在海面气象观测任务中，面对复杂多样的海面场景，无人艇执行任务过程中需要准确识别、分类海面场景，基于识别分类结果实时改变感知决策策略，以保证航行安全和高效作业。本文提出MSSNet场景分类模型，创新性地将MobileNeXt模块与MobileVit模块融合，并引入CA注意力模块高效提取全局语义信息，提高模型识别性能。本文基于艇载多种图像传感器构建无人艇海面场景分类图像数据集，包括雾天、强光、弱光、水渍、盐渍、夜间和正常等7类场景。经试验测试，本文提出的MSSNet模型在海面场景分类图像数据集上的准确率为96.60%，比 MobileNetv3、ViT等主流模型提高了3.53%，满足气象观测任务中无人艇自主航行的需求。
In sea surface meteorological observation tasks, faced with complex and diverse sea surface scenarios, unmanned aerial vehicles (UAVs) need to accurately identify and classify sea surface scenes during the execution process. Based on the recognition and classification results, the perception decision strategy needs to be changed in real time to ensure navigation safety and efficient operation. In response to the problem of sea surface scene recognition, this paper proposes the MSSNet scene classification model, which innovatively integrates the MobileNeXt module with the MobileVit module, and introduces the CA attention module to efficiently extract global semantic information, improving the recognition performance of the model. This article constructs a dataset of unmanned boat sea surface scene classification images based on various onboard image sensors, including seven categories of scenes: foggy, strong light, weak light, waterlogged, saline, nighttime, and normal. After experimental testing, the accuracy of the MSSNet model proposed in this article on the sea scene classification image dataset is 96.60%, which is 3.53% higher than mainstream models such as MobileNetv3 and ViT, and meets the needs of autonomous navigation of unmanned boats in meteorological observation tasks.
2025,47(6): 88-93 收稿日期：2024-5-11
DOI：10.3404/j.issn.1672-7649.2025.06.014
分类号：U661
基金项目：国家重点研发计划重点专项（2021YFC3090200）
作者简介：苏睿涵（1999 – ），男，硕士研究生，研究方向为图像处理和行为意图分析
参考文献：
[1] 胥凤驰, 王伟, 李哲, 等. 水面无人艇系统的设计实现与未来展望[J]. 舰船科学技术, 2019, 41(23): 39-43.
XU F C, WANG W, LI Z, et al. Design and realization of unmanned surface vessel system and its future prospects[J]. Ship Science and Technology, 2019, 41(23): 39-43.
[2] 王博. 无人艇光视觉感知研究发展综述[J]. 舰船科学技术, 2019, 41(23): 44-49.
WANNG B. Review of development in perception of unmanned surface vehicle based on optical vision[J]. Ship Science and Technology, 2019, 41(23): 44-49.
[3] CHEN Z, YANG F, LINDNER A, et al. Howis the weather: Automatic inference from images[C]// 2012 19th IEEE International conference on image processing. 2012.
[4] LECUN Y, BOTTOU L. "Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[5] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words: Transformers for image recognition at scale[C]// International Conference on Learning Representation, 2021.
[6] LI Z, LI Y, ZHONG J, et al. Multi-class weather classification based on multi-feature weighted fusion method[J]. IOP Conference Series: Earth and Environmental Science, 2020, 58(5):38-42.
[7] VASWANI, ASHISH, NOAM S, et al. Attention is all you need[J]. Computer Science, 2023, 7(V1):5-15.
[8] 戴军, 金代中, 高志峰. 基于纹理特征驱动AdaBoost算法的海面场景分类[J]. 激光与红外, 2015, 45(4): 462-466.
DAI J, JIN D Z, GAO Z F. Sea scene classification based on AdaBoost algorithm with texture characteristics[J]. LASER & INFRARED, 2015, 45(4): 462-466.
[9] ZHOU D Q. Rethinking bottleneck structure for efficient mobile network design[J]. Computer Vision-ECCV 2020: 16th European Conference, 2020, 16(3): 23–28.
[10] SANDLER M, HOWARD A, ZHU M, et al. Mobilenetv2: Inverted residuals and linear bottlenecks[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. Interted Residuals, 2018.
[11] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. Computer Vision and Pattern Recognition, 2021.
[12] PASZKE, ADAM, SAM G, et al. Pytorch: An imperative style, high-performance deep learning library[J]. Computer science, 2019, 12(2): 3-15.
[13] 周志华. 机器学习[M]. 北京: 清华大学出版社, 2016.
