Comparison of DEM Super-Resolution Methods Based on Interpolation and Neural Networks

Zhang, Yifan; Yu, Wenhao

doi:10.3390/s22030745

Cited by 31 publications

(8 citation statements)

References 74 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To demonstrate the effectiveness of the proposed DSMSR, it was compared with the traditional bilinear interpolation algorithm (BL) and four widely used deep learning-based SR models (i.e., SRCNN, SRGAN, ESRGAN, and MA-GAN). SRCNN, SRGAN, and ESRGAN have been employed in the DEM super-resolution [9,24,25,27,56], and they were retrained using the DSM data from this experiment. discriminator remain relatively stable, indicating that the training has achieved a favorable Nash equilibrium and resulting in a satisfactory trained DSMSR.…”

Section: A Experimental Designmentioning

confidence: 99%

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Sun,

Chen,

Huang

et al. 2024

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

View full text Add to dashboard Cite

Digital Surface Model (DSM) is the fundamental data in various geoscience applications, such as city 3D modeling and urban environment analysis. The freely available DSM often suffers from limited spatial resolution. Super-resolution (SR) is a promising technique to increase the spatial resolution of DSM. However, most existing SR models struggle to reconstruct spatial details such as buildings, valleys, and ridges. This paper proposes a novel DSM super-resolution (DSMSR) model that integrates high-resolution remote sensing imagery using generative adversarial networks. The generator in DSMSR contains three modules. The first DSM feature extraction module uses the residual-in-residual dense block (RRDB) to extract features from low-resolution DSM. The second multiscale attention feature extraction module employs the pyramid convolutional residual dense (PCRD) blocks to capture spatial details of ground objects at multiple scales from remote sensing imagery. The third DSM reconstruction module uses a squeeze-and-excitation (SE) block to fuse the extracted features from low-resolution DSM and highresolution remote sensing imagery for generating SR DSM. The discriminator of DSMSR uses the relativistic average discriminator (RaD) for adversarial learning. The slope loss is further introduced to ensure the accurate representation of topographic features. We evaluate DSMSR on four different terrain regions in the UK to downscale the 30-m AW3D30 DSM to 5-m DSM. The experimental results indicate that DSMSR outperforms traditional interpolation algorithms and four existing deep learning-based SR models. The DSMSR restores more spatial detail of topographic features and generates more accurate image quality, elevation and terrain metrics.

show abstract

Section: A Experimental Designmentioning

confidence: 99%

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Sun,

Chen,

Huang

et al. 2024

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

View full text Add to dashboard Cite

show abstract

“…More recently, a number of sophisticated deep learning artificial intelligence models, such as deep residual networks [22], Recursive Sub-Pixel Convolutional Neural Networks [23], Laplacian of Gaussian Super-resolution [24], Reconstruction Network Combining Internal and External Learning [25], Super-Resolution with Generative Adversarial Network [26] have been proposed. These methods are mostly based on approaches proposed for image super-resolution [22][23][24][25][26]. The result shows that the DEM accuracy has improved regarding root mean square error (RMSE) and the closeness to the reference data [19,20].…”

Section: Introductionmentioning

confidence: 99%

Impacts of Resampling and Downscaling Digital Elevation Model and Its Morphometric Factors: A Comparison of Hopfield Neural Network, Bilinear, Bicubic, and Kriging Interpolations

