Abstract
A detailed tree inventory is necessary to accurately estimate the ecosystem contributions of urban forests. In this study, we evaluate a novel method for mapping of urban tree species. The method incorporates the fusion of (a) LiDAR data, (b) very-high resolution orthophotos and (c) multi-temporal PlanetScope data within a multi-modal deep learning framework. Early fusion was used to combine the LiDAR data with the orthophotos while intermediate fusion was used to combine both with the PlanetScope data. An ablation study was performed to assess the contribution of each image source. The proposed workflow reached an overall accuracy (OA) of 90.7%. The orthophotos contribute most to the accuracy of the model (80.9% OA) followed by the multi-temporal PlanetScope data (68.2% OA). The early fusion of the LiDAR data and the orthophotos did not prove effective and did not increase model accuracy any further.
Original language | English |
---|---|
Title of host publication | Proceedings 2023 Joint Urban Remote Sensing Event (JURSE) |
Publisher | IEEE Xplore |
Pages | 1-4 |
Number of pages | 4 |
ISBN (Electronic) | 9781665493734 |
DOIs | |
Publication status | Published - 8 Jun 2023 |
Publication series
Name | 2023 Joint Urban Remote Sensing Event, JURSE 2023 |
---|
Bibliographical note
Publisher Copyright:© 2023 IEEE.