【摘要】 运动模式识别是实现船舶智能航行的重要研究方向。针对现有方法中数据更新速度慢、受环境约束大的问题，研究了1种基于九轴惯性测量单元（inertial measurement unit,IMU）的船舶运动模式识别方法。分析了当前船舶运动感知技术的不足，提出利用加速度计、陀螺仪和磁力计构成的九轴IMU来识别船舶运动参数。为处理包含多种运动模式的长时间连续信号，提出了1种基于隐马尔可夫模型的数据分割算法，并采用期望最大化算法估算模型参数，实现按运动模式分割信号，提取单一稳态模式信号。分析分割后的运动模式信号，提取能够表征船舶运动模式的时域特征。为提高识别精度，设计了1种基于二叉树结构的支持向量机（support vector machine,SVM）算法，利用最大割问题构建二叉树结构，在决策节点使用SVM分类器，并通过粒子群优化算法优化模型参数。实验基于实船采集的运动数据进行验证，结果表明：所提识别算法只需训练5个SVM子分类器，能够对6种船舶运动模式进行有效识别，平均识别精度达到96.498%。相比传统的一对一和一对多SVM多分类方法，提出的方法平均识别精度分别提高了13.835%和21.305%，且所需训练的子分类器数量更少，验证了方法的优越性与高效性。更多还原

【Abstract】 Motion pattern recognition is an important issue for achieving intelligent navigation of ships. To address the limitations of existing methods,including slow data update rates and strong environmental constraints,a ship motion pattern recognition method based on nine-axisinertial measurement unit(IMU) is proposed. The shortcomings of current ship motion sensing technologies are analyzed,and a nine-axis IMU consisting of an accelerometer,gyroscope,and magnetometer is utilized to identify ship motion parameters. To process long-duration continuous signals encompassing multiple motion patterns,a data segmentation algorithm based on hidden Markov model is developed. The expectation maximization algorithm is employed to estimate model parameters,enabling signal segmentation according to motion patterns and the extraction of single steady-state motion signals. The time-domain features that characterize ship motion patterns are then extracted from the segmented signals. To improve recognition accuracy,a support vector machine(SVM) algorithm based on a binomial tree structure is designed. The binary tree structure is constructed using the maximum cut problem,with SVM classifiers employed at decision nodes. The particle swarm algorithm is applied to optimize the model parameters. Experiments conducted using ship motion data collected from real ships validate the proposed method. Results show that the proposed recognition algorithm requires training only five SVM sub-classifiers for the recognition of six ship motion patterns,achieving an average recognition accuracy of 96.498%. Compared to traditional one-to-one and one-to-rest SVM multi-classification methods,the proposed method improves average recognition accuracy by 13.835% and 21.305%,respectively,while requiring fewer sub-classifiers for training. These findings demonstrate the superiority and efficiency of the proposed approach.更多还原