摘要： 针对当前海洋牧场人工鱼礁的探测识别以人工识别为主，效率低、成本高的问题，开展了CSF与Ransac两种滤波算法在多波束点云数据识别人工鱼礁的研究和分析。首先介绍两种算法的原理与配置，由NORBIT iWBMS多波束测深仪采集两个试验区域的数据点云，随后在两个鱼礁区域分别进行鱼礁提取实验，并对比CSF与Ransac两种算法的识别精度和识别完整度。结果显示：两种算法对人工鱼礁都有较好的识别效果，由CSF算法自动识别提取到的人工鱼礁正确度为95.88%，完整度为93.94%，而Ransac算法的正确度为93.48%，完整度为90.53%；但CSF算法提取的鱼礁的三维形态更为完整，能够保留单体鱼礁的完整三维信息。本研究方法和成果，为多波束声呐点云数据识别提取人工鱼礁提供了技术路线，为海洋牧场人工鱼礁的科学评估提供了技术保障。

关键词: 鱼礁提取, 多波束点云, Ransac算法, CSF算法, 对比分析

Abstract: In view of the current problem that the detection and identification of artificial reefs in marine ranches is mainly based on manual identification, which is inefficient and costly, the CSF and Ransac filtering algorithms were studied and analyzed in the identification of artificial reefs in multi-beam point cloud data. First, the principles and configurations of the two algorithms were introduced. The data point clouds of the two test areas were collected by the NORBIT iWBMS multi-beam echo sounder. Then, the reef extraction experiments were carried out in the two reef areas, and the recognition accuracy and completeness of the CSF and Ransac algorithms were compared. The results showed that both algorithms had good recognition effects on artificial reefs. The accuracy of the artificial reefs automatically identified and extracted by the CSF algorithm was 95.88%, and the completeness was 93.94%, while the accuracy of the Ransac algorithm was 93.48%, and the completeness was 90.53%. However, the three-dimensional morphology of the reefs extracted by the CSF algorithm was more complete, and the complete three-dimensional information of the single reef could be retained. The research methods and results provide a technical route for the identification and extraction of artificial reefs using multi-beam sonar point cloud data, and provide technical support for the scientific evaluation of artificial reefs in marine ranches.

Key words:  , fish reef extraction, multi-beam point cloud, ransac algorithm, CSF algorithm, comparative analysis