Target Classification Using Kinematic Data and a Recurrent Neural Network
Simon Bakkegaard, Jeppe Blixenkrone-Moller, Jakob Juul Larsen, Lars W. Jochumsen · 2018
We study the performance of a Recurrent Neural Network (RNN) for target classification using kinematic data from vessels sailing in Danish waters.We use data obtained from the Automatic Identification System (AIS) to get labelled data for supervised learning as a proof of concept for later use on 2D radar tracks. The RNN classifier was trained for five classes on five days of AIS data, and tested on data from a separate day. We used five-fold cross validation, achieving a classification accuracy of 78.3%. The results are compared with a random forest classifier using the same dataset. The RNN classifier achieved a classification accuracy 1.9 percent-points higher than the random forest classifier, showing that RNNs have good potential for target classification using kinematic data.