# Deep Convolutional Neural Networks Capabilities for Binary Classification of Polar Mesocyclones in Satellite Mosaics

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Data

## 3. Methodology

#### 3.1. Data Preprocessing

#### 3.2. Formulation of the Problem

#### 3.3. Justification of Using DCNN

#### 3.4. Proposed DCNN Architectures

- CNN #1. This model is built “from scratch” which means we have not used any pre-trained networks. CNN #1 is built in the manner proposed in [35]. We varied sizes of convolutional kernels of each convolutional layers from 3 × 3 to 5 × 5. We also varied sizes of subsampling layers’ receptive fields from 2 × 2 to 3 × 3. For each convolutional layer, we varied the number of convolutional kernels: 8, 16, 32, 64 and 100. The network convolutional core consists of three convolutional layers alternated with subsampling layers. Each pair of convolutional and subsampling layers is followed by a dropout layer. CNN #1 is one-branched, and objects are described by IR 500 × 500 km satellite snapshots only.
- CNN #2. This model is built “from scratch” with two separate branches-for IR and WV data. The convolutional core of each branch is built in the same manner as the convolutional core for CNN #1 and as proposed in [41]. We varied the same parameters of the structure here in the same ranges as for CNN #1.
- CNN #3. This model is built with TL approach. We used VGG16 pre-trained convolutional core to construct this model. None of VGG16 weights were optimized within this model, and only the weights of the FC classifier were trainable. This model is one-branched, and the objects are described by IR 500 × 500 km satellite snapshots only. CNN #3 structure is shown in Figure 3.
- CNN #4. This model is two-branched, and each branch of its convolutional core is built with TL approach, in the same manner as the convolutional core of CNN #3. Input data are IR and WV. None of VGG16 weights of this model in any of the two branches were optimized, and only the weights of the FC classifier were trainable. CNN #4 structure is shown in Figure 4.
- CNN #5 is built with both TL and FT approaches. We built the convolutional core of this model with the use of VGG16 pre-trained network. VGG16 convolutional core consists of five similar blocks of layers. For the CNN #5 we turned the last of these five blocks to be trainable. This model is one-branched, and objects are IR 500 × 500 km satellite snapshots only. CNN #5 structure is shown in Figure 3.
- CNN #6 is two-branched, and branches of its convolutional core are built in the same manner as the convolutional core of CNN #5. For the CNN #6, we turned the last of five blocks of each VGG16 convolutional cores to be trainable. Input data are IR and WV 500 × 500 km satellite snapshots of dataset samples. CNN #6 structure is shown in Figure 4.

#### 3.5. Computational Experiment Design

- Convolutional kernels count for each convolutional layer (only applies to CNN #1 and CNN #2)
- Sizes of convolutional kernels (only applies to CNN #1 and CNN #2)
- Sizes of receptive fields of subsampling layers (only applies to CNN #1 and CNN #2)

## 4. Results

## 5. Conclusions and Outlook

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## Nomenclature

BCE | binary cross-entropy |

CNN | convolutional neural network |

DA | dataset augmentation technique |

DCNN | deep convolutional neural network |

DL | deep learning |

Do | Dropout technique |

FC | fully-connected |

FCNN | fully-connected neural network |

FT | Fine Tuning |

FNR | false negative rate |

FPR | false positive rate |

IR | infrared |

MC | mesocyclone |

NH | Northern Hemisphere |

PL | polar low |

ROC | receiver operator characteristic |

AUC ROC | area under the curve of receiver operator characteristic |

SH | Southern Hemisphere |

SOMC | Shirshov Institute of Oceanology mesocyclone dataset for Southern Ocean |

TL | Transfer Learning |

TNR | true negative rate |

TPR | true positive rate |

VGG16 | the DCNN proposed by Visual Geometry Group (University of Oxford) [45] |

WV | water vapor |

## Appendix A. DCNN Best Practices and Additional Techniques

#### Appendix A.1. Transfer Learning

#### Appendix A.2. Fine Tuning

#### Appendix A.3. Preventing Overfitting

#### Appendix A.4. Preventing Overfitting with Dropout

#### Appendix A.5. Preventing Overfitting with Dataset Augmentation

#### Appendix A.6. Preventing Overfitting with Ensemble Averaging

#### Appendix A.7. Adjustment of the Probability Threshold

## Appendix B. CNN #3 and CNN #5 Best Hyper-Parameters Combinations

Layer (Block) Name | Layer (Block) Nodes Count or Output Dimensions | Connected to |
---|---|---|

