地球科学进展 ›› 2025, Vol. 40 ›› Issue (8): 864 -876. doi: 10.11867/j.issn.1001-8166.2025.063

研究论文 上一篇    

基于U-Net卷积神经网络的海底磁异常条带自动识别方法研究——以沙茨基海隆为例
刘定洲1,2(), 刘虹辰3, 张锦昌1,4(), 张江阳1, 周鹏程1, 郭鑫3, 李璐3, 黄彦铭3   
  1. 1. 热带海洋环境与岛礁生态全国重点实验室,中国科学院南海海洋研究所,广东 广州 510301
    2. 中国科学院大学,北京 100049
    3. 油气资源与勘探技术教育部重点实验室,长江大学,湖北 武汉 430100
    4. 中国—巴基斯坦地球科学研究中心,中国科学院—巴基斯坦 高等教育委员会,伊斯兰堡 45320,巴基斯坦
  • 收稿日期:2025-07-05 修回日期:2025-08-01 出版日期:2025-08-10
  • 通讯作者: 张锦昌
  • 基金资助:
    国家自然科学基金面上项目(42376071); 中国科学院项目(Y4SL021); 中国科学院南海海洋研究所自主部署项目(SCSIO2024QY02)

Automated Identification of Marine Magnetic Anomaly Stripes Using U-Net Convolutional Neural Networks: A Case Study of Shatsky Rise

Dingzhou LIU1,2(), Hongchen LIU3, Jinchang ZHANG1,4(), Jiangyang ZHANG1, Pengcheng ZHOU1, Xin GUO3, Lu LI3, Yanming HUANG3   

  1. 1. State Key Laboratory of Tropical Oceanography, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. Key Laboratory of Exploration Technologies for Oil and Gas Resources, Ministry of Education, Yangtze University, Wuhan 430100, China
    4. China-Pakistan Joint Research Center on Earth Sciences, Chinese Academy of Sciences-Pakistan Higher Education Council, Islamabad 45320, Pakistan
  • Received:2025-07-05 Revised:2025-08-01 Online:2025-08-10 Published:2025-07-10
  • Contact: Jinchang ZHANG
  • Supported by:
    the National Natural Science Foundation of China(42376071); Chinese Academy of Sciences Project(Y4SL021); Development Program of the South China Sea Institute of Oceanology, Chinese Academy of Sciences(SCSIO2024QY02)

海底磁异常条带(磁条带)的发现是地球科学领域的重要突破,为海底扩张学说和板块运动理论提供了决定性证据,磁条带研究对揭示大洋岩石圈形成演化机制以及地球内部动力学过程等具有重要的科学意义。传统研究主要依赖人工识别磁条带并进行地质解释,且面临诸多挑战:磁条带结构复杂、数据量大、识别过程耗时费力;结果易受识别者经验和状态等主观因素影响,难以满足现代地球科学研究对效率和精度的需求。近年来学者们尝试引入大数据和人工智能等先进技术,探索磁条带自动识别的方法来突破这些技术瓶颈。以西太平洋典型的海底构造——沙茨基海隆周围的磁异常数据为基础,采用U-Net卷积神经网络实现磁条带的机器自动识别。具体研究路径为:整合沙茨基海隆区域的海洋磁测数据和磁条带标签数据,构建了高质量的训练数据集;基于U-Net卷积神经网络模型进行数据训练,获得机器预测的磁条带分布;并与人工经验识别的结果进行对比分析,验证了模型的可靠性和准确性。研究取得以下重要成果:以沙茨基海隆为示范区,建立了一套基于U-Net卷积神经网络的磁条带自动识别方法,此方法能够有效提高磁条带的识别效率和精度,减少人工干预带来的主观误差,为磁条带的定量化研究提供了新的技术手段。这不仅为沙茨基海隆区域的磁条带识别提供了科学依据,还为其他类似区域的应用提供了技术支撑和参考范例,对推动磁条带解译的智能化转型具有重要的示范意义。

