基于改进DeepLabV3+模型的海冰提取方法

孙士昌; 王志勇; 李振今; 张保敬; 田康; 赵相禹

doi:10.12284/hyxb2024075

基于改进DeepLabV3+模型的海冰提取方法以北极格陵兰海为例

doi: 10.12284/hyxb2024075

孙士昌^1,,
王志勇^{1, 2, ,},
李振今¹,
张保敬¹,
田康¹,
赵相禹¹

1.
山东科技大学测绘与空间信息学院，山东青岛 266590
2.
测绘工程国家级实验教学示范中心（山东科技大学），山东青岛 266590

基金项目: 国家自然科学基金（41876202）。

详细信息

作者简介:
孙士昌（1999—），男，山东省济宁市人，研究方向为遥感数据处理与应用。E-mail：sunshichang2021@163.com

通讯作者:
王志勇（1978—），男，山东省青岛市人，教授，从事雷达干涉测量、海洋遥感等方面的研究。E-mail：wzywlp@163.com

中图分类号: P731.15
计量
- 文章访问数: 178
- HTML全文浏览量: 60
- PDF下载量: 30
- 被引次数: 0
出版历程
- 收稿日期: 2023-12-12
- 修回日期: 2024-06-05
- 网络出版日期: 2024-08-09
- 刊出日期: 2024-09-26

An extraction method for sea ice based on improved DeepLabV3+ model：Taking the Arctic Greenland Sea as an example

1.
College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao 266590, China
2.
National Demonstration Center for Experimental Surveying and Mapping Education (Shandong University of Science and Technology), Qingdao 266590, China

摘要

摘要: 海冰是全球气候变化的指示剂，北极海冰的变化关系到全球变暖、海平面上升等。针对传统语义分割模型对海冰进行提取时存在细节提取不精确、提取速度慢等问题，构建了一种改进DeepLabV3+的海冰提取方法。首先，将主干网络Xception替换为MobileNetV2，在保证海冰提取精度的同时大幅度降低模型参数量，节约时间；其次，将ASPP改进为DenseASPP，在进行海冰的多尺度特征提取时进一步扩大感受野，获得更为密集的特征；最后，引入坐标注意力机制，同时强化关注通道和空间上的特征，加强海冰边缘细节信息提取。选取北极格陵兰海为实验区，通过对该海域2020–2022年间冬季的10景Sentinel-1A双极化SAR影像进行处理、标注之后形成数据集进行实验，对比U-Net、PSPNet和DeepLabV3+等经典模型。结果表明：本文方法的mIoU达到了88.46%，mPA达到了94.16%。相较于传统DeepLabV3+，mIoU提高了2.35%，mPA提高了2.90%，参数量和GFLOPs分别减少了45.08 M和106.01 G，同时训练模型时间和提取海冰时间分别减少了68%和30%。对比U-Net、PSPNet等模型，同样取得了最优结果。与其他模型相比，本文新构建的模型对海冰特征的学习能力更强，能获取更多海冰细节信息，并大幅度节约用时，能够为研究全球变暖环境下的海冰退化监测问题提供技术支持。
- 海冰提取 /
- 深度学习 /
- MobileNetV2 /
- DenseASPP /
- 坐标注意力
Abstract: Sea ice is an indicator of global climate change, and the change of Arctic sea ice is related to global warming and sea level rise. Aiming at the problems such as inaccuracy and slow speed of extracting details from sea ice by traditional semantic segmentation model, an improved DeepLabV3+ sea ice extraction method was constructed. Firstly, we replaced the Xception backbone network with MobileNetV2, which significantly reduces the network’s parameter count and save time while maintaining the accuracy of sea ice extraction. Secondly, we enhanced the ASPP module to DenseASPP, further expanding the receptive field during multi-scale feature extraction for sea ice, resulting in denser features. Lastly, we introduced a coordinate attention mechanism to strengthen the focus on both channel and spatial features, enhancing the extraction of fine edge details in sea ice. The Greenland Sea in the Arctic is selected as the experimental area, and 10 Sentinel-1A dual-polarization SAR images from the winter of 2020 to 2022 in the sea area are processed and labeled to form a data set for the experiment, we compared our method with classic models such as U-Net, PSPNet and DeepLabV3+. The results showed that our method achieved anmIoU of 88.46% and an mPA of 94.16%. Compared to the traditional DeepLabV3+, mIoU increased by 2.35%, mPA increased by 2.90%, and the parameter count and GFLOPs decreased 45.08 M and 106.01 G, respectively. Meanwhile, the training time and sea ice extraction time decreased by 68% and 30%, respectively. Compared to U-Net、PSPNet and other models, the optimal results are also obtained. Compared with other models, the new model constructed in this paper has a stronger learning ability about sea ice characteristics, can obtain more detailed information of sea ice and greatly saves time, and can provide technical support for the study of sea ice degradation monitoring under global warming environment.
- sea ice extraction /
- deep learning /
- MobileNetV2 /
- DenseASPP /
- coordinate attention

