Ship Target Recognition Based On ISAR

At present,due to the increasing importance of sea power,new requirements are put forward for the recognition of sea targets.Because radar has the advantages of all-weather and multi-direction for target imaging,its importance is becoming increasingly prominent.Inverse synthetic aperture radar(ISAR)imaging is an important branch of radar recognition technology.In this thesis,we will extract texture features,contour features,moment invariant features and other information from ship modeling and imaging,and study ship target recognition method based on ISAR graph using metric learning and decision fusion recognition method.The main work of this thesis is as follows:Firstly,the improved method of ship's three-dimensional scattering point modeling is introduced.Based on the analysis of ship's actual data,we propose a method of obtaining ship's three-dimensional scattering point model by using 3D Max modeling,shrinkage algorithm and particle extraction.Secondly,we mainly introduce the preprocessing algorithm of ISAR image.Firstly,we propose a marking watershed algorithm based on ISAR image gradient.Secondly,we propose a SUSAN algorithm using circular mask to obtain the discrete edge characteristics of ISAR image.On this basis,we introduce a preprocessing algorithm based on image centroid,which smoothes the image and utilizes the target in ISAR image.The edge information extracts the centroid of the image,and then obtains the part of the image that we are interested in.Then we propose a center line extraction algorithm based on multi-frame image marker point processing(FmMPP).This algorithm compensates the missing information of a single image by utilizing the correlation information between multi-frame images.Then we introduce a new texture feature,Tamura texture feature,and change it to better apply it in ISAR images.Finally,we introduce a new texture feature which has the advantages of multi-frame image marker point processing(FmMPP).The Hu moment invariant feature of rotation invariance is used to deal with the influence of ship motion in three dimensions.Finally,we propose a new metric learning method for random system metrics(REMtric).This method obtains the optimal similarity of two vectors by iterating the output of multiple learning classifiers,and completes classification training accordingly.Then,considering the shortcomings of single feature in target recognition,we propose a feature fusion method.We mainly discuss the special features.The fusion strategy of levy-level fusion and decision-level fusion has obtained excellent recognition results.