Enhanced Mapping of Supraglacial Lakes Through Dual-attention Deep Neural Network

• Main Authors: Wang, Haoran, Wei, Jiawei, Gao, Xiaoyong, Sha, Dexuan, Luo, Yiyun, Yang, Junliu
• Other Authors: Botanique et Modélisation de l'Architecture des Plantes et des Végétations (UMR AMAP), Centre de Coopération Internationale en Recherche Agronomique pour le Développement (Cirad)-Centre National de la Recherche Scientifique (CNRS)-Institut de Recherche pour le Développement (IRD France-Sud )-Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement (INRAE)-Université de Montpellier (UM), Association for Computing Machinery, Maria Lui Damiani, Matthias Renz, Ahmed Eldawy, Peer Kröger
• Format: Conference Object
• Language: English
• Published: HAL CCSD 2023
• Subjects: Deep learning; Dual-attention; Object detection; Supraglacial lakes; [SDE]Environmental Sciences; Arctic
• Online Access: https://hal.inrae.fr/hal-04414373; https://doi.org/10.1145/3589132.3629972

International audience Supraglacial lakes in the Arctic undergo seasonal and glacial-activity-induced changes, providing profound insights into ice dynamics and climate changes in these sensitive regions. However, the morphological complexity of these lakes, compounded by the environmental obstructions like clouds and slush fields, poses significant challenges to accurate lake detection. The 31st ACM SIGSPATIAL 2023 initiated a competition, GISCUP 2023, focusing on supraglacial lake detection based on multipart, multi-temporal satellite imagery. This paper, distinguished as the 3rd place winner, introduces a pioneering dual-attention U-net algorithm. This approach synergizes deep learning with spectral and spatial knowledge, ensuring a streamlined pipeline structure that upholds methodological soundness and yields satisfying results.