基于KNN的船舶轨迹分类算法

提出基于KNN（K-Nearest Neighbor）的船舶轨迹分类算法：对轨迹间平均距离、航速距离及航向距离进行融合，构成船舶轨迹间综合距离；通过船舶轨迹初步聚类，得到KNN分类样本轨迹；将综合距离作为KNN分类中轨迹间的距离，最终实现船舶轨迹分类.以长江航道武汉段2017年5月的船舶AIS数据为基础，开展基于轨迹间平均距离、豪斯托夫（Hausdorff）距离以及综合距离的船舶轨迹分类验证.结果表明：轨迹点较多时，轨迹间平均距离较Hausdorff距离具有更好的适用性，且基于KNN的分类方法具有较好的实验结果，可有效应用于实际船舶轨迹分类中.
A ship trajectory classification algorithm based on KNN (K-Nearest Neighbor) was proposed. The average distance, speed distance and heading distance were fused between trajectories to form the integrated distance. By initial clustering of ship trajectories, the trajectory of KNN classification samples was obtained. Taking the integrated distance as the distance between trajectories in KNN classification, trajectories classification was thus achieved. Case study was performed to validate the feasibility of proposed algorithm by using AIS data of Wuhan reach of the Yangtze River in May 2017, and ship trajectory classification verification was carried out based on the mean distance between tracks, Hausdorff distance and the integrated distance. Results indicate that when more trajectories points are engaged, the average distance between the tracks is better than the Hausdorff distance, and the KNNbased classification method has better experimental results, which can be used in the actual ship trajectory classification.