Ship Wake Detection And Estimation Of Ship Parameters In SAR Images

The detection of ship wakes in Synthetic Aperture Radar(SAR)images and estimation of ship parameters by using wakes are of research value and promising.The wake model and the wake detection algorithm of SAR image are mainly studied in this thesis.The Radon transform based enhancement algorithm has limitations on size and scene of the input image.In order to solve the above problems,the deep learning convolutional neural network(CNN)algorithm Faster RCNN is introduced to detect the position of ship and it’s wakes.Besides,ship and it’s wakes are matched,facilitating the estimation of ship parameters.All this work improve the practicability of ship wake.Finally,the method of obtaining ship parameters by means of Kelvin wake and turbulent wake is introduced.The main work and innovation of this thesis are as follows: 1.Based on the ship wake model,the characteristics of wakes in SAR images are discussed.The geometrical features are obtained through the simulations of Kelvin wake,turbulent wake and internal wake.The point source theory is used to simulate ships of different tonnage and ship speed,and the influence of different parameters on the amplitude and wavelength of Kelvin wake is discussed.The internal wake and turbulent wake are simulated by bilinear superposition method,and the morphological characteristics of wake at different ship speeds are discussed.2 The ship wake detection algorithm is deeply discussed.The normalized Radon transform with better detection performance is introduced for detection the linear feature of the wake.In order to improve the positive detection rate of ship wake in SAR images,image preprocessing is carried out firstly,Then the theory of shearlet transform is introduced and the way to implement it.Compared with the traditional edge enhancement algorithm,the shearlet transform performances well in terms of detail integrity and information redundancy.In Radon field,the morphological characteristics of different components of the wake are considered,which improves the efficiency and reliability of the algorithm.3.The Faster RCNN detection algorithm is deeply studied and the purpose of introducing this algorithm is to solve the defects of traditional wake detection algorithm.Firstly,the basic theory of deep learning is introduced,then the development process and corresponding principle of Faster RCNN algorithm are expounded.Before the experiment,the ship’s wake sample data set from Terra SAR-X satellite is firstly created.In order to improve the accuracy of the algorithm,the sample data is expanded and the corresponding creation process is given.Based on the data set,the experiment of Faster RCNN is carried out.After detecting the region of ship and wake,the correspondence between ship and wake is determined by using wake registration principle,which provides convenience for ship parameter inversion.4.The method of obtaining ship parameters by ship wake is studied.Firstly,the frequency domain characteristics of the kelvin wake are studied.The method of estimating course and hull size by using center moment and ship slice pixel is discussed.As the most obvious wake component in a SAR image,The azimuth displacement could be obtained by the ship’s center of gravity and turbulent wake.Finally,the relation between ship size and amplitude function is introduced.