Input_data_IR | 100 × 100 | - |

VGG_16_conv_core | see [45]; output: 3 × 3 × 512 | Input_data_IR |

Reshape_1 | 4608 | VGG_16_conv_core |

Dropout_1 | 4608 | Reshape_1 |

FC1 | 1024 | Dropout_1 |

Dropout_2 | 1024 | FC1 |

FC2 | 512 | Dropout_2 |

Dropout_3 | 512 | FC2 |

FC3 | 256 | Dropout_3 |

Dropout_4 | 256 | FC3 |

FC4 | 128 | Dropout_4 |

FC_output | 1 | FC3 |

## Appendix C. Detailed Performance Metrics of all DCNN Models

**Figure A1.**Confusion matrices for all models and the third-order model CNN #1–6 averaged ensemble, computed on test never-seen subset of data. For each architecture the ensemble averaging technique is applied.

**Figure A2.**Receiver operating characteristic curves computed on test never-seen subset of data for all models. For each architecture the ensemble averaging technique is applied.

## Appendix D. Detailed False Negative Rates of the Third-Order Ensemble-Averaging Model

Cyclogenesis Type | Testing Set, Objects Number | False Negatives, Objects Number | FN Relative Rate, % |
---|---|---|---|

Pressure trough | 841 | 23 | 2.7 |

Centre of mid-latitude cyclone | 147 | 2 | 1.4 |

Low-gradient pressure field | 48 | 2 | 4.2 |

Cold-air outbreak | 45 | 4 | 8.9 |

Low-pressure post-occlusion zone | 84 | 4 | 4.8 |

High pressure gradient field | 61 | 4 | 6.6 |

Orography-induced cyclogenesis | 27 | 4 | 14.8 |

Cloud Vortex Type | Testing Set, Objects Number | False Negatives, Objects Number | FN Relative Rate, % |
---|---|---|---|

Comma cloud | 1006 | 33 | 3.3 |

Spiral cloud | 59 | 5 | 8.5 |

Comma-to-spiral | 177 | 5 | 2.3 |

Merry-go-round | 11 | 0 | 0.0 |

MC Stage | Testing Set, Objects Number | False Negatives, Objects Number | FN Relative Rate, % |
---|---|---|---|

