Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing Image Super-resolution Techniques in Remote Sensing: A Comprehensive Survey

Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, Wen Nie

TL;DR

This comprehensive survey evaluates RSISR across supervised/unsupervised paradigms, detailing traditional and deep-learning approaches, modality-specific strategies, and evaluation frameworks. It emphasizes the ill-posed nature of RSISR and the need for degradation modeling that reflects real-world sensor physics, while highlighting emergent architectures such as Transformers, Diffusion models, and Mamba as drivers of performance gains. The paper underscores the scarcity of realistic benchmarks and the importance of task-oriented evaluation to bridge the gap between SR quality and downstream applicability in land use, monitoring, and disaster response. By proposing an inclusive taxonomy, extensive literature synthesis (over 400 papers), and forward-looking directions including foundation models, multi-modal fusion, and efficiency, it aims to steer RSISR toward robust, real-world deployment.

Abstract

Remote sensing image super-resolution (RSISR) is a crucial task in remote sensing image processing, aiming to reconstruct high-resolution (HR) images from their low-resolution (LR) counterparts. Despite the growing number of RSISR methods proposed in recent years, a systematic and comprehensive review of these methods is still lacking. This paper presents a thorough review of RSISR algorithms, covering methodologies, datasets, and evaluation metrics. We provide an in-depth analysis of RSISR methods, categorizing them into supervised, unsupervised, and quality evaluation approaches, to help researchers understand current trends and challenges. Our review also discusses the strengths, limitations, and inherent challenges of these techniques. Notably, our analysis reveals significant limitations in existing methods, particularly in preserving fine-grained textures and geometric structures under large-scale degradation. Based on these findings, we outline future research directions, highlighting the need for domain-specific architectures and robust evaluation protocols to bridge the gap between synthetic and real-world RSISR scenarios.

Advancing Image Super-resolution Techniques in Remote Sensing: A Comprehensive Survey

TL;DR

This comprehensive survey evaluates RSISR across supervised/unsupervised paradigms, detailing traditional and deep-learning approaches, modality-specific strategies, and evaluation frameworks. It emphasizes the ill-posed nature of RSISR and the need for degradation modeling that reflects real-world sensor physics, while highlighting emergent architectures such as Transformers, Diffusion models, and Mamba as drivers of performance gains. The paper underscores the scarcity of realistic benchmarks and the importance of task-oriented evaluation to bridge the gap between SR quality and downstream applicability in land use, monitoring, and disaster response. By proposing an inclusive taxonomy, extensive literature synthesis (over 400 papers), and forward-looking directions including foundation models, multi-modal fusion, and efficiency, it aims to steer RSISR toward robust, real-world deployment.

Abstract

Remote sensing image super-resolution (RSISR) is a crucial task in remote sensing image processing, aiming to reconstruct high-resolution (HR) images from their low-resolution (LR) counterparts. Despite the growing number of RSISR methods proposed in recent years, a systematic and comprehensive review of these methods is still lacking. This paper presents a thorough review of RSISR algorithms, covering methodologies, datasets, and evaluation metrics. We provide an in-depth analysis of RSISR methods, categorizing them into supervised, unsupervised, and quality evaluation approaches, to help researchers understand current trends and challenges. Our review also discusses the strengths, limitations, and inherent challenges of these techniques. Notably, our analysis reveals significant limitations in existing methods, particularly in preserving fine-grained textures and geometric structures under large-scale degradation. Based on these findings, we outline future research directions, highlighting the need for domain-specific architectures and robust evaluation protocols to bridge the gap between synthetic and real-world RSISR scenarios.

Paper Structure

This paper contains 54 sections, 8 equations, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background and Taxonomy
  3. Background of RSISR
  4. Taxonomy of RSISR Algorithms
  5. Supervised RSISR Methods
  6. Traditional RSISR Methods
  7. Sparse Coding-based Methods
  8. Neighborhood Embedding-based Methods
  9. Regression-based Methods
  10. Deep learning-based RSISR Methods
  11. CNN-based RSISR
  12. GAN-based RSISR
  13. Transformer-based RSISR
  14. Mamba-based RSISR
  15. Diffusion Model-based RSISR
  16. ...and 39 more sections

Figures (8)

  • Figure 1: Several examples of RSISR. The first column shows the LR images, while the fifth column shows the results of 4$\times$ SR using the SRCNN method 711517110.1007/978. The second and third columns show heatmap visualizations for the strongest-response channel of the feature maps generated by the first and second convolutional layers, respectively. The fourth column displays the HR ground truth. As can be seen, SR images contain richer details and texture information. The relevant dataset and SR implementation codes are from Fernandez2017single.
  • Figure 2: The number of papers about SR algorithms for RSI since 2014 (according to Web of Science).
  • Figure 3: Taxonomy of the existing RSISR methods.
  • Figure 4: RSISR framework based on sparse representation.
  • Figure 5: NE-based RSISR method
  • ...and 3 more figures