Sea Ice Concentration Estimation during Freeze-Up from SAR Imagery Using a Convolutional Neural Network

Wang, Lei; Scott, K. Andrea; Clausi, David A.

doi:10.3390/rs9050408

Cited by 70 publications

(38 citation statements)

References 43 publications

(69 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep learning through application of convolutional neural networks (CNN) to SAR images has been successfully used for ice-water classification and to estimate sea ice concentration [18,19]. These studies, however, did not attempt to distinguish different ice types on SAR images and applied CNNs to classify an entire sub-image (45 × 45 pixel in [18]) or pixel-by-pixel classification of relatively large sub-image (250 × 250 pixels, in [19]).…”

Section: Introductionmentioning

confidence: 99%

Classification of Sea Ice Types in Sentinel-1 SAR Data Using Convolutional Neural Networks

Boulze¹,

Korosov²,

Brajard³

2020

Remote Sensing

View full text Add to dashboard Cite

A new algorithm for classification of sea ice types on Sentinel-1 Synthetic Aperture Radar (SAR) data using a convolutional neural network (CNN) is presented. The CNN is trained on reference ice charts produced by human experts and compared with an existing machine learning algorithm based on texture features and random forest classifier. The CNN is trained on two datasets in 2018 and 2020 for retrieval of four classes: ice free, young ice, first-year ice and old ice. The accuracy of our classification is 90.5% for the 2018-dataset and 91.6% for the 2020-dataset. The uncertainty is a bit higher for young ice (85%/76% accuracy in 2018/2020) and first-year ice (86%/84% accuracy in 2018/2020). Our algorithm outperforms the existing random forest product for each ice type. It has also proved to be more efficient in computing time and less sensitive to the noise in SAR data. The code is publicly available.

show abstract

Section: Introductionmentioning

confidence: 99%

Classification of Sea Ice Types in Sentinel-1 SAR Data Using Convolutional Neural Networks

Boulze¹,

Korosov²,

Brajard³

2020

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Operational sea ice monitoring and classification usually relies on SAR data from single-or dual-polarized beam modes, such as the ScanSAR mode of RADARSAT-2 [6][7][8][9][10][11][12][13][14][15][16][17]. However, imagery from such modes provides partial information about the radar target, which could affect the accuracy of sea ice classification.…”

Section: Introductionmentioning

confidence: 99%

Assessment of the High Resolution SAR Mode of the RADARSAT Constellation Mission for First Year Ice and Multiyear Ice Characterization

2018

View full text Add to dashboard Cite

Simulated compact polarimetry from the RADARSAT Constellation Mission (RCM) is evaluated for sea ice classification. Compared to previous studies that evaluated the potential of RCM for sea ice classification, this study focuses on the High Resolution (HR) Synthetic Aperture Radar (SAR) mode of the RCM associated with a higher noise floor (Noise Equivalent Sigma Zero of −19 dB), which can prove challenging for sea ice monitoring. Twenty three Compact Polarimetric (CP) parameters were derived and analyzed for the discrimination between first year ice (FYI) and multiyear ice (MYI). The results of the RCM HR mode are compared with those previously obtained for other RCM SAR modes for possible CP consistency parameters in sea ice classification under different noise floors, spatial resolutions, and radar incidence angles. Finally, effective CP parameters were identified and used for the classification of FYI and MYI using the Random Forest (RF) classification algorithm. This study indicates that, despite the expected high noise floor of the RCM HR mode, CP SAR data from this mode are promising for the classification of FYI and MYI in dry ice winter conditions. The overall classification accuracies of CP SAR data over two test sites (96.13% and 96.84%) were found to be comparable to the accuracies obtained using Full Polarimetric (FP) SAR data (98.99% and 99.20%).

show abstract

“…In general, there will be multiple filters and multiple feature maps are generated and stacked to form a multilayer feature map. 27,28 Each feature map is obtained by dot product between receptive field and kernel in specific rules of scanning the input image. The weights in the kernel will be learned by training.…”

Section: Cnnmentioning

confidence: 99%

Global chlorophyll-a concentration estimation from moderate resolution imaging spectroradiometer using convolutional neural networks

Peng

et al. 2020

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

The accurate estimation of global chlorophyll-a (Chla) concentration from the large remote sensing data in a timely manner is crucial for supporting various applications. Moderate resolution imaging spectroradiometer (MODIS) is one of the most widely used earth observation data sources, which has the characteristics of global coverage, high spectral resolution, and short revisit period. So the estimation of global Chla concentration from MODIS imagery in a fast and accurate manner is significant. Nevertheless, the estimation of Chla concentration from MODIS using traditional machine learning approaches is challenging due to their limited modeling capability to capture the complex relationship between MODIS spatial-spectral observations and the Chla concentration, and also their low computational efficiency to address large MODIS data in a timely manner. We, therefore, explore the potential of deep convolutional neural networks (CNNs) for Chla concentration estimation from MODIS imagery. The Ocean Color Climate Change Initiative (OC-CCI) Chla concentration image is used as ground truth because it is a well-recognized Chla concentration product that is produced by assimilating different satellite data through a complex data processing steps. A total of 12 monthly OC-CCI global Chla concentration maps and the associated MODIS images are used to investigate the CNN approach using a cross-validation approach. The classical machine learning approach, i.e., the supported vector regression (SVR), is used to compare with the proposed CNN approach. Comparing with the SVR, the CNN performs better with the mean log root-mean-square error and R 2 of being 0.129 and 0.901, respectively, indicating that using the MODIS images alone, the CNN approach can achieve results that is close to the OC-CCI Chla concentration images. These results demonstrate that CNNs may provide Chla concentration images that are reliable, stable and timely, and as such CNN constitutes a useful technique for operational Chla concentration estimation from large MODIS data. © The Authors. Published by SPIE under a Creative Commons Attribution 4.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

show abstract

Sea Ice Concentration Estimation during Freeze-Up from SAR Imagery Using a Convolutional Neural Network

Cited by 70 publications

References 43 publications

Classification of Sea Ice Types in Sentinel-1 SAR Data Using Convolutional Neural Networks

Classification of Sea Ice Types in Sentinel-1 SAR Data Using Convolutional Neural Networks

Assessment of the High Resolution SAR Mode of the RADARSAT Constellation Mission for First Year Ice and Multiyear Ice Characterization

Global chlorophyll-a concentration estimation from moderate resolution imaging spectroradiometer using convolutional neural networks

Contact Info

Product

Resources

About