Trawler Fishing Track Interpolation using LSTM for Satellite-based VMS Traces

摘要

Vessel Monitoring System (VMS), based on China Beidou Satellite System, provides a high time resolution of 3 minutes for all trawlers in the East China Sea. However, the data loss is still a common phenomenon, hindering the research on spatial and temporal patterns of fishing activities. Most existing studies apply Cubic Hermite Spline (CHS) to interpolate loss samples. However, CHS interpolation is deficient in precision when the inserted trajectory is composed of turns. This paper proposes a hybrid interpolation scheme, combing Long-Short Term Memory (LSTM) and CHS for loss sample insertion. When the loss samples are at the fishing state and more than seven samples, we apply LSTM for interpolation; otherwise, we use CHS. The proposed hybrid interpolation scheme is evaluated with the VMS dataset of 1689 otter trawlers on the East China Sea during October 2016. The results show that the proposed hybrid interpolation scheme limits the interpolation error under 0.4km for continuous one-hour sample losses, 33% lower than CHS.