地球科学进展 ›› 2026, Vol. 41 ›› Issue (4): 441 -454. doi: 10.11867/j.issn.1001-8166.2026.031   cstr: 32269.14.adearth.CN62-1091/P.2026.031

研究论文 上一篇    下一篇

基于YOLOv8的海底磁异常条带自动识别方法研究
李璐1(), 黄彦铭1(), 张锦昌2,3   
  1. 1.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
    2.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
    3.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,巴基斯坦 伊斯兰堡 45320
  • 收稿日期:2026-02-02 修回日期:2026-04-01 出版日期:2026-04-10
  • 通讯作者: 黄彦铭 E-mail:2257236393@qq.com;ymhuang@yangtzeu.edu.cn
  • 基金资助:
    国家自然科学基金青年科学基金项目(42006056);国家自然科学基金面上项目(42376071)

Research on Automatic Marine Magnetic Stripe Recognition Based on YOLOv8

Lu Li1(), Yanming Huang1(), Jinchang Zhang2,3   

  1. 1.Key Laboratory of Exploration Technologies for Oil and Gas Resources, Ministry of Education, Yangtze University, Wuhan 430100, China
    2.State Key Laboratory of Tropical Oceanography, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China
    3.China -Pakistan Joint Research Center on Earth Sciences, Chinese Academy of Sciences–Higher Education Commission of Pakistan, Islamabad 45320, Pakistan
  • Received:2026-02-02 Revised:2026-04-01 Online:2026-04-10 Published:2026-06-09
  • Contact: Yanming Huang E-mail:2257236393@qq.com;ymhuang@yangtzeu.edu.cn
  • About author:Li Lu, research areas include marine geology and geophysics. E-mail: 2257236393@qq.com
  • Supported by:
    the National Natural Science Foundation of China(42006056)

海底磁异常条带是地磁倒转和海底扩张耦合的产物,其形态和分布特征为重塑大洋中各构造单元的形成和演化提供了宝贵依据,具有重要的地球动力学意义。然而,传统磁条带解释主要依赖专家目视判读与人工处理,在大范围数据处理中普遍存在耗时耗力、主观性强、解释一致性不足等问题。为提升处理效率与结果客观性,将YOLOv8深度学习框架引入海底磁条带自动识别,提出一种从磁异常网格数据输入到分类结果输出的端到端的自动识别方法。研究选取北大西洋两个子区及西太平洋沙茨基海隆周边区域,采用全球磁异常网格数据(EMAG2v3)与高分辨率区域磁异常网格数据,构建了从原始区域磁异常网格到深度学习输入样本的标准化处理流程,通过滑动窗口切片与统一色标渲染生成标准化磁异常图像样本,并基于经纬度映射实现与全球磁条带数据库(GSFML)的精确对齐,建立磁条带/非磁条带(stripe/nonstripe)二分类数据集。基于YOLOv8轻量化分类模型(YOLOv8n-cls)开展监督训练,在北大西洋和西太平洋区域的验证集上分别取得97.58%和96.84%的准确率,测试集F1分数达0.98和0.97,能够稳定区分磁条带和非磁条带纹理,展现出良好的识别精度与跨区域可移植性。此外,该方法可输出类别概率,并将其作为识别结果的置信度表征,支持分级筛查与人工复核,有效提升大范围磁异常数据的处理效率,减少人工判读带来的主观误差,为海底扩张研究和构造演化重建提供了新的技术路径,并具有向其他研究区扩展应用的潜力。

Marine magnetic anomaly stripes result from the coupling of geomagnetic reversals and seafloor spreading. Their morphology and spatial distribution provide valuable evidence for reconstructing the formation and evolution of tectonic units in the oceans and therefore carry important geodynamic significance. However, conventional interpretation of magnetic stripes mainly relies on expert visual inspection and profile comparison, and in processing large-area datasets, it commonly suffers from high labor and time costs, strong subjectivity, and limited interpretive consistency. To improve processing efficiency and result objectivity, this study introduces the YOLOv8 deep-learning framework for automatic marine magnetic stripe recognition and proposes an end-to-end method that spans from magnetic anomaly grid data to classification output. The study selects two subregions in the North Atlantic and the area surrounding Shatsky Rise in the western Pacific and uses global magnetic anomaly grid data (EMAG2v3) together with high-resolution regional magnetic anomaly grid data to establish a standardized workflow that transforms raw regional magnetic anomaly grids into deep-learning input samples. Standardized magnetic anomaly image samples are generated through sliding-window slicing and uniform colormap rendering, and precise alignment with the Global Seafloor Fabric and Magnetic Lineation Data Base Project (GSFML) is achieved through geographic coordinate mapping, thereby constructing a binary classification dataset of stripe and nonstripe samples. This workflow ensures standardized sample generation and preserves the spatial correspondence between anomaly slices and geological labels. Supervised training is then conducted using the lightweight YOLOv8 classification model (YOLOv8n-cls), and the optimal weights are selected. The resulting model achieves validation accuracies of 97.58% and 96.84% in the North Atlantic and western Pacific regions, respectively, while the test-set F1-scores reach 0.98 and 0.97, demonstrating stable discrimination between stripe and nonstripe textures, high recognition accuracy, and good cross-regional transferability. In addition, the method outputs class probabilities, which are used as confidence measures for the recognition results, thereby supporting hierarchical screening and manual review. This effectively improves the processing efficiency of large-scale magnetic anomaly data, reduces subjective errors introduced by manual intervention, provides a new technical pathway for studies of seafloor spreading and tectonic evolution reconstruction, and shows potential for extension to other study areas.

