Research On The Target Detection And Classification In SAR Images

As the typical representative in the microwave remote sensing field, synthetic aperture radar (SAR) has been widely used in the areas of surveying and mapping, military reconnaissance and national economy because of its unique advantages of earth observation. With the growing technologies of SAR, it brings the new challenges in automatic interpretation of SAR images. Besides, the SAR automatic target recognition (ATR) system is the research focus in the current world. Based on the theories and technologies of constant false alarm rate (CFAR) detection, fuzzy rule, level set and support vector machine (SVM), we have done the study on the detection and classification of military targets on ground in SAR images, such as vehicles and tanks. The main contents are as follows:1. A new algorithm for target detection in SAR images is proposed based on fuzzy CFAR. Based on Weibull distribution, the membership functions of fuzzy CA-CFAR and fuzzy OS-CFAR were obtained and combined via fuzzy fusion rules to produce a threshold of fusion centre. The simulation results show great improvement on efficiency of the proposed algorithm without decreasing detection performance in the non-homogenous situations.2. Through in-depth research and analysis of the level set theory, we have achieved the military targets on ground detection and extraction in SAR images, with the influence of speckle and the features of vehicles and tanks in high resolution SAR images. Combining the Chan-Vese model, and using a higher dimensional implicit method to control the target contour curve evolution via level set, the integral data of targets has been extracted. The proposed method shows the great capabilities of target extraction and strong anti-noise performance.3. A novel method of target classification in SAR images is proposed based on data extraction and SVM. Combining the target information extraction method via level set, we use SVM, which has the classification capability of generalization, to classify the different types of target in SAR images. In the simulation, we use the MSTAR data for test. The results prove that it can remove the interference of background and use the target data in effect to improve the classification and recognition accuracy.