Automating image labeling for remote sensing using cadastral database and video game engine simulation

In the remote sensing field, utilization of deep learning algorithms, such as Convolutional Neural Networks (CNNs) for automated detection is a commonly adopted approach, as reported by [1]. These techniques have demonstrated significant power and efficacy, largely due to the availability of increasingly large datasets and the rapid advancement in computing technology. However, the preparation of these datasets necessitates a substantial amount of manual labor, which is often outsourced to cost-efficient labor forces. In this paper, we present two methods developed to automate the labeling work for semantic segmentation and object detection tasks. We will analyze the results in terms of accuracy and time saved, and show how we've successfully applied them to two real-life projects.