中图分类号: 

图1 全球海底磁条带识别数据18
黑线表示板块边界,红框(NA01,NA02)为北大西洋2个子区位置,黄框(WP)为西太平洋1个子区位置。
Fig. 1 Global marine magnetic anomaly identification data18
The black lines denote plate boundaries, the red boxes (NA01,NA02) indicate the locations of the two North Atlantic subregions, and the yellow box (WP) indicates the location of the Western Pacific study area.
图2 研究区域磁异常图及对应GSFML数据库磁条带拾取点分布图
(a) NA01区域磁异常图;(b) NA01区域磁条带拾取点分布图; (c) NA02区域磁异常图;(d) NA02区域磁条带拾取点分布图; (e) WP区域磁异常图;(f) WP区域磁条带拾取点分布图。
Fig. 2 Magnetic anomaly maps of the study areas and corresponding GSFML magnetic lineation pick distributions
(a) Magnetic anomaly map of the NA01 region; (b) Distribution of magnetic lineation picks in the NA01 region; (c) Magnetic anomaly map of the NA02 region; (d) Distribution of magnetic lineation picks in the NA02 region; (e) Magnetic anomaly map of the WP region; (f) Distribution of magnetic lineation picks in the WP region.
图3 YOLOv8n-cls 网络结构
Fig. 3 YOLOv8n-cls network architecture
图4 滑动窗口切片示意图
Fig. 4 Schematic diagram of sliding-window slicing
图5 网格化数据渲染及切分示例(模型输入图)
(a) 数据渲染及切分stripe示例;(b) 数据渲染及切分nonstripe示例。
Fig. 5 Example of gridded-data rendering and slicingmodel input images
(a) Example of data rendering and slicing for stripe; (b) Example of data rendering and slicing for nonstripe.
表1 磁异常数据集构建及划分比例
Table 1 Magnetic anomaly dataset construction and split ratios
图6 YOLOv8n-cls训练与验证损失收敛曲线
(a) 北大西洋区域训练与验证损失曲线; (b) 西太平洋区域训练与验证损失曲线。
Fig. 6 Training and validation loss convergence of YOLOv8n-cls
(a) Training and validation loss curves for the North Atlantic region; (b) Training and validation loss curves for the West Pacific region.
图7 YOLOv8n-cls训练过程中准确率变化曲线
(a) 北大西洋区域准确率曲线; (b) 西太平洋区域准确率曲线。
Fig. 7 Accuracy curve during training of YOLOv8n-cls
(a) Accuracy curve for the North Atlantic region; (b) Accuracy curve for the West Pacific region.
图8 测试集预测结果的图像切片示例
Fig. 8 Example Image Slices of Test Set
表2 磁条带/非磁条带预测混淆矩阵
Table 2 Confusion matrix for tripe/nonstripe prediction
图9 归一化混淆矩阵
(a) 北大西洋区域归一化混淆矩阵; (b) 西太平洋区域归一化混淆矩阵。
Fig. 9 Normalized confusion matrix
(a) Normalized confusion matrix for the North Atlantic region; (b) Normalized confusion matrix for the West Pacific region.
[1] Gee J, Kent D. Source of oceanic magnetic anomalies and the geomagnetic polarity timescale[M]// Treatise on geophysics, volume 5: geomagnetism. Amsterdam: Elsevier, 2007: 455-507.
[2] Vine F J. Spreading of the ocean floor: new evidence[J]. Science1966154(3 755): 1 405-1 415.
[3] MacDonald K C. Mid-ocean ridges: fine scale tectonic, volcanic and hydrothermal processes within the plate boundary zone[J]. Annual Review of Earth and Planetary Sciences198210: 155-190.
[4] Li Yuanjie, Wei Dongping. Review of research on oceanic striped magnetic anomalies[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.
[5] 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.
[6] Zhang Jinchang, Luo Yiming, Li Haiyong, et al. Structure and formation of oceanic plateaus in west Pacific Ocean[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.
[7] Dramsch J S. 70 years of machine learning in geoscience in review[M]// Machine learning in geosciences. Amsterdam: Elsevier, 2020: 1-55.
[8] Schettino A. Magan: a new approach to the analysis and interpretation of marine magnetic anomalies[J]. Computers & Geosciences201239: 135-144.
[9] Dyer L A M .Identifying marine magnetic anomalies using machine learning[D]. Kent, OH: Kent State University, 2022
[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): e2024JB030682.
[11] Liu Dingzhou, Liu Hongchen, Zhang Jinchang, et al. Automated identification of marine magnetic anomaly stripes using U-Net convolutional neural networks: a case study of Shatsky Rise[J]. Advances in Earth Science202540(8): 864-876.
