基于小样本学习理论的船舶目标检测算法研究  

Research on ship target detection algorithm based on small sample learning theory

作　　者：陈静[1] 刘奥祥 蔡翼枫 CHEN Jing;LIU Aoxiang;CAI Yifeng(Tianjin Research Institute for Water Transport Engineering,M.O.T.,Tianjin 300456,China;Transportation Development Center of Henan Province,Zhengzhou 450003,China;DaGukou Maritime Safety Administration,Tianjin 300211,China)

机构地区：[1]交通运输部天津水运工程科学研究所,天津300456 [2]河南省交通事业发展中心,郑州450003 [3]中华人民共和国大沽口海事局,天津300211

出　　处：《水道港口》2025年第1期150-156,共7页Journal of Waterway and Harbor

基　　金：国家重点研发计划项目(2023YFB2603800);中央级公益性科研院所科研创新基金资助项目(TKS20230203)。

摘　　要：针对船舶视频目标多样性、新增样本有限等问题,提出了一种基于计算机视觉信息小样本学习理论的船舶目标检测技术,通过构建时空对称卷积神经网络,融合时序对称的视觉和语义特征,实现视频质量增强,基于初始化表征学习与随机梯度下降理论,实现边缘极少样本数据的快速训练与迭代,通过自建船舶样本数据集进行模型训练,实现了基于视频的船舶目标检测,并结合AIS数据实现了船舶属性信息与视频的融合,建立了一种不依赖于船载终端的主动式非接触船舶监管系统,面向海事监管人员和水运行业参与人员提供智能化便捷化的监管服务,依托该平台打造海事智慧之眼、建设水运服务大脑。Aiming at the problems of ship video target diversity and limited new samples,a ship target detection technology based on the theory of small sample learning of computer vision information was proposed.By constructing a spatio-temporal symmetric convolutional neural network and fusing temporally symmetric visual and semantic features,video quality enhancement was achieved.Based on the theory of initialization representation learning and stochastic gradient descent,rapid training and iteration of extremely few edge sample data were realized.Through self-built ship sample data sets for model training,ship target detection based on video was realized.Combined with AIS data,the fusion of ship attribute information and video was achieved.An active non-contact ship supervision system that does not rely on shipborne terminals was established,providing intelligent and convenient supervision services for maritime supervisors and participants in the water transport industry.Relying on this platform,the goal of building the eyes of maritime intelligence and developing the brain of water transport services was realized.

关 键 词：小样本学习理论 船舶目标检测 目标识别 船舶数据集 人工智能 深度学习 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]