HTML全文

图 1 海冰提取技术路线

Fig. 1 Sea ice extraction technology roadmap

下载: 全尺寸图片幻灯片

图 2 坐标注意力机制结构

Fig. 2 Structure of coordinate attention mechanism

下载: 全尺寸图片幻灯片

图 3 本文模型结构

Fig. 3 Model structure of this paper

下载: 全尺寸图片幻灯片

图 4 海冰提取结果对比

Fig. 4 Comparison of sea ice extraction results

下载: 全尺寸图片幻灯片

图 5 不同方法的海冰提取结果

Fig. 5 Results of sea ice extraction by different methods

下载: 全尺寸图片幻灯片

图 6 训练过程Loss曲线

Fig. 6 Training process Loss curve

下载: 全尺寸图片幻灯片

图 7 相关指标与下采样因子关系曲线

Fig. 7 Relation curve of correlation index and downsampling factor

下载: 全尺寸图片幻灯片

图 8 相关指标与学习率关系曲线

Fig. 8 Relation curve of correlation index and learning rate

下载: 全尺寸图片幻灯片

表 1 实验平台主要参数信息

Tab. 1 Main parameters of the experimental platform

配置	参数
处理器	i5-12490F CPU @ 3.00 GHz
RAM	16.0 G
GPU	NVIDIA GeForce RTX 3060
显存	12.0 GB
操作系统	Windows 10
编程语言	Python 3.9
深度学习框架	Pytorch 1.13

下载: 导出CSV

表 2 具体实验数据

Tab. 2 Specific experimental data

序号	传感器	成像日期	模式	极化方式	用途
1	Sentinel-1B	2021.12.01	EW	HH、HV	训练
2	Sentinel-1B	2021.12.08	EW	HH、HV	训练
3	Sentinel-1B	2021.12.09	EW	HH、HV	训练
4	Sentinel-1A	2021.12.21	EW	HH、HV	训练
5	Sentinel-1A	2021.12.22	EW	HH、HV	训练
6	Sentinel-1A	2021.12.30	EW	HH、HV	训练
7	Sentinel-1A	2021.12.30	EW	HH、HV	训练
8	Sentinel-1A	2020.02.28	EW	HH、HV	验证
9	Sentinel-1A	2021.12.01	EW	HH、HV	验证
10	Sentinel-1A	2022.12.01	EW	HH、HV	验证

下载: 导出CSV

表 3 对比实验精度评价表

Tab. 3 Comparison of experimental accuracy evaluation table

	mIoU	mPA	mP	Parameters	GFLOPs	训练用时	提取用时
PSPNet	85.43	91.42	93.73	46.71 M	118.43 G	13h 48min	57 s
U-Net	85.69	90.97	94.02	24.89 M	451.71 G	13h 43min	58 s
本文模型 (DeepLabV3+_mbV2_DenseASPP_CA)	88.46	94.16	94.82	9.63 M	60.84 G	6h 40min	53 s