Incipient | 352 | 16 | 4.6 |

Mature | 574 | 23 | 4.0 |

Dissipating | 327 | 4 | 1.2 |

## References

- Thorpe, A.J. Polar Lows: Mesoscale Weather Systems in the Polar Regions; Rasmussen, E.A., Turner, J., Eds.; Cambridge University Press: Cambridge, UK, 2003; ISBN 978-0-511-52497-4. [Google Scholar]
- Marshall, J.; Schott, F. Open-ocean convection: Observations, theory, and models. Rev. Geophys.
**1999**, 37, 1–64. [Google Scholar] [CrossRef][Green Version] - Condron, A.; Renfrew, I.A. The impact of polar mesoscale storms on northeast Atlantic Ocean circulation. Nat. Geosci.
**2013**, 6, 34–37. [Google Scholar] [CrossRef] - Condron, A.; Bigg, G.R.; Renfrew, I.A. Modeling the impact of polar mesocyclones on ocean circulation. J. Geophys. Res. Oceans
**2008**, 113. [Google Scholar] [CrossRef][Green Version] - Verezemskaya, P.; Tilinina, N.; Gulev, S.; Renfrew, I.A.; Lazzara, M. Southern Ocean mesocyclones and polar lows from manually tracked satellite mosaics. Geophys. Res. Lett.
**2017**, 44, 7985–7993. [Google Scholar] [CrossRef] - Laffineur, T.; Claud, C.; Chaboureau, J.-P.; Noer, G. Polar lows over the Nordic Seas: Improved Representation in ERA-Interim Compared to ERA-40 and the Impact on Downscaled Simulations. Mon. Weather Rev.
**2014**, 142, 2271–2289. [Google Scholar] [CrossRef] - Michel, C.; Terpstra, A.; Spengler, T. Polar Mesoscale Cyclone Climatology for the Nordic Seas Based on ERA-Interim. J. Clim.
**2017**, 31, 2511–2532. [Google Scholar] [CrossRef] - Bromwich, D.H.; Wilson, A.B.; Bai, L.-S.; Moore, G.W.K.; Bauer, P. A comparison of the regional Arctic System Reanalysis and the global ERA-Interim Reanalysis for the Arctic. Q. J. R. Meteorol. Soc.
**2016**, 142, 644–658. [Google Scholar] [CrossRef] - Press, W.H.; Teukolsky, S.A.; Vetterling, W.T.; Flannery, B.P. Numerical Recipes 3rd Edition: The Art of Scientific Computing; Cambridge University Press: Cambridge, UK, 2007; ISBN 0-521-88068-8. [Google Scholar]
- Rojo, M.; Claud, C.; Mallet, P.-E.; Noer, G.; Carleton, A.M.; Vicomte, M. Polar low tracks over the Nordic Seas: A 14-winter climatic analysis. Tellus A Dyn. Meteorol. Oceanogr.
**2015**, 67, 24660. [Google Scholar] [CrossRef] - Smirnova, J.; Golubkin, P. Comparing Polar Lows in Atmospheric Reanalyses: Arctic System Reanalysis versus ERA-Interim. Mon. Weather Rev.
**2017**, 145, 2375–2383. [Google Scholar] [CrossRef] - Hines, K.M.; Bromwich, D.H. Development and Testing of Polar Weather Research and Forecasting (WRF) Model. Part I: Greenland Ice Sheet Meteorology. Mon. Weather Rev.
**2008**, 136, 1971–1989. [Google Scholar] [CrossRef] - Zappa, G.; Shaffrey, L.; Hodges, K. Can Polar Lows be Objectively Identified and Tracked in the ECMWF Operational Analysis and the ERA-Interim Reanalysis? Mon. Weather Rev.
**2014**, 142, 2596–2608. [Google Scholar] [CrossRef] - Pezza, A.; Sadler, K.; Uotila, P.; Vihma, T.; Mesquita, M.D.S.; Reid, P. Southern Hemisphere strong polar mesoscale cyclones in high-resolution datasets. Clim. Dyn.
**2016**, 47, 1647–1660. [Google Scholar] [CrossRef] - Xia, L.; Zahn, M.; Hodges, K.; Feser, F.; Storch, H. A comparison of two identification and tracking methods for polar lows. Tellus A Dyn. Meteorol. Oceanogr.
**2012**, 64, 17196. [Google Scholar] [CrossRef] - Harold, J.M.; Bigg, G.R.; Turner, J. Mesocyclone activity over the North-East Atlantic. Part 1: Vortex distribution and variability. Int. J. Climatol.
**1999**, 19, 1187–1204. [Google Scholar] [CrossRef] - Noer, G.; Saetra, Ø.; Lien, T.; Gusdal, Y. A climatological study of polar lows in the Nordic Seas. Q. J. R. Meteorol. Soc.
**2011**, 137, 1762–1772. [Google Scholar] [CrossRef][Green Version] - Smirnova, J.E.; Zabolotskikh, E.V.; Bobylev, L.P.; Chapron, B. Statistical characteristics of polar lows over the Nordic Seas based on satellite passive microwave data. Izv. Atmos. Ocean. Phys.