The discovery of marine magnetic anomaly stripes (magnetic stripes) represents a pivotal breakthrough in Earth sciences, providing decisive evidence for the theories of seafloor spreading and plate tectonics. The study of magnetic stripes is of significant scientific importance for understanding the formation and evolutionary mechanisms of oceanic lithosphere and dynamic processes within the Earth. However, traditional geological research primarily relies on manual identification of magnetic stripes, and faces several challenges: the complex structure and large volume of magnetic stripe data, the time-consuming and labor-intensive identification process, and the susceptibility of results to subjective factors such as the interpreter's experience, making it difficult to meet the efficiency and accuracy demands of modern Earth science research. To address these challenges, researchers have introduced advanced technologies, such as artificial intelligence and big data, to explore methods for the automatic identification of magnetic stripes. In this study, we targeted magnetic anomalies around the Shatsky Rise, a typical submarine feature in the western Pacific Ocean, and employed U-Net convolutional neural networks to achieve machine-automated identification of magnetic stripes. The approach first integrated marine magnetic survey data and magnetic stripe label data from the Shatsky Rise region to construct a high-quality training dataset. Subsequently, the U-Net convolutional neural network model was trained on this dataset to obtain machine-predicted magnetic stripe distributions, which were then compared and analyzed with manually identified results to validate the reliability and accuracy of the model. The study's primary findings are as follows: The study establishes a method for the automatic identification of marine magnetic anomaly stripes based on the U-Net convolutional neural network, using the Shatsky Rise as a demonstration area. It demonstrates that the U-Net-based method for automatic identification of magnetic stripes significantly improves identification efficiency and accuracy, and reduces subjective errors introduced by manual intervention. This method provides a new technical tool for the quantitative study of marine magnetic anomaly stripes. This research not only offers scientific insights for magnetic stripe identification in the Shatsky Rise region, but also provides technical support and reference models for applications in other similar areas. Thus, the findings are of significance for promoting the intelligent transformation of magnetic stripe interpretation.

中图分类号: 