下载: 导出CSV

表 4 消融实验精度评价表

Tab. 4 Ablation experiment accuracy evaluation table

	mIoU	mPA	mP	Parameters	GFLOPs	训练用时	提取用时
传统DeepLabV3+	86.11%	91.26%	94.21%	54.71 M	166.85 G	20h 47min	76 s
DeepLabV3+_mbV2	86.58%	92.61%	92.93%	5.81 M	52.88 G	5h 53min	48 s
DeepLabV3+_DenseASPP	86.82%	92.98%	94.32%	58.51 M	175.76 G	23h 5min	80 s
DeepLabV3+_CA	87.39%	93.60%	94.47%	54.72 M	166.85 G	21h 54min	77 s
本文模型(DeepLabV3+_mbV2_DenseASPP_CA)	88.46%	94.16%	94.82%	9.63 M	60.84 G	6h 40min	53 s

下载: 导出CSV

参考文献(24)

[1]	Serreze M C, Holland M M, Stroeve J. Perspectives on the Arctic’s shrinking sea-ice cover[J]. Science, 2007, 315(5818): 1533−1536. doi: 10.1126/science.1139426
[2]	Trusel L D, Das S B, Osman M B, et al. Nonlinear rise in Greenland runoff in response to post-industrial arctic warming[J]. Nature, 2018, 564(7734): 104−108. doi: 10.1038/s41586-018-0752-4
[3]	葛梦滢, 高稳, 祝敏, 等. 基于SE-ConvLSTM的时空特征融合SAR图像海冰分类[J]. 遥感技术与应用, 2023, 38(6): 1306−1316. Ge Mengying, Gao Wen, Zhu Min, et al. Sea ice classification of SAR images based on SE-ConvLSTM spatial-temporal feature fusion[J]. Remote Sensing Technology and Application, 2023, 38(6): 1306−1316.
[4]	李小娜, 张杰, 戴永寿, 等. 灰度共生矩阵纹理特征对SAR海冰漂移监测的增强性能研究[J]. 海洋科学, 2018, 42(4): 9−17. Li Xiaona, Zhang Jie, Dai Yongshou, et al. Research on the enhanced performance of texture feature for sea ice drift monitoring based on gray level co-occurrence matrices[J]. Marine Sciences, 2018, 42(4): 9−17.
[5]	Wang Bin, Xia Linghui, Song Dongmei, et al. A two-round weight voting strategy-based ensemble learning method for sea ice classification of sentinel-1 imagery[J]. Remote Sensing, 2021, 13(19): 3945. doi: 10.3390/rs13193945
[6]	周颖, 匡定波, 巩彩兰, 等. 风云三号卫星MERSI影像提取北极海冰参数的方法[J]. 红外与毫米波学报, 2017, 36(1): 41−48,126−127. Zhou Ying, Kuang Dingbo, Gong Cailan, et al. A method to extract parameters of Arctic Sea ice from FY-3/MERSI imagery[J]. Journal of Infrared and Millimeter Waves, 2017, 36(1): 41−48,126−127.
[7]	臧金霞, 刘建强, 殷晓斌, 等. 基于最优特征集的HY-1C卫星海岸带成像仪影像海冰分类方法研究[J]. 海洋学报, 2022, 44(5): 35−46. Zang Jinxia, Liu Jianqiang, Yin Xiaobin, et al. Study on sea ice classification of HY-1C satellite coastal zone imager images based on the optimal feature set[J]. Haiyang Xuebao, 2022, 44(5): 35−46.
[8]	朱立先, 惠凤鸣, 张智伦, 等. 基于Sentinel-1A/B SAR数据的西北航道海冰分类研究[J]. 北京师范大学学报(自然科学版), 2019, 55(1): 66−76. Zhu Lixian, Hui Fengming, Zhang Zhilun, et al. Sea ice classification in northwest passage based on Sentinel-1A/B SAR data[J]. Journal of Beijing Normal University (Natural Science), 2019, 55(1): 66−76.
[9]	王志勇, 孙培蕾, 刘健. 一种联合多特征的极化SAR海冰类型提取方法[J]. 遥感信息, 2020, 35(4): 23−29. Wang Zhiyong, Sun Peilei, Liu Jian. A sea ice classification method of polarimetric SAR data by multi-feature combination[J]. Remote Sensing Information, 2020, 35(4): 23−29.
[10]	王志勇, 张梦悦, 于亚冉, 等. 一种融合纹理特征与NDVI的随机森林海冰精细分类方法[J]. 海洋学报, 2021, 43(10): 149−156. Wang Zhiyong, Zhang Mengyue, Yu Yaran, et al. A fine classification method for sea ice based on random forest combining texture feature and NDVI[J]. Haiyang Xuebao, 2021, 43(10): 149−156.
