为提高对本船所处海域环境的感知能力，研究机器学习算法的船舶激光雷达点云数据分类方法。利用多旋翼无人机搭载激光雷达系统，获取船舶所处海域环境的点云数据，将其运用形态学滤波算法进行滤波处理，区分出点云数据中的非海洋点，使用灰度共生矩阵提取非海洋点部分的纹理特征，据其采用机器学习算法中的随机森林算法，实现船舶激光雷达点云数据分类。实验结果表明：为获得较理想的船舶激光雷达点云数据滤波性能，需将该方法的结构元窗口设置为20 m；该方法所提取非海洋点部分的纹理特征具有较好的可分性，并且对非海洋点部分的分类全面性和准确性较高。
In order to improve the perception ability of ship managers to the sea environment where the ship is located, the classification method of ship LIDAR point cloud data based on machine learning algorithm is studied. The point cloud data of the marine environment where the ship is located is obtained by using the lidar system of the multi rotor UAV. The point cloud data is filtered by using the morphological filtering algorithm to distinguish the non ocean points in the point cloud data. The texture features of the non ocean points are extracted by using the gray level co-occurrence matrix. According to the random forest algorithm in the machine learning algorithm, the classification of the ship LIDAR point cloud data is realized. The experimental results show that in order to obtain better filtering performance of ship LIDAR point cloud data, the structural element window of this method should be set to 20 m. The texture features of non ocean points extracted by this method have good separability, and the classification of non ocean points is comprehensive and accurate.
2022,44(17): 140-143 收稿日期：2022-05-20
DOI：10.3404/j.issn.1672-7649.2022.17.028
分类号：TP75
基金项目：江苏省高校优秀中青年骨干教师境外研修计划项目(2018)；江苏商贸职业学院校级课题 (SY20181211-19)；江苏商贸职业学院横向课题 (HX2018013)
作者简介：施亮(1980-)，男，硕士，讲师，研究方向为软硬件设计与测试、计算机仿真及计算机应用技术等
参考文献：
[1] 张文豪, 李松, 马跃, 等. 星载光子计数激光雷达海面点云仿真方法[J]. 红外与毫米波学报, 2020, 39(4): 483–490
[2] 雷相达, 王宏涛, 赵宗泽, 等. 整合迁移学习与全卷积网络的小样本机载激光雷达点云分类[J]. 中国激光, 2021, 48(16): 138–149
[3] 任永梅, 杨杰, 郭志强, 等. 基于三维卷积神经网络的点云图像船舶分类方法[J]. 激光与光电子学进展, 2020, 57(16): 230–238
[4] 陈博伟, 庞勇, 李增元, 等. 基于随机森林的光子计数激光雷达点云滤波[J]. 地球信息科学学报, 2019, 21(6): 898–906
[5] 顾尚泰, 王玲, 马燕新, 等. 基于分层墨卡托投影的激光雷达点云数据局部特征描述[J]. 光学学报, 2020, 40(20): 137–143
