Research On Performance Evaluation Method Of Ship Classification Based On Radial Size

Length is an inherent physical characteristic of a ship.Length-based ship classification is an important preparation for refined ship classification on the sea,which is of great significance for maritime radars.High-Resolution Range Profile(HRRP)of a ship in maritime radars reflects its size and structure information in the sight line of the radar,which is easy to be obtained and processed.Thence,HRRP-based ship radial length extraction,classification,and recognition have become into a hotspot of research in signal processing of high-resolution maritime radars.In this thesis,the performance analysis of HRPP-based ship classification in high-resolution maritime radars is investigated under the Bayesian statistical analysis framework.Under this framework,the prior probability models of ship lengths in the offshore waters of our country must be established firstly.On the basis of collecting a mass of ship length data on the offshore waters of our country,the probability models of ship lengths are established.Based on these models,the quantitative relationship between the correct classification probability of ships and the estimation error of their radial sizes are given under the Bayesian statistical framework.The content of this thesis is organized as follows.The 1st chapter introduces the research background as well as domestic and foreign research status of this thesis,and gives the work arrangement of this thesis.In the 2nd chapter,the division of radar bands and the influence of wavelength on the backscatter characteristics of the target are firstly reviewed.Then,the scattering center model of the target in the optical region is introduced,and the imaging principle of HRRP is analyzed.Finally,the radial length extraction method of target based on HRRP is proposed.Aiming at the problem that traditional threshold method is sensitive to noise,two rangeprofile processing algorithms based on adaptive difference operator are introduced.In the 3rd chapter,the probability models of ship lengths on the offshore waters of our country are established.Firstly,the offshore waters of our country are divided into several waters,and a mass of Automatic Identification System(AIS)data on these waters are collected,the statistical characteristics of ship length distribution are also analyzed.Then,the probability density functions and parameter estimation methods of Exponential distribution,Rayleigh distribution,Weibull distribution and K-distribution are introduced.Based on the above four models,the probability model parameters of ship length sample data are estimated respectively.Finally,the fitting performance of each model are quantitatively analyzed by comparing K-S Distance(KSD)and Root Mean Square Error(RMSE).The Weibull distribution model is selected as the probability models of ship lengths on the four offshore waters,inland sea and territorial sea of China.The Rayleigh distribution model and exponential distribution model are respectively selected as the probability models of ship lengths on contiguous zone and exclusive economic zone of China.In the 4th chapter,the performance evaluation method of ship classification based on radial size is researched under the Bayesian statistical analysis framework.Under this framework,for a given radial size estimation error,a conditional probability function is constructed to indicate the correct classification probability when the ship length is assumed to be known.Then,combining with prior probability models of ship lengths in the specified area,the quantitative relationship between the correct classification probability of ships and the estimation error of their radial sizes in this area is obtained.The results indicate that the condition for the big-moderate-small correct classification probability in the offshore areas of China to be up to 90% is that the estimate errors of the ship radial size estimates fall into the interval(-11.89 m,8.84 m)when the heading angle of the ship is between ±75 degrees.The 5th chapter summarizes the work of this thesis and looks forward to the future research direction.