Satellite Mapping of Urban Settlements in a Large City using Deep Learning

Keywords: Deep learning, Urban settlements, Land cover, Urban environments
Abstract Type: Paper Abstract

Abstract

Deep learning (DL) has become a trending topic among the remote sensing community because of its great potential for image analysis. Some DL models can learn high-level unknown features and explore implicit patterns from remotely sensed data, which can provide analytic solutions for complex environments such as urban areas. However, training DL models usually require a large number of samples. Designing DL models that can work with small trainsets can be critical for deep learning applied in various remote sensing efforts. This study aims to improve the mapping accuracy for two complex built-up land types in a large urban area using a DL model with limited training samples. We have demonstrated the great potential of a DL model with limited training samples in addressing some land cover mapping challenges in a complex urban environment.