冰山是威胁极地船舶航行安全的重要原因之一，其漂移轨迹难以预测，在能见度不良的条件下极易与船舶碰撞导致船舶破损、沉没等严重事故。现有的冰山检测方法受环境因素影响较大，难以同时满足冰山检测高精度和实时性的需求。为实现极地能见度不良条件下的冰山检测，首先，对YOLO系列模型进行归纳总结，提出一种融合注意力机制的改进YOLO v7冰山检测算法；其次，改进k-means算法，提高了先验框与冰山的匹配度；最后，引入CA（Coordinate Attention）注意力机制，提升模型对冰山关键特征识别能力。研究结果表明，相比于YOLO v7原始模型，改进后的模型在维持计算性能的同时mAP（mean Average Precision）提升了5.41%。模型对于极地环境下的冰山具有较好的检测能力，可为极地航行中避免船冰碰撞提供参考。
Icebergs pose a significant threat to polar navigation due to unpredictable drift and collision risks in low visibility, leading to severe accidents. Existing detection methods are affected by environmental factors and struggle with accuracy and real-time requirements. This study proposes an improved YOLO v7 iceberg detection algorithm with an attention mechanism, enhances the k-means algorithm for better anchor-iceberg matching, and introduces the Coordinate Attention (CA) mechanism to improve feature recognition. Results show the improved model increases mean Average Precision (mAP) by 5.41% over the original YOLO v7 while maintaining computational performance, aiding in preventing ship-iceberg collisions in polar navigation.
2025,47(12): 175-180 收稿日期：2024-7-22
DOI：10.3404/j.issn.1672-7649.2025.12.031
分类号：U661.4
基金项目：国家重点研发计划项目（2021YFC2801001）
作者简介：章文俊(1977-),男,博士,教授,研究方向为交通信息工程及控制
参考文献：
[1] MUDRYK L R, DAWSON J, HOWELL S E L, et al. Impact of 1, 2 and 4 C of global warming on ship navigation in the Canadian Arctic[J]. Nature Climate Change, 2021, 11(8): 673-679.
[2] 孙鹤泉, 李春花, 张志刚. 基于遥感图像分析的极区海冰漂移研究[J]. 海洋技术学报, 2015, 34(1): 10-14.
[3] 倪汉健, 吴建华. 极地航行中海冰目标的雷达图像识别方法研究[J]. 武汉理工大学学报(交通科学与工程版), 2015, 39(4): 793-796.
[4] 屈猛, 赵羲, 庞小平, 等. 北极冰间水道区域的物理过程和遥感观测研究进展[J]. 地球科学进展, 2022, 37(4): 382-391.
[5] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[6] WEISSLING B, ACKLEY S, WAGNER P, et al. EISCAM—Digital image acquisition and processing for sea ice parameters from ships[J]. Cold Regions Science and Technology, 2009, 57(1): 49-60.
[7] BARBAT M M, WESCHE C, WERHLI A V, et al. An adaptive machine learning approach to improve automatic iceberg detection from SAR images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 156(62): 247-259.
[8] RAMASUBIRAMANIAN S S, SIVASUBRAMANIYAN S, MOHAMED M F P. Aggregate channel features and fast regions CNN approach for classification of ship and iceberg[J]. Applied Sciences-Basel, 2023, 13(12): 72-92.
[9] PANCHI N, KIM E, BHATTACHARYYA A. Supplementing remote sensing of ice: Deep learning-based image segmentation system for automatic detection and localization of sea-ice formations from close-range optical images[J]. IEEE Sensors Journal, 2021, 21(16): 18004-18019.
[10] DOWDEN B, DE SILVA O, HUANG W, et al. Sea ice classification via deep neural network semantic segmentation[J]. IEEE Sensors Journal, 2020, 21(10): 11879-11888.
[11] 肖湘文, 沈校熠, 柯长青, 等. 基于Sentinel-1A数据的多种机器学习算法识别冰山的比较[J]. 测绘学报, 2020, 49(4): 509-521.
[12] 舒苏, 柯长青, 周兴华, 等. 基于“双冒泡法”的SAR影像冰山识别[J]. 海洋科学进展, 2019, 37(1): 92-101.
[13] 胡冰, 于淼, 李志远, 等. 基于实船观测的北极东北航线窗口期海冰冰情研究[J]. 船舶力学, 2021, 25(8): 1002-1009.
[14] 邓亚平, 李迎江. YOLO算法及其在自动驾驶场景中目标检测研究综述[J]. 计算机应用, 2024, 44(6): 1949–1958.
[15] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]// Computer Vision and Pattern Recognition. IEEE, 2016.
[16] 中国船级社. 极地水域操作手册编写指南2017[M]. 北京: 中国船级社, 2016.
[17] 张雪薇, 韩震, 郭鑫. 深度学习在海洋信息探测中的应用: 现状与展望[J]. 海洋科学, 2022, 46(2): 145-155.
[18] 朱张莉, 饶元, 吴渊, 等. 注意力机制在深度学习中的研究进展[J]. 中文信息学报, 2019, 33(6): 1-11.
[19] 任欢, 王旭光. 注意力机制综述[J]. 计算机应用, 2021, 41(S1): 1-6.
[20] HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, USA. IEEE, 2021.
