http://vigir.missouri.edu/~gdesouza/Research/Conference_CDs/IGARSS_2010/pdfs/3544.pdf | No | No | No | No | - spectral unmixing (Vertex Component Analysis), Simulated Annealing

• not a lot of info in general | Jude | | | | | | | Joint Spectral-Spatial Sub-Pixel Mapping Model | Hyperspectral image super resolution reconstruction with a joint spectral-spatial sub-pixel mapping model, IGARSS2016, X. Xu et al.

https://ieeexplore.ieee.org/abstract/document/7730601 | No | No | No | No | - spectral unmixing, sub-pixel mapping, sub-pixel probability map

• similar accuracy to SRSR and SRCNN, but faster compute
• lots of math | Jude | | | | | | | Sparse Representation | Hyperspectral image super-resolution by spectral mixture analysis and spatial–spectral group sparsity, GRSL2016, J. Li et al.

https://ieeexplore.ieee.org/abstract/document/7517322 | No | No | No | No | - design novel SR model for hyperspectral images. usually SR is used for single band

• trained on only images, no external data needed
• lots of math
• Spectral-Mixture-Analysis-Based HSI SR Framework, Mixture-Analysis-Based Spectral Sparsity, Spatial–Spectral Group Sparsity | Jude | | | | | | | Low Rank Tensor, LLA, ADMM | Super-resolution reconstruction of hyperspectral images via low rank tensor modeling and total variation regularization, IGARSS2016, S. He et al. [PDF]

https://arxiv.org/pdf/1601.06243 | No | No | No | No | - represent the image as a tensor, treat is as an optimization problem, use LLA and ADMM to optimize | Jude | | | | | | | combination deep Spectral Difference Convolutional Neural Network and Spatial-Error-Correction model | Hyperspectral image super-resolution by spectral difference learning and spatial error correction, GRSL2017, J. Hu et al.

https://ieeexplore.ieee.org/abstract/document/8019794 | No | No | No | No | - SDCNN used to learn mapping between HR and LR, SEC used to adjust error

• apply very deep convolutional network super-resolution by Kim et al. on key band, SDCNN/SEC on non-key bands -pseudocode for training step
• only requires image
• fast compute time | Jude Matthew | | | | | | | 3D-RUNet (Deep Convolutional Neural Network) | SISR of Hyperspectral Remote Sensing Imagery Using 3D Encoder-Decoder RUNet Architecture

https://ieeexplore.ieee.org/document/9883578 | https://github.com/NourO93/SISR_Library/blob/main/comp_vis/srcnn.py | Tensorflow | N/A | Yes (trained over two datasets, no coefficients stated) | - SISR Approach - using DCNNs

• Best PSNR, SSIM, SAM compared to 2D-RUNet, Bicubic, 3D-FCNN models
• Tested using scale factor of 2x, 4x | Josh | Salinas: 2x: 39.15 4x: 34.65

Botswana: 2x: 36.32 4x: 33.07 | Salinas: 2x: 0.97 4x: 0.92

Botswana: 2x: 0.93 4x: 0.80 | N/A | Josh | Yes | | SSPSR (spatial-spectral prior network) | Learning Spatial-Spectral Prior for Super-Resolution of Hyperspectral Imagery

https://arxiv.org/pdf/2005.08752 | https://github.com/junjun-jiang/SSPSR | PyTorch | N/A | Yes, but no coefficients | • Hyperspectral SISR paper for RBG/single gray using DCNN (model used is called SSPSR) • Multiple public datasets tested in 4x and 8x factor, SSPSR is compared with other models and always has highest PSNR and SSIM values (30+ usually, depends) | Matthew | Cave: 4x: 39.0892 8x: 33.4340

Pavia: 4x: 29.1581 8x: 25.1985

Chikusei: 4x: 40.3612 8x: 35.8368 | Cave: 4x: 0.9553 8x: 0.9010

Pavia: 4x: 0.7903 8x: 0.5365

Chikusei: 4x: 0.9413 8x: 0.8624 | N/A | | | | 3D-FRCNN (3D Full Convolutional Neural Network) | Hyperspectral Image Spatial Super-Resolution via 3D Full Convolutional Neural Network, Remote Sensing, 2017, Saohui Mei et al.

