Snow cover detection over forested and mountainous regions from remote sensing imagery using convolutional neural networks

Date

2024-8-28

Author

Özen, Sadettin

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

49
views

98
downloads

Cite This

Studies on hydrology, ecology, environment, and climate greatly benefit from the monitoring of snow cover in mountainous and forested areas. In such varied terrains, traditional snow cover detection approaches frequently struggle with coverage and precision. On the other hand, deep learning methods have revolutionized remote sensing and provide a viable method for accurate and effective snow cover identification. This work offers a comprehensive exploration of using deep learning techniques for this purpose, in particular, using Sentinel-2 multispectral imagery. Utilizing cutting-edge deep learning methods, our investigation shows significant gains in performance. As an example, our model performs well on a test set with temporal variability, yielding a dice score of 0.805. Moreover, the model obtains a dice score of 0.928 on another test set, demonstrating its capacity to precisely define the extent of snow cover. This study also explores the transfer learning strategies, which minimize the reliance on labeled data by fine-tuning pre-trained models on large datasets. Overall, using deep learning techniques and customized band combinations of Sentinel-2 data, this research represents a major increase in remote sensing capabilities for snow cover identification in complex and heterogeneous forested and mountainous environments.

Subject Keywords

Sentinel-2, Remote sensing, Classification, Deep learning, Snow

URI

https://hdl.handle.net/11511/111050

Collections

Graduate School of Natural and Applied Sciences, Thesis