TY - GEN
T1 - Exploring GeoAI Methods for Supraglacial Lake Mapping on Greenland Ice Sheet
AU - Luo, Xuanshu
AU - Walther, Paul
AU - Mansour, Wejdene
AU - Teuscher, Balthasar
AU - Zollner, Johann Maximilian
AU - Li, Hao
AU - Werner, Martin
N1 - Publisher Copyright:
© 2023 ACM.
PY - 2023/11/13
Y1 - 2023/11/13
N2 - The ACM SIGSPATIAL Cup 2023 proposed the challenge to identify and map supraglacial lakes in Greenland in satellite imagery. The peculiarities of supraglacial lakes pose a hard problem for semantic segmentation and object detection tasks because the definition of a lake is ill-fitted to the inner workings of such approaches. For example, lakes are often covered by ice and snow and narrow streams can connect distinct lakes, which is not directly translatable to the semantic segmentation of water. It is also not well-posed for object detection, especially the identity relation - what is a lake, what is not (yet) a lake, and what are two lakes is challenging. In this context, we worked on adapting semantic segmentation using the Segment Anything Model and instance segmentation using Mask R-CNN to the setting. The latter ended up superior in our own evaluation and even got ranked second among all participants. We are proud that our approach has led to competitive performance. The source code is available from https://github.com/tum-bgd/GISCup23.
AB - The ACM SIGSPATIAL Cup 2023 proposed the challenge to identify and map supraglacial lakes in Greenland in satellite imagery. The peculiarities of supraglacial lakes pose a hard problem for semantic segmentation and object detection tasks because the definition of a lake is ill-fitted to the inner workings of such approaches. For example, lakes are often covered by ice and snow and narrow streams can connect distinct lakes, which is not directly translatable to the semantic segmentation of water. It is also not well-posed for object detection, especially the identity relation - what is a lake, what is not (yet) a lake, and what are two lakes is challenging. In this context, we worked on adapting semantic segmentation using the Segment Anything Model and instance segmentation using Mask R-CNN to the setting. The latter ended up superior in our own evaluation and even got ranked second among all participants. We are proud that our approach has led to competitive performance. The source code is available from https://github.com/tum-bgd/GISCup23.
KW - computer vision
KW - image segmentation
KW - mask R-CNN
KW - satellite imagery
KW - segment anything model
KW - supraglacial lakes
UR - http://www.scopus.com/inward/record.url?scp=85182514073&partnerID=8YFLogxK
U2 - 10.1145/3589132.3629971
DO - 10.1145/3589132.3629971
M3 - Conference contribution
AN - SCOPUS:85182514073
T3 - GIS: Proceedings of the ACM International Symposium on Advances in Geographic Information Systems
BT - 31st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, ACM SIGSPATIAL GIS 2023
A2 - Damiani, Maria Luisa
A2 - Renz, Matthias
A2 - Eldawy, Ahmed
A2 - Kroger, Peer
A2 - Nascimento, Mario A.
PB - Association for Computing Machinery
T2 - 31st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, ACM SIGSPATIAL GIS 2023
Y2 - 13 November 2023 through 16 November 2023
ER -