Identifying Fishing Activities from AIS Data with Conditional Random Fields

Abstract

Fishing activity detection is important for fishery management to maintain abundant oceans. The rising demand for fish and advanced fishing technologies has led to overfishing, species endangerment and marine habitat destruction. Illegal, unreported and unregulated (IUU) is one example of an important ecological and economic issue that requires the understanding of fishing behavior of ships. Our proposed approach to detecting fishing activities uses Conditional Random Fields (CRFs) on Automatic Identification System (AIS) data. We generate features from selected attributes and combine different features based on their relationships and dependencies. We present three experiments on trawlers and longliners respectively as well as comparisons between CRFs and methods such as Autoencoder and Hidden Markov Model (HMM) to demonstrate the stability and effectiveness of the CRF models. Furthermore, we develop a geo-visualization with interaction and animation of these AIS data and our experimental results.