Non-Linear Modelling of Detectability of Ship Wakes

Kurzfassung

On PORSEC 2016 a linear model for the detectability of ship signatures and besides also wake signatures was presented. The published linear model is capable of either calculating the probability of detection for ships or of their wakes in dependency to three parameters out of a set of parameters, which quantify environmental conditions, image acquisition parameters and ship's properties. Due to the linear basis of the model not all parameters are considered appropriately. Therefore the existing model has been extended by a non-linear basis. The new model takes more than three parameters and also additional parameters into account. This time the focus of the paper lies on the detectability of ship's wake signatures visible on TerraSAR-X/TanDEM-X high resolution mode acquisitions. Information about the influence of the respective parameters to the detectability of wakes can be retrieved from the model. Further, in case of non-sparse input data, the model can be used for estimation of missing parameters. The final goal of the model is the control of sensitivity of an automatic wake detection algorithm. The model is based on machine learning; therefore the session "Machine learning applications to ocean satellite remote sensing" would match. As the focus lies on wake detection and TerraSAR-X' high resolution modes, also the session "Sea surface roughness from high resolution SAR" would be appropriate. Last but not least, due to the development goal of an automatic wake detection algorithm, the presentation can also be assigned to the "Operational oceanography" session.