**2016**, 52, 1128–1136. [Google Scholar] [CrossRef] - McMurdie, L.A.; Claud, C.; Atakturk, S. Satellite-derived atmospheric characteristics of spiral and comma-shaped southern hemisphere mesocyclones. J. Geophys. Res. Atmos.
**1997**, 102, 13889–13905. [Google Scholar] [CrossRef][Green Version] - Irving, D.; Simmonds, I.; Keay, K. Mesoscale Cyclone Activity over the Ice-Free Southern Ocean: 1999–2008. J. Clim.
**2010**, 23, 5404–5420. [Google Scholar] [CrossRef] - Neu, U.; Akperov, M.G.; Bellenbaum, N.; Benestad, R.; Blender, R.; Caballero, R.; Cocozza, A.; Dacre, H.F.; Feng, Y.; Fraedrich, K.; et al. IMILAST: A Community Effort to Intercompare Extratropical Cyclone Detection and Tracking Algorithms. Bull. Am. Meteorol. Soc.
**2013**, 94, 529–547. [Google Scholar] [CrossRef] - Wilhelmsen, K. Climatological study of gale-producing polar lows near Norway. Tellus A Dyn. Meteorol. Oceanogr.
**1985**, 37, 451–459. [Google Scholar] [CrossRef] - Carrasco, J.F.; Bromwich, D.H. Mesoscale cyclogenesis dynamics over the southwestern Ross Sea, Antarctica. J. Geophys. Res. Atmos.
**1993**, 98, 12973–12995. [Google Scholar] [CrossRef] - Carrasco, J.F.; Bromwich, D.H.; Liu, Z. Mesoscale cyclone activity over Antarctica during 1991: 1. Marie Byrd Land. J. Geophys. Res. Atmos.
**1997**, 102, 13923–13937. [Google Scholar] [CrossRef][Green Version] - Turner, J.; Thomas, J.P. Summer-season mesoscale cyclones in the bellingshausen-weddell region of the antarctic and links with the synoptic-scale environment. Int. J. Climatol.
**1994**, 14, 871–894. [Google Scholar] [CrossRef] - Harold, J.M.; Bigg, G.R.; Turner, J. Mesocyclone activity over the Northeast Atlantic. Part 2: An investigation of causal mechanisms. Int. J. Climatol.
**1999**, 19, 1283–1299. [Google Scholar] [CrossRef][Green Version] - Carleton, A.M. On the interpretation and classification of mesoscale cyclones from satellite infrared imagery. Int. J. Remote Sens.
**1995**, 16, 2457–2485. [Google Scholar] [CrossRef] - Claud, C.; Katsaros, K.B.; Mognard, N.M.; Scott, N.A. Comparative satellite study of mesoscale disturbances in polar regions. Glob. Atmos. Ocean Syst.
**1996**, 4, 233–273. [Google Scholar] - Gang, F.; Qin-yu, L.; Zeng-mao, W. General features of polar lows over the Japan Sea and the Northwestern Pacific. Chin. J. Ocean. Limnol.
**1999**, 17, 300–307. [Google Scholar] [CrossRef] - Gurvich, I.A.; Pichugin, M.K. Study of the compara tive characteristics of typical mesoscale cyclones over Far Eastern seas on the basis of satellite multisensory sounding. Sovrem. Probl. Distantsionnogo Zondirovaniya Zemli Kosmosa
**2013**, 10, 51–59. [Google Scholar] - Claud, C.; Carleton, A.M.; Duchiron, B.; Terray, P. Southern hemisphere winter cold-air mesocyclones: Climatic environments and associations with teleconnections. Clim. Dyn.
**2009**, 33, 383–408. [Google Scholar] [CrossRef] - Blechschmidt, A.-M. A 2-year climatology of polar low events over the Nordic Seas from satellite remote sensing. Geophys. Res. Lett.
**2008**, 35. [Google Scholar] [CrossRef][Green Version] - Heinle, A.; Macke, A.; Srivastav, A. Automatic cloud classification of whole sky images. Atmos. Meas. Tech.
**2010**, 3, 557–567. [Google Scholar] [CrossRef][Green Version] - Taravat, A.; Frate, F.D.; Cornaro, C.; Vergari, S. Neural Networks and Support Vector Machine Algorithms for Automatic Cloud Classification of Whole-Sky Ground-Based Images. IEEE Geosci. Remote Sens. Lett.
**2015**, 12, 666–670. [Google Scholar] [CrossRef] - Onishi, R.; Sugiyama, D. Deep Convolutional Neural Network for Cloud Coverage Estimation from Snapshot Camera Images. SOLA
**2017**, 13, 235–239. [Google Scholar] [CrossRef][Green Version] - Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA, USA, 2012; pp. 1097–1105. [Google Scholar]
- Shin, H.-C.; Roth, H.R.; Gao, M.; Lu, L.; Xu, Z.; Nogues, I.; Yao, J.; Mollura, D.; Summers, R.M. Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning. IEEE Trans. Med. Imaging
