Authors: Zhe Wang*, University of Idaho, chao fan, Assistant Professor
Topics: Remote Sensing, Urban Geography, Land Use
Keywords: Urban tree canopy, U-net, deep learning, scale
Session Type: Virtual Paper
Start / End Time: 3:05 PM / 4:20 PM
Room: Virtual 43
Presentation File: No File Uploaded
In this study, we applied the U-net architecture to the urban tree mapping using the high-resolution aerial images. We assessed the performance of the U-net at four different scales and make comparisons with other approaches. Compared with the 8-cm ground truth in evaluation 1, the highest metric score is 0.9914 (OA). In evaluation 2, compared the predicted results with the ground truth images after adjusting to the spatial resolution of the predicted output, the highest metric score is 0.9984 (OA) and all metrics were above 0.99 except for the 16-cm experiment. The U-net architecture was proved to be exceptionally effective on the extraction of the urban tree canopy. To further compare with the conventional method, we employed both the U-net and OBIA over a selected area. The OBIA produced significantly lower metric scores. Even compared with other deep learning approaches implemented in previous studies, the U-net architecture used in this study presented a better performance on the urban tree extraction.