图1 全球海底磁条带识别数据1-2 黑线表示板块边界,红框为文中研究区位置。
Fig. 1 Global marine magnetic anomaly stripe identification data1-2 The black lines denote plate boundaries, the red box indicates the location of the study area discussed in the text.
图2 沙茨基海隆区域24-27 (a)位置图:红线是磁条带,黄色虚线是断裂带,绿色框为沙茨海隆上最大的高地大塔穆火山的分布区域;(b)地磁异常图:黑色框指示图4切割的局部区域所在位置(29°4′~33°20′N,150°40′~154°56′E)。
Fig. 2 Shatsky Rise24-27 (a) Location map: the red lines represent magnetic stripes, the yellow dashed lines denote fracture zones, the green box marks the distribution area of the Tamu Massif, the largest edifice within Shatsky Rise; (b) Magnetic anomaly map: the black box indicates the location of Figure 4 (29°4′~33°20′N,150°40′~154°56′E).
图3 磁条带完全已知的数据集分布图 (a)子区域一模型log_11;(b)子区域二模型log_12;(c)子区域三模型log_13;(d)子区域四模型log_17;(e)子区域五模型log_25。蓝框为训练集区域,5个模型的训练集均为沙茨基海隆整体区域,红框为预测集区域,随着子区域的变化,预测集区域在逐渐向下平移,黑框为分割窗口大小,选取不同子区域黑框所在行的模型预测结果整合成最终磁条带预测图,横纵坐标网格单位长度为经纬度的1'。
Fig. 3 Distribution map of the dataset with fully known magnetic stripes (a) Subregion 1 model log_11; (b) Subregion 2 model log_12; (c) Subregion 3 model log_13; (d) Subregion 4 model log_17; (e) Subregion 5 model log_25. The blue boxes indicate the training set areas—the training set for all five models covers the entire Shatsky Rise region. The red boxes denote the prediction set areas, which shift gradually downward across the subregions. The black boxes represent the segmentation window size. Predictive results from models corresponding to the row in which the black box is located in each subregion are integrated to generate the final magnetic stripe prediction map. The grid units along the horizontal and vertical axes represent 1 arc-minute of latitude and longitude.
图4 沙茨基海隆局部区域 (a)地磁异常网格图;(b)人工识别磁条带标签;(c)模型log_25预测结果。图(a)具体位置为29°4′~33°20′N,150°40′~154°56′E,与图2(b)中的黑色方框所在区域相对应,横纵坐标的单位长度为经纬度的1'。
Fig. 4 Local area of Shatsky Rise (a) Gridded magnetic anomaly map; (b) Manually identified magnetic stripe labels; (c) Prediction results from model log_25. The area shown in (a) corresponds to 29°4′~33°20′N,150°40′~154°56′E, marked by the black box in Figure 2(b). The grid units along both axes represent 1 arc-minute in latitude and longitude.
图5 沙茨基海隆磁条带完全已知对比图 (a)地磁异常网格图;(b)人工识别磁条带标签;(c) 模型预测结果;(d)不同模型预测准确率。
Fig. 5 Comparative chart of the fully known magnetic stripes in Shatsky Rise (a) Gridded magnetic anomaly map; (b) Manually interpreted magnetic stripe labels; (c) Model prediction results; (d) Prediction accuracy of different models.
图6 磁条带部分已知的数据集分布图 (a)子区域一模型log_30;(b)子区域二模型log_31;(c)子区域三模型log_32;(d)子区域四模型log_33; 蓝框为训练集区域,红框为预测集区域,黑框为分割窗口大小,选取不同子区域黑框所在行的模型预测结果整合成最终磁条带预测图,横纵坐标网格单位长度为经纬度的1'。
Fig. 6 Distribution map of the dataset with partially known magnetic stripes (a)Subregion 1 model log_30; (b)Subregion 2 model log_31; (c) Subregion 3 model log_32; (d) Subregion 4 model log_33. The blue boxes indicate the training set areas, the red boxes represent the prediction set areas, and the black boxes denote the segmentation window size. The model prediction results from the row where the black box is located in different subregions are integrated to form the final magnetic stripe prediction map. The grid units along the horizontal and vertical axes represent 1 arc-minute of latitude and longitude.
图7 沙茨基海隆磁条带部分已知对比图 (a)地磁异常网格图;(b)人工识别磁条带标签;(c) 模型预测结果;(d)不同模型预测准确率。
Fig. 7 Comparative chart of the partially known magnetic stripes in Shatsky Rise (a) Gridded magnetic anomaly map; (b) Manually interpreted magnetic stripe labels; (c) Model prediction results; (d) Prediction accuracy of different models.
图8 磁条带完全未知的数据集分布图 (a)子区域一模型log_21;(b)子区域二模型log_22;(c)子区域三模型log_23;(d)子区域四模型log_24; 蓝框为训练集区域,红框为预测集区域,黑框为分割窗口大小,选取不同子区域黑框所在行的模型预测结果整合成最终磁条带预测图,横纵坐标网格单位长度为经纬度的1'。
Fig. 8 Distribution map of the dataset with completely unknown magnetic stripes (a) Subregion 1 model log_21; (b) Subregion 2 model log_22; (c) Subregion 3 model log_23; (d) Subregion 4 model log_24. The blue boxes indicate the training set areas, the red boxes represent the prediction set areas, and the black boxes denote the segmentation window size. The model prediction results from the row where the black box is located in different subregions are integrated to form the final magnetic stripe prediction map. The grid units along the horizontal and vertical axes represent 1 arc-minute of latitude and longitude.
图9 沙茨基海隆磁条带完全未知对比图 (a)地磁异常网格图;(b)人工识别磁条带标签;(c)模型预测结果;(d)不同模型预测准确率。