刘定洲, 刘虹辰, 张锦昌, 等. 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[J]. 地球科学进展202540(8): 864-876.
[12] Xu Degang, Wang Lu, Li Fan. Review of typical object detection algorithms for deep learning[J]. Computer Engineering and Applications202157(8): 10-25.
许德刚, 王露, 李凡. 深度学习的典型目标检测算法研究综述[J]. 计算机工程与应用202157(8): 10-25.
[13] Liu Ying, Liu Hongyan, Fan Jiulun, et al. A survey of research and application of small object detection based on deep learning[J]. Acta Electronica Sinica202048(3): 590-601.
刘颖, 刘红燕, 范九伦, 等. 基于深度学习的小目标检测研究与应用综述[J]. 电子学报202048(3): 590-601.
[14] Al-Zihairy A K, Abdelkareem A E. Optimizing YOLOv8-cls: a step towards smarter edge environments[C]// 2024 1st International Conference on Emerging Technologies for Dependable Internet of Things (ICETI). Sana’a, Yemen: IEEE, 2024: 1-6.
[15] Saha U, Ahamed I U, Imran M A, et al. YOLOv8-based deep learning approach for real-time skin lesion classification using the HAM10000 dataset[C]// 2024 IEEE International Conference on E-health Networking, Application & Services (HealthCom). Nara, Japan: IEEE, 2024: 1-4.
[16] 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.
[17] Huang Y M, Sager W 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 Earth2021126(2): e2019JB019116.
[18] 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.
[19] Wessel P, Luis J F, Uieda L, et al. The generic mapping tools version 6[J]. Geochemistry, Geophysics, Geosystems201920(11): 5 556-5 564.
[20] Jiang P Y, Ergu D J, Liu F Y, et al. A review of yolo algorithm developments[J]. Procedia Computer Science2022199: 1 066-1 073.
[21] Terven J, Córdova-Esparza D M, Romero-González J A. A comprehensive review of YOLO architectures in computer vision: from YOLOv1 to YOLOv8 and YOLO-NAS[J]. Machine Learning and Knowledge Extraction20235(4): 1 680-1 716.
[22] Hussain M. YOLO-v1 to YOLO-v8, the rise of YOLO and its complementary nature toward digital manufacturing and industrial defect detection[J]. Machines202311(7): 677.
[23] Zhou Feiyan, Jin Linpeng, Dong Jun. Review of convolutional neural network[J]. Chinese Journal of Computers201740(6): 1 229-1 251.
周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报201740(6): 1 229-1 251.
[24] Liu N N, Liu S Q, Feng K, et al. A classification method for winter wheat growth stages based on an improved version 8 of the you only look once[J]. Engineering Applications of Artificial Intelligence2026163: 113091.
[25] Pidchayathanakorn P, Prayoonchan P. YOLOv8-based grape leaf disease classification with feature enhancement[C]// 2025 9th International Conference on Information Technology (InCIT). Phuket, Thailand: IEEE, 2025: 729-734.
[26] Deng X H, Zhou Z W, Yang S. An enhanced lightweight YOLOv8n-cls model for rapid and accurate detection of mpox skin lesions[J]. Digital Medicine202511(4): e25-00009.
[27] Wu P Y, Lin Y J, Chang Y J, et al. Deep learning-assisted diagnostic system: apices and odontogenic sinus floor level analysis in dental panoramic radiographs[J]. Bioengineering202512(2): 134.
[28] Xu Zhaohui, Liu Yuming, Zhou Xinmao, et al. An experiment in automatic stratigraphic correlation using convolutional neural networks[J]. Petroleum Science Bulletin20194(1): 1-10.
徐朝晖, 刘钰铭, 周新茂, 等. 基于卷积神经网络算法的自动地层对比实验[J]. 石油科学通报20194(1): 1-10.
[1] 潘林林, 靳双龙, 宋宗朋, 丁煌, 胡睿, 肖子牛, 杜杰, 杨静, 包庆, 王勃. 风能和太阳能短中期气象预报技术及其最新进展[J]. 地球科学进展, 2026, 41(1): 73-86.
[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] 邓乃尔, 徐浩, 周文, 唐小川, 陈雨露, 刘永旸, 刘绍军, 张益, 蒋柯, 刘瑞崟, 宋威国. 基于深度学习的页岩黄铁矿扫描电镜图像分割及环境指示意义:以四川盆地泸州Ι区为例[J]. 地球科学进展, 2024, 39(5): 476-488.
[4] 黄春林, 侯金亮, 李维德, 顾娟, 张莹, 韩伟孝, 王维真, 温小虎, 朱高峰. 深度学习融合遥感大数据的陆地水文数据同化:进展与关键科学问题[J]. 地球科学进展, 2023, 38(5): 441-452.
阅读次数
全文


摘要