Sea Clutter Simulation And Research On ISAR Image Recognition For Ship Target

Inverse Synthetic Aperture is a kind of 2-D imaging radar,It can monitor the target continuously in a variety of weather conditions.It has an important value in the wide area detection,environmental remote sensing,air defense.The main research of this thesis is ISAR image recognition,the main contents are as follows:1.The mathematical model of sea clutter is studied in this thesis,the common sea clutter model(lognormal distribution,Weibull distribution,K distribution,Inverse Gaussian Complex Gaussian)are simulated and analyzed by the method of zero memory nonlinear transformation and Fourier series expansion.The results show that the shape parameter of clutter affects the distribution of the clutter,and it can control the shape of the clutter’s tail,and the velocity variance has an effect on internal fluctuation of clutter,it does not contribute to the tail of the clutter.2.The imaging and rotation model of ship target are introduced.By using the range Doppler imaging algorithm,the imaging of three small angle rotation modes(roll、pitch、yaw)of the target lattice model is studied and getting the ISAR images.It provides data support of the follow up research.3.The feature extraction is very import to the imaging recognition.Firstly,According to the clutter suppression method,this paper designs whitening filter to filter the ISAR image,and then enhances the image,which is helpful for feature learning.This paper using the convolutional neural network to image feature learning.It does not need to extract the features by human,and it can carry out the characteristic by studying independently.And even the image has the translation,the telescopic change,it also has a strong robustness.Through the experiment,the network feature learning structure is optimized from the network layer,the number of convolution kernels and the size of the filter.4.This paper focuses on the Support vector machine and three types of AdaBoost(Gentle AdaBoost,Real AdaBoost,Modest AdaBoost).AdaBoost makes use of the iteration and weight update to transform the classifier into a strong classifier to improve the recognition effect.The results show that the MAB algorithm is the worst and unstable in the three kind of AdaBoost.RAB is better than GAB and MAB,and it has good stability.The result of the intersection method is better the reservation method.Support vector machine algorithm is better than AdaBoost algorithm,moreover,it is more stable to the recognition results of different angle distribution.