Semi-Automatic Data Enrichment for Open Street Map (Osm) Using Deep Learning Algorithms

Abstract

It is crucial to obtain continuous data on unplanned urbanization regions in order to develop precise plans for future studies in these regions. An unplanned urbanization area was selected for analysis, and road extraction was performed using very high-resolution unmanned aerial vehicle (UAV) images. In this regard, the Sat2Graph deep learning model was employed, utilizing the object detection tool integrated within the deep learning package published by ArcGIS Pro software, for the purpose of road extraction from a very high-resolution UAV image. The high-resolution UAV images were subjected to analysis using the photogrammetry method, with the results obtained through the application of the Sat2Graph deep learning model. The resulting road extraction was employed for the purpose of data enhancement on OpenStreetMap (OSM). This will facilitate the expeditious and precise implementation of data updates conducted by volunteers. It should be noted that the recall, F1 score, precision ratio/uncertainty accuracy, average producer accuracy, and intersection over union of products were automatically extracted with the algorithm and determined to be 0.816, 0.827, 0.838, 0.792, and 0.597, respectively.