Study On Target Detection And Discrimination For SAR Images In Complex Scenes

Synthetic aperture radar(SAR)automatic target recognition(ATR)technology is an critical way to realize SAR image interpretation.The typical SAR ATR system consists of three stages: detection stage,discrimination stage,and classification/recognition stage.The detection and discrimination stages are the foundation of the whole SAR ATR system,and many problems have still not been solved in these two stages;thus it is of great significance to study the target detection and discrimination algorithms for SAR images.In order to solve the target detection and discrimination problem for SAR images in complex scenes,this paper proposes two target detection methods and two target discrimination methods.The main contents of this paper can be summarized into the following four aspects:1.To improve the accuracy and efficiency of the target detection algorithm for the large SAR images,we study the fast detection method for SAR images.Specifically,we propose a fast target detection method for SAR images based on multi-scale saliency(MSS).According to the prior knowledge of targets to be detected,the proposed MSS detection method selects the task-dependent scales from the Gaussian pyramid of the original SAR image to construct the saliency map,which can highlight the targets of interest and suppress the background clutter,and thus improve the detection accuracy.In addition,the proposed MSS detection method uses the time-saving center-surround mechanism to construct the saliency map,instead of using the time-consuming sliding window mechanism as the two-parameter CFAR does,which can greatly reduce the computational load of the target detection method for the large SAR images.The experiments based on the real SAR data demonstrate that the proposed MSS detection method is much better than the traditional detection methods in terms of the accuracy and speed,however,the effectiveness of the proposed MSS method in eliminating the strong clutter false alarms in heterogeneous background is not ideal.2.To eliminate the strong clutter false alarms generated by the target detection method for the SAR images with heterogeneous background,we study the target detection method for the SAR images with heterogeneous background.Specifically,we propose a new target detection method for SAR images based on Bayesian-morphological saliency(BMS).The proposed BMS detection method mainly contains two stages: Bayesian saliency map construction and morphological saliency map construction.The Bayesian saliency map can obtain the complete structures of the bright objects including the targets of interest and the strong clutter,which facilitate the utilization of the prior information of the targets of interest;the morphological saliency map can highlight the targets of interest while suppressing both the natural and manmade strong clutter by combining the size and shape prior information of the targets.The experiments based on the real SAR data demonstrate that the proposed BMS detection method has better detection performance than the traditional detection methods in the heterogeneous background.3.To solve the problem that some traditional discrimination features depend on the target segmentation preprocessing and have slow feature extraction speed,we study the discrimination feature extraction method which do not depend on the target segmentation preprocessing and have fast feature extraction speed.Specifically,we propose a new chip-level discrimination method based on the modified saliency and gist(MSG)features.It is difficult to acquire the accurate target-shaped blob via segmentation preprocessing,thus some classical discrimination features which are extracted based on target segmentation may lose effectiveness.To solve the above problem,we propose the MSG features based on the saliency and gist(SG)features for the optical images.The MSG features are complementary to each other and can provide a more complete description of the extracted SAR image chips without segmentation,which also reduces the computation burden.The experiments based on the real SAR data demonstrate that the MSG has higher discrimination accuracy than the traditional discrimination features.In addition,the feature extraction speed of the MSG features is also faster than those of most of the traditional discrimination features.4.To solve the problem that the chip-level target discrimination method has low accuracy in the multiple target environments,we study the discrimination method to improve the discrimination accuracy in the multiple target environments.Specifically,we propose a superpixel-level target discrimination method based on the multi-level and multi-domain(MLMD)feature descriptor.The proposed discrimination method mainly contains three stages.Firstly,based on the superpixel-level target detection results,we describe each superpixel via the MLMD feature descriptor,which can reflect the differences between targets and clutter meticulously and comprehensively.Compared with the chips,the superpixels have more flexible sizes and shapes,which can adhere well to object boundaries;and each superpixel belongs to only one target or part of the background,not stretching over both of them.Thus the proposed method which takes the superpixel as the elementary units for SAR target discrimination can avoid the occurrence of the multiple/partial target chip cases in the multiple target areas.Secondly,we employ the support vector machine(SVM)as the discriminator to obtain the discriminated target superpixels.Finally,we cluster the discriminated target superpixels and extract the target chips from the original SAR image based on the clustering results.The experimental results based on the real SAR data show that the performance of the proposed superpixel-level discrimination method is much better than those of the traditional discrimination methods.