Proof of Concept for Sea Ice Stage of Development Classification Using Deep Learning
"> Figure 1
<p>(<b>a</b>) Hudson Bay region. Input-output data sources: (<b>b</b>) Single raw SCW image. Input data consist of 350 RADARSAT-2 ScanSAR Wide images. (<b>c</b>) the scheme in <a href="#remotesensing-12-02486-t001" class="html-table">Table 1</a>. Output data consist of 172 image analysis charts.</p> "> Figure 2
<p>Candidate dataset samples consist of SAR image sub-regions, highlighted in the figure, centred about ice chart image analysis sample coordinates. Background SAR image showing the southwestern region of the Hudson Bay and Churchill, Manitoba, indicated by the blue dot. The red hatched area denotes land.</p> "> Figure 3
<p>U-Net based model: The model consists of the encoder (down-sampling) portion of a U-Net model that maps a SAR sub-image to a prediction label. A representative model used in this work consists of approximately 1.2 million parameters.</p> "> Figure 4
<p>DenseNet model: The model consists of a number of dense blocks connected by transition blocks that maps a SAR sub-image to a predication label. A representative model used in this work consists of approximately 7 million parameters.</p> "> Figure 5
<p>A visual representation of training and testing samples from three SAR scenes and the predictions from DenseNet. (<b>a</b>) Training samples from 2018/10/16 11:27:17. (<b>b</b>) Testing samples from 2018/10/16 11:27:17. (<b>c</b>) Training samples from 2018/10/17 22:29:32. (<b>d</b>) Testing samples from 2018/10/17 22:29:32. (<b>e</b>) Training samples from 2018/11/04 12:14:50. (<b>f</b>) Testing samples from 2018/11/04 12:14:50.</p> ">
Abstract
:1. Introduction
2. Materials and Methods
2.1. Study Region
2.2. Description of Data Products
2.3. Dataset Labelling
2.4. SAR Sub-Region Extraction
2.5. Model Setup
2.6. Training Setup
3. Results
3.1. Experimental Configurations
- Data batch—indicates the SAR scenes and CIS data source files to use. In this work, only Batch 4 is considered, whereas previous iterations of data batches consisted of different subsets of source files leading up to Batch 4.
- DEX type—determines the type of CIS data to use, DEXA being a code for daily ice charts and DEXI being a code for image analysis charts. In this work, only DEXI is used.
- Label type—specifies the egg code labelling type used. Only Label Types 4 and 5 are used in this work. Label Type 4 refers to the set of labels defined in Table 1. Label Type 5 modifies the labels in Table 1 to group new ice (label 1) and first-year ice (label 2) into a single class, producing a two class experiment used to benchmark the Label Type 4 experiments.
- Polarization—specifies the SAR polarizations included as input channels in the dataset. Only two channel input samples with HH and HV polarizations and three channel input samples with HH, HV, and the HV/HH ratio are used in this study.
- Total concentration (Ct)—indicates the minimum Ct required from the egg code for a sample to be included in the dataset. Only samples with concentrations ≥ 9+ were considered in this study.
- Pre-processing—determines the pre-processing steps to use on the SAR samples. No pre-processing was considered in this study.
3.2. Experimental Results—Summary and Interpretation
3.3. Experimental Conclusions
4. Discussion
4.1. The Utility of DenseNet for Sea Ice Mapping
4.2. Future Work
5. Conclusions
Author Contributions
Funding
Acknowledgments
Conflicts of Interest
Abbreviations
CIS | Canadian Ice Service |
SAR | Synthetic Aperture Radar |
ASIS | Automated Sea Ice Segmentation |
ERS | European Remote Sensing |
AMSR-E | Advanced Microwave Scanning Radiometer-Earth Observing System |
MODIS | Moderate Resolution Imaging Spectroradiometer |
SSM/I | Special Sensor Microwave Imager |
ASI | Artist Sea Ice |
PCNN | Pulse-Coupled Neural Network |
WMO | World Meteorological Organization |
R2 | RADARSAT-2 |
SCW | ScanSAR Wide |
CSA | Canadian Space Agency |
PIF | Product Information File |
CNN | Convolutional Neural Network |
HH | Horizontal transmit Horizontal receive |
HV | Horizontal transmit Vertical receive |
References
- Komarov, A.S.; Caya, A.; Buehner, M.; Pogson, L. Assimilation of SAR Ice and Open Water Retrievals in Environment and Climate Change Canada Regional Ice-Ocean Prediction System. IEEE Trans. Geosci. Remote Sens. 2020, 58, 4290–4303. [Google Scholar] [CrossRef]
- Carter, N.A.; Dawson, J.; Joyce, J.; Ogilvie, A. Arctic Corridors and Northern Voices: Governing Marine Transportation in the Canadian Arctic (Arviat, Nunavut Community Report). Available online: https://ruor.uottawa.ca/handle/10393/36924 (accessed on 31 July 2020).
