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

Area of Science:

• Geoinformatics
• Remote Sensing
• Geomorphometry

Background:

• High-resolution Digital Elevation Models (DEMs) are vital for geospatial analysis but challenging to acquire.
• Traditional DEM super-resolution (SR) methods like bicubic interpolation often over-smooth terrain features.
• Applying advanced image SR techniques to DEM SR is underexplored due to terrain complexity.

Purpose of the Study:

• To evaluate the suitability of state-of-the-art image super-resolution (SR) methods for Digital Elevation Model (DEM) SR.
• To compare the performance of traditional interpolation against neural network-based SR techniques for DEM data.

Main Methods:

• Comparative analysis of DEM super-resolution techniques.
• Implementation and evaluation of bicubic interpolation.
• Assessment of three neural network-based SR methods, including Super-Resolution with Generative Adversarial Network (SRGAN).

Main Results:

• SRGAN demonstrated superior accuracy in DEM super-resolution tasks compared to other methods.
• Traditional interpolation methods showed limitations in preserving high-frequency terrain details.
• Neural network approaches show promise for enhancing DEM resolution.

Conclusions:

• SRGAN is a highly effective method for DEM super-resolution.
• Advanced machine learning techniques, particularly SRGAN, offer significant improvements over traditional interpolation for DEM enhancement.
• Further research into adapting image SR methods for geospatial data is warranted.