Tuning Detection of Ship Wakes by Detectability Modelling

The appearance and detectability of ship wakes in SAR imagery has been investigated for decades. Various wake components have been identified in the complex appearance of wake signatures. Wake signatures in SAR imagery can also be exploited for indirect ship detection if ship signatures are weak or absent. A systematic investigation on dependencies between wake detectability and physical parameters affecting the detectability of wake components, i.e. influencing parameters, has recently been published. Recently also machine learning has been applied for wake detection, making the wake detection performance more robust. This study presents a novel approach to increase detection accuracy during post-processing. A combination of the systematic wake detectability modelling with a robust wake detector based on machine learning is proposed. The results show that the dynamic filtering on the basis of wake detectability models can increase the precision while maintaining the recall.