- Barber, D.; Babb, D.; Ehn, J.; Chan, W.; Matthes, L.; Dalman, L.; Campbell, Y.; Harasyn, M.; Firoozy, N.; Theriault, N.; et al. Increasing mobility of high Arctic sea ice increases marine hazards off the east coast of Newfoundland. Geophys. Res. Lett. 2018, 45, 2370–2379. [Google Scholar] [CrossRef]
- Ulaby, F.; Long, D. Microwave Radar and Radiometric Remote Sensing; Artech House: Norwood, MA, USA, 2015. [Google Scholar]
- Barber, D.G. Microwave remote sensing, sea ice and Arctic climate. Phys. Can. 2005, 61, 105–111. [Google Scholar]
- World Meteorological Organization. WMO Sea-Ice Nomenclature: Terminology, Codes, Illustrated Glossary and Symbols; Secretariat of the World Meteorological Organization: Geneva, Switzerland, 1970. [Google Scholar]
- Howell, S.E.; Small, D.; Rohner, C.; Mahmud, M.S.; Yackel, J.J.; Brady, M. Estimating melt onset over Arctic sea ice from time series multi-sensor Sentinel-1 and RADARSAT-2 backscatter. Remote Sens. Environ. 2019, 229, 48–59. [Google Scholar] [CrossRef]
- Yackel, J.J.; Gill, J.P.; Geldsetzer, T.; Fuller, M.C.; Nandan, V. Diurnal scale controls on C-band microwave backscatter from snow-covered first-year sea ice during the transition from late winter to early melt. IEEE Trans. Geosci. Remote Sens. 2017, 55, 3860–3874. [Google Scholar] [CrossRef]
- Isleifson, D.; Hwang, B.; Barber, D.G.; Scharien, R.K.; Shafai, L. C-Band Polarimetric Backscattering Signatures of Newly Formed Sea Ice During Fall Freeze-Up. IEEE Trans. Geosci. Remote Sens. 2010, 48, 3256–3267. [Google Scholar] [CrossRef]
- Soh, L.; Tsatsoulis, C. Unsupervised segmentation of ERS and radarsat sea ice images using multiresolution peak detection and aggregated population equalization. Int. J. Remote Sens. 1999, 20, 3087–3109. [Google Scholar] [CrossRef]
- Scheuchl, B.; Cumming, I.; Hajnsek, I. Classification of fully polarimetric single- and dual-frequency SAR data of sea ice using the Wishart statistics. Can. J. Remote Sens. 2005, 31, 61–72. [Google Scholar] [CrossRef]
- Kim, M.; Im, J.; Han, H.; Kim, J.; Lee, S.; Shin, M.; Kim, H.C. Landfast sea ice monitoring using multisensor fusion in the Antarctic. GISci. Remote Sens. 2015, 52, 239–256. [Google Scholar] [CrossRef]
- Huang, G.; Liu, Z.; Maaten, L.V.D.; Weinberger, K.Q. Densely Connected Convolutional Networks. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017. [Google Scholar] [CrossRef] [Green Version]
- Cooke, C.L.V.; Scott, K.A. Estimating Sea Ice Concentration From SAR: Training Convolutional Neural Networks With Passive Microwave Data. IEEE Trans. Geosci. Remote Sens. 2019, 57, 4735–4747. [Google Scholar] [CrossRef]
- Spreen, G.; Kaleschke, L.; Heygster, G. Sea ice remote sensing using AMSR-E 89-GHz channels. J. Geophys. Res. Ocean. 2008, 113, C02S03. [Google Scholar] [CrossRef] [Green Version]
- Wang, L.; Scott, K.A.; Xu, L.; Clausi, D.A. Sea Ice Concentration Estimation During Melt From Dual-Pol SAR Scenes Using Deep Convolutional Neural Networks: A Case Study. IEEE Trans. Geosci. Remote Sens. 2016, 54, 4524–4533. [Google Scholar] [CrossRef]
- Karvonen, J.A. Baltic Sea ice SAR segmentation and classification using modified pulse-coupled neural networks. IEEE Trans. Geosci. Remote Sens. 2004, 42, 1566–1574. [Google Scholar] [CrossRef]
- Bogdanov, A.V.; Sandven, S.; Johannessen, O.M.; Alexandrov, V.Y.; Bobylev, L.P. Multisensor approach to automated classification of sea ice image data. IEEE Trans. Geosci. Remote Sens. 2005, 43, 1648–1664. [Google Scholar] [CrossRef] [Green Version]
