摘要
针对船舶定位单纯依赖全球导航卫星系统(global navigation satellite system,GNSS)易遭受欺骗攻击的问题,提出了一种非周期特征卷积循环编解码器(convolutional recurrent encoder-decoder,CRED)算法,用于船舶导航欺骗检测。将来自GNSS、罗经、计程仪等传感器非周期特征数据的协方差矩阵作为输入,应用自动编码器、卷积长短记忆神经网络和注意力机制构建三层神经网络,通过累加残差矩阵和输入矩阵中具有相关性的元素值,计算检测统计量以识别欺骗攻击。实验表明,该方法在船舶不同航行状态下,针对跳变攻击和慢攻击,在多种欺骗程度下均具有良好的欺骗检测性能。该算法可用于处理船舶导航多源非周期特征数据,在船舶导航欺骗检测方面表现出检测效果好、泛化能力强的特点,具有很好的应用价值。
To address the issue of ship positioning being solely dependent on the Global Navigation Satellite System(GNSS)and itsvulnerability for spoofing attacks,thispaper proposesaNon-Periodic Feature Convolutional Recurrent Encoder-Decoder(CRED)algorithm for ship navigation spoofing detection.The covariance matrix of non-periodic feature data from sensors such as GNSS,compass,and log is used as input.A three-layer neural network is constructed using autoencoders,convolutional long-term and short-term memory networks,and attention mechanisms.By accumulating the residual matrices and adding the values of correlated elements in the input matrix,the detection statistic is calculated to identify spoofing attacks.Experiments show that the method performs well in detecting spoofing under various degrees of deception and different sailing states of the ship,including jump attacks and slow attacks.This algorithm can process multi-source non-periodic feature data for ship navigation and exhibits good detection performance and strong generalization capabilities in ship navigation spoofing detection,demonstrating significant practical value.
作者
姜毅
宫起正
向进
JIANG Yi;GONG Qizheng;XIANG Jin(School of Information Science and Technology,Dalian Maritime University,Dalian 116026,China)
出处
《海洋测绘》
CSCD
北大核心
2024年第6期54-59,共6页
Hydrographic Surveying and Charting
基金
国家重点研发计划(2021YFB3901502)
国家自然科学基金(52071047)
辽宁省教育厅高等学校基本科研项目(LJKZ0061)
辽宁省科学技术计划项目(2021JH1/10400008)。
关键词
卫星导航
欺骗检测
深度神经网络
自动编码器
卷积长短记忆神经网络
注意力机制
satellite navigation
spoofing detection
deep neural networks
autoencoder
convolutional long-term and short-term memory neural network
attention mechanism