Fig. 9 Comparative chart of the completely unknown magnetic stripes in Shatsky Rise (a) Gridded magnetic anomaly map; (b) Manually interpreted magnetic stripe labels; (c) Model prediction results; (d) Prediction accuracy of different models.
[1]
SETON M WHITTAKER J M WESSEL P, et al. Community infrastructure and repository for marine magnetic identifications[J]. Geochemistry, Geophysics, Geosystems201415(4): 1 629-1 641.
[2]
SETON M MÜLLER R D ZAHIROVIC S, et al. A global data set of present-day oceanic crustal age and seafloor spreading parameters[J]. Geochemistry, Geophysics, Geosystems202021(10). DOI:10.1029/2020GC009214 .
[3]
VINE F J. Spreading of the ocean floor: new evidence[J]. Science1966154(3 755): 1 405-1 415.
[4]
RAFF A D MASON R G. Magnetic survey off the west coast of North America, 40°N. latitude to 52°N. latitude[J]. Geological Society of America Bulletin196172(8). DOI:10.1130/0016-7606(1961)72 [1267:MSOTWC]2.0.CO;2.
[5]
MASON R G RAFF A D. Magnetic survey off the west coast of North America, 32°N. latitude to 42°N. latitude[J]. Geological Society of America Bulletin196172(8). DOI:10.1130/0016-7606(1961)72 [1259:MSOTWC]2.0.CO;2.
[6]
VINE F J MATTHEWS D H. Magnetic anomalies over oceanic ridges[J]. Nature1963199(4 897): 947-949.
[7]
LI Yuanjie WEI Dongping. A comprehensive review of marine magnetic anomaly stripes[J]. Progress in Geophysics201631(3): 949-959.
李园洁,魏东平.海底磁异常条带研究综述[J].地球物理学进展201631(3): 949-959.
[8]
BUKRY D BRAMLETTE M N. Coccolith age determinations leg 3, deep sea drilling project[R] Initial reports of the deep sea drilling project, 3. U.S. Government Printing Office, 1970.
[9]
HONSHO C, URA T, KIM K, et al. Postcaldera volcanism and hydrothermal activity revealed by autonomous underwater vehicle surveys in Myojin Knoll caldera, Izu-Ogasawara arc[J]. Journal of Geophysical Research: Solid Earth2016121(6): 4 085-4 102.
[10]
WU S THORAM S SUN J, et al. Characterizing marine magnetic anomalies: a machine learning approach to advancing the understanding of oceanic crust formation[J]. Journal of Geophysical Research: Solid Earth2025130(2). DOI: 10.1029/2024JB030682 .
[11]
HARSHVARDHAN G GOURISARIA M K PANDEY M, et al. A comprehensive survey and analysis of generative models in machine learning[J]. Computer Science Review2020, 38. DOI:10.1016/j.cosrev.2020.100285 .
[12]
LECUN Y BENGIO Y HINTON G. Deep learning[J]. Nature2015521(7 553): 436-444.
[13]
MENDEL V MUNSCHY M SAUTER D. MODMAG, a MATLAB program to model marine magnetic anomalies[J]. Computers & Geosciences200531(5): 589-597.
[14]
SCHETTINO A. Magan: a new approach to the analysis and interpretation of marine magnetic anomalies[J]. Computers & Geosciences201239: 135-144.
[15]
IVANOV S A MERKUR’EV S A. Interpretation of marine magnetic anomalies. part I. a survey of existing methods and analysis of the analytic signal method[J]. Geomagnetism and Aeronomy201454(3): 388-396.
[16]
IVANOV S A MERKUR’EV S A. Interpretation of marine magnetic anomalies. part II. analysis of the new method and algorithm based on the least squares method[J]. Geomagnetism and Aeronomy201454(4): 530-536.
[17]
MEYER B CHULLIAT A SALTUS R. Derivation and error analysis of the Earth magnetic anomaly grid at 2 arc Min resolution version 3 (EMAG2v3)[J]. Geochemistry, Geophysics, Geosystems201718(12): 4 522-4 537.
[18]
DYER L. Identifying marine magnetic anomalies using machine learning[D]. Kent: Kent State University, 2022.
[19]
WANG S G ZHANG X Y QIN Y Q, et al. Marine target magnetic anomaly detection based on multitask deep transfer learning[J]. IEEE Geoscience and Remote Sensing Letters2023, 20. DOI: 10.1109/LGRS.2023.3273722 .
[20]
WANG C PENG G WANG H, et al. Multi-feature fusion and intelligent iterative optimization algorithm based magnetic anomaly detecting method, involves inputting magnetic signal sample into magnetic anomaly detection model to obtain magnetic anomaly detection result [P]. 2021, CN112633147-A.
[21]
SAGER W W KIM J KLAUS A, et al. Bathymetry of shatsky rise, northwest Pacific Ocean: implications for ocean plateau development at a triple junction[J]. Journal of Geophysical Research: Solid Earth1999104(B4): 7 557-7 576.
[22]
ZHANG J C SAGER W W KORENAGA J. The seismic Moho structure of Shatsky Rise oceanic plateau, northwest Pacific Ocean[J]. Earth and Planetary Science Letters2016441: 143-154.
[23]
