Estimating land covers from street-level images
Abstract
Land cover mapping is a process to characterize terrestrial surfaces into defined thematic classes across space and is often achieved by supervised classification modeling of satellite and/or aero images. However, it is challenging to build rich-quantity and high-quality reference data sets for the model due to the manual check of the ground information. To help understand land cover on the ground, we developed a deep learning model to estimate land cover from street-level photos. We automatically build geo-referenced point data sets with land cover estimation from Mapillary images with a classification accuracy of 0.84. This model contributes to the subsequent study of building a semi-automatic reference database from geo-tagged street-level photos.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMGC45I0919T