Lake Ice Monitoring with Webcams and Crowd-Sourced Images

摘要: Lake ice is a strong climate indicator and has been recognised as part of the Essential Climate Variables (ECV) by Global Observing System (GCOS). The dynamics freezing thawing, possible shifts patterns over time, can help in understanding local global systems. One way to acquire spatio-temporal information about lake formation, independent clouds, analyse webcam images. This paper intends move towards universal model for monitoring with freely available data. We demonstrate good performance, including ability generalise across different winters lakes, state-of-the-art Convolutional Neural Network (CNN) semantic image segmentation, Deeplab v3+. Moreover, we design variant that model, termed Deep-U-Lab, which predicts sharper, more correct segmentation boundaries. have tested model's data from multiple camera views two winters. On average, it achieves intersection-over-union (IoU) values ~71% cameras ~69% winters, greatly outperforming prior work. Going even further, show 60% IoU on arbitrary images scraped photo-sharing web sites. As work, introduce new benchmark dataset images, Photi-LakeIce, along pixel-wise ground truth annotations.