SAR Ship Target Discrimination Based On Scene Priori Knowledge And Image Feature Extraction

Synthetic Aperture Radar(SAR)is an active microwave sensor,which has been widely used in military target detection,marine research,resource exploration and other fields,thanks to its all-day and all-weather working ability and its strong penetrability.With the development of SAR technology,the quantity of available SAR image data is increased day by day and,meanwhile,the resolution of the acquired image data is improved.Therefore,how to boost the ship target detection performance using SAR data has attracted more and more attentions.Although many advanced SAR ship target detection algorithms have been proposed,false alarms still inevitably exist in the detection results.The target discrimination algorithms can process the target chips extracted by the detection stage,further removing the false alarms while retaining the real targets.This dissertation mainly focuses on the ship target discrimination algorithms to improve the final ship target detection accuracy by further discriminating the target chips extracted after the detection stage.Current SAR ship target discrimination algorithms are not good enough for SAR images of complex scenes,and it is hard to distinguish the strong clutter on land and the ship targets using these algorithms.In practice,the discrimination of ship targets and clutter can be achieved by using scene priori knowledge,or by feature extraction of the image.In this dissertation,we focus on the SAR ship target discrimination task in complex scenes,which mainly includes two aspects,the first aspect is to achieve ship target discrimination using sea-land segmentation based on scene priori knowledge in SAR images,and the second aspect is to discriminate ship targets using feature extraction from intensity information as well as amplitude and phase information of SAR images.The specific work done is as follows:1.The ship target discrimination methods in SAR images based on scene priori knowledge are studied.The main idea is to obtain more accurate sea-land segmentation results by combining the sea and land priori knowledge,and then achieve the discrimination of ship targets and strong clutter on land with sea-land segmentation results.First,two techniques for acquiring sea and land priori knowledge of SAR images without Internet connection are investigated.In order to meet the demand for real-time processing of images after satellite imaging,it is necessary to acquire scene priori knowledge of SAR images offline.The M_Map package provided by MATLAB software can provide sea and land priori knowledge in the scenario,but requires the support of the MATLAB software environment.However,it is difficult to create an environment with MATLAB software in satellite.Therefore,this paper firstly investigates how to produce a database that can be carried on satellites and contains global land and sea information,and how to obtain sea and land priori knowledge of SAR images in real time using this database.Then,a sea-land segmentation method based on adaptive threshold and a sea-land segmentation method based on clustering and adaptive region merging are studied,and a sea-land segmentation method based on scene priori knowledge of sea-land database and region merging is proposed.Finally,the results on the actual measured SAR image data show that the produced global sea and land database can effectively provide the priori knowledge of SAR images,and the proposed segmentation method can well segment land and sea regions with similar grayscale features and similar texture features,improving the quality of image segmentation as well as helping to discriminate the ship targets and strong clutter on land more accurately.2.The ship target discrimination methods based on intensity features of SAR images are studied.First,two local features,namely,Scale-Invariant Feature Transform(SIFT)feature based on gradient information and Dense Local Self-Similarities(DLSS)feature based on shape information are studied.Then,Spatial Pyramid Matching(SPM)algorithm based on Bag of Words(BOW)model is studied,and the two local features are combined with SPM algorithm to extract global discriminative features.Finally,the two discriminative features are classified to discriminate the ship targets and strong clutter by linear Support Vector Machine(SVM)classifier.The experimental results show that SIFT feature achieves better discrimination results.3.The ship target discrimination methods based on amplitude and phase features of complex SAR images are studied.Firstly,the ship target discrimination algorithm based on multiple spectral features and feature selection is studied.Secondly,the spatial domain and frequency domain of the images are linked,and the radar spectrogram feature is extracted based on the joint time-frequency analysis algorithm.Thirdly,the SPM algorithm and the radar spectrogram feature are combined to extract global discriminative feature to discriminate ship targets.Finally,we combine the radar spectrogram feature and SIFT feature to discriminate ship targets,and improved results are obtained.