Minh,

Huong,

Khanh

et al. 2024

Remote Sensing

View full text Add to dashboard Cite

The digital elevation model (DEM) and its derived morphometric factors, i.e., slope, aspect, profile and plan curvatures, and topographic wetness index (TWI), are essential for natural hazard modeling and prediction as they provide critical information about the terrain’s characteristics that can influence the likelihood and severity of natural hazards. Therefore, increasing the accuracy of the DEM and its derived factors plays a critical role. The primary aim of this study is to evaluate and compare the effects of resampling and downscaling the DEM from low to medium resolution and from medium to high resolutions using four methods: namely the Hopfield Neural Network (HNN), Bilinear, Bicubic, and Kriging, on five morphometric factors derived from it. A geospatial database was established, comprising five DEMs with different resolutions: specifically, a SRTM DEM with 30 m resolution, a 20 m resolution DEM derived from topographic maps at a scale of 50,000, a 10 m resolution DEM generated from topographic maps at a scale of 10,000, a 5 m resolution DEM created using surveying points with total stations, and a 5 m resolution DEM constructed through drone photogrammetry. The accuracy of the resampling and downscaling was assessed using Root Mean Square Error (RMSE) and mean absolute error (MAE) as statistical metrics. The results indicate that, in the case of downscaling from low to medium resolution, all four methods—HNN, Bilinear, Bicubic, and Kriging—significantly improve the accuracy of slope, aspect, profile and plan curvatures, and TWI. However, for the case of medium to high resolutions, further investigations are needed as the improvement in accuracy observed in the DEMs does not necessarily translate to the improvement of the second derivative morphometric factors such as plan and profile curvatures and TWI. While RMSEs of the first derivatives of DEMs, such as slope and aspect, reduced in a range of 8% to 55% in all five datasets, the RMSEs of curvatures and TWI slightly increased in cases of downscaling and resampling of Dataset 4. Among the four methods, the HNN method provides the highest accuracy, followed by the bicubic method. The statistics showed that in all five cases of the experiment, the HNN downscaling reduced the RMSE and MAE by 55% for the best case and 10% for the worst case for slope, and it reduced the RMSE by 50% for the best case of aspect. Both the HNN and the bicubic methods outperform the Kriging and bilinear methods. Therefore, we highly recommend using the HNN method for downscaling DEMs to produce more accurate morphometric factors, slope, aspect, profile and plan curvatures, and TWI.

show abstract

“…In order to enhance perception, Ledig et al [20] integrated a perceptual loss function with a generative adversarial network for SR. Given the distinct characteristics of DEM data compared to natural images, Zhang and Yu [21] applied SRGAN, ESRGAN, and CEDGAN to DEM SR, and the experimental results showed that SRGAN was the most effective in terms of extracting terrain features, outperforming other GANs. This result suggests that neural networks with superior performance in the CV domain might not be equally effective when extracting topographic features.…”

Section: Introductionmentioning

confidence: 99%

A Seabed Terrain Feature Extraction Transformer for the Super-Resolution of the Digital Bathymetric Model

Cai,

Liu,

Chen

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

The acquisition of high-resolution (HR) digital bathymetric models (DBMs) is crucial for oceanic research activities. However, obtaining HR DBM data is challenging, which has led to the use of super-resolution (SR) methods to improve the DBM’s resolution, as, unfortunately, existing interpolation methods for DBMs suffer from low precision, which limits their practicality. To address this issue, we propose a seabed terrain feature extraction transform model that combines the seabed terrain feature extraction module with the efficient transform module, focusing on the terrain characteristics of DBMs. By taking advantage of these two modules, we improved the efficient extraction of seabed terrain features both locally and globally, and as a result, we obtained a highly accurate SR reconstruction of DBM data within the study area, including the Mariana Trench in the Pacific Ocean and the adjacent sea. A comparative analysis with bicubic interpolation, SRCNN, SRGAN, and SRResNet shows that the proposed method decreases the root mean square error (RMSE) by 16%, 10%, 13%, and 12%, respectively. These experimental results confirm the high accuracy of the proposed method in terms of reconstructing HR DBMs.

show abstract

Comparison of DEM Super-Resolution Methods Based on Interpolation and Neural Networks

Cited by 31 publications

References 74 publications

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Digital Surface Model Super-Resolution by Integrating High-Resolution Remote Sensing Imagery Using Generative Adversarial Networks

Impacts of Resampling and Downscaling Digital Elevation Model and Its Morphometric Factors: A Comparison of Hopfield Neural Network, Bilinear, Bicubic, and Kriging Interpolations

A Seabed Terrain Feature Extraction Transformer for the Super-Resolution of the Digital Bathymetric Model

Contact Info

Product

Resources

About