Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic
Deep learning has been used for mapping retrogressive thaw slumps and other periglacial landforms but its application is still limited to local study areas. To understand the accuracy, efficiency, and transferability of a deep learning model (i.e., DeepLabv3+) when applied to large areas or multiple...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-06-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/14/12/2747 |
| _version_ | 1797482751424724992 |
|---|---|
| author | Lingcao Huang Trevor C. Lantz Robert H. Fraser Kristy F. Tiampo Michael J. Willis Kevin Schaefer |
| author_facet | Lingcao Huang Trevor C. Lantz Robert H. Fraser Kristy F. Tiampo Michael J. Willis Kevin Schaefer |
| author_sort | Lingcao Huang |
| collection | DOAJ |
| description | Deep learning has been used for mapping retrogressive thaw slumps and other periglacial landforms but its application is still limited to local study areas. To understand the accuracy, efficiency, and transferability of a deep learning model (i.e., DeepLabv3+) when applied to large areas or multiple regions, we conducted several experiments using training data from three different regions across the Canadian Arctic. To overcome the main challenge of transferability, we used a generative adversarial network (GAN) called CycleGAN to produce new training data in an attempt to improve transferability. The results show that (1) data augmentation can improve the accuracy of the deep learning model but does not guarantee transferability, (2) it is necessary to choose a good combination of hyper-parameters (e.g., backbones and learning rate) to achieve an optimal trade-off between accuracy and efficiency, and (3) a GAN can significantly improve the transferability if the variation between source and target is dominated by color or general texture. Our results suggest that future mapping of retrogressive thaw slumps should prioritize the collection of training data from regions where a GAN cannot improve the transferability. |
| first_indexed | 2024-03-09T22:37:58Z |
| format | Article |
| id | doaj.art-a68c0037dafe47798a3d3caae0c53ee4 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-09T22:37:58Z |
| publishDate | 2022-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-a68c0037dafe47798a3d3caae0c53ee42023-11-23T18:46:09ZengMDPI AGRemote Sensing2072-42922022-06-011412274710.3390/rs14122747Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian ArcticLingcao Huang0Trevor C. Lantz1Robert H. Fraser2Kristy F. Tiampo3Michael J. Willis4Kevin Schaefer5Earth Science and Observation Center, Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO 80309, USASchool of Environmental Studies, University of Victoria, Victoria, BC V8P 5C2, CanadaCanada Centre for Mapping and Earth Observation, Natural Resources Canada, 560 Rochester Street, Ottawa, ON K1S 5K2, CanadaCooperative Institute for Research in Environmental Sciences and Geological Sciences, University of Colorado Boulder, Boulder, CO 80309, USACooperative Institute for Research in Environmental Sciences and Geological Sciences, University of Colorado Boulder, Boulder, CO 80309, USANational Snow and Ice Data Center, Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, CO 80309, USADeep learning has been used for mapping retrogressive thaw slumps and other periglacial landforms but its application is still limited to local study areas. To understand the accuracy, efficiency, and transferability of a deep learning model (i.e., DeepLabv3+) when applied to large areas or multiple regions, we conducted several experiments using training data from three different regions across the Canadian Arctic. To overcome the main challenge of transferability, we used a generative adversarial network (GAN) called CycleGAN to produce new training data in an attempt to improve transferability. The results show that (1) data augmentation can improve the accuracy of the deep learning model but does not guarantee transferability, (2) it is necessary to choose a good combination of hyper-parameters (e.g., backbones and learning rate) to achieve an optimal trade-off between accuracy and efficiency, and (3) a GAN can significantly improve the transferability if the variation between source and target is dominated by color or general texture. Our results suggest that future mapping of retrogressive thaw slumps should prioritize the collection of training data from regions where a GAN cannot improve the transferability.https://www.mdpi.com/2072-4292/14/12/2747DeepLabdomain adaptationgenerative adversarial networkpermafrostthermokarst |
| spellingShingle | Lingcao Huang Trevor C. Lantz Robert H. Fraser Kristy F. Tiampo Michael J. Willis Kevin Schaefer Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic Remote Sensing DeepLab domain adaptation generative adversarial network permafrost thermokarst |
| title | Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic |
| title_full | Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic |
| title_fullStr | Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic |
| title_full_unstemmed | Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic |
| title_short | Accuracy, Efficiency, and Transferability of a Deep Learning Model for Mapping Retrogressive Thaw Slumps across the Canadian Arctic |
| title_sort | accuracy efficiency and transferability of a deep learning model for mapping retrogressive thaw slumps across the canadian arctic |
| topic | DeepLab domain adaptation generative adversarial network permafrost thermokarst |
| url | https://www.mdpi.com/2072-4292/14/12/2747 |
| work_keys_str_mv | AT lingcaohuang accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic AT trevorclantz accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic AT roberthfraser accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic AT kristyftiampo accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic AT michaeljwillis accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic AT kevinschaefer accuracyefficiencyandtransferabilityofadeeplearningmodelformappingretrogressivethawslumpsacrossthecanadianarctic |