Correcting Spectra Outside the Backbone: A Model-Agnostic Rectifier for Hyperspectral Image Super-Resolution

Ji-Xuan He; Guohang Zhuang; Bo Junge; Chen Ling; Tingyi Li; Yanan Qiao; Miaomiao Cai; Jungfeng Fang

cs.CVJan 29, 2026

Correcting Spectra Outside the Backbone: A Model-Agnostic Rectifier for Hyperspectral Image Super-Resolution

Authors: Ji-Xuan He, Guohang Zhuang, Bo Junge, Chen Ling, Tingyi Li, Yanan Qiao, Miaomiao Cai, Jungfeng Fang

Organizations: School of Computer Science and Technology, Xi’an Jiaotong University · School of Computer Science and Information Engineering, Hefei University of Technology

Abstract

Hyperspectral image super-resolution (HSI-SR) aims to recover spatial detail while preserving the spectral shape on which quantitative analysis relies. Recent HSI-SR methods, from repurposed RGB super-resolution backbones to dedicated spectral-spatial architectures, have greatly improved spatial reconstruction. However, overlooking the compact spectral structure of hyperspectral data leaves residual spectral errors, while binding the spectral treatment to each architecture forces it to be rebuilt for every new backbone. Yet the low-dimensional structure of spectra belongs to the data, not to any backbone. Backbones differ in the errors they leave, but not in the structure of the true spectra. One rectifier design can therefore serve any backbone. Building on this insight, we propose the \textbf{S}pectral \textbf{R}ectification \textbf{S}uper-\textbf{R}esolution Network (\textbf{SR $^{2}$ -Net}), a model-agnostic rectifier that needs nothing from the backbone but its output, leaves its internal architecture untouched, and is trained per backbone. SR $^{2}$ -Net follows an \emph{enhance-then-rectify} pipeline in which Hierarchical Spectral-Spatial Synergy Attention (\textbf{H-S $^{3}$ A}) reinforces cross-band interactions, while Mode-Constrained Rectification (\textbf{MCR}) confines the correction to a learned compact spectral subspace. A degradation-consistency constraint further ties the output to the observed low-resolution input. Experiments with five backbones spanning CNN, Transformer, and diffusion families show that one fixed configuration improves spectral fidelity in every reported setting while preserving or improving spatial quality. Averaged over thirty in-domain settings, SR $^{2}$ -Net removes 22.6% of the residual spectral error at a backbone-independent cost of 0.048M parameters.

Figures & tables

Figure 1: Motivation. (a) Strong SR models, general-purpose and HSI-specific alike, introduce spectral distortion during reconstruction; (b) the model-agnostic SR 2 -Net rectifier restores spectral consistency.

Figure 2: Overview of SR 2 -Net . Given the LR input $\textit{I}_{\mathrm{LR}}$ , an SR backbone produces an initial reconstruction $\tilde{\textit{I}}_{\mathrm{SR}}$ , which SR 2 -Net rectifies into the output $\hat{\textit{I}}_{\mathrm{SR}}$ . (a) H-S 3 A enhances cross-band interaction via spectral grouping and adjacent shuffle; (b) TSA fuses spectral, height, and width views to recalibrate features; (c) MCR, drawn with $N$ stages ( $N{=}1$ in our configuration), projects features onto a compact spectral subspace and aggregates the back-projected corrections.

Figure 3: Error maps at $\times 4$ (in-domain, CAVE and ARAD-1K). Pseudo-RGB uses bands $25$ / $15$ / $5$ as R–G–B, and band-averaged absolute error maps (darker is better) pair each +Ours result with its Base. Red boxes mark where SR 2 -Net sharpens edges and fine structures; the cross-domain ICVL comparison and more galleries are in Appendix I .

Figure 4: Per-pixel spectra at $\times 4$ on CAVE across CNN-, Transformer-, and diffusion-based backbones. For two representative pixels (marked in the pseudo-RGB images), the baseline spectra oscillate away from the ground truth while SR 2 -Net follows it; additional scenes and datasets are in Appendix J .