ZHANG Jinchang LUO Yiming LI Haiyong, et al. Internal structure and evolutionary mechanisms of west Pacific oceanic plateaus[J]. Science & Technology Review202341(2): 65-79.
张锦昌,罗怡鸣,李海勇,等.西太平洋洋底高原内部结构与形成演化[J].科技导报202341(2): 65-79.
[24]
HUANG Y M SAGER W ZHANG J C, et al. Magnetic anomaly map of shatsky rise and its implications for oceanic plateau formation[J]. Journal of Geophysical Research: Solid Earth2018. DOI: 10.1029/2019JB019116 .
[25]
THORAM S SAGER W W REED W, et al. Improved high-resolution bathymetry map of Tamu massif and southern shatsky rise and its geologic implications[J]. Journal of Geophysical Research: Solid Earth2022127(11). DOI: 10.1029/2022JB024304 .
[26]
HUANG Y M SAGER W W TOMINAGA M, et al. Magnetic anomaly map of Ori Massif and its implications for oceanic plateau formation[J]. Earth and Planetary Science Letters2018501: 46-55.
[27]
SAGER W W HUANG Y M TOMINAGA M, et al. Oceanic plateau formation by seafloor spreading implied by Tamu Massif magnetic anomalies[J]. Nature Geoscience201912: 661-666.
[28]
SAGER W W ZHANG J C KORENAGA J, et al. An immense shield volcano within the Shatsky Rise oceanic plateau, northwest Pacific Ocean[J]. Nature Geoscience20136(11): 976-981.
[29]
NAKANISHI M SAGER W W KLAUS A. Magnetic lineations within shatsky rise, northwest Pacific Ocean: implications for hot spot-triple junction interaction and oceanic plateau formation[J]. Journal of Geophysical Research: Solid Earth1999104(B4): 7 539-7 556.
[30]
NAKANISHI M TAMAKI K KOBAYASHI K. Mesozoic magnetic anomaly lineations and seafloor spreading history of the northwestern Pacific[J]. Journal of Geophysical Research: Solid Earth198994(B11): 15 437-15 462.
[31]
NAKANISHI M TAMAKI K KOBAYASHI K. Magnetic anomaly lineations from Late Jurassic to Early Cretaceous in the west-central Pacific Ocean[J]. Geophysical Journal International1992109(3): 701-719.
[32]
TOMINAGA M SAGER W W TIVEY M A, et al. Deep-tow magnetic anomaly study of the Pacific Jurassic Quiet Zone and implications for the geomagnetic polarity reversal timescale and geomagnetic field behavior[J]. Journal of Geophysical Research: Solid Earth2008113(B7). DOI: 10.1029/2007JB005527 .
[33]
KOPPERS A A P STAUDIGEL H WIJBRANS J R, et al. The Magellan seamount trail: implications for Cretaceous hotspot volcanism and absolute Pacific plate motion[J]. Earth and Planetary Science Letters1998163(1/2/3/4): 53-68.
[34]
ZHAO Guochun ZHANG Guowei. Origin of continents[J]. Acta Geologica Sinica202195(1): 1-19.
赵国春, 张国伟. 大陆的起源[J]. 地质学报202195(1): 1-19.
[35]
ZHANG X B BROWN E L ZHANG J C, et al. Magmatism of Shatsky Rise controlled by plume-ridge interaction[J]. Nature Geoscience202316(11): 1 061-1 069.
[36]
MEYER B SALTUS R CHULLIAT A. EMAG2v3: Earth magnetic anomaly grid (2-arc-minute resolution) Version 3 [DS]. NOAA National Centers for Environmental Information2017.
[37]
RONNEBERGER O FISCHER P BROX T. U-Net: convolutional networks for biomedical image segmentation[M]// Medical image computing and computer-assisted intervention-MICCAI 2015. Cham: Springer International Publishing, 2015: 234-241.
[38]
WANG F JIANG M Q QIAN C, et al. Residual attention network for image classification[C]// 2017 IEEE conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA: IEEE, 2017: 6 450-6 458.
[39]
CANNY J. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence1986, PAMI-8(6): 679-698.
[40]
BARASH G CASTILLO-EFFEN M CHHAYA N, et al. Reports of the Workshops Held at the 2019 AAAI Conference on Artificial Intelligence [R]. AI Magazine, 2019, 40(3).
[41]
YANG B BENDER G LE Q V, et al. CondConv: conditionally parameterized convolutions for efficient Inference 33rd Conference on Neural Information Processing Systems [C]. Vancouver, Canada, 2019.
[42]
ESTEVES C SLOTINE J J MAKADIA A. Scaling spherical CNNs[C]. 40th International Conference on Machine Learning, 2023.
[1] 唐学远. 中国极地冰冻圈人工智能技术应用的进展与展望[J]. 地球科学进展, 2025, 40(8): 778-794.
[2] 周杰, 巨能攀, 张燕, 田华兵, 何朝阳. 泥石流视频图像跟踪检测方法研究[J]. 地球科学进展, 2025, 40(4): 388-400.
[3] 王凡, 张旭东, 任沂斌, 刘颖洁, 王浩宇, 李晓峰. 基于人工智能技术的海洋智能预报研究进展[J]. 地球科学进展, 2025, 40(2): 111-125.
[4] 张井勇. 面向碳中和的气候变化预估与风险研究新框架[J]. 地球科学进展, 2025, 40(1): 15-20.
[5] 马雷鸣. 天气预报中的人工智能技术进展[J]. 地球科学进展, 2020, 35(6): 551-560.
[6] 李春峰, 汪品先, Dieter Franke, 李家彪, Randell Stephenson, 许树坤,Peter Michael, 周祖翼,等. 南海张裂过程及其对晚中生代以来东南亚构造的启示-IODP建议书735-Full介绍[J]. 地球科学进展, 2009, 24(12): 1339-1351.
[7] 曹洁冰,周祖翼. 被动大陆边缘:从大陆张裂到海底扩张[J]. 地球科学进展, 2003, 18(5): 730-736.
[8] 肖义越; 戴桦; 陈辉. 专家系统在地球科学中的应用[J]. 地球科学进展, 1993, 8(1): 49-54.
阅读次数
全文


摘要