**2016**, 35, 1285–1298. [Google Scholar] [CrossRef] [PubMed] - Krinitskiy, M.A. Application of machine learning methods to the solar disk state detection by all-sky images over the ocean. Oceanology
**2017**, 57, 265–269. [Google Scholar] [CrossRef] - Liu, Y.; Racah, E.; Correa, J.; Khosrowshahi, A.; Lavers, D.; Kunkel, K.; Wehner, M.; Collins, W. Application of deep convolutional neural networks for detecting extreme weather in climate datasets. arXiv, 2016; arXiv:1605.01156. [Google Scholar]
- Huang, D.; Du, Y.; He, Q.; Song, W.; Liotta, A. DeepEddy: A simple deep architecture for mesoscale oceanic eddy detection in SAR images. In Proceedings of the 2017 IEEE 14th International Conference on Networking, Sensing and Control (ICNSC), Calabria, Italy, 16–18 May 2017; pp. 673–678. [Google Scholar]
- Lecun, Y.; Bottou, L.; Bengio, Y.; Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE
**1998**, 86, 2278–2324. [Google Scholar] [CrossRef][Green Version] - LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature
**2015**, 521, 436–444. [Google Scholar] [CrossRef] [PubMed] - Liu, W.; Wang, Z.; Liu, X.; Zeng, N.; Liu, Y.; Alsaadi, F.E. A survey of deep neural network architectures and their applications. Neurocomputing
**2017**, 234, 11–26. [Google Scholar] [CrossRef] - Guo, Y.; Liu, Y.; Oerlemans, A.; Lao, S.; Wu, S.; Lew, M.S. Deep learning for visual understanding: A review. Neurocomputing
**2016**, 187, 27–48. [Google Scholar] [CrossRef] - Simonyan, K.; Zisserman, A. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv, 2014; arXiv:1409.1556. [Google Scholar]
- Szegedy, C.; Liu, W.; Jia, Y.; Sermanet, P.; Reed, S.; Anguelov, D.; Erhan, D.; Vanhoucke, V.; Rabinovich, A. Going deeper with convolutions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 7–12 June 2015; pp. 1–9. [Google Scholar]
- Chollet, F. Xception: Deep Learning with Depthwise Separable Convolutions, CoRR abs/1610.02357. 2016. Available online: http://arxiv.org/abs/1610.02357 (accessed on 16 August 2018).
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Miami, FL, USA, 20–25 June 2009; pp. 248–255. [Google Scholar]
- Eckersley, P.; Nasser, Y. AI Progress Measurement. Available online: https://www.eff.org/ai/metrics (accessed on 13 August 2018).
- Deng, L.; Yu, D. Deep Learning: Methods and Applications. SIG
**2014**, 7, 197–387. [Google Scholar] [CrossRef][Green Version] - Deng, L. A tutorial survey of architectures, algorithms, and applications for deep learning. APSIPA Trans. Signal Inf. Process.
**2014**, 3. [Google Scholar] [CrossRef][Green Version] - Schmidhuber, J. Deep learning in neural networks: An overview. Neural Netw.
**2015**, 61, 85–117. [Google Scholar] [CrossRef] [PubMed][Green Version] - Lazzara, M.A.; Keller, L.M.; Stearns, C.R.; Thom, J.E.; Weidner, G.A. Antarctic satellite meteorology: Applications for weather forecasting. Mon. Weather Rev.
**2003**, 131, 371–383. [Google Scholar] [CrossRef] - Kohrs, R.A.; Lazzara, M.A.; Robaidek, J.O.; Santek, D.A.; Knuth, S.L. Global satellite composites—20 years of evolution. Atmos. Res.
**2014**, 135–136, 8–34. [Google Scholar] [CrossRef] - Simard, P.Y.; Steinkraus, D.; Platt, J.C. Best practices for convolutional neural networks applied to visual document analysis. In Proceedings of the Seventh International Conference on Document Analysis and Recognition, Washington, DC, USA, 3–6 August 2003; IEEE: Edinburgh, Scotland, UK, 2003; p. 958. [Google Scholar]
- Pratt, L.Y.; Mostow, J.; Kamm, C.A.; Kamm, A.A. Direct Transfer of Learned Information Among Neural Networks. In Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA, USA, 1992; Volume 5, pp. 204–211. [Google Scholar]
