Good practice in training set preparation for marine navigation systems

Abstract

The automation of navigation systems for marine vessels has been gaining momentum recently. Such systems require the ability to perform object detection on images received by the vessel’s sensors, which in turn requires machine learning models and training data. This document addresses good practice in training set preparation for these object detection algorithms. We give an overview of a typical marine navigation system and describe the accompanying object detection task. We highlight three main areas of good practice: data transferability, data coverage and data accuracy, and we give concrete recommendations on good practice.

External Links