Research On ISAR Imaging And Rotation Information Estimation Algorithm Of Ship Target

Marine ship target classification is of great significance for the national strategic defense and the shipping traffic administration.With the capability of high-resolution imaging in all-day,all-weather conditions,Inverse Synthetic Aperture Radar(ISAR)plays an important role in the classification of ship targets.However,there are still several problems to be solved.Firstly,high-quality ISAR image and reasonable feature description is the key of ship target recognition.Compared with the images of other-views,the top-view or side-view ISAR image of ship target can provide more structured information on the target.Therefore,selecting a proper imaging instant and accumulated time from radar echoes to obtain high-quality side-view or top-view image is the focus of this article.Secondly,the obtained images only represent the distribution of the target scattering points in the Distance-Doppler domain,which does not show the actual size of the target.In order to extract the features more accurately,it is necessary to carry out ISAR cross-range scaling.Thirdly,owing to the addition of time dimension,ISAR image time-series are effective carriers about rotational information which is an important feature in ship target classification.Improving the extraction accuracy of rotational information is an important technology urgently needed to be solved in the field of ISAR ship target recognition.The above issues are not only the difficulties of ISAR ship target recognition,but also restrict the development of ISAR system in practical application.Aiming at improving the accuracy of ship target feature extraction,this paper has carried out the researches on optimum imaging time selection,ISAR cross-range scaling and rotational information extraction,which are helpful to the application of ISAR ship target classification.The reminder of this paper is summarized as follows.Optimum imaging time selection algorithm is introduced in the first part,which aims to select a proper time from radar echoes to generate high-quality side-view or top-view image of the ship target.This issue consists of two parts: optimum imaging instant and optimum imaing accumulated time.Firstly,a mathematical analysis of radar imaging is present,which illustrates the relationship between the Doppler frequency of scattering points and the 3D rotational components of ship target.Therefore,the vertical and horizontal rotational components of ship target can be estimated by detecting the slope of ship center-line and the height information from the image time-series.Then,we analyze the reason why traditional center-line extraction algorithm fails and propose a novel algorithm to improve the accuracy of center-line and ship height estimation at different images.After that,the image gradient is used as the criterion for selecting the optimum imaging accumulated time.The advantage is that the image gradient itself is the standard for evaluating image focus.ISAR cross-range scaling algorithm is researched in the second part,aiming at obtaining the accurate size information of ship target.In the case of constant velocity,the reason for the failure of image rotation correlation algorithm is analyzed,which the inconsistency between the range dimension and the cross-range dimension of ISAR image leads to not only the angular transformation between two adjacent sub-aperture images,but also the difference in scale.Then,this difference can be regarded as affine transformation,and an algorithm based on affine projection matrix is proposed in this article.The new method can eliminate the negative effects caused by ‘scaled–stretched’ transformation between the two images and keep the rotation transformation.In the case of non-constant velocity,a new algorithm based on weighted average method is introduced to rescale the ISAR image.The new method can obtain the rotational velocity of the intermediate image in the ISAR time-series.Ship target rotation information extraction is introduced in the last part,aiming at improving the accuracy of rotation information estimation.The focus of research is composed of three steps: feature point extraction,tracking and rotation center search.In our study,the interest points are first extracted from ISAR image time-series by Speeded-up Robust Feature(SURF).Different from traditional feature point extraction methods,SURF-based feature points are not only invariant to scattering intensity,target rotation and image size,but also extracted from different scales of the image.Then,a new feature registering model combining the bilateral feature registration and Random Sample Consensus(RANSAC)method is employed to track these feature points.The bilateral feature registration is used to perform a rough search,most of the unsatisfied feature points are eliminated,and then the remaining feature points are purified by the RANSAC method.The new feature registering model improves not only the efficiency,but also the accuracy.After that,the rotation center is detected by Fourier transform(FT).In frequency domain,the rotation center is not needed to be found,which it is at the frequency(0,0)Hz.