https://www.mdpi.com/2072-4292/9/11/1139 | https://github.com/MeiShaohui/Hyperspectral-Image-Spatial-Super-Resolution-via-3D-Full-Convolutional-Neural-Network | Ternsorflow/Keras | Python 2.7 | Yes, only on one dataset | - Tested on 30, 60 dB/SNR

• Higher than 3B-SRCNN, Bicubic, but less than both models above | Josh | Pavia: 2x: 33.916 3x: 29.853 4: 27.59 | Pavia: 2x: 0.969 3x: 0.921 4: 0.862 | N/A | | | | NLRTATV | Hyperspectral image super-resolution via nonlocal low-rank tensor approximation and total variation regularization, Remote Sensing, 2017, Yao Wang et al.

https://www.mdpi.com/2072-4292/9/12/1286 | N/A | N/A | N/A | N/A | - Tensor Approximation Model for Hyperpsectral

• No code, or information on training, tested on different datasets than other models, not too good comparison | Josh | Urban: 2x: 35.2754

Moffett Field: 2x: 35.9816 3x: 33.1308

DC Mall: 2x: 34.3935 | Urban: 2x: 0.9663

Moffett Field: 2x: 0.9545 3x: 0.9301

DC Mall: 2x: 0.9665 | N/A | | | | SCT_SDCNN Model | Hyperspectral image super-resolution using deep convolutional neural network, Neurocomputing, 2017, Sen Lei et al.

https://www.researchgate.net/publication/317024713_Hyperspectral_image_super-resolution_using_deep_convolutional_neural_network | N/A | N/A | N/A | Yes | - Spatial constraint strategy (SCT) combined with DCNN (SDCNN)

• Worse PSNR, SSIM, and SAM than SSPSR model above on Cave dataset, not too good to use + no code | Josh | Cave: 2x: 39.95 4x: 35.83 8x: 31.60

Harvard: 2x: 40.65 4x: 37.82 8x: 34.82

Foster: 2x: 46.31 4x: 40.77 8x: 34.54 | Cave: 2x: 0.9804 4x: 0.9546 8x: 0.9074

Harvard: 2x: 0.9620 4x: 0.9406 8x: 0.9119

Foster: 2x: 0.9905 4x: 0.9746 8x: 0.9329 | N/A | | | | Collaborative nonnegative matrix factorization (CNMF) Model | Hyperspectral image superresolution by transfer learning, Jstars2017, Y. Yuan et al.

http://ieeexplore.ieee.org/iel7/4609443/4609444/07855724.pdf | N/A | N/A | N/A | Yes | - Transfer learning approach using CNN with collaborative nonnegative matrix factorization (CNMF)

• Worse PSNIR, SSIM, and SAM than previous models, not too good | Josh | Cave: 37.17

Nature: 49.01

Pavia: 26.57 | Cave: 0.951

Nature: 0.984

Pavia: 0.753 | N/A | | | | Hyperspectral Image Acquisition Model** | Super-resolution reconstruction of hyperspectral images, TIP2005, T. Akgun et al.

https://ieeexplore.ieee.org/document/1518950 | No | No | No | No | - A lot of math: also includes results but no code for the model used.

• represent hyperspectral observations from different wavelengths as weighted linear combinations of a small number of aliased and blurred basis image planes.
• reconstruction process as the inverse problem of finding a hi-res target hyperspectral image that agrees best with the observations under the proposed model. | Hari | | | | | | | Enhanced Self-Training Superresolution Mapping | Enhanced self-training superresolution mapping technique for hyperspectral imagery, GRSL2011, F. A. Mianji et al.

https://ieeexplore.ieee.org/document/5706437 | No | No | No | No | - provides the layout of the model with some specifics. Uses many different techniques.