[11]	吴斌, 王志勇, 李兴, 等. CryoSat-2雷达高度计海冰波形优选特征分类[J]. 测绘通报, 2023(5): 164−169. Wu Bin, Wang Zhiyong, Li Xing, et al. CryoSat-2 radar altimeter sea ice waveform preferred feature classification[J]. Bulletin of Surveying and Mapping, 2023(5): 164−169.
[12]	Zakhvatkina N, Korosov A, Muckenhuber S, et al. Operational algorithm for ice–water classification on dual-polarized RADARSAT-2 images[J]. The Cryosphere, 2017, 11(1): 33−46. doi: 10.5194/tc-11-33-2017
[13]	Li Xiaoming, Sun Yan, Zhang Qiang. Extraction of sea ice cover by sentinel-1 SAR based on support vector machine with unsupervised generation of training data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(4): 3040−3053. doi: 10.1109/TGRS.2020.3007789
[14]	Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431−3440.
[15]	崔艳荣, 邹斌, 韩震, 等. 卷积神经网络在卫星遥感海冰图像分类中的应用探究——以渤海海冰为例[J]. 海洋学报, 2020, 42(9): 100−109. Cui Yanrong, Zou Bin, Han Zhen, et al. Application of convolutional neural networks in satellite remote sensing sea ice image classification: a case study of sea ice in the Bohai Sea[J]. Haiyang Xuebao, 2020, 42(9): 100−109.
[16]	郑付强, 匡定波, 胡勇, 等. 基于U-ASPP-Net的北极独立海冰精细识别方法[J]. 红外与毫米波学报, 2021, 40(6): 798−808. Zheng Fuqiang, Kuang Dingbo, Hu Yong, et al. Refined segmentation method based on U-ASPP-Net for Arctic independent sea ice[J]. Journal of Infrared and Millimeter Waves, 2021, 40(6): 798−808.
[17]	黄冬梅, 李明慧, 宋巍, 等. 卷积神经网络和深度置信网络在SAR影像冰水分类的性能评估[J]. 中国图象图形学报, 2018, 23(11): 1720−1732. doi: 10.11834/jig.180226 Huang Dongmei, Li Minghui, Song Wei, et al. Performance of convolutional neural network and deep belief network in sea ice-water classification using SAR imagery[J]. Journal of Image and Graphics, 2018, 23(11): 1720−1732. doi: 10.11834/jig.180226
[18]	徐欢, 任沂斌. 基于混合损失U-Net的SAR图像渤海海冰检测研究[J]. 海洋学报, 2021, 43(6): 157−170. Xu Huan, Ren Yibin. Detecting sea ice of Bohai Sea using SAR images based on a hybrid loss U-Net model[J]. Haiyang Xuebao, 2021, 43(6): 157−170.
[19]	Sandler M, Howard A, Zhu Menglong, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE/CVFConference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 4510−4520.
[20]	Huang Gao, Liu Zhuang, Van Der MaatenL, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 2261−2269.
[21]	Yang Maoke, Yu Kun, Zhang Chi, et al. DenseASPP for semantic segmentation in street scenes[C]//Proceedings of the IEEE/CVFConference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 3684−3692.
[22]	Hu Jie, Shen Li, Sun Gang. Squeeze-and-excitation networks[C]//Proceedings of the IEEE/CVFConference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132−7141.
[23]	Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the 15th European Conference on Computer Vision. Munich: Springer, 2018: 3−19.
[24]	Hou Qibin, Zhou Daquan, Feng Jiashi. Coordinate attention for efficient mobile network design[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021: 13708−13717.