基于滑动窗口和LSTM自动编码器的渔船作业类型识别  

Fishing Vessel Operation Type Identification Based on Sliding Window and LSTM Auto-encoder

作　　者：徐文进[1] 董少康 XU Wen-Jin;DONG Shao-Kang(College of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061,China)

机构地区：[1]青岛科技大学信息科学技术学院,青岛266061

出　　处：《计算机系统应用》2022年第6期287-293,共7页Computer Systems & Applications

基　　金：国家自然科学基金(61806107)。

摘　　要：过度捕捞和非法捕捞给海洋生态造成严重破坏,随着船舶自动识别系统(AIS)的发展,国内外学者基于AIS轨迹数据提出了许多算法进行渔船作业类型识别,但是这些算法忽视了轨迹的时域特征.因此,本文提出了一种基于滑动窗口和LSTM自动编码器的识别算法,该算法首先使用滑动窗口提取轨迹特征,再通过LSTM自动编码器去学习轨迹的时域特征和潜在的高级特征,最后在LSTM自动编码器中嵌入Softmax分类器,联合优化损失函数,使分类效果达到最优.在浙江海域的渔船AIS轨迹数据上进行了实验,结果表明所提方法的准确率为95.82%,证明了本方法的有效性和可靠性,算法可用于辅助拖网、围网作业类型的判断.Overfishing and illegal fishing have caused serious damage to marine ecology. With the development of the automatic identification system(AIS) on vessels, scholars have proposed plenty of algorithms based on AIS trajectory data to identify the operation types of fishing vessels. However, these algorithms ignore the temporal features of the trajectory. Therefore, this study puts forward the identification of operation type based on the sliding window and LSTM auto-encoder. Firstly, it utilizes the sliding window to extract trajectory features and then uses an LSTM auto-encoder to learn the temporal features and potential advanced features of trajectories. Finally, the Softmax classifier is embedded in the LSTM auto-encoder to jointly optimize the cost function, achieving the best classification. The algorithm is verified based on AIS trajectory data of fishing vessels in the Zhejiang sea area, China. The results show that the accuracy is95.82%, which proves the effectiveness and reliability of the proposed algorithm. The algorithm can be used to assist in judging the operation type of trawl and purse seine.

关 键 词：LSTM自动编码器 滑动窗口 深度表征学习 AIS 拖网 围网 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] S972.7[自动化与计算机技术—控制科学与工程]