- Caruana, R. Learning Many Related Tasks at the Same Time with Backpropagation. In Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA, USA, 1995; Volume 7. [Google Scholar]
- Collobert, R.; Weston, J. A unified architecture for natural language processing: Deep neural networks with multitask learning. In Proceedings of the 25th International Conference on Machine Learning, Helsinki, Finland, 5–9 July 2008; ACM: New York, NY, USA, 2008; pp. 160–167. [Google Scholar]
- Pan, S.J.; Yang, Q. A Survey on Transfer Learning. IEEE Trans. Knowl. Data Eng.
**2010**, 22, 1345–1359. [Google Scholar] [CrossRef][Green Version] - Mesnil, G.; Dauphin, Y.; Glorot, X.; Rifai, S.; Bengio, Y.; Goodfellow, I.; Lavoie, E.; Muller, X.; Desjardins, G.; Warde-Farley, D. Unsupervised and transfer learning challenge: A deep learning approach. Proc. Int. Conf. Unsupervised Transf. Learn. Workshop
**2011**, 27, 97–111. [Google Scholar] - Oquab, M.; Bottou, L.; Laptev, I.; Sivic, J. Learning and Transferring Mid-Level Image Representations using Convolutional Neural Networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 23–28 June 2014; pp. 1717–1724. [Google Scholar]
- Maclin, R.; Shavlik, J.W. Combining the predictions of multiple classifiers: Using competitive learning to initialize neural networks. In Proceedings of the 1995 International Joint Conference on AI, Montreal, QC, Canada, 20–25 August 1995; Citeseer: Montreal, QC, Canada, 1995; pp. 524–531. [Google Scholar]
- Srivastava, N.; Hinton, G.; Krizhevsky, A.; Sutskever, I.; Salakhutdinov, R. Dropout: A simple way to prevent neural networks from overfitting. J. Mach. Learn. Res.
**2014**, 15, 1929–1958. [Google Scholar] - Agakov, F.V.; Barber, D. An Auxiliary Variational Method. In Neural Information Processing; Lecture Notes in Computer Science; Springer: Berlin/Heidelberg, Germany, 2004; pp. 561–566. [Google Scholar]
- Rumelhart, D.E.; Hinton, G.E.; Williams, R.J. Learning representations by back-propagating errors. Nature
**1986**, 323, 533–536. [Google Scholar] [CrossRef] - Sorokin, A.A.; Makogonov, S.V.; Korolev, S.P. The Information Infrastructure for Collective Scientific Work in the Far East of Russia. Sci. Tech. Inf. Proc.
**2017**, 44, 302–304. [Google Scholar] [CrossRef] - Carleton, A.M.; Carpenter, D.A. Satellite climatology of ‘polar lows’ and broadscale climatic associations for the Southern Hemisphere. Int. J. Climatol.
**1990**, 10, 219–246. [Google Scholar] [CrossRef] - Carleton, A.M.; Fitch, M. Synoptic aspects of Antarctic mesocyclones. J. Geophys. Res. Atmos.
**1993**, 98, 12997–13018. [Google Scholar] [CrossRef] - Ronneberger, O.; Fischer, P.; Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015; Lecture Notes in Computer Science; Springer: Cham, Switzerland, 2015; pp. 234–241. [Google Scholar]
- Badrinarayanan, V.; Kendall, A.; Cipolla, R. SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. arXiv, 2015; arXiv:1511.00561. [Google Scholar] [CrossRef] [PubMed]
- Parkhi, O.M.; Vedaldi, A.; Zisserman, A. Deep Face Recognition. In Proceedings of the British Machine Vision Conference (BMVC); BMVA Press: Durham, UK, 2015; pp. 41.1–41.12. [Google Scholar]
- Schroff, F.; Kalenichenko, D.; Philbin, J. FaceNet: A Unified Embedding for Face Recognition and Clustering. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, USA, 7–12 June 2015; pp. 815–823. [Google Scholar]
- Minsky, M.; Papert, S.A. Perceptrons; MIT Press: Cambridge, MA, USA, 1969; ISBN 978-0-262-13043-1. [Google Scholar]
- Breiman, L. Bagging predictors. Mach. Learn.
**1996**, 24, 123–140. [Google Scholar] [CrossRef][Green Version] - Lincoln, W.P.; Skrzypek, J. Synergy of clustering multiple back propagation networks. In Advances in Neural Information Processing Systems; MIT Press: Cambridge, MA, USA, 1990; pp. 650–657. [Google Scholar]