• Uses SRM to extrapolate the simulated subsampled LR.
• Uses the input data as the source of training
• Endmember exraction using MNF and PPI | Hari | | | | | | | SRR method | A super-resolution reconstruction algorithm for hyperspectral images. Signal Process. 2012, H. Zhang et al.

https://www-sciencedirect-com.myaccess.library.utoronto.ca/science/article/pii/S0165168412000345?via%3Dihub | No | No | No | No | - proposes a maximum a posteriori (MAP) based multi-frame super-resolution algorithm for hyperspectral images. Principal component analysis (PCA) is utilized in both parts of the proposed algorithm: motion estimation and image reconstruction.

• PCA Based resolution enhancement | Hari | | | | | | | BSR_LR_GS (Blind Super-Resolution with Low-Rank and Group-Sparse models) | Super-resolution hyperspectral imaging with unknown blurring by low-rank and group-sparse modeling, ICIP2014, H. Huang et al.

https://ieeexplore.ieee.org/document/7025432 | No | No | No | No | In the proposed blind super-resolution method, the intrinsic low-rank structure is imposed with a predefined spectral basis by SVD. Group-sparse model is developed on different types of high frequency components, containing finite difference, wavelet coefficients, and contourlet coefficients. The desired high spatial resolution HSI and blurring kernel is alternatively optimized according to the proposed cost function. | Hari | | | | | | | SRM method via multi-dictionary based sparse representation (MSRSM) | Super-resolution mapping via multi-dictionary based sparse representation, ICASSP2016, H. Huang et al.

https://ieeexplore.ieee.org/document/6854256 | No | No | No | No | the proposed feature vector can capture the significant information about spatial dependence, and multiple distribution dictionaries are learned via sparse representation. In the class allocation process, the feature vector of the underlying subpixel is reconstructed by every dictionary, and is assigned to a class according to the reconstruction errors and the introduced spectrum distortions. | Hari | | | | | | | Local–Global Combined Network (LGCNet) | Super-Resolution for Remote Sensing Images via Local–Global Combined Network
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7937881&tag=1 | N/A | CNNs Using Coffe | N/A | N/A | -breaking down images into multiple layers. Lower layers will include local features(edges and contour). Higher layers will include global features (environmental type). -The Convoluted results of each layer will be combined together by an additional layer. High-frequency components (residuals) need to be added to the low-resolution images to generate the high-resolution image. | Denny | | | | | | | No specific name mentioned | Hyperspectral Image Superresolution by Transfer Learning https://ieeexplore.ieee.org/document/7855724 | No code, but algorithm was provided | CNNs and transfer learning | N/A | Fine-tuned a pretrained convolutional neural network (CNN) | - uses deep convolutional neural networks (CNNs) and transfer learning. -Collaborative nonnegative matrix factorization (CNMF) is introduced to enforce collaboration between low- and high-resolution HSIs. | Denny | | | | | | | Spatial Constrained Spectral Difference Convolutional Neural Network (SCT SDCNN) | Hyperspectral image super-resolution using deep convolutional neural network https://www.researchgate.net/publication/317024713_Hyperspectral_image_super-resolution_using_deep_convolutional_neural_network | N/A | CNN | N/A | N/A | -combines a spatial constraint strategy (SCT) with a deep spectral difference convolutional neural network (SDCNN) -SCT involves iteratively updating the HR HSI estimate to reduce the difference between the down-sampled version of the HR HSI and the input LR HSI. -SDCNN model is designed to learn an end-to-end mapping between the spectral differences of the LR HSI and the HR HSI

