Constructing a Dataset for House Detection in Baghdad Boundaries Using Satellite Imagery and Deep Learning

Abstract

Unplanned residential expansion into agricultural areas has been driven by factors like population growth, internal displacement, and expensive housing options. This article presents the Baghdad Houses detection dataset. The dataset includes 1627 manually annotated images of the Dorra area in Baghdad, along with bounding box annotations that define houses in the satellite image. This dataset will be used by deep learning algorithms to understand and analyze human population expansion by detecting the spread of houses in the newly developed area in the boundaries of Baghdad (object detection). After the training and evaluation process, the chosen algorithms were able to detect houses and visualize residential expansion with a reasonable degree of success. This method can be used to monitor and understand the population expansion in Iraq.