Fishing Gear Classification from Vessel Trajectories and Velocity Profiles: Database and Benchmark

Abstract

International Organizations demand to take care of our oceans and their ecosystems since they are of incalculable value to humanity. The illegal fishing activity does irreparable damage to these ecosystems and these organism are pushing to detect and combat illegal fishing activities. Fishing vessels are equipped with a radio frequency beacon that emits their GPS position and other information relevant to the Automatic Identification System (AIS). The GPS positions can be used to infer the vessel trajectories and detect illegal fishing activities. In this study we present a new database ( https://github.com/BiDAlab/TrFGdb ) including trajectories representing 5 different fishing gears, and analyze them as in a problem of time sequence analysis. We extract global and local features from the trajectories of vessels, and propose several supervised learning algorithms to classify the kinematics of vessels according to different fishing gears. Compared to previous works, we highlight the importance of considering trajectories with sampling period in the order of minutes instead of hours, to detect activities carried out in a short time that could help to distinguish fishing gears. A considerable effort has been dedicated to pre-processing the real data at our disposal, to generate a quality dataset with highly reliable labels. The best classification accuracy obtained in this study is 90%. We expect to improve it if more trajectories describing the different fishing gears were available.