Ship Wake Detectability and Classification on TerraSAR-X high resolution data

This study elaborates on the detection of ship wake signatures on high resolution TerraSAR-X images. A new data-driven detectability model based on a binary logistic regression classifier is proposed. Additionally, to de-tect small and incompletely imaged wake signatures, a second machine learning approach is investigated. The second approach uses feature extraction and binary classification to classify the area around ships. By applying the detectability model before the classification, the sensibility of the resulting wake detection can be adjusted. The detectability model shows expected dependencies between wake visibility and the affecting parameters. Ini-tial results reach above 70% accuracy for wake classification