In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution surface models are precious, as they can provide detailed information for a specific area. However, measurements with a high resolution are time-consuming and costly. Only a few approaches exist to create high-resolution normalized digital surface models for extensive areas. This paper explores approaches to extract high-resolution nDSMs from low-resolution Sentinel-2 data, allowing us to derive large-scale models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, and providing steady updates through a high repetition rate. Several deep-learning models are trained to overcome the gap in producing high-resolution surface maps from low-resolution input data. With U-Net as a base architecture, we extend the capabilities of our model by integrating tailored multiscale encoders with differently sized kernels in the convolution as well as conformed self-attention inside the skip connection gates. Using pixel-wise regression, our U-Net base models can achieve a mean height error of approximately two meters. Moreover, through our enhancements to the model architecture, we reduce the model error by more than seven percent.
%0 Journal Article
%1 10189905
%A Müller, Konstantin
%A Leppich, Robert
%A Geiß, Christian
%A Borst, Vanessa
%A Pelizari, Patrick Aravena
%A Kounev, Samuel
%D 2023
%J IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
%K descartes t_interdisciplinary t_journalmagazine myown
%P 1-14
%T Deep Neural Network Regression for Normalized Digital Surface Model Generation with Sentinel-2 Imagery
%X In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution surface models are precious, as they can provide detailed information for a specific area. However, measurements with a high resolution are time-consuming and costly. Only a few approaches exist to create high-resolution normalized digital surface models for extensive areas. This paper explores approaches to extract high-resolution nDSMs from low-resolution Sentinel-2 data, allowing us to derive large-scale models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, and providing steady updates through a high repetition rate. Several deep-learning models are trained to overcome the gap in producing high-resolution surface maps from low-resolution input data. With U-Net as a base architecture, we extend the capabilities of our model by integrating tailored multiscale encoders with differently sized kernels in the convolution as well as conformed self-attention inside the skip connection gates. Using pixel-wise regression, our U-Net base models can achieve a mean height error of approximately two meters. Moreover, through our enhancements to the model architecture, we reduce the model error by more than seven percent.
@article{10189905,
abstract = {In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution surface models are precious, as they can provide detailed information for a specific area. However, measurements with a high resolution are time-consuming and costly. Only a few approaches exist to create high-resolution normalized digital surface models for extensive areas. This paper explores approaches to extract high-resolution nDSMs from low-resolution Sentinel-2 data, allowing us to derive large-scale models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, and providing steady updates through a high repetition rate. Several deep-learning models are trained to overcome the gap in producing high-resolution surface maps from low-resolution input data. With U-Net as a base architecture, we extend the capabilities of our model by integrating tailored multiscale encoders with differently sized kernels in the convolution as well as conformed self-attention inside the skip connection gates. Using pixel-wise regression, our U-Net base models can achieve a mean height error of approximately two meters. Moreover, through our enhancements to the model architecture, we reduce the model error by more than seven percent.},
added-at = {2023-07-25T01:04:42.000+0200},
author = {Müller, Konstantin and Leppich, Robert and Geiß, Christian and Borst, Vanessa and Pelizari, Patrick Aravena and Kounev, Samuel},
biburl = {https://www.bibsonomy.org/bibtex/21cb66ed1a5f48ed7e014fd25f552757f/vanessa_borst},
interhash = {9324f74c2a1ee548007c384406d821a9},
intrahash = {1cb66ed1a5f48ed7e014fd25f552757f},
journal = {IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
keywords = {descartes t_interdisciplinary t_journalmagazine myown},
pages = {1-14},
timestamp = {2023-07-25T01:04:42.000+0200},
title = {Deep Neural Network Regression for Normalized Digital Surface Model Generation with Sentinel-2 Imagery},
year = 2023
}