- Andrews, J.; Babb, D.; Barber, D.G. Climate change and sea ice: Shipping accessibility on the marine transportation corridor through Hudson Bay and Hudson Strait (1980–2014). Elem. Sci. Anthr. 2017, 5, 15. [Google Scholar] [CrossRef] [Green Version]
- Martini, I. Coastal features of Canadian inland seas. In Elsevier Oceanography Series; Elsevier: Amsterdam, The Netherlands, 1986; Volume 44, pp. 117–142. [Google Scholar]
- Hochheim, K.P.; Barber, D.G. An update on the ice climatology of the Hudson Bay system. Arctic Antarct. Alp. Res. 2014, 46, 66–83. [Google Scholar] [CrossRef] [Green Version]
- Harasyn, M.L.; Isleifson, D.; Chan, W.; Barber, D.G. Multi-scale observations of the co-evolution of sea ice thermophysical properties and microwave brightness temperatures during the summer melt period in Hudson Bay. Elem. Sci. Anthr. 2020, 8, 16. [Google Scholar] [CrossRef] [Green Version]
- Parkinson, C.L.; Cavalieri, D.J.; Gloersen, P.; Zwally, H.J.; Comiso, J.C. Arctic sea ice extents, areas, and trends, 1978–1996. J. Geophys. Res. Ocean. 1999, 104, 20837–20856. [Google Scholar] [CrossRef]
- Slade, B. RADARSAT-2 Product Description. RN-SP-52-1238 2018. Available online: https://mdacorporation.com/docs/default-source/technical-documents/geospatial-services/52-1238_rs2_product_description.pdf (accessed on 31 July 2020).
- MANICE, CIS. Manual of Standard Procedures for Observing and Reporting Ice Conditions, 9th ed.; Canadian Ice Service (CIS), Meteorological Service of Canada: Ottawa, ON, Canada, 2005. [Google Scholar]
- MacDonald, D. Ltd. Radarsat-2 Product Format Definition; RN-RP-51-2713; MDA Corporation: Richmond, BC, Canada, 2008; Available online: https://mdacorporation.com/docs/default-source/technical-documents/geospatial-services/radarsat-2-product-format-definition.pdf?sfvrsn=4 (accessed on 31 July 2020).
- MacDonald, D. Ltd. Geolocation of RADARSAT-2 Georeferenced Products; Technical Report, Tech. Rep. RN-TN-53-0076; MDA Corporation: Richmond, BC, Canada, 2010; Available online: https://mdacorporation.com/geospatial/international/satellites/RADARSAT-2/docs/default-source/technical-documents/geospatial-services/rn-tn-53-0076-geolocation-tn-1-3.pdf?sfvrsn=4 (accessed on 31 July 2020).
- Ronneberger, O.; Fischer, P.; Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, 5–9 October 2015; Springer: Cham, Switzerland, 2015; pp. 234–241. [Google Scholar]
- Siam, M.; Gamal, M.; Abdel-Razek, M.; Yogamani, S.; Jagersand, M. RTSeg: Real-time Semantic Segmentation Comparative Study. arXiv 2018, arXiv:cs.CV/1803.02758. [Google Scholar]
- Singh, A.; Kalke, H.; Loewen, M.; Ray, N. River Ice Segmentation with Deep Learning. arXiv 2019, arXiv:cs.CV/1901.04412. [Google Scholar] [CrossRef] [Green Version]
- Wang, S.; Liu, W.; Wu, J.; Cao, L.; Meng, Q.; Kennedy, P.J. Training deep neural networks on imbalanced datasets. In Proceedings of the 2016 International Joint Conference on Neural Networks (IJCNN), Vancouver, BC, Canada, 24–29 July 2016; pp. 4368–4374. [Google Scholar]
- Pleiss, G.; Chen, D.; Huang, G.; Li, T.; van der Maaten, L.; Weinberger, K.Q. Memory-Efficient Implementation of DenseNets. arXiv 2017, arXiv:cs.CV/1707.06990. [Google Scholar]
- Masters, D.; Luschi, C. Revisiting Small Batch Training for Deep Neural Networks. arXiv 2018, arXiv:cs.LG/1804.07612. [Google Scholar]
Category | Label |
---|---|
Ice-free | 0 |
New ice corresponds to the stage of development codes 1–5 | 1 |
First-year ice corresponds to the stage of development codes 6–4• | 2 |
