An automatic marine mesoscale eddy detection model based on improved U-Net network
-
摘要: 海洋中尺度涡对浮游生物的分布、能量和盐分的输送具有非常重要的影响，海洋中尺度涡的自动检测是监测、分析中尺度涡时空变化的重要基础。针对传统基于物理特征检测海洋中尺度涡的方法存在受限于人工设计参数导致精度不高的问题，本文依据海洋卫星反演的海表面高度图，提出了一种基于改进U-Net网络的海洋中尺度涡自动检测模型。该模型在海洋中尺度涡的特征提取阶段嵌入了卷积注意力机制，使得模型能够关注于海表面高度图中最具有类别区分度的区域，同时引入了残差学习机制解决了网络过深导致模型难以训练的问题。本文以南大西洋的卫星海表面高度数据集为例开展实验验证，结果表明，本文提出的模型海洋中尺度涡检测准确率达到了93.28%，显著优于EddyNet等现有模型。模型可为海洋学家通过海表面高度探测中尺度涡提供可靠技术方法。Abstract: Marine mesoscale eddies play an important role in plankton distribution, energy and salt transport. The automatic detection of marine mesoscale eddies is a basis for monitoring and analyzing their spatiotemporal variations. Traditional physical characteristics-based methods depend on artificially designing parameters, and result in low accuracy of mesoscale eddy extraction. Therefore, a marine mesoscale eddy automatic detection model based on improved U-Net network according to the marine satellite sea surface height images is proposed in this paper. The proposed model embeds the convolution attention modules in the feature extraction stage, which enables the model focuses on the most relevent area. Meanwhile, the model introduces the residual learning module to solve the problem that the network is too deep to train the model. In this paper, the satellite sea surface height dataset in the South Atlantic Ocean is taken as an example to carry out experiments. The results show that the proposed model achieves a high accuracy of 93.28% when detecting the marine mesoscale eddies, which is significantly better than EddyNet and other models. The model can provide a reliable technology for oceanographers to detect marine mesoscale eddies through the satellite sea surface height.
-
图 7 不同方法在EddyNet-Data数据集上的分割结果
Fig. 7 Segmentation results on EddyNet-Data dataset of different methods
图 8 2011年南大西洋海域中尺度涡空间分布
黄色区域表示气旋涡；绿色区域表示反气旋涡
Fig. 8 Distribution of mesoscale eddies in the South Atlantic Ocean in 2011
Yellow areas represent cyclonic eddies； green areas represent anticyclonic eddies
图 9 2011年南大西洋海域中尺度涡个数、面积的周变化
Fig. 9 Weekly variation of mesoscale eddies number and area in the South Atlantic Ocean in 2011
表 1 基础系统平台配置
Tab. 1 Basic system platform configuration
系统 CPU 内存 硬盘 显卡 Ubuntu
16.04Intel Xeon E5-2637 64 GB 2 TB Titan XP 
下载: 导出CSV
表 2 重要软件配置
Tab. 2 Important software configuration
GPU-Driver CUDA Python Keras Tensorflow-gpu 384 9.0 3.6 2.2.4 1.4.0
下载: 导出CSV
表 3 6种算法在EddyNet-Data数据集上的分割性能对比
Tab. 3 Comparison of segmentation performance of six algorithms on EddyNet-Data dataset
模型 交叉验证结果 准确率 查准率 查全率 F1-Score EddyNet 0.881 5 0.886 4 0.881 5 0.883 9 SegNet 0.876 8 0.880 1 0.876 8 0.878 4 经典U-Net 0.889 1 0.900 4 0.889 2 0.894 8 经典U-Net+残差 0.903 0 0.914 4 0.903 0 0.908 7 经典U-Net+CBAM 0.914 4 0.924 5 0.914 4 0.919 4 本文模型 0.932 8 0.941 7 0.931 8 0.936 7
下载: 导出CSV
-
[1] Ikeda M, Mysak L A, Emery W J. Observation and modeling of satellite-sensed meanders and eddies off vancouver island[J]. Journal of Physical Oceanography, 1984, 14(1): 3−21. doi: 10.1175/1520-0485(1984)014<0003:OAMOSS>2.0.CO;2 [2] Chelton D B, Schlax M G, Samelson R M, et al. Global observations of large oceanic eddies[J]. Geophysical Research Letters, 2007, 34(15): L15606. [3] Chelton D B, Gaube P, Schlax M G, et al. The influence of nonlinear mesoscale eddies on near-surface oceanic chlorophyll[J]. Science, 2011, 334(6054): 328−332. doi: 10.1126/science.1208897 [4] 杜艳玲. 基于海洋遥感影像的中尺度涡自动识别及与渔场动态关系研究[D]. 