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Setting	Value
Framework	PyTorch
Hardware	NVIDIA RTX 3090 GPU
Optimizer	AdamW ( $\beta_{1}=0.9$ , $\beta_{2}=0.99$ )
Learning rate	$1\times 10^{-4}\rightarrow 1\times 10^{-5}$ (cosine)
Epochs	1000
Batch size	8

Appendix

Table A1: Training hyperparameters, shared across all backbones.

Component	Configuration
Spectral groups	$4$ contiguous groups of $\lceil S/4\rceil$ bands (bands zero-padded to a multiple of four; padding removed after back-projection)
H-S 3 A blocks	$4$ , applied in sequence with independent parameters and a residual connection around each: $F^{(l)}=F^{(l-1)}+\mathrm{Block}_{l}(F^{(l-1)})$ , where $F^{(0)}$ is the zero-padded input and $F^{(4)}=F_{s}$ enters MCR
$\triangleright$ per-group stems	identity / $3\times 3$ / $5\times 5$ depthwise / $3\times 3$ dilated (one per group); $1\times 1$ mixing after the shuffle
$\triangleright$ TSA views	height, width, spectral ( $h,w,s$ )
MCR blocks (refinement stages $N$ )	$1$ ; ablations that drop MCR remove the module entirely (projection and back-projection included), so the rectified output is $\Pi_{S}(F_{s})$ , giving the H-S 3 A-only variant of the component ablation
$\triangleright$ projection $P(\cdot)$	$1\times 1$ conv, $S^{\prime}\rightarrow r$ , GeLU

Appendix

Table A2: Configuration of the SR 2 -Net rectifier, where

S

is the number of spectral bands,

S^{\prime}=4\lceil S/4\rceil

the zero-padded band count, and

r

the MCR bottleneck rank.

Tag	Type	Setting
K1	Kernel	area-based downsampling
K2	Kernel	alternative bicubic implementation
B1	Blur	Gaussian $\sigma=0.6$ , kernel $5\times 5$
B2	Blur	Gaussian $\sigma=1.0$ , kernel $7\times 7$
N1	Noise	Gaussian $\sigma=1/255$
N2	Noise	Gaussian $\sigma=2/255$

Appendix

Table A3: Test-time degradation settings. Each is applied only at test time to a model trained under standard bicubic downsampling.

Method	Scale	mPSNR $\uparrow$	mSAM $\downarrow$	mSSIM $\uparrow$	CC $\uparrow$
Base (SwinIR)	$\times 2$	50.4662	0.7727	0.9972	0.9990
	$\times 4$	39.5717	1.3950	0.9709	0.9902
	$\times 8$	33.4185	2.4682	0.8869	0.9551
Base + SG	$\times 2$	46.1893	1.9059	0.9959	0.9975
	$\times 4$	38.7834	2.4512	0.9698	0.9889
	$\times 8$	33.2419	3.2638	0.8859	0.9541

Appendix

Table A6: Spectral-rectifier comparison on ARAD-1K with SwinIR. SG, PCA, and IBP are fixed post-hoc methods; MCR and SR 2 -Net are jointly trained. Best in bold , second best underlined .

Variant (SwinIR, $\times 4$ )	mPSNR $\uparrow$	mSAM $\downarrow$	mSSIM $\uparrow$
Base	$39.58_{\pm.03}$	$1.397_{\pm.011}$	$0.9710_{\pm.0005}$
+ $\mathcal{L}_{\mathrm{deg}}$ only	$39.70_{\pm.03}$	$1.371_{\pm.010}$	$0.9710_{\pm.0007}$
+ Generic head	$39.99_{\pm.04}$	$1.362_{\pm.012}$	$0.9713_{\pm.0006}$
+ MCR only	$40.25_{\pm.03}$	$1.319_{\pm.010}$	$0.9716_{\pm.0006}$
+ SR 2 -Net	$\mathbf{40.96}_{\pm.03}$	$\mathbf{1.280}_{\pm.009}$	$\mathbf{0.9753}_{\pm.0006}$

Appendix

Table A7: Seed stability on ARAD-1K (

\times 4

, SwinIR): mean

\,\pm\,

std over three seeds

\{42,123,2024\}

. The architecture gap (cheap controls versus MCR/ SR 2 -Net ) is

0.043

0.091

mSAM, at least

3.5\times

the across-seed std.

