Multi-class Land Use/Cover Change Detection with Deep Learning

Keywords: Land Use/Cover Change; GIS; Deep Learning
Session Type: Virtual Poster Abstract
Day: Friday
Session Start / End Time: 2/25/2022 03:40 PM (Eastern Time (US & Canada)) - 2/25/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 39

Abstract

This research aims to use the image segmentation deep learning model to detect land use/cover change over time. To train the deep learning model, Landsat-8 imagery of two time periods in Springfield, MO were obtained, and a change raster was generated with class information between the two time periods (2014-2020). The training data is used to train the change detection deep learning model UNet. The training change raster is used as labels for constructing the model. The trained model was then applied to Kansas City and S.t. Louis area for inferencing. The results show that the model well captures the land use/cover change between 2014 and 2020 in those two areas. All the data processing and modeling are implemented in ArcGIS Pro.