Ship Detection Method On Polarization Synthetic Aperture Radar

In comparison with single-polarization synthetic aperture radar(SAR),polarimetric synthetic aperture radar(Pol SAR)could enhance the ability of discriminating different types of targets by capturing abundant structural and textural information of the medium.Meanwhile,ship detection is the precondition and foundation of ships recognition and an integral part of ship interpretation.Besides,Pol SAR has got a wide range of application and plays an important role in the seagoing rescue,fishery and traffic services fields.Therefore,ship detection using Pol SAR data has a bright future and the prospects for development.The contents of this paper include two aspects: First,improve the contrast between ship and sea in the Po SAR image by dealing with the polarimetric scattering information;Second,automatic extraction of targets features by applying convolutional neural network(CNN),both of the two methods achieve a good performance on ship detection.The organization of this paper is as follows:(1)Introduction of Pol SAR,including the main data model of polarimetric scattering information and the difference of scattering mechanism between the sea and targets.(2)Analysis the limitation of some common ship detection methods and propose the ship detection algorithm based on the Relative Kurtosis(RK).This algorithm uses the sea clutter statistical distribution,and the relationships between the covariance complex matrix with the RK to translate the Pol SAR data image into the Relative Kurtosis image data.Then,the Canny method is used to obtain the robust ship detection results.(3)CNN is a very important algorithm to achieve accurate classification and recognition in remote sensing area.This part reviews the past research of classification with CNN on the remote sensing,and puts forward a new detection method termed TS-CNN(Two-Step CNN)based on the superiority of the CNN which could learn feature automatically.This method builds a precise CNN architecture and puts the coarse classification result of the trained CNN classifier into the network again to reach the accurate location of ships.(4)Finally,the real measured UAVSAR and AIRSAR datasets are used for validating the two proposed ship detection algorithms.And the detection result shows that RK ship detection method achieves a good detection performance by using the clutter statistical information sufficiently,especially the superiority of detection about the weak and small ship,but it costs much time on detection progress at the same time.The other method based on CNN could achieve the detection result on any size of Pol SAR images with limited time and high accuracy.