Scale	Metric	Base	+Ours	$\Delta$
$\times 4$	Params (M)	8.228	8.276	+0.048
$\times 4$	FLOPs (G)	19.302	20.785	+1.482

Appendix

Table A8: Complexity of SR 2 -Net using SwinIR on ARAD-1K at

\times 4

. Parameters and FLOPs are computed for the

32^{2}

LR training patch (

128^{2}

HR); the rectifier adds

0.048

M parameters and

1.482

G FLOPs.

Residual statistic	Base	+ SR 2 -Net	Improved	$p$
Spectral MAE	0.00783	0.00534	50/50	$<10^{-6}$
Gradient $\|\Delta e\|$	0.000838	0.000781	44/50	$<10^{-3}$
Curvature $\|\Delta^{2}e\|$	0.000697	0.000618	42/50	$<10^{-3}$

Appendix

Table A10: Dataset-level residual spectral-error analysis over 50 ARAD-1K test scenes (SwinIR,

\times 4

). “Improved” counts scenes on which SR 2 -Net lowers the statistic;

p

is from a one-sided Wilcoxon signed-rank test. Per-scene mSAM improves on all 50 scenes (

p<10^{-6}

Explore similar work

Spectral Dynamic Attention Network for Hyperspectral Image Super-Resolution
Apr 30, 2026Tengya Zhang, Feng Gao, Lin Qi +2Hyperspectral Image Super-Resolution Hyperspectral Image
Sanya Oceanographic Institution, Ocean University of China, Sanya, China · State Key Laboratory of Physical Oceanography, Ocean University of China, Qingdao 266100, China · Department of Electrical and Computer Engineering, Mississippi State University, Starkville, MS 39762 USA
HyperBench: Standardizing and Scaling Synthetic Evaluation for Hyperspectral Super-Resolution
May 20, 2026Ritik Shah, Marco F. DuarteHyperspectral Image Super-Resolution Signal Reconstruction
University of Massachusetts Amherst, MA 01003
Super-Resolving Unseen Hyperspectral Sensors at Any Scale via Spatial Operators
Sep 30, 2026Ji-Xuan He, Guohang Zhuang, Bo Junge +6Hyperspectral Image Super-Resolution Upsampling
Xi’an Jiaotong University · Hefei University of Technology · National University of Singapore +1

Apr 30, 2026, eess.IVJ/K move · Enter open · S save

Spectral Dynamic Attention Network for Hyperspectral Image Super-Resolution

Tengya Zhang, Feng Gao, Lin Qi, Junyu Dong+1

Sanya Oceanographic Institution, Ocean University of China, Sanya, China · State Key Laboratory of Physical Oceanography, Ocean University of China, Qingdao 266100, China · Department of Electrical and Computer Engineering, Mississippi State University, Starkville, MS 39762 USA

Hyperspectral Image Super-Resolution Hyperspectral Image Two-Level Spatial-Frequency Fusion Strategy Dynamic Sparse Attention

Open paper AskPDF arXiv

Hyperspectral image super-resolution is essential for enhancing the spatial fidelity of HSI data, yet existing deep learning methods often struggle with substantial spectral redundancy and the limited non-linear modeling capacity of standard feed-forward networks (FFNs). To address these challenges, we propose Spectral Dynamic Attention Network (SDANet), a framework designed to adaptively suppress redundant spectral interactions. SDANet integrates two key components: 1) Dynamic Channel Sparse Attention (DCSA) module that computes channel-wise correlations and selectively preserves the most informative attention responses through dynamic and data-dependent sparsification. 2) Frequency-Enhanced Feed-Forward Network (FE-FFN) that jointly models spatial and frequency-domain representations to enhance non-linear expressiveness. Extensive experiments on two benchmark datasets demonstrate that SDANet achieves state-of-the-art HISR performance while maintaining competitive efficiency. The code will be made publicly available at https://github.com/oucailab/SDANet.

Abstract

Explore similar work

Spectral Dynamic Attention Network for Hyperspectral Image Super-Resolution

HyperBench: Standardizing and Scaling Synthetic Evaluation for Hyperspectral Super-Resolution

Super-Resolving Unseen Hyperspectral Sensors at Any Scale via Spatial Operators