Deep Neural Network Regression for Normalized Digital Surface Model Generation with Sentinel-2 Imagery

In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution are precious, they can provide detailed information for specific area. However, measurements with high resolution time-consuming and costly. Only few approaches exist create high-resolution extensive areas. This paper explores extract nDSMs from low-resolution Sentinel-2 data, allowing us derive models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, providing steady updates through repetition rate. Several deep-learning trained overcome gap in producing maps input data. With U-Net base architecture, we extend capabilities our model by integrating tailored multiscale encoders differently sized kernels convolution well conformed self-attention inside skip connection gates. Using pixel-wise regression, achieve mean height error approximately two meters. Moreover, enhancements reduce more than seven percent.