1 | 2 | 3 | 4 | 5 | 6 | |
---|---|---|---|---|---|---|
Data Batch | DEX Type | Label Type | Polarizations | Total Concentration | Pre-Processing | |
A | - | - | - | - | - | Raw |
B | - | DEXI | - | - | - | - |
C | - | - | - | - | Ct≥ 9+ | - |
D | Batch 4 | - | Label Type 4 | HH, HV | - | - |
E | - | - | Label Type 5 | - | - | - |
F | - | - | - | - | - | - |
G | - | - | - | HH, HV, HV/HH | - | - |
Training | Testing | |||||||
---|---|---|---|---|---|---|---|---|
Config. | # Classes | # Channels | Loss | Acc. % | Ice Acc. % | Loss | Acc. % | Ice Acc. % |
DBEDCA 1 | 2 | 2 | 0.0190 | 99.37 | 99.32 | 0.0337 | 99.05 | 98.99 |
DBEGCA 1 | 2 | 3 | 0.0128 | 99.59 | 99.45 | 0.0356 | 99.01 | 98.90 |
DBDDCA | 3 | 2 | 0.1378 | 94.76 | 92.62 | 0.1742 | 92.99 | 90.51 |
DBDGCA | 3 | 3 | 0.0741 | 97.49 | 96.51 | 0.2371 | 91.75 | 88.48 |
Training | Testing | |||||||
---|---|---|---|---|---|---|---|---|
Config. | # Classes | # Channels | Loss | Acc. % | Ice Acc. % | Loss | Acc. % | Ice Acc. % |
DBEDCA 1 | 2 | 2 | 0.0244 | 99.37 | 99.13 | 0.0329 | 99.07 | 98.65 |
DBEGCA 1 | 2 | 3 | 0.0225 | 99.42 | 99.17 | 0.0309 | 99.10 | 98.65 |
DBDDCA | 3 | 2 | 0.1414 | 95.95 | 94.23 | 0.1872 | 94.02 | 91.75 |
DBDGCA | 3 | 3 | 0.1129 | 97.21 | 96.02 | 0.2008 | 93.12 | 90.34 |
Before | After | |||||
---|---|---|---|---|---|---|
Label Type | Class 0 | Class 1 | Class 2 | Class 0 | Class 1 | Class 2 |
2 Class | 186,673 | 81,928 | - | 81,928 | 81,928 | - |
3 Class | 186,673 | 14,539 | 58,726 | 14,539 | 14,539 | 14,539 |
Training | Testing | |||||||
---|---|---|---|---|---|---|---|---|
Month | Class 0 | Class 1 | Class 2 | % of Total | Class 0 | Class 1 | Class 2 | % of Total |
Jun. | 0 | 0 | 11,600 | 33.24 | 0 | 0 | 1435 | 32.89 |
Oct. | 5017 | 489 | 0 | 15.78 | 644 | 64 | 0 | 16.23 |
Nov. | 6606 | 3946 | 0 | 30.24 | 811 | 474 | 0 | 29.45 |
Dec. | 0 | 7177 | 58 | 20.73 | 0 | 932 | 3 | 21.43 |
Total | 11,623 | 11,612 | 11,658 | 100.00 | 1455 | 1470 | 1438 | 100.00 |
Training | Testing | |||||||
---|---|---|---|---|---|---|---|---|
Month | Class 0 | Class 1 | Class 2 | Acc. % | Class 0 | Class 1 | Class 2 | Acc. % |
Jun. | - | - | 95.54 | 95.54 | - | - | 92.96 | 92.96 |
Oct. | 98.70 | 87.12 | - | 97.68 | 97.20 | 79.69 | - | 95.62 |
Nov. | 99.89 | 93.84 | - | 97.63 | 99.63 | 90.72 | - | 96.34 |
Dec. | - | 93.44 | 17.24 | 92.83 | - | 91.52 | 0.00 | 91.23 |
Total | 99.38 | 93.31 | 95.15 | 97.95 | 98.56 | 90.75 | 92.77 | 94.02 |
True Condition | |||||||
---|---|---|---|---|---|---|---|
Training | Testing | ||||||
Class 0 | Class 1 | Class 2 | Class 0 | Class 1 | Class 2 | ||
Prediction Condition | Class 0 | 11551 | 30 | 76 | 1434 | 5 | 17 |
Class 1 | 8 | 10835 | 489 | 2 | 1334 | 87 | |
Class 2 | 64 | 747 | 11,093 | 19 | 131 | 1334 |
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Kruk, R.; Fuller, M.C.; Komarov, A.S.; Isleifson, D.; Jeffrey, I. Proof of Concept for Sea Ice Stage of Development Classification Using Deep Learning. Remote Sens. 2020, 12, 2486. https://doi.org/10.3390/rs12152486
Kruk R, Fuller MC, Komarov AS, Isleifson D, Jeffrey I. Proof of Concept for Sea Ice Stage of Development Classification Using Deep Learning. Remote Sensing. 2020; 12(15):2486. https://doi.org/10.3390/rs12152486
Chicago/Turabian StyleKruk, Ryan, M. Christopher Fuller, Alexander S. Komarov, Dustin Isleifson, and Ian Jeffrey. 2020. "Proof of Concept for Sea Ice Stage of Development Classification Using Deep Learning" Remote Sensing 12, no. 15: 2486. https://doi.org/10.3390/rs12152486