**Figure 1.**The input for the deep convolutional neural networks (DCNNs). (

**a**) Trajectories of all mesocyclones (MCs) in Southern Ocean MesoCylones (SOMC) dataset, blue dots mark the point of generation of MC. Snapshots of satellite mosaics for SH for (

**b**) InfraRed (IR) and (

**c**) Water Vapor (WV) channels at 00:00 UTC 02/06/2004. The red/blue squares indicate patches centered over the MCs (red squares) and those having no MC cloudiness signature in (blue) being cut from the mosaics for DCNNs training.

**Figure 2.**Examples (IR only) of true and false samples for DCNNs training and testing of DCNNs results assessment. 100 × 100 grid points (500 × 500 km) patches of IR mosaics for (

**a**–

**d**) true samples and false (

**e**–

**h**) samples.

**Figure 3.**Convolutional neural networks (CNN) #3 and CNN #5 structures. The dots along the “reshaped to vector” line denote elements of the convolutional core output reshaped to a vector, which is the fully-connected classifier input data.

**Figure 4.**CNN #4 and CNN #6 structures. The dots along the “reshaped to vector” and “concatenated features vector” lines denote elements of convolutional cores outputs reshaped to vectors, which are, being concatenated to a combined features vector, the fully-connected classifier input data.

**Figure 5.**Examples of false classified objects, for which (

**a**,

**b**) IR satellite data is missing or corrupted, (

**c**) the source satellite data is suspected to be corrupted, (

**d**) the source satellite data is realistic, but the classifier has made a mistake.

**Figure 6.**Confusion matrices and receiver operating characteristic curve for (

**a**,

**b**) CNN #3 and (

**c**,

**d**) CNN #5, both with the ensemble averaging approach applied (second-order models); and (

**e**,

**f**) third-order model CNN #1–6 averaged ensemble.

**Figure 7.**False negatives (FN, which are missed MCs) in the never-seen by the model testing set with respect to (

**a**) lifecycle stages; (

**b**) diameters; (

**c**) cyclogenesis types; and, (

**d**) types of cloud vortex.

True Samples | False Samples | Total Samples | |
---|---|---|---|

IR | 6177 (55%) | 5012 (45%) | 11,189 (100%) |

WV | 6177 (55%) | 5012 (45%) | 11,189 (100%) |

**Table 2.**Accuracy score of each model with the best hyper-parameters combination. BA—basic approach [41], TL—Transfer Learning, FT—Fine Tuning, Do—dropout, DA—dataset augmentation. $Acc$ is the accuracy score averaged across models of the particular architecture. AsEA is the accuracy score of the ensemble averaged models with the optimal probability threshold. ${p}_{th}$ is the optimal probability threshold value.

Model Name | IR | WV | BA | TL | FT | Do | DA | $\mathit{A}\mathit{c}\mathit{c}$ | AsEA | ${\mathit{p}}_{\mathit{t}\mathit{h}}$ |
---|---|---|---|---|---|---|---|---|---|---|

CNN #1 | X | - | X | - | - | X | X | 86.89 ± 1.1% | 89.3% | 0.381 |

CNN #2 | X | X | X | - | - | X | X | 94.1 ± 1.4% | 96.3% | 0.272 |

CNN #3 | X | - | X | X | - | X | X | 95.8 ± 0.1% | 96.6% | 0.556 |

CNN #4 | X | X | X | X | - | X | X | 95.5 ± 0.3% | 96.3% | 0.526 |

CNN #5 | X | - | X | X | X | X | X | 96 ± 0.2% | 96.6% | 0.5715 |

CNN #6 | X | X | X | X | X | X | X | 95.7 ± 0.2% | 96.4% | 0.656 |

Third-order model CNN #1–6 averaged ensemble | 97% | 0.598 |

© 2018 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

**MDPI and ACS Style**

Krinitskiy, M.; Verezemskaya, P.; Grashchenkov, K.; Tilinina, N.; Gulev, S.; Lazzara, M. Deep Convolutional Neural Networks Capabilities for Binary Classification of Polar Mesocyclones in Satellite Mosaics. *Atmosphere* **2018**, *9*, 426.
https://doi.org/10.3390/atmos9110426

**AMA Style**

Krinitskiy M, Verezemskaya P, Grashchenkov K, Tilinina N, Gulev S, Lazzara M. Deep Convolutional Neural Networks Capabilities for Binary Classification of Polar Mesocyclones in Satellite Mosaics. *Atmosphere*. 2018; 9(11):426.
https://doi.org/10.3390/atmos9110426

**Chicago/Turabian Style**

Krinitskiy, Mikhail, Polina Verezemskaya, Kirill Grashchenkov, Natalia Tilinina, Sergey Gulev, and Matthew Lazzara. 2018. "Deep Convolutional Neural Networks Capabilities for Binary Classification of Polar Mesocyclones in Satellite Mosaics" *Atmosphere* 9, no. 11: 426.
https://doi.org/10.3390/atmos9110426