Occluded Target Modeling Simulation And Detection For SAR Imagery

Synthetic Aperture Radar(SAR)is a kind of radar that can work all-time and all-weather as well as generate two-dimensional high-resolution images.It's wildly used in military and civilian fields.In the field of military scouting,when the military targets are partially occluded by the bunkers(such as armoured concrete)that microwaves can hardly penetrate,the performance of traditional detection and recognition algorithms degrade sharply.Therefor,in order to improve the performance of SAR Automatic Target Recognition(ATR)system in the military field,this thesis is focused on the occluded target detection for SAR imagery.The main contents of this thesis are as follows:1.Due to the lack of occluded targets database,by performing target modeling and electromagnetic simulation,we verified it is feasible to generate occluded targets by replacing the targets' pixels on the side of the direction of illumination with the background pixels.Then we use this method to generate enough occluded targets for the study of the detection of occluded target for SAR imagery by using the military targets database from Moving and Stationary Target Acquisition and Recognition(MSTAR)plan.2.The traditional Constant False Alarm Rate(CFAR)detection algorithms have large amount of calculation and perform badly when multiple targets exist.This thesis proposed a fast CFAR detection algorithm based on the0 G distribution.By the usage of the binary potential target areas map which is obtained through the global threshold detection and filtering stage,the proposed algorithm not only get fast detection speed but also high detection rate.However,this algorithm doesn't distinguish the occluded targets from the non-occluded targets.It's necessary to conduct further detection on the fast CFAR detection's results.3.For the issue of distinguish among the occluded targets,the non-occluded targets and a few false alarms,a algorithm based on the Genetic Algorithm(GA)and Support Vector Machine(SVM)is proposed in this thesis.On the one hand,GA reduces the huge amount of calculation in the search for the best subset of features.On the other hand,using SVM to perform fine detection can further eliminate the false alarms and ensure high detection rate occluded targets at the same time.The experiment results validated that the occluded target detection algorithm for SAR imagery proposed in this thesis has high performance which satisfies the application requirement.