the same spatial resolution as the desired HSI.…”

Section: Spatial Observation Modelmentioning

confidence: 99%

Joint Spatial-spectral Resolution Enhancement of Multispectral Images with Spectral Matrix Factorization and Spatial Sparsity Constraints

et al. 2020

View full text Add to dashboard Cite

This paper presents a joint spatial-spectral resolution enhancement technique to improve the resolution of multispectral images in the spatial and spectral domain simultaneously. Reconstructed hyperspectral images (HSIs) from an input multispectral image represent the same scene in higher spatial resolution, with more spectral bands of narrower wavelength width than the input multispectral image. Many existing improvement techniques focus on spatial-or spectral-resolution enhancement, which may cause spectral distortions and spatial inconsistency. The proposed scheme introduces virtual intermediate variables to formulate a spectral observation model and a spatial observation model. The models alternately solve spectral dictionary and abundances to reconstruct desired high-resolution HSIs. An initial spectral dictionary is trained from prior HSIs captured in different landscapes. A spatial dictionary trained from a panchromatic image and its sparse coefficients provide high spatial-resolution information. The sparse coefficients are used as constraints to obtain high spatial-resolution abundances. Experiments performed on simulated datasets from AVIRIS/Landsat 7 and a real Hyperion/ALI dataset demonstrate that the proposed method outperforms the state-of-the-art spatial-and spectral-resolution enhancement methods. The proposed method also worked well for combination of exiting spatial-and spectral-resolution enhancement methods.Remote Sens. 2020, 12, 993 2 of 24 remote sensing data have become increasingly available and the corresponding applications have attracted wide interests, there are existing challenges in acquiring images with simultaneously high spatial resolution and high spectral resolution [6]. Therefore, many research focuses on recovering high quality synthetic image from low resolution (LR) inputs, including spatial resolution improvement approaches [7][8][9][10][11][12][13][14][15][16][17][18][19] and spectral resolution enhancement techniques [20-27].1.1. Spatial Resolution Improvement of HSI Over past decades, several spatial improvement methods have been proposed based on the fusion of MSI and PAN images. Hyperspectral pan-sharpening improves spatial resolution of HSI by fusing LR HSI with a high resolution (HR) PAN image covering a spectral range from the visible to the near infrared ranges [7]. Except for the classical hyperspectral pan-sharpening approaches such as component substitution (CS) methods [8], multi-resolution analysis (MRA) methods [9], and model based optimization methods [10,11], deep learning, particularly convolutional neural network (CNN), has been widely exploited in pan-sharpening tasks. In [12], a residual CNN model is designed to describe the mapping between LR/HR MSI pairs and PAN image. Yuan et al. [13] proposed a multi-scale and multi-depth CNN (MSDCNN) for pan-sharpening, in which multi-scale feature extraction is used to reconstruct the HR MSI.HSI fusion is another popular approach to spatial resolution improvement. A HR HSI is recovered by fusing a LR HSI with a H...

show abstract

“…Hyperspectral images collected by using whisk-broom sensors and push-broom scanners can be degraded by stripes, often caused by calibration error or inconsistent responses between detectors [3]. Such stripes degrade visual quality and pose negative influence on subsequent processing, such as unmixing [4,5], super-resolution [6], classification [7,8], compressive sensing reconstruction [9,10], and recovery [11,12]. In general, three types of stripes exist: horizontal (row-by-row) [13,14], vertical (column-by-column) [15,16], and oblique stripes [17,18].…”

Section: Introductionmentioning

confidence: 99%

Global and Local Tensor Sparse Approximation Models for Hyperspectral Image Destriping

et al. 2020

Self Cite

View full text Add to dashboard Cite

This paper presents a global and local tensor sparse approximation (GLTSA) model for removing the stripes in hyperspectral images (HSIs). HSIs can easily be degraded by unwanted stripes. Two intrinsic characteristics of the stripes are (1) global sparse distribution and (2) local smoothness along the stripe direction. Stripe-free hyperspectral images are smooth in spatial domain, with strong spectral correlation. Existing destriping approaches often do not fully investigate such intrinsic characteristics of the stripes in spatial and spectral domains simultaneously. Those methods may generate new artifacts in extreme areas, causing spectral distortion. The proposed GLTSA model applies two ℓ 0 -norm regularizers to the stripe components and along-stripe gradient to improve the destriping performance. Two ℓ 1 -norm regularizers are applied to the gradients of clean image in spatial and spectral domains. The double non-convex functions in GLTSA are converted to single non-convex function by mathematical program with equilibrium constraints (MPEC). Experiment results demonstrate that GLTSA is effective and outperforms existing competitive matrix-based and tensor-based destriping methods in visual, as well as quantitative, evaluation measures.

show abstract

Nonlocal Tensor Sparse Representation and Low-Rank Regularization for Hyperspectral Image Compressive Sensing Reconstruction

Cited by 60 publications

References 60 publications

Multispectral Transforms Using Convolution Neural Networks for Remote Sensing Multispectral Image Compression

Multispectral Transforms Using Convolution Neural Networks for Remote Sensing Multispectral Image Compression

Joint Spatial-spectral Resolution Enhancement of Multispectral Images with Spectral Matrix Factorization and Spatial Sparsity Constraints

Global and Local Tensor Sparse Approximation Models for Hyperspectral Image Destriping

Contact Info

Product

Resources

About