Research on super-resolution reconstruction of remote sensing images: a comprehensive review

Liu, Hui; Qian, Yurong; Zhong, Xiwu; Chen, Long; Yang, Guangqi

doi:10.1117/1.oe.60.10.100901

Cited by 13 publications

(6 citation statements)

References 85 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The advent of deep learning gave a fresh start to the Single Image Super-Resolution problem, for which convolutional, residual and then generative-adversarial architectures have been proposed with success [11,12]. Provided that enough data are available for training, it is possible to estimate a nonlinear mapping with a few hundred thousand parameters that undoes all the high frequency damping and aliasing occurring at sensor level, and even generates plausible high resolution details past the cut-off frequency of the sensor.…”

Section: Introductionmentioning

confidence: 99%

SEN2VENµS, a Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

Michel

Vinasco-Salinas

Inglada

et al. 2022

Data

View full text Add to dashboard Cite

Boosted by the progress in deep learning, Single Image Super-Resolution (SISR) has gained a lot of interest in the remote sensing community, who sees it as an opportunity to compensate for satellites’ ever-limited spatial resolution with respect to end users’ needs. This is especially true for Sentinel-2 because of its unique combination of resolution, revisit time, global coverage and free and open data policy. While there has been a great amount of work on network architectures in recent years, deep-learning-based SISR in remote sensing is still limited by the availability of the large training sets it requires. The lack of publicly available large datasets with the required variability in terms of landscapes and seasons pushes researchers to simulate their own datasets by means of downsampling. This may impair the applicability of the trained model on real-world data at the target input resolution. This paper presents SEN2VENµS, an open-data licensed dataset composed of 10 m and 20 m cloud-free surface reflectance patches from Sentinel-2, with their reference spatially registered surface reflectance patches at 5 m resolution acquired on the same day by the VENµS satellite. This dataset covers 29 locations on earth with a total of 132,955 patches of 256 × 256 pixels at 5 m resolution and can be used for the training and comparison of super-resolution algorithms to bring the spatial resolution of 8 of the Sentinel-2 bands up to 5 m.

show abstract

Section: Introductionmentioning

confidence: 99%

SEN2VENµS, a Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

Michel

Vinasco-Salinas

Inglada

et al. 2022

Data

View full text Add to dashboard Cite

show abstract

“…We evaluated AutoSR4EO and the baseline methods using two metrics: peak signalto-noise ratio (PSNR), a pixel-wise metric related to the MSE, and the structural similarity index measure (SSIM), a perception-based metric that considers the contrast, luminance and structure of images to better reflect human visual interpretation. Both metrics are widely used for evaluating SR approaches [80,[86][87][88]. The PSNR is given by…”

Section: Discussionmentioning

confidence: 99%

AutoSR4EO: An AutoML Approach to Super-Resolution for Earth Observation Images

Wąsala,

Marselis,

Arp

et al. 2024

Remote Sensing

View full text Add to dashboard Cite

Super-resolution (SR), a technique to increase the resolution of images, is a pre-processing step in the pipelines of applications of Earth observation (EO) data. The manual design and optimisation of SR models that are specific to every possible EO use case is a laborious process that creates a bottleneck for EO analysis. In this work, we develop an automated machine learning (AutoML) method to automate the creation of dataset-specific SR models. AutoML is the study of the automatic design of high-performance machine learning models. We present the following contributions. (i) We propose AutoSR4EO, an AutoML method for automatically constructing neural networks for SR. We design a search space based on state-of-the-art residual neural networks for SR and incorporate transfer learning. Our search space is extendable, making it possible to adapt AutoSR4EO to future developments in the field. (ii) We introduce a new real-world single-image SR (SISR) dataset, called SENT-NICFI. (iii) We evaluate the performance of AutoSR4EO on four different datasets against the performance of four state-of-the-art baselines and a vanilla AutoML SR method, with AutoSR4EO achieving the highest average ranking. Our results show that AutoSR4EO performs consistently well over all datasets, demonstrating that AutoML is a promising method for improving SR techniques for EO images.

show abstract

“…SR in RS data has been studied in e.g. [92]- [94]. For future work on AGB prediction by DL regression, it appears relevant to incorporate ideas from the field of SR and investigate additional architectures and balancing of GAN losses against traditional L 1 and L 2 losses for reconstruction.…”

Section: Discussionmentioning

confidence: 99%

On the Potential of Sequential and Nonsequential Regression Models for Sentinel-1-Based Biomass Prediction in Tanzanian Miombo Forests

Björk

Anfinsen

Næsset

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

This study derives regression models for aboveground biomass (AGB) estimation in miombo woodlands of Tanzania that utilise the high availability and low cost of Sentinel-1 data. The limited forest canopy penetration of C-band SAR sensors along with the sparseness of available ground truth restrict their usefulness in traditional AGB regression models. Therefore, we propose to use AGB predictions based on airborne laser scanning (ALS) data as a surrogate response variable for SAR data. This dramatically increases the available training data and opens for flexible regression models that capture fine-scale AGB dynamics. This becomes a sequential modelling approach, where the first regression stage has linked in situ data to ALS data and produced the AGB prediction map; We perform the subsequent stage, where this map is related to Sentinel-1 data. We develop a traditional, parametric regression model and alternative nonparametric models for this stage. The latter uses a conditional generative adversarial network (cGAN) to translate Sentinel-1 images into ALS-based AGB prediction maps. The convolution filters in the neural networks make them contextual. We compare the sequential models to traditional, non-sequential regression models, all trained on limited AGB ground reference data. Results show that our newly proposed non-sequential Sentinel-1based regression model performs better quantitatively than the sequential models, but achieves less sensitivity to fine-scale AGB dynamics. The contextual cGAN-based sequential models best reproduce the distribution of ALS-based AGB predictions. They also reach a lower RMSE against in situ AGB data than the parametric sequential model, indicating a potential for further development.

show abstract

Research on super-resolution reconstruction of remote sensing images: a comprehensive review

Cited by 13 publications

References 85 publications

SEN2VENµS, a Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

SEN2VENµS, a Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

AutoSR4EO: An AutoML Approach to Super-Resolution for Earth Observation Images

On the Potential of Sequential and Nonsequential Regression Models for Sentinel-1-Based Biomass Prediction in Tanzanian Miombo Forests

Contact Info

Product

Resources

About