Research On Key Technology Of SAR Image Interpretation System For Unmanned Aerial Vehicle Precision Strike Process

Synthetic Aperture Radar(Synthetic Aperture Radar, SAR) has the characteristic of imaging without having impacts of adverse weather and climates. It has the significant strategic value in military and civil uses and has become a worldwide concern especially for the military. SAR has been one of the standard configurations of Unmanned Aerial Vehicle(UAV). The unique coherent imaging modality of SAR which leads to more sophisticated image interpretation than the optical image’s is special compared to the ordinary optical sensor imaging mode. Therefore, researching on the SAR image interpretation system which is used for unmanned aerial vehicle precision strike process is challenging but meaningful. This thesis has carried out a series of research work respectively on SAR image filtering, detection and recognition of ground targets and the image registration between optical images and SAR images. SAR image interpretation system has the ability to interpret images instantly, its also known for its robustness.First of all, the synthesis of existing literature and our previous research results, combined with airborne SAR images speckle noise characteristics, considered the specific application environment, an overview to the principle of SAR imaging sensor and analysis of SAR image data set characteristics are provided. For the purpose of airborne SAR image interpretation system for target positionsing, specific research framework is proposed.Secondly, according to the characteristics of SAR image data, this thesis gives a description on the physical properties of speckle noise. SAR image filtering algorithms research based on statistics and wavelet domain is conduct, and objective evaluation index is proposed to perform quantitative analysis and comparison. Simulation on filtering algorithms is conducted. By observing effect of filtering algorithms through the ratio image, this thesis gives the scope of SAR image filtering algorithms targeting different conditions, and established the foundation for follow-up work.Thirdly, for the target detection module in the SAR image interpretation system, a ground vehicle target detection algorithm based on robust principal component analysis is proposed. By utilizing multi-frame SAR images and conducting an observation matrix, this thesis reconstructs the low-rank matrix and the sparse matrix through solving a convex optimization problem, and gives the constraints. Furthermore, for the purpose of conducting accurate detection to SAR image vehicle target and voiding false segmentation, a threshold selection method based on sparsity is designed and to realize the segmentation of single vehicle target. Simulation results show that, for multi-target detection problem in the circumstance of multi-frame substantially SAR scenes, this algorithm can effectively separate the background and targets, and achieve the desired test results.In the following, given the huge amount of data often brings great difficulties to the target recognition in SAR image interpretation system, this thesis proposes SAR image target recognition algorithms based on compressed sensing. In accordance with the over-complete dictionary in compressed sensing theory, it designs the dimension reduction method on the basis of principal component analysis and kernel principal component analysis. The test samples after dimensionality reduction are projected on over-complete dictionary after dimensionality reduction, and reconstructed with the sparse coefficient. The 2-norm is then used to obtain the category of the test samples. Simulation results suggest that, when the target image is impacted by strong noise, this recognition algorithm can achieve accurate target identification, with a strong adaptability to the attitude change of SAR target images.Finally, in order to study UAV target positioning during precision strike, considering the heterologous image registration under the affine transformation model, this thesis proposed a registration strategy of optical image and SAR image based on a control line method. In order to improve the real-time and robustness of scene matching navigation, improved line segment detector is utilized to detect feature lines of image, and Hough transform is used to link line segments, and then intersected control line pairs are constructed according to certain constraints, thus a corresponding points matching function based on the intersection of control lines is designed. Precise automatic image registration is achieved based on the registration parameters derived from the affine transformation model. Experimental results show that, the proposed method has high registration accuracy for the SAR image and optical image, which are largely different in intensive, rotation and translation. The computation time is substantially reduced, and it is possible to meet requirements of some real-time applications.