基于DBSCAN的VMS数据定置刺网渔船网次提取方法

网次是捕捞努力量计算、渔业资源调查和渔业生产管理中的重要统计参数,定置刺网作业过程中起网状态下的航速与航行、放网时相差较大,可通过阈值进行提取。利用2017年浙临渔12870和浙三渔66666定置刺网渔船VMS数据,首先对其航速及相邻作业点时间差进行统计分析,得到合理的聚类参数,然后采用DBSCAN(Density-Based Spatial Clustering of Application With Noises)算法对定置刺网周围作业点进行聚类,提取作业网次,继而统计各网次作业持续时间与距离。以15 min作为合理误差范围,将各网次起止时间与日志起网时起止时间进行比较验证。结果表明:该方法在网次识别上效果良好,聚类提取的各个网次起网时的起止时间准确率达到80%以上。

Method of set gillnet hauls extraction based on DBSCAN and VMS data

Vessel monitoring system (VMS), as a comprehensive application system integrating global satellite positioning, electronic map, network communication and database technology, aims to acquire the ship''s position and operating status information in near-real time, and transmit them to the onshore monitoring center to realize the information interaction. More than sixty thousand fishing vessels have installed VMS so far. The VMS data mainly contains time, position, speed, direction and rate of turn with a temporal resolution of 3 minutes and a spatial resolution of 10 meters which can be analyzed deeply using data mining technology to identify the status of vessels, calculate fishing effort, search fishing ground and so on. DBSCAN is a density-based spatial clustering algorithm designed to find high-density regions segmented by low-density. Compared with distance-based clustering algorithms, it can identify noise data and find clusters of arbitrary shapes. In this paper, we used VMS data of Zhelinyu 12870 and Zhesanyu 66666 in 2017 and DBSCAN to extract the hauls of set gillnet. The speed of hauling gillnet is quite different from that during tracking or setting the net, so they could be extracted them by threshold. Firstly, we obtained the speed threshold of vessel''s fishing condition by the statistics of navigational speed and extracted the points whose speed is in the range of threshold value. Secondly, the time difference of adjacent points was carried out to obtain time interval between hauls. Thirdly, the DBSCAN algorithm was used to cluster the fishing points around the set gillnet, which determines the starting time and the ending time of each fishing haul. The difference of the starting time and the ending time of each haul between clustering result and fishing log were calculated, and 15 minutes were token as a reasonable error-tolerant range. The comparisons showed that the accuracy of the hauls was above 80%.