| Denny | | | | | | | NLRTATV (Nonlocal Low-Rank Tensor Approximation and Total Variation Regularization) | https://www.researchgate.net/publication/317024713_Hyperspectral_image_super-resolution_using_deep_convolutional | N/A | tensor-based modeling techniques and optimization methods | N/A | N/A | Steps involves: -Converting the image to a tensor -add noise -find and group similar patches -upsample initial image -apply low-rank tensor approximation -smooth the image -optimize using ADMM. -Alternating Direction Method of Multipliers (ADMM): This optimization method is used to solve the resulting nonconvex optimization problem effectively. | Denny | | | | | | | MSDformer (Multiscale Deformable Transformer for Hyperspectral Image Super-Resolution) | https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10252045 | N/A | PyTorch | N/A | N/A | -Combines Convolutional Neural Networks (CNN) and Transformer structures. -Captures local multiscale spatial-spectral features with dilated convolutions (MSAM). -Extracts global spatial-spectral features using deformable convolutions and Transformers (DCTM). | Denny | Chikusei 2x: 43.5016 dB 4x: 38.9173 dB 8x: 34.7037 dB | | | | | | PDE-Net | Deep Posterior Distribution-based Embedding for Hyperspectral Image Super-resolution

https://arxiv.org/abs/2205.14887 | https://github.com/jinnh/PDE-Net | Pytorch | N/A | N/A | -Uses Source-Consistent Reconstruction -Posterior Distribution-based Embedding (new) -Lots of math | | | | | | | | Applies Deep Image Prior Strategy to Hyperspectral Images, CNN | Deep Hyperspectral Prior: Denoising, Inpainting, Super-Resolution

https://arxiv.org/abs/1902.00301 | https://github.com/acecreamu/deep-hs-prior?tab=readme-ov-file | pytorch | python 3.6 | yes | -No new model, just using inherent characteristics of CNNs applied to HSI -Should work out of the box | | | | | | Yes, needs cuda but maybe can be bypassed? | | | Unsupervised Sparse Dirichlet-Net for Hyperspectral Image Super-Resolution

https://arxiv.org/abs/1804.05042 | https://github.com/aicip/uSDN | tensorflow | python 3.5 | yes | | | | | | | | | HSI-RefSR | Hyperspectral Image Super Resolution with Real Unaligned RGB Guidance

https://arxiv.org/abs/2302.06298 | https://github.com/Zeqiang-Lai/HSI-RefSR?tab=readme-ov-file | pytorch | python 3.7 | yes | | | | | | Josh | | | | Model-Guided Deep Hyperspectral Image Super-Resolution

https://ieeexplore.ieee.org/document/9429905 | https://github.com/chengerr/Model-Guided-Deep-Hyperspectral-Image-Super-resolution | pytorch | unsure | no | | | | | | | | | HSRnet | Hyperspectral Image Super-resolution via Deep Spatio-spectral Convolutional Neural Networks

https://arxiv.org/abs/2005.14400 | https://github.com/liangjiandeng/HSRnet | tensorflow | unsure | no | tensorflow version is too old, not good to use | | | | | Josh | | | | Hyperspectral Image Spatial Super-Resolution via 3D Full Convolutional Neural Network

https://www.mdpi.com/2072-4292/9/11/1139 | https://github.com/MeiShaohui/Hyperspectral-Image-Spatial-Super-Resolution-via-3D-Full-Convolutional-Neural-Network?tab=readme-ov-file | tensorflow | python 2.7 | yes | | | | | | | | | | https://ieeexplore.ieee.org/abstract/document/8954237 | https://github.com/ColinTaoZhang/HSI-SR | | c++ | | | | | | | | | | | https://arxiv.org/abs/2201.09851 | https://github.com/xiuheng-wang/Deep_gradient_HSI_superresolution | | matlab | | | | | | | | | | Fusformer | https://arxiv.org/abs/2109.02079 | https://github.com/J-FHu/Fusformer | pytorch | python 3.8 | yes | | | | | | Josh | | | Sylvester Model | https://arxiv.org/abs/2009.04237 | https://github.com/xiuheng-wang/Sylvester_TSFN_MDC_HSI_superresolution | | | | | | | | | Josh | Yes, WIP | | Implementation of https://ieeexplore.ieee.org/document/9334383 | https://ieeexplore.ieee.org/document/9334383 | https://github.com/qianngli/ERCSR | pytorch | python 2.7 | no | | | | | | | |

• https://github.com/junjun-jiang/Hyperspectral-Image-Super-Resolution-Benchmark?tab=readme-ov-file
• https://github.com/topics/hyperspectral-image
• https://github.com/topics/hyperspectral-imaging