SAR Target Detection And False Alarm Reduction Based On Priori Knowledge About The Imaged Scene

Synthetic aperture radar(SAR) is an active microwave perception sensor, which has been an important tool of earth observation and military surveillance, and has the characteristics of working under all weather, and time conditions, with high resolution, and strong penetrability. Since the 1950 s, SAR image target detection and recognition has attracted more and more attentions as the radar imaging technology matures and the imaging resolution improves.Current SAR target detection algorithms are not good enough for SAR images of complex scenes, and can detect a large number of false alarms that reduce accuracy of the subsequent discrimination and recognition. By using priori knowledges, taking specific relation of target and environment into consideration, and detecting specific target in the specific environment, this thesis forcuses on the study of target detection and false alarm reduction. Meanwhile in this thesis, detection, discrimination and recognition algorithms are embeded into a human-machine interface. Thus the whole process looks more obvious and visible. The main contents in this thesis are as follows:1.Algorithm of plane detection in SAR image based on priori scene knowledge is studied. Firstly, edge detectors appropriate for SAR image are introduced. Secondly, introduction on Hough transform is made. For the result of edge detection, Hough transform is used to detect lines to acquire the locations of airport runway. Finally, region growing algorithm is used to grow airport runway regions, and CFAR algorithm is used to detect plane targets in the grown airport runway regions.2.Method about reducing false alarms on the result of vehicle target detection based on prior knowledge is discussed. In general, vehicle targets are parked in the open area that is regarded as the main background region in a common SAR image. Hence, the initial seed can be selected from the area according to the result of histogram statistics for the amplitude SAR image, and then the region vehicle parked is available by using region growing method. The ungrown regions probably are regions of trees, buildings and shadows, in which vehicle targets do not exist. For these regions, we use the one-class classification method to obtain the real trees, buildings and shadow regions. Finally, the targets detected in these regions of image are removed after doing vehicle target detection by super-pixel CFAR detector.3.Software on SAR image target detection and recognition is designed. Matlab&C mixed programming technology is used to embed detection, discrimination and recognition algorithms into a human-machine interface, which makes the process visible. The whole software includes reading data, preprocessing, detection, discrimination, recognition and assistant module, and the modules are both dependent and independent among each other. In the meanwhile, the other algorithm modules are reserved in the software, which strengthen its extensibility.