Research On SAR Image Target Recognition Based On Multi-source Data Fusion

Synthetic Aperture Radar(SAR) based Automatic Target Recognition(ATR) algorithm has become a hot topic of research in recent years. The research achievements have been widely used in military and civilian domains, such as homeland security, video surveillance, intelligent home, etc. However, it has limitations like the imaging results are sensitive to the change of target perspective and the results are relatively abstract. Meanwhile, multi-source data fusion can grab more information of target by fusing data from different sources to improve the robust to several parameters and recognition rate. Therefore, this thesis will adopt target recognition algorithm based on multi-sources data fusion, which will waken impact on recognition result caused by the target bearing dependency of SAR images, and improve the stability of recognition results, better fulfilling the requirements for automatic target recognition to targets on ground in the battlefield environment.The thesis will do multi-view image fusion, which will enlarge the information about recognition target. The thesis, on the basis of Joint Sparse Representation algorithm, will promote an improved method to control the affect of azimuth information on recognition result. The contents include:1. Research feature extraction method suitable for SAR images. The thesis will adopt Independent and Identically Distributed(IID) Gaussian Random Projection method to reduce dimensions of the image data, and demonstrate the efficiency and effectiveness of the dimension reduction method. At the same time, the thesis will study the principle of Sparse Representation method and Joint Sparse Representation method to extract features, and the application in SAR and ATR fields.2. In consideration of the SAR image features, the thesis will analyze the limitations if the traditional Sparse Representation algorithm is directly used in target recognition of SAR images. It will promote a certain sub-dictionary based Joint Sparse Representation algorithm, which can control the affect of multi-view information to recognition result by integrating multi-view information. By designing experiment analog and adopting MSTAR and experiment measured data, the thesis will demonstrate that the certain sub-dictionary Based Joint Sparse Representation method can improve the accuracy of target recognition.3. As for the above analysis on target multi-view information fusion algorithm, the certain sub_dictionary based ATR method can increase the amount of target information and control the amount of disturbance information in the meantime. But it has requirement for the data, needing that the data distribution is methodic and the label of data type is known. For the above issues, the thesis promotes a Joint Sparse Representation algorithm over locally adaptive dictionary, which has a higher adaptability and keeps the advantages of the above mentioned algorithms. By testifying via experiment analog, the multi-view target image fusion method based on Joint Sparse Representation over locally adaptive dictionary has better results in such aspects as recognition rate, robustness, and convergence.