上海: 上海海洋大学, 2017.Du Yanling. Automatic recognition of mesoscale eddy and dynamic relation with fishing ground based on ocean remote sensing images[D]. Shanghai: Shanghai Ocean University, 2017. [5] Faghmous J H, Le M, Uluyol M, et al. A parameter-free spatio-temporal pattern mining model to catalog global ocean dynamics[C]// 2013 IEEE 13th International Conference on Data Mining. Dallas, USA: IEEE, 2013. [6] Kersalé M, Doglioli A M, Petrenko A A. Sensitivity study of the generation of mesoscale eddies in a numerical model of Hawaii islands[J]. Ocean Science, 2011, 7(3): 277−291. doi: 10.5194/os-7-277-2011 [7] Chen Yushi, Lin Zhouhan, Zhao Xing, et al. Deep learning-based classification of hyperspectral data[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7(6): 2094−2107. doi: 10.1109/JSTARS.2014.2329330 [8] Lguensat R, Sun M, Fablet R, et al. EddyNet: a deep neural network for pixel-wise classification of oceanic eddies[C]//2018 IEEE International Geoscience and Remote Sensing Symposium. Valencia, Spain: IEEE, 2018. [9] Franz K, Roscher R, Milioto A, et al. Ocean eddy identification and tracking using neural networks[C]//2018 IEEE International Geoscience and Remote Sensing Symposium. Valencia, Spain: IEEE, 2018. [10] Xu Guangjun, Cheng Cheng, Yang Wenxian, et al. Oceanic eddy identification using an AI Scheme[J]. Remote Sensing, 2019, 11(11): 1349. doi: 10.3390/rs11111349 [11] 芦旭熠, 单桂华, 李观. 基于深度学习的海洋中尺度涡识别与可视化[J]. 计算机系统应用, 2020, 29(4): 65−75.Lu Xuyi, Shan Guihua, Li Guan. Oceanic mesoscale eddy detection and visualization based on deep learning[J]. Computer Systems & Applications, 2020, 29(4): 65−75. [12] Woo S, Park J, Lee J, et al. CBAM: convolutional block attention module[C]//European Conference on Computer Vision. Munich, Germany: ECCV, 2018 . [13] He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA: IEEE, 2016: 770−778. [14] 何红术, 黄晓霞, 李红旮, 等. 基于改进U-Net网络的高分遥感影像水体提取[J]. 地球信息科学学报, 2020, 22(10): 2010−2022. doi: 10.12082/dqxxkx.2020.190622He Hongshu, Huang Xiaoxia, Li Hongga, et al. Water body extraction of high resolution remote sensing image based on improved U-Net network[J]. Journal of Geo-Information Science, 2020, 22(10): 2010−2022. doi: 10.12082/dqxxkx.2020.190622 [15] Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[M]//Fleet D, Pajdla T, Schiele B, et al. Computer Vision – ECCV 2014. Cham: Springer International Publishing, 2014. [16] 李其. 基于深度特征的SAR图像舰船目标检测方法研究[D]. 成都: 电子科技大学, 2020.Li Qi. Research on ship detection in SAR images based on depth feature[D]. Chengdu: University of Electronic Science and Technology of China, 2020. [17] Mason E, Pascual A, McWilliams J C. A new sea surface height-based code for oceanic mesoscale eddy tracking[J]. Journal of Atmospheric and Oceanic Technology, 2014, 31(5): 1181−1188. doi: 10.1175/JTECH-D-14-00019.1 [18] 侯向丹, 赵一浩, 刘洪普, 等. 融合残差注意力机制的U-Net视盘分割[J]. 中国图象图形学报, 2020, 25(9): 1915−1929.Hou Xiangdan, Zhao Yihao, Liu Hongpu, et al. Optic disk segmentation by combining U-Net and residual attention mechanism[J]. Journal of Image and Graphics, 2020, 25(9): 1915−1929. [19] 杨彬. 基于深度学习的高分辨率遥感影像变化检测[D]. 徐州: 中国矿业大学, 2019.Yang Bin. Change detection of high resolution remote sensing image based on deep learning[D]. Xuzhou: China University of Mining and Technology, 2019. [20] 张盟, 杨玉婷, 孙鑫, 等. 基于深度卷积网络的海洋涡旋检测模型[J]. 南京航空航天大学学报, 2020, 52(5): 708−713.Zhang Meng, Yang Yuting, Sun Xin, et al. Ocean eddy detection model based on deep convolution neural network[J]. Journal of Nanjing University of Aeronautics & Astronautics, 2020, 52(5): 708−713. [21] Badrinarayanan V, Kendall A, Cipolla R. SegNet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481−2495. doi: 10.1109/TPAMI.2016.2644615 [22] 黎安舟. 南海中尺度特征遥感提取与时空分布研究[D]. 上海: 上海海洋大学, 2018.Li Anzhou. Extraction of mesoscale features based on remote sensing and spatio-temporal distribution of lighting fishing boat in South China Sea[D]. Shanghai: Shanghai Ocean University, 2018. -
图(9) / 表(3)
计量
- 文章访问数: 712
- HTML全文浏览量: 196
- PDF下载量: 78
- 被引次数: 0
