Registration of Interferometric DEM by Deep Artificial Neural Networks Using GPS Control Points’ Coordinates as Network Target

Abstract

The Shuttle Radar Topography Mission (SRTM) satellite’s digital elevation model (DEM) is an important tool for studying topographic features on a medium-spacing scale. Data were collected and processed using the satellite’s orbital and navigation parameters with selected global GPS stations for verification. Distortion may be expressed by surveying measurements, such as position, distance, area, and shape. This study focuses on this distortion and proposes a new registration method to reduce its effect. Because of generality, the purpose shapes were excluded from this study. The proposed registration method depends on precise GPS control points that act as the ground truth for describing the considered surveying measurements. The processing was carried out using deep artificial neural networks (DANN) to produce a new registered DEM. A comparison was made between the original DEM and the new one, focusing on the selected surveying measurements. Another comparison was made between the GPS coordinates and SRTM polynomials to determine the potential of the proposed system. Some statistical investigations were applied to determine the level of significance of the distortion in each surveying measurement. The study shows that the distortion is highly significant; therefore, the proposed registration method is recommended to fix the distortion.